GB2390154A

GB2390154A - Method of volume manufacture of a product in a staged production process

Info

Publication number: GB2390154A
Application number: GB0223360A
Authority: GB
Inventors: Marie Rosalie Dalziel; William Frederick Clocksin; Alistair Stray; Christopher Davies
Original assignee: MILLENNIUM VENTURE HOLDINGS LT
Current assignee: MILLENNIUM VENTURE HOLDINGS LT
Priority date: 2001-10-08
Filing date: 2002-10-08
Publication date: 2003-12-31
Also published as: WO2003031370A3; AU2002329482A1; WO2003032096A2; WO2003032096A3; WO2003031370A2; GB0223360D0; GB0124130D0; AU2002329476A1

Abstract

This invention concerns improvements relating to staged production in volume manufacture of products such as ceramics tiles. Parameters are measured at multiple points along the process and correlations are identified between input and output parameters. Interdependencies between parameters are stored and the process is controlled at multiple stages according to a stored interdependency in response to a given parameter change. Output parameters may include: density of or defects in products, parameters of glaze applied in a glazing station or moisture content or grain size distribution of particulate material from storage. Input parameters may include press pressure, press speed, pump pressure, air temperature and air flow rate. Aspects include: monitoring the density of "green" tiles on-line: detecting defects in "green" tiles on-line using infrared imaging; monitoring on-line the colour in raw material and measuring the temperature profile of tiles using an infra-red camera. A non-contact ultrasonic density sensor is also described, and a belt conveyor having a drive speed output signal and a sensor for sensing product movement along the belt. Predicted product movement along the belt is computed based on the drive speed output signal and compared with actual product movement. An alert is provided if actual product movement departs from predicted product movement by more than a preset threshold.

Description

1 2390154

t METHOD OF VOLUME MANUFACTURE OF A PRODUCT IN A STAGED

PRODUCTION PROCESS

5 Field of the Invention

This invention concerns improvements relating to staged production in volume manufacture, namely the production of finished products in high volume by processes involving multi-stage, non-continuous operations. The invention will be particularly 10 described herein by reference to the manufacture of ceramics tiles, but is applicable also to other multistage process industries including, for example, the food, leather, timber, plastics and textiles industries.

BackEround of the Invention Continuous production processes, for example in an oil refinery, are relatively simple to control, but the control of staged production is much more complex and traditionally it has not been possible to control staged production processes in volume manufacture in real time. The cost of this to the relevant manufacturing industries is 20 immense. In the ceramics tile industry, for example, some 30% of production is typically rejected and goes to land-fill sites. When it is considered that the annual world production of ceramics tiles currently exceeds four thousand million square metres of tiles with a retail value of US$40 billion and is expected to double within the next five to ten years, the volume of lost production is staggering and its cost to the 25 manufacturers and to their customers, for it is the customers who pay the price, is likewise enormous.

Current multi-stage manufacturing processes are characterized by interruptions and off-line storage requirements between successive production process stages. This 30 leads to the key problems of interrupted production flow, information fragmentation, post-mortem analysis, process bottlenecks and high in-process losses as outlined above.

In the ceramics tile industry, the current modern production process is a batch

production process consisting of a number of serial and independent manufacturing stages including batching and milling of raw materials to produce a slurry which is then stored, the spray drying of the slurry to form a granulated product which is then stored, the supply of the granulated product to the tile presses, the supply of the "green" tiles 5 from the presses to driers, the printing and glazing of the dried tiles, subsequent firing of the tiles, and sorting, stacking and batching of the tiles at the end of their production.

While it is known for some level of automation to have been provided within a given sub-process, and while it is also known in other industries (for example in the contact lens industry - see US Patent No S 461 570) to correlate product defects with a 10 particular single input parameter to optimise that parameter for minimum defects, no attempt has hitherto been made to provide the industry with a unified view of how various sub-processes link together and to provide knowledge and control of the production parameters which influence overall product quality and production efficiency. Obiects and Summarv of the Invention It is the object of the present invention to overcome or at least substantially reduce the abovementioned problems.

According to a first aspect of the present invention, a method of volume manufacture of a product in a staged production process is provided comprising the steps of: measuring parameters at multiple points along the process and identifying correlations between input and output parameters, including correlations between an 25 output parameter and a plurality of input parameters or an input parameter and a plurality of output parameters; storing interdependencies between parameters according to the identified correlations; and controlling the process at multiple stages according to a stored interdependency in response to a given parameter change (e.g. a single parameter change or a given set of parameter changes). An example of such a process 30 is a ceramic product (e.g. tile) manufacturing process, in which case the multiple stages include a drier, a press and a kiln.

The step of identifying correlations is preferably performed for different product data sets, such as different types of tile, or different batches of the same type or different production lines or even different work shifts or machine operators, and different interdependencies between parameters are stored for different product data 5 sets, whereby an interdependency is recalled from storage when the data set (e.g. the type of product to be made) is changed. Output parameters may include one or more of: density of products emerging from the press, defects in products emerging from the press; parameters of glaze to be applied to the products; moisture content of particulate material entering the press; and grain size distribution in particulate material entering 10 the press. The input parameters may include one or more of: press pressure, press speed, pump pressure, air temperature and air flow rate.

In accordance with a second aspect of the invention, a facility for a staged production, volume process of manufacture of products is provided comprising means 15 for feeding material into a stage of the process at a stage input and means for feeding out of that stage at a stage output, means for measuring a plurality of input parameters of the material entering the stage and/or conditions within the stage, means for measuring at least one output parameter of a product exiting the stage; and means for automatically correlating the at least one output parameter with the plurality of input 20 parameters to infer the influence the input parameters have on the output parameter and the relationship therebetween.

The means for automatically correlating the at least one output parameter with the input parameters preferably perform the correlation product-byproduct using the 25 input parameters and the output parameter specific to individual products.

It is a further preferred feature that the correlation means use, as input parameters, parameters measured in relation to a preceding product or a subsequent product passing through the stage. Gap measuring means may be provided for 30 measuring the gap between products passing through the stage, wherein the means for correlating uses, the gap as an input parameter. The reasons for correlating output parameters with input parameters of preceding and subsequent products and for

measuring gaps between products are that in a tile manufacturing process where tiles pass through a kiln, the particular temperature profile and other aspects to which a tile is dependent will be influenced by adjacent tiles and by the closeness of tiles passing through. s A ceramic product press moisture content measuring means, for example an infra-red sensor, may be provided for measuring moisture content of material to be fed into the press. Grain size distribution monitoring means may be provided for monitoring grain size distribution of particulate material to be fed into the stage.

According to a third aspect of the invention, a facility for a staged production, volume manufacture of products is provided comprising: a first process stage having an associated controller for controlling the first process stage, a second process stage having an associated controller for controlling the second process stage, first feeder 1 S means for feeding first material into the first process stage, second feeder means for feeding second material from the first process stage into the second process stage, second measuring means for measuring second parameters of the second material and/or the second process stage, and third measuring means for measuring third parameters of products emerging from the second process stage, and feedback means for feeding back 20 to the controller of the first process stage signals responsive to the second and third parameters. Correlation means are preferably provided for correlating measured third parameters with measured first parameters and cause the feedback means to provide 25 signals to the controller of at least one of the first and second stages responsive to results of correlation. The correlation means may further correlate measured third parameters with measured second parameters and cause the feedback means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation. The correlation means may further correlate measured second 30 parameters with measured first parameters and cause the feedback means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation.

s According to a fourth aspect of the invention a facility for a staged production, volume manufacture of products is provided comprising: a first process stage having an associated controller for controlling the first process stage, a second process stage 5 having an associated controller for controlling the second process stage, first feeder means for feeding first material into the first process stage, second feeder means for feeding second material from the first process stage into the second process stage, first measuring means for measuring first parameters of the first material and/or the first process stage, second measuring means for measuring second parameters of the second 10 material andlor the second process stage, and feedforward means for feeding forward to the controller of the second process stage signals responsive to the first and second parameters. The correlation means preferably correlate measured third parameters with l 5 measured first parameters and cause the feedforward means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation. The correlation means may further correlate measured third parameters with measured second parameters and cause the feedforvvard means to provide signals to the controller of at least one of the first and second stages responsive to results of 20 correlation. The correlation means may further correlate measured second parameters with measured first parameters and cause the feedforward means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation. 25 It is particularly preferred that the correlating means performs a correlation product-by-product over a plurality of identified products passing through production, correlating at least one measured parameter at one stage in the production with at least one measured parameter at another stage in the production of each identified product.

30 According to a fifth aspect of the invention, a facility for a staged production, volume process for the manufacture of ceramic tiles is provided, comprising means for processing of raw materials to produce a raw material for tile production, one or more

4 6 presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for monitoring the density of "green" tiles on-line.

5 The means for monitoring the density of "green" tiles preferably comprise a non-contact ultrasound sensor. Further means may be provided for detecting defects in "green" tiles.

According to a sixth aspect of the invention, a facility for a staged production, 10 volume process for the manufacture of ceramic tiles is provided, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and infrared imaging means for detecting defects in "green" tiles on-line.

According to a seventh aspect of the invention, a facility for a staged production, volume process for the manufacture of ceramic tiles is provided, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the 20 "green" tiles, their printing and/or glazing and their firing, and means for monitoring on line the colour in the raw material that is produced for tile production and/or in glaze preparation. The colour monitoring means may include a spectrophotometer.

According to a seventh aspect of the invention, a facility for a staged 25 production, volume process for the manufacture of ceramic tiles is provided, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for monitoring parameters of glaze material on-line as it is being applied to the tiles. The means for 30 monitoring glaze quality preferably include means for measuring glaze material viscosity.

According to an eighth aspect of the invention, a facility for a staged production, volume process of manufacture of ceramic tiles is provided comprising means for processing of raw materials to produce a raw material for tile production, batching and milling means for producing a water-based slurry, and spray drying means 5 for producing from said slurry a particulate raw material for tile production, comprising one or more presses for the production of "green" tiles, means for subsequent drying of i the "green" tiles, their printing and/or glazing and their firing, and means for inspection of the finished tiles and sorting and batching of the same, further comprising means for monitoring moisture content of said particulate material. The means for monitoring 10 moisture content preferably comprise an infra-red sensor.

According to a ninth aspect of the invention a facility for a staged production, volume process of manufacture of ceramic tiles is provided, comprising means for processing of raw materials to produce a raw material for tile production batching and 15 milling means for producing a water-based slurry, and spray drying means for producing from said slurry a particulate raw material for tile production, comprising one or more presses for the production of "green" tile, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for inspection of the finished tiles and sorting and batching of the same, further comprising means for 20 monitoring grain size distribution in said particulate material. The grain size distribution measurement is preferably effected by a process of computer- aided image! analysis. In each of the various aspects of the invention, tracking means may be 25 provided for tracking tiles through their manufacturing process, for example means for marking the tiles with machine-readable identifiers such as barcodes and/or datamatrix codes. According to a tenth aspect of the invention, a method of monitoring of tiles in 30 a tile manufacturing facility is provided which comprises measuring the temperature profile of tiles using an infra-red camera. Preferably tile moisture content is inferred

from the temperature profile. The infra-red tile moisture content may be used as a parameter for correlation against a measured output parameter of the facility.

According to a further aspect of the invention, apparatus is provided for 5 detection of inclusions in a workpiece comprising means for subjecting the workplace to spatially-uniform heat flux input, an infra-red camera for acquiring an image of the workpiece surface and means for analysing said image to detect features indicative of inclusions. 10 According to another aspect of the invention, apparatus for density measurement are provided comprising a non-contact ultrasound sensor.

According to another aspect of the invention, a conveyor is provided for conveying unit products along a belt, comprising drive means providing a drive speed 15 output signal and a sensor for sensing product movement along the belt and computation means for comparing predicted product movement along the belt, based on the drive speed output signal, with actual product movement and providing a signal (e.g. an alert) if actual product movement departs from predicted product movement by more than a pre-set threshold.

The sensor may comprise a first sensor for sensing a product passing a first point along the belt and a second sensor for sensing the product passing a second point along the belt. The sensor may be placed relatively close to the drive means and the second sensor may be located relatively remote from the drive means.

The computation means preferably includes modelling means for modelling a predicted progress of movement of a product along the belt and compares a measured progress of movement with the predicted progress of movement. The modelling means preferably take into account stoppages in belt movement as reported by the drive speed 30 output signal. Using this conveyor, predicted product movement along the belt is computed based on the drive speed output signal and compared with actual product

movement. If actual product movement departs from predicted product movement by more than a pre-set threshold, an alert can be generated.

The invention proposes to provide the tools and the means to monitor online 5 process variables and control real-time the product quality at the output of each production process. This functionality is provided by the combination of intelligent sensors, a powerful architecture for the acquisition and traceability of complete process data and software for adaptive, knowledge-based control of processes. The user interfaces preferably and advantageously are internet enabled and portable for remote 10 production control and information analysis.

As indicated hereinbefore, the present invention is particularly, but not exclusively, useful in the ceramic tile industry where it enables integrated manufacturing production automation. The invention will hereinafter be described in 15 detail by reference to an exemplary system for integrated manufacturing production automation for the ceramic tile industry. The hereinafter described system incorporates raw materials storage and processing (batching and milling) to produce a slurry, storage of the slurry, spray drying of the slurry as required to produce a particulate raw material for tile production, storage of the particulate and supply on demand to presses for the 20 formation of "green" tiles, the subsequent drying of the "green" tiles, their printing and glazing, firing of the tiles and buffer storage of fired tiles, inspection of the finished tiles and sorting and batching of the same. These multiple manufacturing stages are largely separate and the present invention proposes for each stage a comprehensive before-and-

after sensory/monitoring agenda providing full information relevant to the respective 25 stage. Controls are effected in each stage, and overall, based on sensed/monitored production process variables, with data feedback throughout to operating and system management personnel to the ultimate degree.

The system, in prototype studies that we have performed, has proven itself 30 capable of achieving an 80% reduction in machine down time through improvements inter alla in intelligent, information-driven, preventative maintenance, a reduction of 25% in in-process losses achieved through intelligent, information-driven quality

- 10 control procedures throughout the multiple stages of the tile manufacturing process, recycling of 90% of waste process materials and residual water, reduction of scrap to less than 2%, and reduction of second and third quality products to less than 10%. Any one of these advantages might be considered significant, in combination they are very 5 substantial. The above and further features of the present invention are set forth in the appended claims and will be made clear, by way of example only, in the following detailed description.

Brief Descrintion of the Drawings Figure 1 is a diagrammatic overview of a typical tile manufacturing plant showing the stages in manufacture from clay storage through to fmal quality control and 1 5 classification.

Figure 2 illustrates a belt tracking subsystem according to a preferred embodiment of one aspect of the invention.

Figure 3 is a graph illustrating belt slippage.

Figure 4 illustrates certain control loops implemented in the integrated control 20 system of the preferred embodiment of the invention.

Figure 5 is a diagrammatic representation of a generic stage in the process of Figure 1.

Figure 6 is a block diagram of the porcelain tile manufacturing process of Figure 1.

25 Figure 7 is a generic scheme of a spray drying process.

Figure 8 illustrates a control scheme for spray drying.

Figure 9 illustrates a general press model in accordance with the preferred embodiment. Figure 10 illustrates two stations in a staged production process.

30 Figure 1 1 is an illustration of computer-aided image analysis apparatus for granulometry.

Figure 13 is an illustration of an instrumentation rack, system controller and transducers for the apparatus of Figure 9.

Figure 14 is a neural network for sensing of defects in the preferred embodiment of the invention.

5 Figure 15 is an illustration of an infra-red camera station for detecting delamination in tiles.

Detailed Description of the Preferred Embodiment

10 Referring to Figure 1, a ceramic product manufacturing plant, in particular a tile manufacturing plant is illustrated, comprising a series of stations for the manufacture of powder in a continuous manufacturing process and a series of stations for the pressing of the powder into ceramic products such as tiles and the firing of those products to create the finished product. The powder manufacturing facility can be co 15 located with the product manufacturing facility, but is often located separately.

The powder manufacturing facility comprises a clay store 10, a pregrinding station 12, dosing vats 14, a milling machine 16, one or more slip tanks 18, a spray drier 20, a conveyor 22 and a set of powder silos 24. Each of the stations has a local 20 controller (e.g. a programmable logic array or controller) and each may have a server computer. Spray drier 20 is shown as having an associated server computer 21. The ceramic product manufacturing facility has powder silos 24 which may be the same as the silos used by the powder manufacturing facility or may be separate silos to receive powder transported from the powder manufacturing facility and it further has a press 26 25 (with associated server computer 27), a drier 28 (with server computer 29), a glazing station 30 (with server computer 31), intermediate storage 32, a kiln 34 (with server computer 34) and a quality control and classification station 36 (with server computer 37). Running between the various stations forming the process is a bus 50 having a central controller 52 connected thereto. The bus 50 provides control to various stations 30 as indicated by arrows pointing towards those stations and accepts sensor data from the various stations either directly or via the server computers, as indicated by the arrows pointing towards the bus. The bus SO, the controller 52 and the server computer

communicate through intelligent field bus cards produced by Applicom International

which are capable of many field bus protocols including Profibus, ModBus, luterbus-S

and others. Java is used at the language for user interfaces. Cormected to the controller 52 is an Oracle datastore 54 for storing production data to permit the controller 52 to 5 perform analysis of collected data and to allow charts of historical trends to be displayed to users. The data store 54 is capable of continuously accepting data from any part of the factory while simultaneously servicing queries from multiple clients. An application program interface (API) is provided to allow the controller 52 to access data which is acquired from the production line. Preventative maintenance events such as l 0 alarm conditions and scheduled maintenance events and tools for manipulating these are provided and described in greater detail below.

In operation, powder from the manufacture of ceramic products is made from the clay store lo and is preground, dosed and milled by techniques known in the art, and l 5 a slurry is created in a slip tank 18. This slurry is sprayed into spray drier 20, where it is dried at a carefully controlled temperature to produce powder which is conveyed along conveyor 22 into powder silos 24. When a batch of tiles is to be made, powder from the powder silos 24 is conveyed into the press 26 which presses the powder (having a certain instantaneous moisture content) into the "green" tiles or other ceramic products.

20 The green tiles are conveyed into drying racks 28 before passing along a further conveyor through the glazing station 30. At the glazing station 30, a glaze is applied onto the surface of the tiles (e.g. by spraying) and the tiles are then conveyed through intermediate storage 32 to the oven or kiln 34. On passing through the kiln 34, a carefully controlled temperature profile is maintained to allow the tiles to heat up to 25 their firing temperature and remain at that temperature for the necessary firing time as tiles continuously move through the kiln. On emerging from the kiln 34, the finished tiles are passed through quality control and classification before packing and storing.

At different stages along the production line, various parameters are measured, 30 including the following. In the clay mill 16, the flow rate and machine state are monitored. In the slip tank l 8, viscosity, density, pump pressure and process loss are monitored, as well as water flow into the slip tank 18. In the spray drier 20, slip particle

size, machine state, feed flow rate, burner temperature, moisture content, exhaust temperature, outflow rate and local feedback are monitored. In the powder silos 24, powder humidity and powder particle size are monitored. In the press 26, the machine state, humidity, operating pressure, product size/rate, tile thickness and process loss are S measured. Also provided are a vision system inspection and a delaTnination check.

In the drying rack 28, the air temperature, the tile temperature, tile count, machine state, drying cycle and process loss are monitored. Also measured are the bulk density and humidity and a check is made of size and planarity. In the glazing line 30, 10 the disc speed, glaze viscosity or density and glazing line speed are measured. Also, the weight of tiles entering the glazing line and the weight of tiles leaving the glazing line are measured. At the exit to the glazing line 30, colour defects are identified. In the kiln 34, the gas and air pressure is measured at each burner in each group. The section firing time is measured as well as the section temperatures and the tile count and I 5 process loss. In the quality control and classification section 36, texture analysis is performed, surface defects are identified, planarity defects are identified and a size check is performed.

Each of the active stages in production has its own local controller (typically a 20 programmable logic controller) for controlling parameters inherent to that station. For example, in the spray drier 20, the feed flow rate, burner temperature, moisture content, exhaust temperature and output flow rate are input variables to the local controller to control that drier. As well as being input variables to the local controller, these variables are provided onto databus SO for retrieval by central controller 52. In the press 25 26, humidity, operating pressure, size check, stroke rate, tile thickness and process loss are all input variables to the local controller. In the drier 28, the local controller input variables include air temperature, tile temperature, tile counter, machine state, drying cycle and process loss. In the glazing station 30, the local controller input variables include disc RPM, glaze viscosity/density and glazing line speed. In the oven 34, local 30controller input variables include gas and electric pressure at each burner in each group, section firing time, section temperatures, tile counter and process loss.

In order that data from the production process can be used for analysis of tile defects, it is necessary to be able to identify, for any given tile, the conditions it experienced during its production. In practice, this means that information on the exact time that the tile passed through each stage of the production must be known, so that the 5 conditions at that time can be ascertained by querying the database 54. There are two possible methods of achieving this end, either physically track tiles as they move along the conveyor belts, or individually mark tiles and then read the identification back from the tile at the relevant reference points. Each of these techniques has different merits and both techniques are used to take advantage of their special characteristics. Details 10 of the preferred method of individually marking tiles are described later below.

A description is now given of the data acquisition architecture, in which the

term "variable" will be used to refer to a single datum from a production process - a temperature, tile count, humidity value etc. Variables have various properties including 15 a name and a current value, an up-to-date period and an access control list which defines which users may read or change that variable. The variables are grouped together into "machines". A machine generally represents a single process in the production area, for example a press or a kiln. The primary purpose of the "machine" grouping is to allow data from similar locations (in particular from adjacent locations in a single 20 programmable logic controller's memory) to be grouped together and fetched simultaneously across the field bus network. Each "machine" is managed by an

individual process on an industrial personal computer running under Linux (trademark) operating system located in the factory.

25 A scalable model is provided for managing and accessing the data acquisition processes. Although all the data acquisition is run on Linux, it is also useful to be able to connect higher-level devices directly into the system architecture. CORBA (trademark) is provided for this purpose. To implement variables individually as CORBA objects would have been inefficient, so the smallest object presented at the 30 CORBA level is the machine. Each machine provides methods for access to its variables, and also basic identification. The distributed nature of the CORBA object model allows 'machines' to be run on multiple computers throughout the factory, and

hence allows the design to scale as it grows from the prototype on a single line to control the whole of the production area. To keep track of the logical machines that are distributed throughout the factory, an extra CORBA object is used with a known Object Reference, which is called the "marshal". This allows all users to obtain the Object 5 Reference required to locate the particular machine in which they are interested, or to list all the available machines in the network, for generic browsing.

The variables that are monitored fall into three classes of storage requirements.

The majority of the measurements taken are not specific to any particular tile; rather 10 they give an indication of the conditions for a number of tiles i.e. production trends.

Therefore, these variables are stored chronologically. By looking at the raw data the controller 52 can tell what the conditions (e.g. temperature) were in a given location at a given moment in time. When coupled with the tile tracking techniques, which allow determination of the time a given tile passed through each machine, this means that the 15 conditions which any given tile experienced as it passed through the machine, can easily be ascertained. The majority of the data is stored in this form, so as to enable the acquisition of simple data to be de-coupled from the operations of code reading and printing equipment - the correlation can be perfonned at a later time / date, when the information is required by the users. This feature is extremely desirable, because it 20 means that the complexity of the highest priority task - the data acquisition - is reduced.

Some variables, however, are highly specific to the particular tile on which they were measured - for example, the classification produced by the ATIS (trademark): Integrate. The measurements of these variables can not be stored chronologically, and 25 these are stored against the particular tile ID for which they were measured.

The third classification of data storage, if it can be considered similar to the others, is the requirements of the code readers themselves. They must record the time and place at which each valid code was read from a tile, in order that the correlation as 30 mentioned above can later be performed.

In order to provide accountability for changes introduced, there is a need to have a concept of 'users', each of whom must enter a password in order to log on to the system. However, it should not be necessary for each "machine" object to continually verify the credentials of each user who attempts to make changes. Therefore, the 5 security model implemented it uses a central user authentication server to provide tickets' which are cryptographically signed and verify the identity of the user who bears them. Because the "ticket" is signed by the authentication server with a known key, the machine can be sure that the user bearing the key really is who they claim to be. The authentication works as follows. During installation, a private/public key pair is 10 generated for the authentication server. All the machines are configured with the public half of the authentication server's key. When a client program is started, it generates a "session" key pair, a private and a public key that are used for the lifetime of that client.

It initiates an encrypted connection to the authentication server, and provides the public version of the session key, along with the username and password of the user who is to 15 be authenticated. The authentication server then verifies the username and password given, and creates a "ticket" containing a timestamp, the public part of the session key which was provided, and a numeric user identification and list of "groups" to which the user belongs. The whole of this "ticket" is then cryptographically signed with the authentication server's private key.

In future communication with machines on the network, the client presents this "ticket" to the machine from which it is requesting services. Because the ticket is signed with the key of the authentication server, the machine knows that it can trust the ticket.

In addition to providing the ticket, the client also signs each request which it makes, 25 using the private half of the session key. In this way, the machine which is providing the service can be sure that the ticket is being used correctly, and that it has not been "stolen" by a third party who is capable of "snooping" the network.

The database 54 provides fast and reliable means to back statistical analysis 30 tools. Table designs described below take into account these requirements for optimal performance. The database provides the ability to tap into the vast amount of data inside all the tables to glean the information that facilitates query analysis.

Apart from updates, typical queries from users are the retrieval of data over a period of time for a particular variable. For a constantly changing variable, updates every 10 seconds leads to over 8,000 continuous values per day to fetch. Database 5 performance will depend on table structure and analysis techniques to keep data transfer and computation as low as possible.

With regard to statistical analysis tools, most anticipated queries will require to join several tables for a principal component analysis (PCA) or other decision making 10 tools - estimated average between 15 and 20 tables to a machine - that leads to a total of 150.000 rows. As most tables will be accessed to retrieve continuous data, careful design of the database is necessary.

The database is able to cope with an important flow of data coming from the 15 acquisition processes. The following characteristics of this database are similar to those of data warehouses: large table size, large table row count, low number of online users, typical operation is to produce reports, analytical requirements are medium to high, regular fulltable scans, historical data of 1 year or more.

20 A variable is referenced according to its properties: machine type, machine number, variable name and variable index. Relevant variables are selected for storage.

Each variable has its own dedicated table to store the data. Coupled with the features of Oracle 8i (trademark), it is possible to ensure access control of specific 25 tables by authorised processes as discussed above. All tables recording variables have the same structure: (<match Type>, <match ID>, <bar name>, <var index>: date, value). They have the attributes: Indexed by (date) and Primary Key (date). This will facilitate their management and offer plenty of flexibility for all types of access.

Internally to the database, data for each variable is sorted by date since all queries are 30 date related. "Value" can be of any type.

An acquisition has several ways to submit data to the database. The relative merits of the various techniques are discussed here.

An acquisition process may have direct threaded connection to the database.

5 For a smooth run, the main loop doing data retrieval from acquisition processes will be on a different thread than the one performing the database submission. While this technique makes the acquisition process development more complex, it ensures maximum reliability on the data delivery to the database.

10 In the direct threaded connection case, the acquisition process needs to have access to the database shared libraries or to use jdbc/odbc to communicate. This has the drawback of requiring a permanent connection to the database for each acquisition process on the system and thus increasing licensing costs dramatically. Another solution is to make the submission thread separate as a CORBA object. Access to this object is 15 done via a push/pull mechanism. A push mechanism takes advantage of CORBA broadcast ability to distribute data efficiently, instead of a pull mechanism that requires regular checks from the third party. Although acquisition processes are easier to produce because the third party is generic, in some circumstances, data delivery may not be insured. Both designs are implemented and the choice depends on the level of service 20 required.

End-users do not have use for display of raw data, but instead they use one of the following classes. The Java MinMaxAvg class provides a simple average value calculation over a set of (Time, Value) data. The set is spread over a period of time 25 divided into a number of sub-periods summarising the evolution of the value. For each sub-period, an average value is generated with timestamp set to the meantime of the sub-period. This produces a fairly good estimation because values are not evolving very fast. However, the variance value should be incorporated to report the average value result accuracy. The Java Interpol class provides nonlinear interpolation to the set of 30 points. Given a set of (Time, Value) data, it uses a Lagrange 4th order algorithm to interpolate existing points over a period of time to produce regularly time spaced values.

Ideally, all the conditions that have affected each tile during its manufacturing process should be recorded. These conditions are reconstituted by correlation between data from submission processes and the time a tile passed strategic checkpoints. This 5 ability to track tiles not only gives accurate feedback on production performance, but also enables end-users to obtain information about product/customer correlation. Along with the tile ID issued by the printer, additional ID data is stored in a tile tracking table having: id, classification, press time, press 1D, drier time, drier ID, glazer ID. The table is indexed by ID. (This table is an example and does not include all machine types 1 0 required).

The monitoring system allows users to be notified when preventative maintenance (PM) is scheduled and due. These events are at present entered in one of two ways: either directly into the user interface, or directly from the monitoring system.

15 They are stored in a standalone table of events in the database. Like variables, events have time stamps to enable the user to determine when events have occurred. Multiple events of the same ID can be recorded in order to allow tracking of their history over time. For example if an event has failed to be cleared then it is important to be able to ascertain the reasons for such delays. Each record contains the location of the machine 20 where PM is due (machine_type and machine_number) and a description of the PM task

(title and message). In addition it contains information for notification of the event directly to the person responsible (responsibility), and the notification method to be used (notification), via email, SMS or other message services. The relative importance (gravity) of the PM task is also stored. Classification by gravity includes the following 25 states: preventive, warning, critical. Provision for user defined states is also included.

Status indicates if the event has been started/completed/delayed. The status of a PM event changes as time progresses, starting with "started", possibly moving through "pending" or "delayed" and finally on to "completed". Each event is given a unique identification (ID) so that the system can track the progress of the event.

Whenever new information about an event is passed to the database a new record is entered into the database table thus allowing the user to track the history and

evaluate progress. An important feature includes the grouping of related events per task.

A "group id" field in the "events" table can be used to display this information.

The controller 52 is modular in the sense that it has a software module for each 5 of the machines it controls.

The operation of the modular controller 52 is determined by two sets of factors.

The overall control scheme used is generated by the graphical design tool, and is read in from a configuration file when the modular controller 52 is started. In addition to this, it 10 is possible to modify the parameters of the control blocks at runtime, and it was necessary to allow this configuration to be performed through a user interface integrated with the remainder of the monitoring system.

Parameters of the controller 52 are presented as extra variables in the machine 15 for which the module of the controller 52 was designed. The modular controller user interface is combined into the existing user interface for that machine.

As an example of a monitor controller interface, a press monitor controller interface can be provided with fields into which data can be entered. Table 1 gives an

20 example:

Density at S le 2.08 km3 amp Humidity at Sample 4.5 % Pressure at Sample 300 kg/cm3 Target Density 2.1 kg/m Density Deviation 2 % Controller loopmode 0.0 Density limits (min/max) 2.0 2.3 km3z Specific pressure limited (min/max) 270.0 330.0 kg/cm Humidity limits (min/max) 4. 9 5.8 % New specific pressure 0.0 bar Table I

25 The preventative maintenance (PM) scheduler allows an integrated way of predicting, notifying and controlling PM events. An event is simply some task, which

should be performed by factory staff in order to allow production at the highest level of efficiency possible. Events have a unique identifier, an associated date, a textual description, a severity level, a record of the machine to which they refer, and various

other status information. The scheduler is implemented within the database 54. The 5 architecture allows for monitoring alarms to be treated in the same way as other PM events. Typically these would carry a higher severity level, and may be notified to the end user differently, but the underlying database in used in all cases.

Events can be defined either by entering a priori knowledge into the user 10 interface, or by automatic notification of an event from the monitoring system itself.

Full automatic notification of PM events has been implemented in the ATISTM: Integrale product, and these events are passed into the PM scheduler.

End users can be notified of pending events by a "maintenance" button in the 15 interface. Clicking this button launches a separate window giving a short overview of pending preventative maintenance events. The end user may then zoom in on each event to see more detail, and to change the event's status. Hyperlinks to online documentation documenting the precise procedures required may be included.

20 Events are notified to the end user not only by means of the display screen user interface, but also by any method of communication which the underlying computer 52 has access to. This includes activating audio or visual alarms, sending GSM SMS text messages or faxes to operators, or sending popup messages to managerial staffs computers (irrespective of whether they are running an Impact display screen at the 25 time). PM is a key element in a manufacturing enterprise resource management. To make full use of its capabilities it requires a comprehensive database on maintenance events associated with all plant equipment, in order to implement an expert predictive 30 behaviour. This can only be achieved after all logged data from the monitoring system is analysed and trends established and correlated with the relevant equipment.

Two types of user interface are provided. "Display screens" are highlevel overview screens which will allow access not only to the current monitoring status, but also access to the historical data and inference aspects of the user interface, by leveraging the database connection. "Local clients" are terminals designed to be 5 installed within production areas, for use by factory staff. They will facilitate the manual entry of data that is not measured online for example, measuring the breaking load of a tile at pressing requires a manual amine destructive test along with providing an instant snapshot of monitoring. All data displayed by local clients will come directly from the Corba- based acquisition system.

Starting the display screen user interface launches a window, which provides a graphical, clickable, map overview of manufacturing sites, initially globally, and then in the specific country, factory, line and individual process of interest. Clicking on a manufacturing location launches the factory overview screen, which takes the form of a 15 floor plan. Each factory has several production lines, and clicking on the production line of interest launches the line overview screen. Key parameters of interest for each manufacturing process, usually those which indicate quality and production throughput, within that line are displayed, with arrows indicating how the processes fit together.

Again, clicking on a process name launches the relevant process display screen. A 20 Process Display Screen gives a graphical overview of the current status of a given instance of a manufacturing process, such as spray drying, pressing or firing. The current value of all variables to be monitored is displayed, current production total information is displayed, and historical values of variables are displayed graphically.

The historical data will be obtained by querying the factory-wide database, while, in 25 most cases, the current values of data will be displayed using the variable server of the acquisition system.

Local clients are interfaces to allow display and entry of information on the factory floor, using a high-level protocol over the factory field bus network. Examples

30 are summary screens for shift supervisors, and data entry terminals. These terminals

need have no interface to the database system; rather they communicate with acquisition processes over the network. Local clients can run on basic hardware in text mode, or

can be written in Java to run on new or existing PC hardware in a graphical mode. A glazing terminal Local Client shows a breakdown of the production line into positions, at each of which the following variables are recorded: time, glaze code, silkscreen code, weight, viscosity, density, lateral density and whether this represents a change on 5 the previous measurements.

Tile marking and tracking will now be described in greater detail.

There are three reasons for wanting to track the movement of individual tiles 10 through the various production stages, each requiring a different quality of information.

The first level is tracking tiles over short lengths of belt. The information needed is simply equivalent to a malfunction alarm connected to but not limited to tile jams.

When a tile jams, it is most likely that the subsequent tiles along the belt will also either Jason or be forced off the belt with consequent loss of production and / or shop-floor staff 1 S injury. The second level of information relates to the cross-referencing of tile quality-

state to machine status. This information can be used to monitor production and focus more closely on the combinations of machine parameters causing defective tiles. To obtain enough information for 'trend analysis' it is sufficient to collect data on a representative sample of the particular production batch. During its progress along a -

20 production line, any given tile can be removed from the belt and replaced in a different position andlor orientation, or be rejected. Between processes, tiles may be stacked for storage or temporarily removed from the line to balance flow rates. For this reason, it is impractical to attempt to identify individual tiles accurately by means of counting tiles passing key points. To achieve reliable identification we need to physically distinguish 25 each individual tile using an ID mechanism. This can be achieved by marking each tile at the exit of the press and then reading its ID as it passes through each production process. The third level, aims to provide all the benefits mentioned for levels one and two, and in addition the benefit of ensuring that every tile made can be uniquely identified. The most significant benefit of this is that each individual tile can be traced 30 back to the particular press, die and the time it was produced. To achieve this an ID format is needed that can provide enough combinations to mark each individual tile

produced by a factory over a required period of time. In addition, the marking method must ensure 100% reliability.

Tile tracking of unmarked tiles involves creating a model of the flow of tiles S along a belt that allows the prediction of the position of all the tiles on the monitored section of belt. This is explained with reference to Figure 2 (which is not to scale). A number of proximity sensors 80 and 82, a shaft encoder 84 and a small micro-controller 86. The micro-controller can be trained so as to know the length of the section of belt 88 to be monitored and the size of the tiles 90 passing along the belt. This information 10 allows two features ofthe tile flow to be monitored.

One feature allows for the detection of missing tiles from the belt. This feature resolves an important production problem that arises when tiles become raised clothe belt and get caught in the rollers 92 and 98. In such a case, subsequent tiles passing 15 under the raised tile get damaged resulting in large number of rejects when this condition remains undetected over a period of time. The second feature allows monitoring of tile separation in order to avoid tile jams. This is monitored at each proximity sensor. The tracking unit needs to know the tile size and the belt length (abutting tiles cannot be distinguished from single tiles unless the approximate length of 20 the tiles passing along the belt is known).

This type of tracking system allows tolerances to be set. For each feature there are two tolerances - a warning and an alarm tolerance. On a warning, a visible, and possibly an audible, notification can be made. On an alarm being raised, visible and 25 audible notifications can similarly be made. Also, a signal can be sent to the belt controller 86 to stop the belt, allowing for manual intervention. This tracking subsystem can be presented to the overall monitoring system as another machine. It can report transport statistics and alarms. It can also be informed of the tolerances to apply to the system. The subsystem has both a CAN field bus and a serial line over which this

30 communication can occur. The system is independent of any tile code reading machines. While these may be used in conjunction with the tracking sub-system, it is

possible to fit the sub-system into factories as a stand alone solution to provide a full belt monitoring system.

The belts 88 on which the tracking subsystem is designed to run are not 5 accurate belts. Their main aim is to transport tiles 90 across the factory as cheaply as possible. As a result, tiles always slip on the belt to some extent and the belt slips against the motor. For this reason it is important that tolerances of the subsystem can easily be adjusted. On a belt where tile slippage is high, it would be expected that a large amount of slippage would be allowed and tiles would be spaced out so that they 10 did not run into each other. Figure 3 shows an example of slippage that can occur due to temperature. As the belt warms up the belt slippage against the motor decreases and so the effective slippage of tiles increases. Large jumps occur, as shown in the figure above, when the belt stops and is allowed to cool for about half an hour.

15 Aspects of the tile marking will now be described in greater detail.

Each ceramic tile exits the press face down, and is immediately flipped face up thereafter maintaining this orientation throughout all production processes. The face of the tile may be subjected to one or more surface treatments such as glazing or polishing 20 before firing. On its passage through these stages, the tile may be rotated many times.

Marking the sides of the tile is impractical because it is impossible to fix the orientation of tiles in a horizontal axis. The only place to mark the tile is on the bottom surface, in the centre. It is only in this position that the location of the ID can be guaranteed relative to the leading edge of the tile (for a specific tile size), although the orientation 25 of the ID will still be unpredictable.

Whatever ID system is used must fulfil several important criteria. Most significantly, the marking and reading processes must not damage the tile - this criterion makes a contact solution much less attractive. The ID mark must survive the full 30 production process and be readable once the tile has left the factory. This means that it must be unaffected by both the engobing and firing processes. Laser etching is a possibility, using the laser to burn away a thin layer of the tile in a specific pattern.

However, laser systems do not work well in a dusty atmosphere, and this is a significant problem within a ceramic factory. Adhesive labels are available but are unsuitable for temperatures within a ceramic kiln. The preferred option is therefore an inkjet printed ID code, using ink that contrasts to the tile body and remains intact through the firing 5 process. The full traceability offered by allocating each tile produced with a unique ID is very cost effective for the higher quality end of the tile market whilst still offering cost savings in the low quality market. To determine the necessary ID format the information content had to be exactly specified. The simplest method is to give each tile a different serial number. Allocation of sequential numbers based on time-stamping information in the

database has the problem that two sequential lD numbers could be found on completely different types of tile pressed from different presses, merely because they were pressed one after the other. This is not necessarily unacceptable, but it would be preferable to 15 be able to extract more information directly from the tile ID itself, without reference to a database. For this reason, it is preferred that the ID numbers contain a reference to the press and the die from which they were pressed. Allocating 6 bits of data to specify 20 which of up to 64 presses the tile came from, and 5 bits to specify which of up to 32 dies within that press, being able to identify all tiles within one pressing only requires an additional single number. Based on a press working at 100 strokes per minute and never stopping, 33 bits of data would give enough numbers to identify each pressing for approximately 163 years. However, it only requires an additional 2 bits of data to store 25 a timestamp instead, showing precisely when the tile was pressed to a resolution of Cloth of a second for 108 years of constant production from the same press. The additional benefit of being able to see directly where and when the tile was pressed is why we adopted 46 bits to be the minimum amount of data for a tile code.

30 The code reader is based around a 'smart camera' system. This is a compact, black-and-white video camera on a single PCB with video memory and an image processor. It integrates a 500x582 pixel Sony ICX055AL CCD sensor with an Analog

Devices ADSP2 181 signal processor. The camera has an RS232 serial port, a 24V process control interface and an analogue video signal output. The advantages are that all the hardware required to acquire and store the image, locate the data matrix, decode it and transmit the decode ID number are present on a single board which is both 5 cheaper and smaller than a traditional system based on separate camera, frame-grabber and image processing system.

The ideal data matrix is a black on white image of cells, each of which is a perfect square, and in which each square is touching the next. The outermost cells of a 10 data matrix are used as a finder pattem; this is used for finding the position, orientation and number of cells in each dimension of the matrix.

The above-described system, permits the rigorous definition of products and automation requirements, provides various measurements and critical process variables 15 and allows the monitoring of their effects on product quality by means of process variable product quality correlation. A production-wide infommation system is provided for product tracking and process loss tracking. By an iterative process, the functional dependence of product parameters on boundaries of process variables is established and e quantitative knowledge is stored in the database 54.

To facilitate the iterative process of correlating process variables to product quality and identifying appropriate boundaries for product process variables and different functional dependencies between product parameters and those boundaries, the infrastructure is scalable, allowing full connectivity and monitoring of, in the first 25 instance, sensors that are standard in equipment of this nature and, additionally, further sensors described below that permit further correlations to be explored.

The system has a resilient, distributed architecture for acquisition of process data and has capabilities to support: complete product traceability; unifomn access to 30 production variables defined at levels of ISO stack; wide level of field bus support;

internet enabled portable user interfaces; database functionality providing complete access to historical data and high level of security for intemet enabled access.

At user level, the system provides hierarchical displays of process and product information configured to meet the requirements of the following remote clients. The hierarchical displays provide: (i) an overview of quantity and quality produced over 5 different timescales, product type and production process involved for the purposes of a general plant manager; (ii) complete process information overall machines involved in a given sub-process for the purposes of a sub-process controller; (iii) complete information an actual machine and all sensors monitoring and controlling the particular machine, for the benefit of an operator; and (iv) predictive information on machine 10 status for a maintenance manager.

Pseudocode is now provided summarising the information requirements for on-

line measurement and control of process variables and product parameters and for calculating in-process loss and total production losses for each type of product.

For each type of product (Tp) List total processes (pi) {for each (p,) List materials-in (Vj) [(V') (V.)] {for each (V;) List variables (V') n requiring monitoring {for each (Vj) n state measurement frequency state accepted limits (A) state critical values (K) state required ACTION if (V; ) n. > K (if (V;) n. > K, previous process defective) state required ACTION if (V;) n. < K but not in limit A, <(V)n c A2 } Measure Q(V')= quantity of material (V;) in } List process parameters Pn [Pl P {for each Pa state sampling frequency state operational constant (B) state critical limits Bn state required ACTION if P n is outside limits (if P n is not within limits B' <Pn< B2 machine malfunction) } List machine states M s {for each M s state required information output } List materials out (VO) I(VI) (VO)] {for each VO List variables (VO) n requiring monitoring {for each (VO) n state measurement frequency state accepted limits (C) state critical number (L) for rejects (trend monitoring) state required ACTION if number of rejects R>L } Measure (Vo) n = Quantity of material (VO) out } } Calculate P. {Q(Vj) - Q(Vo)} = in-process loss Calculate Tp {{Q(V;) Q(Vo)}} = total production losses for each type of product.

PI

Referring again to Figure 1, the press 26 has various sensors connected to the bus 50. Among these are sensors at the input to the press 26, which include a clay sensor and a belt sensor. At the output of the drier 28 (which together with the press 26 forms the press zone) there are the following sensors 62: a tile temperature sensor, a tile 5 counter, a belt sensor and a shaft encoder photocell. The press 26 has a programmable logic controller, as shown at connection 64 and this is connected to the bus SO via server 27. The drier also has a programmable logic controller connected at connection 66.

The drier and press together form a pressing zone which has a programmable logic controller (not shown) connected to the bus 50. This programmable logic controller can 10 also receive an input from a laboratory, testing the bulk density of samples of powder from the powder silos 24.

In addition, the following novel parameters that affect the quality of the pressed tile body are measured: clay humidity and bulk density. The monitoring system 15 presents the values of these parameters to the controller 52, which then determines the correct value of the pressing pressure as a function of humidity and density. The clay humidity is measured at position 68 in the process of Figure 1, but may alternatively be derived from the exhaust humidity of the spray drier 20 derived from position 70. The bulk density is also measured at position 68. Bulk density is the density of the powder 20 when taking into account any cavities in the powder or non-homogenous "clumping" of the powder. The bulk density can be measured by sampling off-line, but is preferably measured on-line.

Figure 4 illustrates an adaptive control loop for the pressing zone integrated 25 into the overall monitoring system. In the control loop, step 100 represents the sensing of the clay humidity at position 68 in the process. The output parameter H (instantaneous moisture value) is input to a filter 102 as well as a model 104 simulating a press and the programmable logic controller running the control program 106 that controls the actual press. The output of the filter 102 is a value HE, which is the filtered 30 moisture value. This is input into a pressure adjusting algorithm 110. The pressure adjusting algorithm l l O receives, as an input, the required density Dr and gives as an output the pressing pressure P. The pressing pressure P is input into the simulated press

model 104 which provides an output Ds which is the simulated density to a filter 112.

If the output of the filter 1 12 is out of range in step 1 14, the result is input to the pressure adjusting algorithm 1 10, which causes a change of pressure. If it is within range, the process returns to step 100. The output of the pressure adjusting algorithm 5 1 10 is fed to the real press in step 106 and the resulting tile is subjected to a density measurement giving a real density D. If the real density D differs from simulated density by more than a certain tolerance (step I 18), a check is made in step 120 to determine that the sensor H is properly calibrated and if not, it is calibrated in step 122 If the sensor H is properly calibrated, a set of adjustment curves is adjusted in step 124 10 These adjustment curves give pressure as a function of density and instantaneous moisture value and density as a function of pressure and instantaneous moisture value.

These curves are used in step I 10 to calculate the desired pressure for a given required density. 5 In the glazing zone, which encompasses the glazing line 30, a tile counter is provided at location 72 and a second tile counter at location 74. A direct comparator may be provided comparing the tile counter at the exit 74 with the tile counter at the entrance 72 to provide an indication if tiles are lost in the glazing process or if tiles separate into tile fragments that are counted as separate tiles. Additionally, weighing 20 means may be provided at these locations and a comparator to measure the increase in weight through the glazing process and thereby indicate the thickness of glaze deposited. The glazing zone 30 also has an output to the bulk density lab.

Turning to the zone of the kiln 34, a tile counter is provided at the entrance at 25 location 76 and the tile counter and the exit at location 78. These counters provide signals to the programmable logical array of the kiln and its associated server 35 and are made available to the bus 50. Additionally, at the exit 78 or in the quality control and classification section 36, a novel tile surface-temperature sensor is provided as well as an automatic tile size sensor and an ATIS (trademark) quality assurance system.

Thus, it has been described how the following additional novel sensors have been incorporated into the system to provide corresponding parameters: infra-red

humidity sensor, tile counters, bulk density sensor, tile surfacetemperature sensor, automatic tile size sensor and ATIS quality assurance system.

Technical, production, maintenance and section management are implemented S and further levels can readily be added, such as corporate management levels. The information provided can be for current data and also for historical data. The "historical period" available to each management level is different and is password controlled.

Historical data is mainly used for auditing performance for management decisions and for establishing process transport control strategies.

Referring now to Figure S. this Figure shows each of the stages in the manufacturing process of a porcelain tile. The process is further divided into raw material storage 200, batching process 202, grinding process 204, spray drying process 206, atomised powder storage 208, mixing process 210, pressing step 212, drying 15 process 214, glazing process 216 with its associated soluble salt and glaze preparation process 218, unfired tiles storage 220, firing 222, fired tiles storage 224, sorting and packing process 226 with its associated polishing step 228 and sorting and packing at step 230. The final step in the process is warehousing 232.

20 As shown in Figure 6, any one ofthe processes 202, 204, 206, 210, 212, 214, 216, 218, 222 and 226 can be considered to be an operation having an input and an output and input material variables, non-measurable process variables, output material variables and a control point 260 at which the operation can be controlled using measurable process parameters and output material variables. The input/output 25 variables and the process parameter are now described for each process.

In the raw material preparation and batching, input material variables are: material appearance, chemical composition, mineralogical composition, residue, humidity content, loss on ignition and rhetorical behaviour. Non-measurable process 30 parameters include composition of rejects. Measurable process boundaries include component's weight (kg).

In the grinding process, input material variables include: solids in recycled water (%), deflocculant and oxides weight, and water weight. Input materials include: the mill pre-load, oxides and pigments, deflocculant, pure water, recycled water and rejects. The non-measurable process parameters include: charge weight/dry material 5 weight ratio and grinding charge size distribution. The measurable process parameters include: speed (rpm), grinding time (h) and grinding curve. The output material variables include: slip water content (%w/w), residue at 42 rim (%w/w), density (g/1), viscosity ( E), temperature ( C), pH, colour and grain size distribution. The measurable process parameters include: speed (rpm), grinding time (h) and grinding curve.

Note that in each of these examples, the output material variables can be used to control the measurable process parameters using the programmable logic array of the machine in question.

15 In the spray drier, the input materials comprise a slip from the homogenization tanks and the input material variables are: density (%), grain size distribution, viscosity ( E) and colour. The non-measurable process parameters include: humidity in the combustion air, natural gas pressure (pci), jet wear and jet distribution. The output is the spraydried powder and the measurable process parameters are temperature ( C), 20 pump pressure (bar), slip flow (kglh), exhausted air pressure (bar) and gas flow (N m3/h). In the mixing step prior to pressing, the input material is the spray dried powders and the input material variables are the humidity content (%), the grain size 25 distribution, presence of lumps and colour. The non-measurable process parameters are the mixing stages and transfer devices cleaning. The output material is the powdered mixture and the output material variables are the humidity content (%), particle size distribution, colour and intrusions from other colours. The measurable process parameters include the coloured powders' flow (kg/h).

In the press, the input material is the powder mixture and the input material variables are the humidity content (HO), the grain size distribution, the presence of lumps

and the colour. The non-measurable process parameters are the die status and the machine status. The output material is the green tiles and the output material variables are the tile size (mm), tile thickness (mm), tile weight (g), delaminations, surface defects, breaking load (N/mm2), withdrawing expansion, colour and bulk density 5 (g/cm3). The measurable process parameters are the pressure (bar), pressing time (s), air removal time (s), tile thickness (mm), feeder speed (m/s), die temperature ( C) and cycles per minute.

In the soluble salt and glaze's preparation, the input materials are the raw 10 materials, the Grits, pigments, soluble salts, water and additives and the input material variables are the weight (g) and chemical composition. The non-measurable process parameters are charge size distribution, charge weight and mill geometry. The output product is the glaze suspension, and the output material variables are the residue at 42 rim (%), density (g/1), viscosity ( E), theological characteristics, colour and grain 15 size distribution. The measurable process parameters are speed (rpm) time (h) and grinding curve.

In the glazing process, the input materials are the glaze, englobe, additives and tile. The non-measurable process parameters are the status of silk screen, aerograph's 20 status (rpm) and tile porosity. The output material is glazed tile and the output material variables are the applied weight (g), appearance, defects, integrity, back tile cleaning and engobe application. The measurable process parameters are tile speed, glaze flow, glaze density and viscosity.

25 In the firing process, the input material is the unfired tiles. The non measurable process parameters are the firing atmosphere 02, C02, CO, temperature gradients, combustion air, exhausted air and gas flow. The output material is the fired tiles and the output material variables are the integrity, orthogonality, planarity, dimensions (mm), water adsorption (%), shrinkage (%), breaking load (N/mm2), defects 30 and shade. The measurable process parameters are the temperature/time curve ( C/mins), gas flow (Nm3/H), air pressure (bar) and tile speed.

In the sorting and packaging process, the input material is the fired tiles, the output material is the finished product and the measurable process parameters are: squareness, flatness, size and defect and shades analysis.

5 Referring again to Figure 5, novel sensors installed in the production line to complete the monitoring for process control are illustrated. These are: APH and conductivity meter 231 at the output from the raw material storage 200, a spectrophotometer 233 in the grinding process 204, a slip flow pump 235 at the exit to the grinding process 204, and air pressure sensor 237 in the spray drying process 206, a 10 humidity sensor 239 at the exit to the spray drying process 206, a tile thickness measuring device 241 in the pressing process 212, a tile weighing device 243 in the pressing process 212, a bulk density measuring device 245 at the output to the drying process 214, a tile temperature measuring device 247 at the same position, a glaze weight measuring device 249 in the glazing process 216, an air pressure measuring 15 instrument 251 in the kiln or the firing process 222, a roller speed measuring element 253 at about the same position and tile defects and shades monitoring apparatus 253 in the sorting and packing process 226.

Table 2 gives a list of new sensors that are installed on-line to enable 20 monitoring of variables for closed loop control at the various production stages illustrated in Figure 5. I

Sensor Measured Variable Electromagnetic flow meter Slip flow (Spray drier-inlet) Massic flow meter Slip flow (Spray drier-inlet) Differential pressure transmitter Air pressure in the spray drying unit _. _ Differential pressure transmitter Exhausted air pressure in the kiln preheating Double encoder Pressed tile thickness IR pyrometer Dried tile temperature (glazing line) Encoder Kiln roller speed - _ _

pH-meter Recycled water quality Conductivity-meter Recycled water quality Table 2

Table 3 lists the measuring techniques developed and implemented in the 5 system, further details of which are described below.

Measured Variable Measurement PrinciDIe Humidity of atomised powder 1R absorption Particle size of atomised powder Image acquisition Delamination after the press IR camera Density after the press Non- contact ultrasound Humidity after the drier IR absorption Cracks and broken corners at drier exit Image acquisition Structural defects Acoustic and vibration analysis Slip and glazes colour Spectrophotometer Tile surface defects at sorting line Optics / image acquisition Tile Texture at sorting line Optics / Image acquisition Table 3

The moisture content in the spray dried powder is an important output variable.

10 It strongly influences the behaviour of the powder at the press. A high humidity usually results in contaminating the press dies, a low humidity results in problems of bad de-

aeration and as a consequence defects of delaminations and / or low compaction.

Moisture control is traditionally done by an operator, off-line, using a thermal balance.

The frequency of this measurement is approximately once or twice an hour. The 15 temperature ofthe spray crier burner is set as a function ofthe value of humidity, in order to keep it within a specific range (usually from 5.0% to 5.5%).

An on-line sensor of good sensitivity, gives a constant and accurate measurement and enables on-line control. In addition the measurement gives also the ability to automatically set the burner temperature.

5 Two different sensors have been tested an infrared sensor and a microwave one. It was found that the IR sensor fits the needs of ceramic production and the error is about +0.2%wlw. Particular attention must be paid however to the powder for porcelain tiles, as different colour powders are used. Some colours require very particular calibration because they absorb electromagnetic radiation in the same wavelengths as 10 used by the instrument.

It is necessary to be able to control accurately the colour of slip in comparison with a standard and automatically formulate the corrections for colour formulation.

Instrumental colour formulation is a common procedure in paints, inks and other 15 industries, but it is not used in the ceramic production. Colour formulation instrumentation is to obtain a mixture of pigments that allows reproduction with sufficient precision to certain reference colour. Information derived by the spectrophotometer is used for pigment correction.

20 Once the characterization of a single component of pigments or oxides has been done it is possible to start the instrumental formulation. It is based on the principle that for a mixture of pigments K and S correspond to the weighted average of a single component (as function of the wavelength) so that it is possible to calculate for each wavelength the resulting reflectance and therefore the colour resulting from the mixture.

25 The calculation of a formula involves two aspects: a qualitative one, relative to the selection ofthe pigments and a quantitative one, relative to the choice ofthe dosing.

After having defined a rough initial formula the corresponding colour and the difference with respect to the reference colour is calculated, in terms of chromatic co ordinates. The derivatives of these co-ordinates with respect to the concentration of the 30 pigments and the variations required for the amount of the single component is then calculated. Because the functions are not linear the correct formula is not immediately obtained, but only a better approximation. The procedure is repeated iteratively until the

residual error is negligible. The method of instrumental colour formulation has been applied to the correction of the slip tanks with satisfactory results. The total colour difference is generally kept within tolerances with only one correction, and with a second correction a very good reproduction of the colour (DE<0.5) is achieved. The 5 only problems have been found in the characterization of some pigments, which are Fe-

based. These have shown an anomalous behaviour of concentration/developed colour with respect to the other pigments.

The methodology has the advantages of reduction of laboratory work and 10 reduction of the number of corrections. Additional advantages are the improvement of the quality and a rationalized choice of the components. The method is also used in the control of glaze preparation.

Supervision sub-systems for the following specific areas of the process are 15 provided: (i) raw materials for storage and grinding, (ii) spray drying, mixing and press-feeding, (iii) pressing, (iv) unfired storage, firing and fired storage, and (v) sorting and finished product storage.

In accordance with a method of the present invention, a mathematical 20 correlation between process variables is first obtained and from these a model of sub-

processes is derived. The modular controller in the controller 52 for the particular sub-

process is then implemented to regulate the stages within the sub-process automatically.

For example, in closed loop control of the spray drying burner, the following 25 sequence was used. The communications between the controller 52 and the spray drier programmable logic controller is fixed and that a model is identified for the spray drier.

This model is tested with simulations, the modular controller is applied to the spray drier in open loop to check the mathematical model and the model is applied by the modular controller on the spray drier in closed loop. In this manner, the automatic 30 control of the burner can significantly improve the consistency of moisture.

As a further example, closed loop control of air pressure in the spray drier is implemented to obtain a constant value of depressure in the drying unit. The pressure is measured by a differential pressure-cell transducer 237 and a programmable logic controller regulates the pressurization baffle, that is the baffle of inlet air. The main 5 baffle (outlet air) is usually kept open at a constant value. This control mechanism has given very good results in terms of constancy of internal pressure.

As a further example, closed loop control of air aspiration in the firing process 222 is provided.

The exhausted air coming from the firing and heating zone is aspirated by a fan towards the pre-heating, then it goes through a depurator that provides for the abatement of dust and fluorides. The regulation of the amount of aspirated air is very important for the firing process because it affects the quality of the tile, in fact the aspirated air has the 15 function of supplying the heat for the pre-heating zone, in which the residual water evaporation takes place (too strong aspiration would carry too much hot air from the heating-firing zone, leading to problems of cracks) and it affects the depurator functioning. 20 It was found possible to control the aspiration by the measurement of the decompression inside the kiln. To measure the value of air pressure a differential pressure transducer (Druckaufnehmer, mod. 230 VAC) was introduced in the pre heating zone. The measured pressure is then compared to the pressure set point and a PID controller regulates the aspirator fan by inverter. The set point is an important 25 parameter of the firing of each product. This kind of control can assume great importance when there are empty spaces between tiles, in absence of regulation, the aspiration is too strong and this leads to problems with successive products.

The advantages of closed loop relative to an open loop process are lower 30 consumption of fuel, better operating of the scrubber, reduction of dimensional defects.

The kiln infeed roller speed is controlled by closed loop in the firing process.

The speed measurements are performed by encoder and the regulation is made by inverters. The infeed-roller speed is then adjusted by the kiln PC 35, which also regulates the roller speed and the temperatures inside the kiln. This control allows 5 automatic change of the tile size (i.e. from 30X30 to 40X40), without manual intervention. The modelling of the spray drying process is now described in greater detail with reference to Figure 7, which shows a generic scheme of spray drying, for which 10 three main elements can be observed: pump system 300' heat system 302 and the drying unit 304.

Milling and spray drying are processes that transform raw materials to powder.

The milling process mixes the raw materials (water, clays, etc) making clay slip. The I S clay slip's main characteristics are density, viscosity, water content and particle size distribution. The spray drying process atomises the clay slip to powder, which should have a percentage of water (moisture) and specific particle size distribution. The significant elements of the system are a slip supply pump, a slip pump

20 control unit, filters and slip piping, a nozzle holder, a drying unit, a discharger, separator cyclones, a combustion system, a pressurization system, a hot air generator, a hot air duct, a distributor, a wet separator, a main extractor fan, a flue and electronic control equipment. According to one aspect of the preferred process of the invention, a model needs to be obtained identifying the correlations between the variables which affect the 25 spray drying process. The model will identify the required control of the variables to give the desired output variables for given disturbances i.e. variables which affect the output variables but cannot be controlled. The main output from the spray drying process is the powder, which is qualified according to its grain size distribution and its residual moisture. These variables have to be controlled within their set points. For the 30 grain size distribution, it can be measured using several sieves of different diameters.

For the residual moisture, a desired residual moisture of between 4 to 6% is needed for the next process (i.e. depressing).

The control inputs that determine the set point of the outputs are the pump pressures, hot air temperature, hot air flow and number of nozzles. The pump pressure influences all output variables. If the pump pressure increases more clay slip is 5 introduced, and with all other conditions constant, production and residual moisture increases. If the pump pressure increases clay slip spreads out in a more uniform way, and the grain sizes decrease. The heat contribution is defined by two variables - the flow and temperature of the air. The heat only affects the residual moisture, which goes down, as heat increases. The number of nozzles is used to control production. When the 10 number of nozzles is increased the clay slip flow must increase to keep pressure constant. Thus more clay slip is entering the spray drier, producing a higher residual moisture. Generally, when the number of nozzles is increased all parameters must increase too.

15 Possible disturbances in the process are nozzle wear and clay slip characteristics (density, viscosity and residue). The nozzle wear increases the diameter thus producing a lower pressure at the nozzle output, giving an increased residual moisture and grain size. At some unspecified time, nozzles get replaced. The only way to know the current state of each nozzle, is to monitor the time from installation. Clay 20 slip density is also a disturbance because it cannot be controlled at the spray drying system and can only be modified at the milling process. Clayslip density and water content are highly correlated. Lower water content produces a higher density. If the density goes down the resulting flow from the pump system increases, hence residual moisture increases too. Higher viscosity decreases the clay slip flow, therefore residual 25 moisture decreases. If the viscosity is too high the pumping system can be damaged.

Residue is the percentage of the higher size particles of the clay slip. It influences the gram size.

There are other variables related to control and disturbance variables, but 30 because they can be locally controlled by the control variables and are not related to quality variables, they are not discussed in greater detail here. The variables in question are the clay slip flow, which is a consequence of the pumping pressure (higher pressure

produces higher clay slip flow) and the output air flow and temperature, which indicates the volume of atomised water removed through the output air flue.

Referring to Figure 8, a control process for the spray drier consists of a feed 5 forward control 350 compensating for the effect of pressure on moisture and a Smith predictor-based control of moisture (351 and 352). The feed forward control 350 uses a reference pressure 355 and the Smith predictor-based control 351, 352 takes a reference humidity 357 compares this with an output humidity 359 and controls an input air temperature 361.

Referring to Figure 9, a schematic diagram of the press 26 is shown, with input material variables on the lefthand side, control input variables along the top and output variables on the righthand side.

15 Atomised powder from the silos 24 is transported on special belts to the press 26 and reaches the press with a humidity between 4.8 and 6%. Taking into account this humidity value and the apparent density required in the tile at the press output, the pressing pressure can be adjusted.

20 A model of the operation of the press was obtained relating the humidity of the powder to the press input and the pressing pressure with the apparent density of tile at the press output. By adjusting the powder moisture and pressing pressure, the powder particle size distribution, the slip rejects, the dye position and the pressing pressure and the deaeration time, it was concluded that the pressing pressure is directly proportional 25 to the bulk density, the SDP moisture is directly proportional to the bulk density and the slip reject is directly proportional to the bulk density, while the particle size distribution does not affect the bulk density. The die position is directly proportional to the thickness and the pressing pressure is inversely proportional to the thickness. It has further been observed that delamination occurs when all de-aeration parameters are 30 small. In normal conditions, these parameters are high. If delamination is observed at a later stage, it is caused by the kiln conditions further down the production line.

The principal control variable is the bulk density. This variable is mainly affected by the pressing pressure and the SDP moisture. The bulk density (d = f(P, H)) can be obtained as a function of the pressing pressure and SDP moisture values as a result of an experimental curve fitting. The slip reject influence can be modelled as a 5 disturbance that modifies these curves.

Having established the relationships between the press variables, these are integrated in a general press model and this model is used in the controller 52 to control the press in a manner similar to the control of the drying process.

It has been explained how a method is provided for volume manufacture of a product in staged production process such as the manufacture of porcelain tiles, in which parameters are measured at multiple points along the process and correlations are identified between the input and output parameters, including correlations between an 15 output parameter and a plurality of input parameters (e.g. tile density being a function of pressing pressure and instantaneous moisture value) or an input parameter and a plurality of output parameters (e.g.: in the spray dryer - volume of spray dried powder, moisture content and grain size distribution as functions of pump pressure of clay slip; in the press - green tile bulk density and strength being functions of moisture content of 20 powder; in the glazing process - glaze penetration into tile and quality of glaze application being functions of moisture distribution in green tiles; or in the kiln - quality of glaze finish and tile body integrity being functions of moisture content of tile body).

* Interdependencies between these parameters according to the identified correlations are stored in controller 52 and the overall process is controlled by the controller 52 at 25 multiple stages according to the stored interdependency in response to a single parameter change.

It is a further feature of the invention that a given stage in the process is controlled by the control controller 52 responsive to input parameters from another 30 stage in the process, i.e. parameters that are not inherent to the stage under control.

Parameters that are inherent to a stage under control are those parameters that are used by the programmable logic array of the stage in question while parameters that are not

inherent to that stage are those that are provided by some other stage in the process.

This is illustrated in Figure to.

Figure 10 shows two stages, 400 and 401. The first stage (stage A) may, for 5 example, be the spray drier and the second stage (stage B) may, for example, be the press. It will, however, be understood that there may be other stages in between these.

As a further example, stage A could be the spray drier and stage B could be the kiln. As illustrated in Figure 10, input parameters to stage A may also be provided in feed-

forward manner as input parameters to stage B. An example of this would be the use of 10 the density or grain size distribution, viscosity of the material entering the spray drier 20 as an input variable to the press 26. Another example would be the use of the moisture content of the powder from the powder silos 24 being used as an input variable to the kiln 34. In each of these examples, there is no intuitive reason why the input variable to stage A may be relevant to the process in stage B. but there may be hidden correlations 15 by which an output from stage B is indeed highly dependent upon an input of a stage upstream of the stage in question.

In the preferred embodiment, the controller 52 uses correlation identification algorithms (such as Winters-Holt and other variations on the exponential sliding 20 formula or by means of Chemometrics) to identify these hidden correlations. Similarly, an output from stage B indicating shrinkage in fired tiles or defects in fired tiles or defects in glazed tiles can be used in feed-back manner as an input to stage A. An example of such a correlation would be experience that loss of planarity in fired tiles had a strong correlation in certain conditions with pressing pressure or die temperature 25 or air removal time in the press, or indeed a correlation to some combination of these or other variables. Correlations identified by the controller 52 and stored in a database 54 need not be limited to the parameters of a given stage, because the controller 52 has overall view of all the control parameters across the entire process and is able to control one stage in response to parameters measured in another stage upstream or downstream 30 of the stage under control.

The on-line humidity sensor 239 of Figure 5 for measuring the humidity of powder entering or leaving the powder storage 208 is now described in greater detail.

The sensor 239 is a near-infra-red (NIR) sensor. Correct calibration of such a sensor is particularly difficult, but very important for reliable results.

NIR calibration accuracy can be no better than the sampling and the reference methods used to characterize the samples. The quality of sampling techniques, the laboratory references methods utilised and the range of samples available are critical in gaining the best possible accuracy in an NIR application. It has been found that the 10 calibration accuracy improves if performed by increasing the number of reference samples, but dispersion increases. This is due to the deterioration of reference samples, whose humidity changes during the long time needed for an accurate calibration. It is therefore necessary to either re-measure the reference samples during the calibration or to keep the calibration time as short as possible. This allows a reliable high sampling 15 frequency. The best result achieved is a calibration with an accuracy of about 0.60 % of humidity (+ 3 SD(HR)).

The incidence of interference typical of the on-line installation was evaluated by repeating the calibration procedure on-line. The calibration was performed with a 20 reduced number of samples, to provide a quick procedure suitable for industrial applications. It was found that an acceptable uncertainty (m = 0.96, b = 0. 17, SD(HM) = 0.26%, SD(HR) = 0. 26%) could be achieved using a minimum of 7 samples. The correction required for the application on different powder colours or for interference due to the movement of the powder on the conveyor belt was also considered.

The sensor developed was applied to a real-time control loop on the atomiser.

Results show that the implemented procedure is capable of regulating the input air temperature, in such a way as to significantly reduce the variations of the moisture content with respect to the desired set point.

Using the described NIR sensor, it is possible to control the spray drier using closed-loop process control. The spray drying process consists of drying the ceramic

slip by spraying it into a hot air stream, by means of which the water undergoes rapid evaporation. The product obtained is made up of round granules and the closed-loop process controls the particle size or size distribution and the humidity. These output variables are the most important output variables for influencing the compacting 5 behaviour of the powder in the next stage (the pressing stage). The manipulated input variable is also in this case the hot air temperature, T. which is regulated as a function of the difference d H = HM-HSP between required and measured humidity. This closed-

loop control is implemented in a programmable logic controller in the spray drier 20.

10 The time delay between temperature regulation and humidity variation (measured by the AIR sensor) is about 3 minutes. This is a measure of the inertia of the spray drying process. Results show that automatic, closedloop control provides much lower variation in output humidity than in the case using manual control.

15 A second parameter of the output of the spray drier, measured by sensor 239, is the ceramic particle size or granularity. On-line ceramic particle size measuring apparatus is now described with reference to Figure 11. The apparatus comprises a high resolution monochrome charge- coupled device (CCD) camera 500 having I k x 1 k pixels and equipped with a 16 mm macro-objective mounted on extension bellows. A 20 metallic surface 502 is provided for receiving the powder sample 504. Means for automatically depositing the sample on the metallic surface and removing it are not shown. The metallic surface 502 is illuminated with lamps 506 and 507 and an acoustic excitation source 506 driven by a signal generator and amplifier 508 is provided positioned beneath the metallic surface 502. The camera 500 is connected to a 25 computer 510 having image processing software. The apparatus is positioned near the belt at the exit of the atomiser (the drier 20), the colour of the metallic surface 502 is preferably chosen to give a high contrast against the particles 504.

In operation, the metallic plate 502 is made to vibrate at its first resonance 30 frequency, using the acoustic excitation source (loudspeaker) 506. This vibration causes the particles to separate and disperse within the sample, until each particle is clearly distinguishable on the plate. An image of the particles is acquired by the camera

500. This is then processed by the software in computer 510. The software used is developed under the National Instruments LabView (trademark) environment, using the Imaq-Vision (trademark) image processing library. It is convenient that the image is acquired and processed before any measurement is done, because the particles must be 5 highlighted with respect to the background. Once the measurement is performed, the

particles are returned on the belt and a new sample, extracted from the line, is analysed.

The dimensional range of the particles being measured (90-600 mm) necessitates measurements close to the powder in order to reduce uncertainty. At the 10 minimal focal distance a pixel is equivalent to 9 mm, and this is the maximum resolution of the particle's size. At this minimal distance however only a few particles at a time can be measured. The best compromise between measurement precision and statistical reliability of the results is used. Each measurement, taking a few seconds to complete, should be continuously repeated in such a way as to generate, every few 15 minutes, results which are statistically significant.

The main specifications of the acquired image are shown in Table 4.

Type CCD Progressive scan interline Number ofelements 1008x 1018 Colour Black and white Pixel pitch 9 x 9 mm Array of micro lens Standard, fill factor 60% Output digital Parallel bit Active area 9.1 x 9.2 mm Frequency of maximum acquisition 30 Hz Range of temperature 0 40 C Table 4

It is convenient to have the measurement and the data processing steps separated, as this improves flexibility. The measurement range depends only on the magnification factor of the employed optics. Starting from a single image, it is possible to extract several other important features of the sample, simply by modifying the 25 processing routines. For example, parameters related to the morphology of the articles could be extracted, as well as other typologies of characteristic diameters.

During the atomising process, two or more slip drops may agglomerate. It is preferred that the image processing software, during the separation step, is able to distinguish between agglomerates and adjacent articles.

The objects in the binary image after a number of erosions are labelled according to the specified connectivity and finally the original image is reconstructed without touching regions. This provides distinct objects without any significant loss of information compared with the original image. In addition, particles on the edge of the 10 measurement area, which may not be completely acquired, are also eliminated.

A sequence of deinterlating, hole filling and separation filters can be applied.

The described method gives results which are highly comparable, or even 15 better than more cumbersome sieving methods or more expensive laser methods of particle size measurement.

A description is now given of one aspect of the sensor 243 at the output to the

press 26. It is important to measure the green tile density in order to control the 20 pressing process. Prior art methods of measuring green tile density are based on

mercury absorption, which is not only dangerous for the operator's health, but is cumbersome in an off-line process and not at all suitable for online measurement.

It has been known in the past to use ultrasound analysis for measurement of 25 material properties, but prior art equipment has required contact between the sensor and

the object to be measured, making it limited in its usefulness for online measurement.

Ultran Laboratories Inc have introduced a non-contact ultrasound sensor (NCA1000, trademark), which claims to perform accurate, non-contact ultrasound 30 analysis with reduced signal attenuation and, therefore, a limited loss of information. In accordance with an aspect of the present invention, a non-contact ultrasound system is used in sensor 243 for online quality control and monitoring of green ceramic tiles.

Referring to Figure 12, the non-contact ultrasound sensor consists of three distinct components: an instrumentation rack 600 having a system controller 602 and first and second transducers 604 and 605. The instrumentation rack 600 has the S following plug-in modules: a video module 610, a keyboard 611, a mouse 612, various interface cards 614, first and second power amplifiers 620 and 621, first and second analogue I/O cards 622 and 623 and a digital signal processor 624. Connecting the power amplifiers and their respective analogue I/O cards with their respective transducers are diplexers 626 and 627. The power amplifiers 620 and 621 are 10 broadband amplifiers.

The object of the equipment is to establish the density-velocity relationship for green ceramic tile samples and to determine densities of tiles. The system is operated in direct transmission mode, i.e. ultrasound travels from the transmitting transducer 604 to 15 the receiving transducer 605 through ambient air and through the sample.

Initially the system is calibrated without a sample between the transducers. In this mode, the ultrasound velocity in air (Va), which is a function of temperature, is fixed, and the distance between the transducers (D) is measured. When the sample is 20 interposed, the system can extract the new time between the transducers (tc) and thus the time of flight through the sample (tm). Ultrasound (or any sound wave) velocity is a function of the density of the material through which the sound wave is travelling. Thus, the time taken to traverse a given sample is a function of path length and sample density. Knowing the transit time through the tile and the tile thickness (path length), 25 the density can be calculated.

Under laboratory conditions, the accuracy of the velocity measurement was found to be +/-0.5%. This value is higher for on-line tests. The stability and repeatability of the measurement is affected by the following factors: temperature, 30 variation due to the drier, dust, vibrations, humidity content of the tile, superficial roughness of the tile and movement of the tile.

so A further aspect of the sensor 243 of Figure 5 is detection of defects in green tiles. Such defects include delamination, cracks and broken corners. Delamination (voids) can originate at the slip preparation and at the pressing stages. For dust preparation, the most critical issues are variable humidity, low humidity or excessively 5 fine particles in the dust. For the pressing process, the critical issues are the velocity of the press during the first pressings, insufficient time between the first and the second pressing and accuracy of the manufactured die. Delamination inspection before firing is important, because materials can be recycled without any modification.

10 A variety of non-destructive detection techniques are applicable to ceramics and their composites. These include optical examination, liquid penetration and inspection, radiography and ultrasound methods. The preferred technique is an optical inspection technique based on infra-red imaging and uses a multilayer perceptron artificial neutral network as shown in Figure 13. The neural network has twenty-six 15 input neurons 700, four hidden neurons, 730 and one output neuron 740. In testing mode, the inputs to the twenty-six input neurons 700 are the features extracted Mom the measurement on the tiles.

The extracted features are any property of the image that can be represented as 20 a 256 point line profile. An example is a one dimensional projection along the Y axis of an infra red image, that shows the average cross sectional profile of the heat distribution of the surface. However the approach is not limited to IR images. Images obtained from visible or W light, radiographic and ultrasonic methods can also be used as inputs. Other examples of extracted features include the image histogram, cross 25 sectional profile, and Fourier or other transformed images.

The output is a single number between - l and 1, which classifies the tile as good or bad.

30 For the learning algorithm, a particular version of the backpropagation algorithm has been used, the "resilient backpropagation". This training algorithm is employed to lower the time necessary for the training cycles, which is an important parameter for an industrial application. Sigmoid functions have as a characteristic that

their slope must approach zero as the inputs increase. This causes problems when using steepest descent to train a multilayer network with Sigmoid functions, since the gradient can have a very small magnitude, and therefore cause small changes in the weights and biases, even though the weights and biases are far from their optimal values. The 5 purpose of the resilient backpropagation training algorithm is to eliminate these hannful effects of the magnitudes of the partial derivatives. Only the sign of the derivative is used to determine the direction of the weight update - the magnitude of the derivative has no effect on the weight update. The size of the weight change is determined by a separate update value.

By way of example of a defect measured in a green tile, the detection of delamination shall be described using infra-red imaging.

Referring to Figure 14, an infra-red camera 800 is positioned above a tile 90, 15 illuminated by a 2 kilowatt infra-red lamp (not shown). The camera 800 is a sterling cooled IR camera with a 256 x 256 CCD array of photovoltaic InSb detectors working in the 3 to 5 lam speckal range (shortwave). Means for moving the tile 90 into the field

of vision of the camera 800 and removing it are not shown. In operation, heat from the lamp flows towards the surface of the tile 90 and is reflected and captured by the camera 20 800. Feature extraction is used to extract from the data from the camera, features which better highlight the differences between two classes to be separated (in this case non defective tiles and defective tiles). The 1 6-bits grey scale infra-red image of the tile is converted into an 8-bit grey image and the contrast of the image is enhanced. i 25 The image is integrated along the Y direction, orthogonal to the direction of the tile movement on the production line. This operation transforms a 256x256 pixel image into a 256 point line profile. The Y direction has been chosen because, if the line is momentarily stopped, the presence of the rollers under the tile will cause temperature gradients to correspond to the temperature of the rollers. Integrating in a direction 30 orthogonal to the rollers, this effect will be reduced and it will just appear in the line profile as a slight increase of the general temperature level, without introduction of local

gradients. The line profile is normalised between 0 and 1. This eliminates the

dependence of the results from the temperature level, which could vary significantly for the different types of tiles and during the different seasons of the year. The resulting curve is filtered using a linear digital filter in order to eliminate the noise. From the 256 points line profile, a 26 points line profile is sampled in such a way as to reduce the 5 amount of data to be processed.

As shown in Figure 14, the defective tile having the delamination in its centre, has a superficial temperature in the delamination zone which is elevated by an amount DT relative to the superficial temperature of the integral zone, which would be the 10 profile for a non-defective tile.

The irregularity of the profiles achieved by processing the images of the defective tiles allows automatic recognition of delaminations. This behaviour was found to be extremely repeatable. In tests, a set of 33 samples were employed for ANN 15 testing. Among these 33 images, I 1 were used as a training set and 22 for classification. The classifier system performance was 100% correct. This system has a Noise Equivalent Temperature Difference (NEAT) of about 0.025 K. In practice, in on-line operation, the temperature of the matrix is higher than 20 the temperature of the powder because the tile is heated by the pressing process. This enables the IR image acquisition to be done without the need for the infra-red lamps.

Results show that the profiles of good tiles have higher temperatures at the edges and lower temperatures in the centre. By contrast, the peak temperature for a defective tile is in the centre. By projecting the profiles in the Y direction of the images, image 25 processing software can readily distinguish between good tiles and defective tiles, for example by projecting the IR thermal image to one axis and comparing the location of the peak of the image with the expected location for a nondefective tile, or by comparing the level of IR measured at the centre of the tile relative to the edge with the expected level for a non-defective tile.

Referring now to sensor 253 in Figure S. for sensing quality of finished tiles, two modules are used for this sensor. One analyses and classifies tiles in terms of their

textural and colour characteristics in "grades" (Textone) and the other in terms of frequency and severity of their surface defects (QS) in "classes". The combined results guides the tile in the designated stackingbay for final packaging. The Textone module is described in greater detail in GB Patent Application No 01 23984.7 filed on 5 5 October 2001, refiled as an International Patent Application on Monday, 7 October 2002. The QS module is the subject of British Patent Application No 01 21288.5, refiled as International Patent Application No PCT/GB 02/04028 filed on 3 September 2002. The contents of these applications are incorporated herein by reference.

l 0 The system runs on proprietary hardware and uses feedback control to maintain a constant environment. It has been integrated with the monitoring system and can generate auto-diagnostics for preventative maintenance. It inspects surfaces of width in the range of 100 mm to 600 mm and any length, at maximum belt speed of I m/sec.

The number of grades in a given production batch is defined by the user during a 15 training phase but these can be changed, if required, on-line during the sorting mode.

The number of grades and the criteria for the selection are fully user configurable.

The classification of tiles in terms of quality is based in a combination of defect type, severity and frequency. Tiles selected by grade are classified in terms of quality, 20 each grade having in addition to first, second, reject etc. classes a're-work' crass to enable manufactures to recycle tiles if defects can be eliminated by reprocessing the tiles. This is important for high quality products, which have a high value added.

Surface defects detected are scratches, cracks, bumps, uneven surface gloss, etc. as well as mechanically damaged tiles. The system can also detect fine and shallow scratches on 25 high gloss or polished surfaces that humans cannot detect at typical belt speeds and standard viewing angles. For each type of defect the user can define a class in terms of the defect's frequency and severity. The system is then able to combine the specifications to one classification criterion. The user interacts with the system via a

touch screen, and can override system decisions. Selection and classification results are 30 colour coded and me screen simulates the process in the various bays of the stacking machine.

Validation results for the described equipment indicate that it can sort 90% of production in terms of tile types, with consistency in access of 98%. It can select at twice the speed of humans and with an accuracy of 99.9% as compared to 75% being obtained, at best, from human selection online. The technology can easily be applied 5 to other types of materials such as natural building materials, wood and board, leather and others.

The Textone module is located along the conveyor, downstream of a climate control module.

The position where tiles enter the system has an adjustable height bar (the tile jam bar) which is used to physically trap over-height objects (such as stacked tiles) which could damage something inside the Textone module. A sensor is used to stop the belt if an over-height object tries to enter the system, the bar is precautionary.

Signal inputs are opto-isolated and can be configured to accept 24V NPN or PNP signals. A tile output sensor is mounted over the end of the Textone module conveyor. It determines the position at which the Textone module generates a sort code for the packaging machine. The Textone module is preferably designed to generate the 20 sort code at the end of the conveyor that passes through the system. A shaft encoder is mounted onto the conveyor motor spindle and measures rotation of the conveyor belt pulley. The sensor is used to monitor the movement of tiles through the Textone module. 25 A sort code is generated to indicate the grade of the sorted tile. This code is timed to arrive as the tile passes under a tile output sensor. The sort code is connected to the packaging machine. The Textone module generates 7 bits plus a strobe timed to appear similar to the tile output sensor. It has an automatic assignment of sort codes for grades which can be overridden. By default, it generates code l for the first grade 30 entered, code 2 for the second, code 3 for the third, etc. A reject is indicated by setting all bits high, i.e. code 127 if all seven bits are used. Table 5 indicates the number of grades that can be indicated from the number of bits connected:

I Reject onl Sort code: None Re ect code: 1 _ Y 2 2 grades plus reject Sort code: 1, 2 Reject code: 3 3 6 grades plus reject Sort code: 1, 2, 3, 4, 5, 6 Reiect code: 7 _. 4 14 grades plus reject Sort code: 1, 2, 3 12, 13, 14 Reject code: 15 5 30 grades plus reject Sort code: 1, 2, 3 28, 29, 30 Reject code: 31 6 62 grades plus reject Sort code: 1, 2, 3 60, 61, 62 Reject code: 63 7 126 grades plus reject Sort code: 1, 2, 3 124, 125, 126 Reject code 127 Table 5

The Textone module drives a simple remote display unit that can be placed 5 near an operator to indicate whether tiles are being graded or rejected.

It is important to control illumination very accurately, in order to optimise image details. Light sensors are mounted on brackets over fluorescent tubes. Calibration involves setting the lights to maximum brightness and adjusting their height until the 10 sensor output is at a predetermined level. As the system warms up, the maximum light level changes quite considerably and it is important that the sensors are calibrated at the same time so that lights are at the same point in their warm up curve.

It was found that the electronic circuitry (including the camera) had 15 unacceptable tolerance variations, so a special frame grabber with an analogue front end is provided consisting of a digitally adjustable gain stage followed by an offset stage.

The digitally controlled gains and offsets are implemented to 1 6-bits resolution.

The analogue gain and offset are used to compensate tolerance variations 20 during initial calibration (it is a two- stage process starting at the last stage of framestore assembly and completed at the full system installation and commissioning stage). A table of gain and offset values is generated for a set of different grey scale reference intensities. This table is later used to select the appropriate gains and offsets for the reflectivities of the tiles under inspection.

There is a 12-bit ADC output and then a PRNU is applied and a flat field

correction digitally in hardware. A 16X16 bit multiplier is used followed by an adder

for this purpose. The upper 16-bits of the multiplier inputs are used. The 12-bit ADC output is multiplied by a 10-bit constant (one per pixel). The upper 10-bits of the multiplier output is then combined with an 8-bit signed offset value. The result is a 10-

bit word of which bits 1-8 are extracted. These form the output graylevel. Bit 9 is 5 regarded as an overflow flag, and bit 0 is 1/2 LSB.

Shaft encoder and Input and Output sensor signals are also carried to QS straight from the Textone panel. The QS has two cameras, one vertical and one oblique.

The vertical camera is a TDI and the oblique a line scan. The viewing angles of the 10 oblique camera are a function of the surface characteristics of the inspected object.

Each camera has an associated illumination source, equipped with a photoresistor to control brightness. This is adjusted to give a fixed value when the light is operated at full power. The optical set-up is critical for the QS if it is to provide accurate and consistent results.

Various types of ceramic production and factory layout are possible. These include a single fired process having two driers followed by a glazing line, followed by a kiln, quality control and packaging. Alternatively, a twice-fired process uses two driers and a biscuit kiln followed by storage and thereafter from the storage a glazing 20 line, a further kiln, quality control and packaging. Quality control modules can be inserted at additional positions to minimise in-process losses and reduce scrap (rejects that cannot be recycled).

For inspection of embossed surfaces, a three-channel telecentric flying scanner 25 is used. The optical system contains a rotating multi-facet mirror and a large F#4 parabolic mirror of diameter 500 mm. This scanner is capable of producing high-

resolution (2048x2048 pixels) pseudo-colour images through a linear combination of scattered and specular light signals observed at three different angles. The grabbed image size is mainly limited by the DT3152 card video register. Image sizes as large as 30 lOOOOx 400 pixels are also feasible. Ceramic tiles having different surface characteristics and various degrees of embossed features were scanned. Their composite images showed a much greater amount of surface information than those obtained using

a normal colour CCD camera, or even a single channel FSS. Surface defects such as scratches and finger marks could be identified when the observation was made at non-

specular angles. Subtle variations of the surface profiles can be seen with the multichannel scanner. The variation of the surface texture is easily identified in the S pseudo-colour images. Finer details can be seen when the illumination is not normal to the surface.

Image analysis of embossed tiles for defect detection demands preprocessing because the elevated embossed profiles may not be easily distinguishable from the 10 surface blemishes or patterns. If the height information for each pixel is calculated then the elevated profiles may be extracted thus leaving out only the flat surface details, which would be much easier to analyse. A 3-D light scattering model has been developed to obtain the height information.

15 A multichannel 3-D Lambertian model is used in a simulation/image processing program. Parameters such as the gain and offset can be manipulated to analyse the model. Profiles providing various degrees of slopes and heights may be tested. Saturated regions may be introduced to simulate the specular effect and random noise components may be added to simulate the FSS noise. Four different profiles 20 namely the ramp, pyramid, cone and the circle are used in the analysis.

Light intensities as observed at three different angles viz., 40 , 45 and 50 from the normal vector are calculated using the Lambertian model. Pseudo-colour images comprising these three intensity components assigned respectively to Red, Green and 25 Blue channels are obtained. Intensities for any pair of these three channels are used to calculate the slope along the y-direction. Histograms of the angle image show the same values provided the gain(Gs), offsets(Cs) and observations angles (qs) used in the calculations are the simulated values. The angle images clearly show the developed 3-D model works satisfactorily for simulated profiles. Each pixel angle may be integrated 30 along the x-axis (scan direction) to obtain the height profile.

Accordingly, a scaleable, resilient and distributed architecture has been described for uniform acquisition of data from diverse equipment manufactures including process monitoring, closed loop process control, intelligent modular controller, enterprise-wide information system connectivity, product tracking and 5 process loss tracking.

Certain product defects manifest themselves at specific stages of production but their cause may be due to process malfunction of previous stages. For some such defects the exact quantitative relationship to process variable is known because of 10 laboratory tests. For most cases however, laboratory tests are several hours or days later when process conditions have already changed. In addition certain defects are caused by the interaction of different process, and the exact relationship is not known. One-to-one relationships are deduced by controlled experiments and the subsequent analysis of the observations, especially if the cause of the defect and its manifestation occur in the IS same process or in consecutive processes. Many-to-one relationships and relationships between non consecutive events can be deduced using complete product tracking through all stage process with complete time records relating each product (or small batches of products) to a specific production process.

20 The methods and results described have wider applications in other traditional industries beyond the industries of ceramic tiles and other ceramic products, which are characterized by batch processes and require complex quality control and multilevel monitoring. 25 It will be understood that the above description of the preferred embodiment of

the invention has been given by way of example only and that modifications of detail of the preferred embodiment and the various aspects of the invention can be made within the scope of the claims.

Claims

C L A I M S

1. A method of volume manufacture of a product in a staged production process, . compnsmg: 5 measuring parameters at multiple points along the process and identifying correlations between input and output parameters, including correlations between an output parameter and a plurality of input parameters or an input parameter and a plurality of output parameters; storing interdependencies between parameters according to the identified 10 correlations; and controlling the process at multiple stages according to a stored interdependency in response to a given parameter change.

2. A method according to claim I, wherein the step of and identifying correlations 15 is performed for different product data sets, and different interdependencies between parameters are stored for different product data sets, whereby an interdependency is recalled from storage when the type of product to be made is changed.

3. A method according to claim I or 2, wherein the process is a ceramic product 20 manufacturing process and the multiple stages include a drier, a press and a kiln. D

4. A method according to claim 3, wherein the output parameters include density of products emerging from the press.

25

5. A method according to claim 3 or 4, wherein the output parameters include defects in products emerging from the press. D

6. A method according to claim 3, 4 or 5, wherein the output parameters include parameters of glaze to be applied to the products.

7. A method according to any one of claims 3 to 6, wherein the output parameters include moisture content of particulate material entering the press.

8. A method according to any one of claims 3 to 7, wherein the output parameters include grain size distribution in particulate material entering the press.

5

9. A method according to any one of claims 3 to 8, wherein the input parameters include one or more of: press pressure, press speed, pump pressure, air temperature and air flow rate.

10. A facility for a staged production, volume process of manufacture of products 10 comprising means for feeding material into a stage of the process at a stage input and means for feeding out of that stage at a stage output, means for measuring a plurality of input parameters of the material entering the stage and/or conditions within the stage, means for measuring at least one output parameter of a product exiting the 15 stage; and means for automatically correlating the at least one output parameter with the plurality of input parameters to infer the influence the input parameters have on the output parameter and the relationship therebetween.

20

11. A facility according to claim 10, wherein the means for automatically correlating the at least one output parameter with the input parameters perform the correlation product-by-product using the input parameters and the output parameter specific to individual products.

25

12. A facility according to claim 11, wherein the means for automatically correlating the at least one output parameter of a given product use, as input parameters, parameters measured in relation to a preceding product passing through the stage.

13. A facility according to claim 11 or 12, wherein the means for automatically 30 correlating the at least one output parameter of a given product use, as input parameters, parameters measured from a subsequent product passing through the stage.

14. A facility according to any one of claims 11, 12 or 13, further comprising gap measuring means for measuring the gap between products passing through the stage, wherein the means for correlating uses, the gap as an input parameter.

5 1S. A facility according to any one of claims 10 to 14, comprising a ceramic product press moisture content measuring means for measuring moisture content of material to be fed into the press.

16. A facility according to claim 15, wherein the moisture content measuring 10 means comprises an infra-red sensor.

17. A facility according to any one of claims 10 to 16, further comprising means for monitoring grain size distribution of particulate material to be fed into the stage.

15

18. A facility for a staged production, volume manufacture of products comprising: a first process stage having an associated controller for controlling the first process stage, a second process stage having an associated controller for controlling the second process stage, 20 first feeder means for feeding first material into the first process stage, second feeder means for feeding second material from the first process stage into the second process stage, second measuring means for measuring second parameters of the second material and/or the second process stage, and 25 third measuring means for measuring third parameters of products emerging from the second process stage, and feedback means for feeding back to the controller of the first process stage signals responsive to the second and third parameters.

30

19. A facility according to claim 18, further comprising correlation means for correlating measured third parameters with measured first parameters and causing the

feedback means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation.

20. A facility according to claim 18 or 19, further comprising correlation means for S correlating measured third parameters with measured second parameters and causing the feedback means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation.

21. A facility according to claim 18, 19 or 20, further comprising correlation 10 means for correlating measured second parameters with measured first parameters and causing the feedback means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation.

22. A facility for a staged production, volume manufacture of products comprising: 15 a first process stage having an associated controller for controlling the first process stage, a second process stage having an associated controller for controlling the second process stage, first feeder means for feeding first material into the first process stage, 20 second feeder means for feeding second material from the first process stage into the second process stage, first measuring means for measuring first parameters of the first material and/or the first process stage, second measuring means for measuring second parameters of the second 25 material and/or the second process stage, and feedforward means for feeding forward to the controller of the second process stage signals responsive to the first and second parameters.

23. A facility according to claim 22, further comprising correlation means for 30 correlating measured third parameters with measured first parameters and causing the feedforward means to provide signals to the controller of at least one of the first and second stages responsive to results of correlation.

24. A facility according to claim 22 or 23, further comprising correlation means for correlating measured third parameters with measured second parameters and causing the feedforward means to provide signals to the controller of at least one of the first and 5 second stages responsive to results of correlation.

25. A facility according to claim 22, 23 or 24, further comprising correlation means for correlating measured second parameters with measured first parameters and causing the feedforvard means to provide signals to the controller of at least one of the 10 first and second stages responsive to results of correlation.

26. A facility according to any one of claims 18 to 25, wherein the correlating means performs a correlation product-by-product over a plurality of identified products passing through production, correlating at least one measured parameter at one stage in 15 the production with at least one measured parameter at another stage in the production of each identified product.

27. A facility for a staged production, volume process for the manufacture of ceramic tiles, comprising means for processing of raw materials to produce a raw 20 material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for monitoring the density of "green" tiles on-line.

28. A facility as claimed in claim 27, wherein said means for monitoring the 25 density of "green" tiles comprises a non-contact ultrasound sensor.

29. A facility as claimed in claim 27 or 28, including means for detecting defects in "green" tiles.

30 30. A facility for a staged production, volume process for the manufacture of ceramic tiles, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles,

means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and infrared imaging means for detecting defects in "green" tiles on-line.

31. A facility for a staged production, volume process for the manufacture of S ceramic tiles, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for monitoring on-line the colour in the raw material that is produced for tile production and/or in glaze preparation.

32. A facility as claimed in claim 31, wherein said colour monitoring means includes a spectrophotometer.

33. A facility for a staged production, volume process for the manufacture of 15 ceramic tiles, comprising means for processing of raw materials to produce a raw material for tile production, one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and means for monitoring parameters of glaze material on-line as it is being applied to the tiles.

34. A facility as claimed in claim 33, wherein said means for monitoring glaze quality comprises means for measuring glaze material viscosity.

35. A facility for a staged production, volume process of manufacture of ceramic 25 tiles comprising means for processing of raw materials to produce a raw material for tile production batching and milling means for producing a water-based slurry, and spray drying means for producing from said slurry a particulate raw material for tile production, comprising one or more presses for the production of "green" tiles, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, 30 and means for inspection of the finished tiles and sorting and batching of the same, further comprising means for monitoring moisture content of said particulate material.

36. A facility as claimed in claim 35, wherein said means for monitoring moisture content comprises an infra-red sensor.

37. A facility for a staged production, volume process of manufacture of ceramic 5 tiles, comprising means for processing of raw materials to produce a raw material for tile production batching and milling means for producing a water-based slurry, and spray drying means for producing from said slurry a particulate raw material for tile production, comprising one or more presses for the production of "green" tile, means for subsequent drying of the "green" tiles, their printing and/or glazing and their firing, and 10 means for inspection of the finished tiles and sorting and batching of the same, further comprising means for monitoring grain size distribution in said particulate material.

38. A facility as claimed in claim 37, wherein grain size distribution monitoring in said particulate material is effected by a process of computer-aided image analysis.

39. A facility as claimed in any one of claims 10 to 38, comprising means for tracking tiles through their manufacturing process.

40. A facility as claimed in claim 39, wherein said tile tracking means includes 20 means for marking the tiles with machine-readable identifiers.

41. A facility as claimed in claim 40, wherein said machine-readable identifiers comprise barcodes and/or datamatrix codes.

25

42. A method of monitoring of tiles in a tile manufacturing facility comprising measuring the temperature profile of tiles using an infra-red camera.

43. The method of claim 42, wherein tile moisture content is inferred from the temperature profile.

44. The method of claim 43, further comprising using the inferred tile moisture content as a parameter for correlation against a measured output parameter of the facility. 5

45. Apparatus for detection of inclusions in a workpiece comprising means for subjecting the workplace to spatiallyuniform heat flux input, an infra-red camera for acquiring an image of the workpiece surface and means for analysing said image to detect features indicative of inclusions.

10

46. Apparatus for density measurement comprising a non-contact ultrasound sensor.

47. A conveyor for conveying unit products along a belt, comprising drive means providing a drive speed output signal and a sensor for sensing product movement along 15 the belt and computation means for comparing predicted product movement along the belt, based on the drive speed output signal, with actual product movement and providing a signal if actual product movement departs from predicted product movement by more than a pre-set threshold.

20

48. A conveyor according to claim 47, wherein the sensor comprises a first sensor for sensing a product passing a first point along the belt and a second sensor for sensing the product passing a second point along the belt.

49. A conveyor according to claim 48, wherein the first sensor is placed relatively 25 close to the drive means and the second sensor is located relatively remote from the drive means.

50. A conveyor according to claim 47, 48 or 49, wherein the computation means includes modelling means for modelling a predicted progress of movement of a product 30 along the belt and compares a measured progress of movement with the predicted progress of movement.

51. A conveyor according to claim SO, wherein the modelling means takes into account stoppages in belt movement as reported by the drive speed output signal.

52. A conveyor according to any one of claims 57 to S 1, comprising two parallel 5 tracks of elastomeric material suitable for conveying tiles.

53. A method of conveying unit products, comprising: driving a belt and providing a speed output signal; sensing product movement along the belt; 10 computing predicted product movement along the belt based on the drive speed output signal; comparing actual product movement with predicted product movement; and providing an alert if actual product movement departs from predicted product movement by more than a pre-set threshold.

54. A facility according to any one of claims 10 to 41, further comprising a conveyor according to any one of claims 48 to 52.