CN112328703B - Numerical control equipment health state diagnosis device and method based on incremental learning - Google Patents
Numerical control equipment health state diagnosis device and method based on incremental learning Download PDFInfo
- Publication number
- CN112328703B CN112328703B CN202011195855.6A CN202011195855A CN112328703B CN 112328703 B CN112328703 B CN 112328703B CN 202011195855 A CN202011195855 A CN 202011195855A CN 112328703 B CN112328703 B CN 112328703B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- cloud platform
- numerical control
- edge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/283—Multi-dimensional databases or data warehouses, e.g. MOLAP or ROLAP
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to a numerical control equipment health state diagnosis device and method based on incremental learning, wherein the device comprises a cloud platform, edge equipment and an intelligent sensor which are sequentially connected, and the cloud platform comprises a data warehouse and a model library management module; the edge device comprises a concept drift detection module, an edge model module and a health state diagnosis module, wherein the edge model module is used for storing a base model, when the concept drift is detected, an updated base model is obtained from a cloud platform, and a historical model is uploaded to the cloud platform, otherwise, the local base model is updated by adopting an incremental learning method based on real-time acquired data of an intelligent sensor; and the health state diagnosis module is used for carrying out health state diagnosis on the numerical control equipment based on the output result of the current base model. Compared with the prior art, the method has the advantages of avoiding repeated learning of data, being high in training speed and the like.
Description
Technical Field
The invention relates to the technical field of fault diagnosis of numerical control equipment, in particular to a health state diagnosis device and method of numerical control equipment based on incremental learning.
Background
With the development of modern industrial and scientific technology, the structure of numerical control equipment becomes more and more complex. In the operation process of the numerical control equipment, the performance of the numerical control equipment is degraded along with the operation time and fault problems are generated at times due to the influence of various operation factors of the equipment or the outside, such as component abrasion, high temperature and high pressure, chemical corrosion, external impact and the like. Once equipment fails, the equipment affects industrial production slightly, and safety accidents happen seriously. Therefore, various fault diagnosis and prediction technologies are comprehensively utilized to monitor and early warn the state of the equipment, so that problems are found and treated in time, decision basis is provided for maintenance plans, and the reliability of the operation of the equipment is very necessary to be ensured.
Faults of numerical control equipment generally include faults of three subsystems, faults of an electrical system, faults of a numerical control system and faults of a mechanical system. The method for processing the faults and the exceptions also comprises the steps of passive reaction from the beginning, active prevention, and prediction and planning management in advance, namely, after-maintenance development, regular maintenance and on-condition maintenance. Meanwhile, with the development of intelligent manufacturing, industrial internet and internet of things, the acquisition and storage of industrial big data become more and more convenient, and under the background of industrial big data, how to utilize the data well and obtain valuable knowledge from the data has important application value.
The development of artificial intelligence provides an effective solution, machine learning has strong learning capability on data, the data-driven fault diagnosis and prediction technology does not need prior knowledge of a research object, and only needs to utilize a related method of data analysis to mine implicit information in the data and establish a machine learning model to realize fault diagnosis and prediction on equipment, for example, the mechanical equipment fault diagnosis method based on the machine learning classification algorithm disclosed in patent application CN 110108431A. However, most of the existing data-driven fault diagnosis and prediction models adopt an off-line working mode, which is generally established on the basis of comprehensive off-line analysis modeling of a large amount of historical data, and the models have poor dynamic updating capability and cannot utilize real-time data flow of industrial sites in real time.
For artificial intelligence fault diagnosis of numerical control equipment, patent application CN109933004A introduces a method and a system for predicting and diagnosing machine tool faults based on edge computing and cloud cooperation, which is mainly characterized in that a cloud platform returns results, and after receiving the results, a client needs an expert to feed back and label fault diagnosis and send the results to the cloud platform, and the cloud platform trains and updates a model again based on labeled data. Data communication is basically used as the main part between the edge equipment and the cloud platform, main computing functions are completed by the cloud platform, data need to be manually marked at the edge section, and certain use cost is increased. In addition, aiming at the problem that the data distribution possibly changes in the performance degradation of the equipment, the corresponding coping strategy is lacked when the model is in failure prediction, and a method of complete retraining is adopted in the updating of the model, so that the resource occupation is large, the training is slow, and the cost is high.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide the health state diagnosis device and method of the numerical control equipment based on incremental learning, which avoid repeated learning of data and have high training speed.
The purpose of the invention can be realized by the following technical scheme:
a numerical control equipment health state diagnosis device based on incremental learning comprises a cloud platform, edge equipment and an intelligent sensor which are connected in sequence, wherein the intelligent sensor is connected with the numerical control equipment,
the cloud platform includes:
a data warehouse for storing training data in a data block manner;
the model library management module is used for storing an integrated model, the integrated model consists of a plurality of base models, the base models are used for carrying out batch processing training by using data blocks in the data warehouse, and the integrated model is updated in an incremental learning mode based on the data blocks;
the edge device includes:
the concept drift detection module is used for judging whether to detect the concept drift or not based on the real-time collected data of the intelligent sensor;
the edge model module is used for storing a base model, acquiring an updated base model from the cloud platform when the result of the concept drift detection module is 'yes', uploading a historical model to the cloud platform, and updating the local base model by adopting an incremental learning method based on real-time acquired data of the intelligent sensor when the result of the concept drift detection module is 'no';
and the health state diagnosis module is used for diagnosing the health state of the numerical control equipment based on the output result of the current base model.
Further, the plurality of base models form the integrated model through a model difference maximization principle.
Further, the process of updating the integration model includes:
receiving a new data block, extracting the characteristics of the data block, and constructing a new base model ft;
and judging whether the number of the base models in the current integrated model reaches a set value, if so, temporarily adding ft into the integrated model, discarding one base model from the integrated model according to a model difference maximization principle, and if not, directly adding ft into the integrated model.
Further, when the cloud platform receives the historical model uploaded by the edge device, the model library management module performs the following operations:
and judging whether the number of the base models in the current integrated model reaches a set value, if so, temporarily adding the historical models into the integrated model, discarding one base model from the integrated model according to a model difference maximization principle, and if not, directly adding the historical models into the integrated model.
Further, the model difference is measured through the Q statistic, specifically:
wherein, Q (f) i ,f j ) Is a model f i And f j M is a set value.
Further, the obtaining the updated base model from the cloud platform specifically includes:
when the result of the concept drift detection module is yes, the edge device sends a cooperative work request to the cloud platform;
the cloud platform responds to the cooperative work request and acquires a feature set corresponding to the current real-time acquired data from the edge equipment;
and learning the feature set by using the integrated model, selecting a base model which has better performance on the current feature set as the updated base model, and sending the updated base model to the edge equipment.
Further, the edge device further includes:
and the data processing module is used for carrying out data cleaning on the data acquired by the intelligent sensor.
Further, the concept drift detection performed by the concept drift detection module specifically includes:
and (3) performing feature extraction on the real-time acquired data of the intelligent sensor, judging whether the data distribution change degree exceeds a set value or not based on the feature set, and if so, judging that the concept drift is detected.
Further, the health state diagnosis module also predicts the health state of the numerical control equipment based on the output result of the base model and generates corresponding early warning maintenance information.
The invention also provides a numerical control equipment health state diagnosis method based on incremental learning, which comprises the following steps:
1) the method comprises the steps that edge equipment acquires real-time acquisition data of numerical control equipment;
2) extracting features of the real-time acquired data, judging whether the data distribution change degree exceeds a set value or not based on the feature set, if so, judging that concept drift is detected, sending a cooperative work request to the cloud platform, executing step 3), otherwise, updating a local base model of the real-time acquired data by adopting an incremental learning method, and executing step 5);
3) the cloud platform responds to the cooperative work request, acquires a feature set corresponding to the current real-time acquired data from the edge equipment, learns the feature set by using the integrated model, selects a base model which is better represented on the current feature set, and sends the base model serving as an updated base model to the edge equipment;
4) the edge device replaces the local base model with the updated base model, uploads the historical model to the cloud platform, and executes the step 5);
5) and performing health state diagnosis on the numerical control equipment based on the output result of the current base model.
Compared with the prior art, the invention utilizes the machine learning method to carry out data mining on the operation monitoring data of the equipment, realizes intelligent fault diagnosis and health prediction, has important significance for the development of manufacturing intelligence, and has the following beneficial effects:
1. a side cloud cooperative numerical control machine tool key component fault prediction device is constructed, and the advantages of the side cloud cooperative numerical control machine tool key component fault prediction device and the side cloud cooperative numerical control machine tool key component fault prediction device are fully exerted: the edge equipment is close to the equipment, so that real-time data can be timely utilized, and the updating is realized in an incremental learning mode; the cloud platform is strong in computing power, can perform centralized management on the model, and avoids repeated learning on data through cooperative work with the edge.
2. The model base is organized by adopting an incremental learning mode based on data blocks, the learning of each data block can be regarded as a subtask, a corresponding base model is obtained by learning the data blocks, the base model forms an integrated model by a model selection strategy based on difference, the integrated model is stored in the model base, and the integrated model can better cope with the data distribution change condition when equipment is degraded.
3. Aiming at the equipment state monitoring data flow, because the computing power of the edge equipment is weak and the storage capacity is small, generally in the cooperation of the edge and the cloud platform, only the general functions of data acquisition, data preprocessing, feature extraction and the like are borne, and the training and updating of the model are borne by the cloud platform.
4. Aiming at the problem that the data distribution possibly changes in the equipment performance degradation, the concept drift detection method can be used for accurately capturing the data, not only adapting to the general degradation process, but also adapting to the concept drift problem of the mutant type.
5. When the edge cloud works cooperatively, the cloud platform selects a base model which is better represented on the current feature set to update the model in the edge device, and the fault diagnosis precision is effectively improved.
Drawings
FIG. 1 is an overall architecture diagram of the present invention;
FIG. 2 is a flow chart of the cloud platform building a model based on new data blocks;
FIG. 3 is a flowchart illustrating updating of a cloud platform based on a history model uploaded by an edge device;
FIG. 4 is a flow diagram of incremental learning based device health assessment;
FIG. 5 is a flow chart of the cooperative work of the edge and cloud platform.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
1. Noun explanation
(1) Edge model
An edge model is a model arranged on an edge device, which may be considered a base model, which is generally applicable to the same type of device, but not necessarily to exactly the same device; or to the same type of component, but not to the exact same component. The algorithms used by the various base models may be the same or different, depending on the particular device.
Firstly, constructing an edge model: when a certain numerical control device is accessed to the cloud platform through the edge for the first time, the cloud platform issues a corresponding base model and initialization parameters through learning of state monitoring data uploaded by the edge device, and construction of the edge model is completed.
Updating the edge model: the updating of the edge model is divided into two conditions, one is that the numerical control equipment normally operates, the state monitoring data slowly changes, and the edge model is updated in a local incremental learning mode; and the other is a concept drift condition of a sudden change type of state monitoring data, and the edge model is updated in a mode of cooperation of the edge and the cloud platform.
(2) Cloud platform model
The cloud platform model can be regarded as an integrated model, is constructed in an incremental learning mode based on data blocks, and is integrated by a plurality of base models according to a certain strategy.
Firstly, constructing a cloud platform model: the initial integrated model is generally obtained by a mode of batch training of historical data in blocks, the historical data is divided into data blocks according to a time sequence, learning of each data block is regarded as a subtask, a base model is obtained from a learning result of each subtask, and the base model forms the integrated model through a model difference maximization principle.
Updating the cloud platform model: the updating of the cloud platform model is divided into two cases, wherein one case is that a new base model is constructed on the cloud platform through learning a new data block, and then the integrated model is updated through a corresponding strategy; in another case, the edge device uploads the edge model to be added as a base model to the update of the integrated model through cooperation with the edge device.
2. System architecture
The invention provides a numerical control equipment health state diagnosis device based on incremental learning, which comprises a cloud platform, edge equipment and an intelligent sensor which are sequentially connected, wherein the intelligent sensor is connected with the numerical control equipment, uploads real-time acquired data to the edge equipment and then uploads the data to the cloud platform, the edge equipment realizes intelligent health state diagnosis based on the incremental learning and the cooperative work of the edge equipment and the cloud platform, and early warning information can be generated based on a diagnosis prediction result.
(1) Cloud platform
The cloud platform is a service platform running on the Internet and can be composed of 1 or more servers. The cloud platform mainly comprises a data warehouse and a model library management module. In a preferred embodiment, a feature building module and a communication module may also be included.
The functions of the modules are described as follows:
the characteristic construction module: and the system is responsible for carrying out operations such as screening, fusion, recombination and the like on the feature set uploaded by the edge equipment.
Communication module: and the cloud platform is responsible for communication between the edge device and the cloud platform, such as receiving data sent by the edge device, and sending the model and the prediction result thereof to the edge device.
Model library management module: the model library module is responsible for management and use of the cloud platform model and interaction and cooperation with the edge model. The model library management module stores an integrated model, the integrated model is composed of a plurality of base models, the base models perform batch processing training by using data blocks in the data warehouse, and the integrated model is updated in an incremental learning mode based on the data blocks.
Fourthly, a data warehouse: the data warehouse is responsible for the functions of historical data storage, addition, deletion, check, modification and the like, and stores training data in a data block mode.
(2) Edge device
The edge device mainly comprises a concept drift detection module, an edge model module and a health state diagnosis module, and in a preferred embodiment, the edge device also comprises a communication module, a data processing module and a feature construction module.
The functions of the modules are described as follows:
a communication module: the data communication between the edge and the intelligent sensor and the data communication between the edge and the cloud platform are carried out, for example, the data collected by the sensor are received and preprocessed, and then the preprocessed data are uploaded to the cloud platform; and receiving the sending parameters or the prediction results of the cloud platform and the like.
The data processing module: and performing data cleaning on the data acquired by the sensor. Incomplete data, erroneous data, repeated data, etc. are mainly processed.
Characteristics construction module: and (3) performing feature extraction on the cleaned data by adopting a certain method to form a feature data set, and performing simple extraction methods such as time domain, frequency domain analysis, time-frequency domain analysis and the like.
Fourthly, a concept drift detection module: and the system is responsible for detecting whether concept drift occurs or not and whether data distribution changes or not. In machine learning, the assumption of data independent and distributed is the premise of learning, but in large-scale data and data streams, the phenomenon that the data distribution changes often occurs, which is the concept drift problem.
An edge model module: and the system is responsible for the use and management of the edge model and the interaction and the cooperative work with the cloud platform model. The edge model module stores a base model, when the result of the concept drift detection module is yes, the updated base model is obtained from the cloud platform, the historical model is uploaded to the cloud platform, and when the result of the concept drift detection module is no, the local base model is updated by adopting an incremental learning method based on real-time data collected by the intelligent sensor.
Sixthly, a health state diagnosis module: and evaluating and predicting the health state of the equipment based on the learning result of the edge model module, and providing corresponding early warning and maintenance suggestions and the like.
The applicable object of the health state diagnosis device of the numerical control equipment can be any numerical control equipment, including various numerical control machines, such as lathes, planers, milling machines, punching machines, grinding machines, drilling machines, boring machines and the like. Each numerical control device comprises one or more key components, and one or more intelligent sensors are arranged on the key components, so that the operating states of the components are monitored, and the operating state data of the components, including but not limited to amplitude, speed, acceleration, rotating speed, temperature, humidity and the like, are collected.
3. Related procedures
The process of building and updating the integration model of the cloud platform based on the new data blocks is shown in fig. 2, and comprises the following steps:
s101: the data warehouse of the cloud platform updates the training data in the form of data blocks.
S102: and on the cloud platform, the data block is subjected to feature extraction and construction by using a feature construction module.
S103: in a model library management module of the cloud platform, a historical model (base model) maintained in the system is tested by using a current data block.
S104: and in a model library management module of the cloud platform, establishing a new model ft according to the current data block.
S105: assuming that the number of maintainable base models in the integrated model is limited to m, determining whether the number of base models in the integrated model reaches m, if yes, executing S106, and if no, executing S107.
S106: ft is temporarily added to the integration model, and S108 is performed.
S107: and directly adding ft into the integrated model to complete the updating of the cloud platform model.
S108: and selecting one model fi to move out of the integrated model by using a mode of maximizing the difference of the models, and finishing the updating of the cloud platform model. A common measure of model dissimilarity is the Q statistic.
Using the Q statistic as an example, some base model in the integrated model is shifted out to maximize the following:
wherein Q (f) i ,f j ) Is two models f i And f j Q statistic in between. The specific calculation of this value is as follows.
Wherein N is ab Representation model f i Classification as a and f j The data amount classified as b, with 1 indicating a correct classification and 0 indicating an incorrect classification.
For example, when m is 4, Q (f1, f2), Q (f1, f3), Q (f1, f4), Q (f2, f3), Q (f2, f4), Q (f3, f4), i.e., 3+2+1, are 6 types, and thus, Q is shared
When the cloud platform receives the historical model uploaded by the edge device, the model can be regarded as a base model of the system, and is added into the update of the integrated model, and the process is shown in fig. 3 and includes the following steps:
s201: and a model library management module of the cloud platform receives the historical model fx uploaded by the edge device.
S202: in the model library management module, the number of base models which can be maintained by the integrated model is m, whether the number of the current base models reaches m is judged, if yes, S203 is executed, and if not, S204 is executed.
S203: fx is temporarily added to the integration model, and S205 is performed.
S204: and f, directly adding fx into the integrated model to complete model updating of the cloud platform.
S205: and selecting a model fx to move out of the integrated model by using a mode of maximizing the difference of the models, and completing model updating of the cloud platform.
When the numerical control equipment health state diagnosis device based on incremental learning is used for real-time numerical control equipment health state assessment, the method mainly comprises the steps of collecting and preprocessing numerical control equipment state data, extracting characteristics of a state data set, constructing a model, detecting conceptual drift of new data, updating an edge model, predicting and assessing equipment health on the basis of an edge model learning result and the like. The detailed flow of the evaluation process is shown in fig. 4, and includes the following steps:
s301: the numerical control equipment is started to operate, the intelligent sensor monitors in real time, and the edge equipment acquires state monitoring data of the equipment.
S302: in the edge device, a data processing module is used for cleaning data, and the original data acquired by a sensor has the conditions of missing, abnormity and the like, and the data is processed by methods such as filling, deleting and the like.
S303: in the edge device, a feature signal is obtained from the data using a feature construction module.
Taking a vibration signal as an example, time domain analysis, frequency domain analysis and time-frequency analysis technologies are utilized to perform corresponding analysis on the signal, a high-dimensional feature set comprising time domain, frequency domain and time-frequency domain features is constructed, and then a feature reduction method is utilized to perform dimensionality reduction and reduction on the feature set, wherein the common feature reduction method comprises correlation analysis, Principal Component Analysis (PCA) and the like.
S304: in the edge device, a concept drift detection module is used for carrying out concept drift detection on a feature set formed by data at the time t, and whether new data distribution changes or not is judged.
A general Method such as Drift Detection Method (DDM) is based on conceptual Drift Detection of model angles. The DDM algorithm mainly uses the classification error p and the standard deviation s of the error of the model for judgment.
After learning t samples, if
p t +s t ≥p min +2*s min
Wherein p is t And s t Classification error and standard deviation, p, for the model after learning t samples min And s min The concept drift is predicted to occur for the minimum dynamic classification error and standard deviation of the model in the incremental learning; if it is
p t +s t ≥p min +3*s min
Then a concept drift is considered to be detected and the data distribution changes.
S305: and judging whether the degree of concept drift and the data distribution change or not according to the concept drift detection result, if so, executing S307, and if not, executing S306.
S306: if no concept drift is detected or the drift degree is within an acceptable range (i.e. in the above condition range, it can be considered that although there is concept drift, the data distribution does not change greatly), the edge device locally performs dynamic update of the edge model by using an incremental learning method.
Taking a Support Vector Regression (SVR) model in machine learning as an example, adding a new sample into a training set, adjusting a current support vector sample, and adjusting a weight coefficient of a non-support vector sample to a kernel function in an iterative computation or optimization manner to make the model still conform to the KKT condition, i.e., dynamically updating the SVR model without retraining.
S307: and if the concept drift or the drift degree changes greatly and the edge model considers that the data distribution changes, a request is made to the cloud platform, and the cooperative working stage of the edge and the cloud platform is started.
S308: the updating of the local edge model is done according to the methods of S306 and S307.
S309: and the equipment health evaluation and prediction module evaluates the current health state, updates the equipment performance degradation curve and predicts the change trend of the health state according to the calculation result of the local edge model. For example, a method of evaluating Remaining Useful Life (RUL).
RUL generally refers to the time difference from the current operating time of a device to the time when the device fails due to a failure. Due to uncertainty caused by the influence of the operation environment and the operation condition, the RUL is a condition random variable which depends on time. The definition formula is as follows:
RUL(t)＝{T f (t)-t|T f (t)＞t,M(t)}
wherein T represents the current operation time of the current equipment, and a random variable T f (t) represents the equipment failure time, and M (t) represents all state information of the machine equipment before the time t.
S310: and providing basis for maintaining the maintenance system, for example, providing early warning and other functions according to the current state and the future change trend of the equipment.
The edge model can process the gradual conceptual drift by means of local incremental learning. When the edge model encounters the concept drift of the mutation, the edge and the cloud platform need to perform certain cooperative work, the flow is shown in fig. 5, and the steps are described in detail as follows:
s30701: and the concept drift detection module in the edge equipment considers that the data distribution is changed, and the edge equipment provides a cooperative work request to the cloud platform through the communication module.
S30702: the cloud platform agrees to the collaborative work request and notifies the edge device through the communication module.
S30703: and the edge device receives the notification and uploads the feature set constructed by the feature construction module in the edge device to the cloud platform aiming at the current data.
S30704: in a model library management module in the cloud platform, an integrated model learns the current feature set, and the change condition of data distribution is analyzed by using the advantages of multi-model integration.
S30705: and a model library management module of the cloud platform selects a base model which is better represented on the current feature set according to the learning result, and sends the base model as a new model to the edge equipment.
S30706: after the edge device obtains the new model, the historical model is uploaded to the cloud platform for updating the cloud platform model.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A numerical control equipment health state diagnostic device based on incremental learning comprises a cloud platform, edge equipment and an intelligent sensor which are connected in sequence, wherein the intelligent sensor is connected with the numerical control equipment,
the cloud platform includes:
a data warehouse for storing training data in a data block manner;
the model library management module is used for storing an integrated model, the integrated model consists of a plurality of base models, the base models are used for carrying out batch processing training by using data blocks in the data warehouse, and the integrated model is updated in an incremental learning mode based on the data blocks;
the edge device includes:
the concept drift detection module is used for judging whether to detect the concept drift or not based on the real-time collected data of the intelligent sensor;
the edge model module is used for storing a base model, acquiring an updated base model from the cloud platform when the result of the concept drift detection module is yes, uploading the historical model to the cloud platform, and updating the local base model by adopting an incremental learning method based on real-time data acquired by the intelligent sensor when the result of the concept drift detection module is no;
and the health state diagnosis module is used for carrying out health state diagnosis on the numerical control equipment based on the output result of the current base model.
2. The incremental learning-based health status diagnostic apparatus for numerical control equipment according to claim 1, wherein a plurality of base models constitute the integrated model by a model difference maximization principle.
3. The incremental learning-based health status diagnostic apparatus for a numerical control device according to claim 1, wherein the process of updating the integrated model comprises:
receiving a new data block, extracting the characteristics of the data block, and constructing a new base model ft;
and judging whether the number of the base models in the current integrated model reaches a set value, if so, temporarily adding ft into the integrated model, discarding one base model from the integrated model according to a model difference maximization principle, and if not, directly adding ft into the integrated model.
4. The incremental learning-based health state diagnosis device for the numerical control equipment according to claim 1, wherein when the cloud platform receives the historical model uploaded by the edge device, the model library management module performs the following operations:
and judging whether the number of the base models in the current integrated model reaches a set value, if so, temporarily adding the historical models into the integrated model, discarding one base model from the integrated model according to a model difference maximization principle, and if not, directly adding the historical models into the integrated model.
5. The incremental learning-based health status diagnostic apparatus for numerical control equipment according to any one of claims 2 to 4, wherein the model variability is measured by Q statistic, specifically:
wherein, Q (f) i ,f j ) Is a model f i And f j M is a set value.
6. The incremental learning-based health state diagnostic apparatus for numerical control equipment according to claim 1, wherein the obtaining of the updated base model from the cloud platform specifically comprises:
when the result of the concept drift detection module is yes, the edge device sends a cooperative work request to the cloud platform;
the cloud platform responds to the cooperative work request and acquires a feature set corresponding to the current real-time acquired data from the edge equipment;
and learning the feature set by using the integrated model, selecting a base model which has better performance on the current feature set, taking the base model as the updated base model, and sending the updated base model to the edge equipment.
7. The incremental learning-based numerical control equipment health state diagnosis device according to claim 1, wherein the edge device further comprises:
and the data processing module is used for carrying out data cleaning on the data acquired by the intelligent sensor.
8. The incremental learning-based health state diagnostic device for the numerical control equipment according to claim 1, wherein the concept drift detection module is specifically configured to perform the concept drift detection by:
and (3) performing feature extraction on the real-time acquired data of the intelligent sensor, judging whether the data distribution change degree exceeds a set value or not based on the feature set, and if so, judging that the concept drift is detected.
9. The incremental learning-based health state diagnosis device for the numerical control equipment as claimed in claim 1, wherein the health state diagnosis module is further configured to predict the health state of the numerical control equipment based on the output result of the base model and generate corresponding early warning maintenance information.
10. A numerical control equipment health state diagnosis method based on incremental learning is characterized by comprising the following steps:
1) the method comprises the steps that edge equipment acquires real-time acquisition data of numerical control equipment;
2) extracting features of the real-time acquired data, judging whether the data distribution change degree exceeds a set value or not based on the feature set, if so, judging that concept drift is detected, sending a cooperative work request to the cloud platform, executing step 3), otherwise, updating a local base model of the real-time acquired data by adopting an incremental learning method, and executing step 5);
3) the cloud platform responds to the cooperative work request, acquires a feature set corresponding to the current real-time acquired data from the edge equipment, learns the feature set by using an integrated model, selects a base model which is better represented on the current feature set, and sends the base model serving as an updated base model to the edge equipment;
4) the edge device replaces the local base model with the updated base model, uploads the historical model to the cloud platform, and executes the step 5);
5) and performing health state diagnosis on the numerical control equipment based on the output result of the current base model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011195855.6A CN112328703B (en) | 2020-10-31 | 2020-10-31 | Numerical control equipment health state diagnosis device and method based on incremental learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011195855.6A CN112328703B (en) | 2020-10-31 | 2020-10-31 | Numerical control equipment health state diagnosis device and method based on incremental learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112328703A CN112328703A (en) | 2021-02-05 |
| CN112328703B true CN112328703B (en) | 2022-09-16 |
Family
ID=74323210
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011195855.6A Active CN112328703B (en) | 2020-10-31 | 2020-10-31 | Numerical control equipment health state diagnosis device and method based on incremental learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112328703B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113159566B (en) * | 2021-04-19 | 2022-10-25 | 华南理工大学 | Adaptive collection method, system, apparatus and medium for equipment health stage detection |
| CN113473515A (en) * | 2021-05-31 | 2021-10-01 | 杭州电子科技大学 | Industrial wireless network reliability assessment method |
| CN113778028B (en) * | 2021-08-12 | 2023-09-26 | 西安交通大学 | Bian Yun cooperation and double increment based intelligent coagulation algorithm |
| CN116300690B (en) * | 2023-05-17 | 2023-07-25 | 济宁联威车轮制造有限公司 | Radial drilling machine fault monitoring and early warning system based on edge calculation |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110792A (en) * | 2019-05-09 | 2019-08-09 | 南京大学 | A kind of multi-tag method for classifying data stream based on incremental learning |
| CN110719210B (en) * | 2019-12-05 | 2020-08-28 | 赣江新区智慧物联研究院有限公司 | Industrial equipment predictive maintenance method based on cloud edge cooperation |
| CN111507489B (en) * | 2020-04-20 | 2023-04-18 | 电子科技大学中山学院 | Cloud-edge-coordinated amusement equipment fault prediction and health management system and method |
-
2020
- 2020-10-31 CN CN202011195855.6A patent/CN112328703B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112328703A (en) | 2021-02-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112328703B (en) | Numerical control equipment health state diagnosis device and method based on incremental learning | |
| Cao et al. | An ontology-based approach for failure classification in predictive maintenance using fuzzy C-means and SWRL rules | |
| CN110806743A (en) | Equipment fault detection and early warning system and method based on artificial intelligence | |
| Pinto et al. | Robot fault detection and remaining life estimation for predictive maintenance | |
| NO343417B1 (en) | System and method for condition assessment of downhole tools | |
| CN101859128A (en) | Knowledge-based fault prediction expert system for complex milling machine tool | |
| KR102373655B1 (en) | Apparatus and method for diagnosing trouble of machine tool | |
| Wöstmann et al. | A retrofit approach for predictive maintenance | |
| Gopalakrishnan et al. | IIoT Framework Based ML Model to Improve Automobile Industry Product. | |
| CN113395182B (en) | Intelligent network equipment management system and method with fault prediction | |
| KR102474332B1 (en) | Prediction method of machine health stability in smart factory system, and recording medium thereof | |
| CN117592975A (en) | Operation and maintenance decision processing method and system for electromechanical equipment of expressway based on cloud computing | |
| De Vita et al. | A Semi-Supervised Bayesian Anomaly Detection Technique for Diagnosing Faults in Industrial IoT Systems | |
| Xin et al. | Dynamic probabilistic model checking for sensor validation in Industry 4.0 applications | |
| KR102340249B1 (en) | Apparatus and method for diagnosing and predicting faults of plc using data manning | |
| CN112966785B (en) | Intelligent constellation state identification method and system | |
| CN115469643A (en) | Nuclear power station rotating machinery health management method, system and medium | |
| KR20220097252A (en) | Method and system for managing equipment of smart plant using machine-learning | |
| El Kihel et al. | Maintenance 4.0 Model Development for Production Lines in Industry 4.0 Using a Deep Learning Approach and IoT Data in Real-Time: an Experimental Case Study | |
| Karagiorgou et al. | On making factories smarter through actionable predictions based on time-series data | |
| Olalere et al. | Optimising production through intelligent manufacturing | |
| Martins et al. | Predictive Maintenance of Mining Machines Applying Advanced Data Analysis | |
| CN117527859B (en) | Equipment monitoring method and system based on industrial Internet | |
| CN110263949B (en) | Data processing method and system fusing machine mechanism and artificial intelligence algorithm system | |
| CN117272844B (en) | Method and system for predicting service life of distribution board |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |