WO1999014641A1

WO1999014641A1 - Method for a bending procedure

Info

Publication number: WO1999014641A1
Application number: PCT/SE1998/001561
Authority: WO
Inventors: Ulf RÖNNMARK; Kenneth Berglund
Original assignee: Pullmax Ursviken Ab
Priority date: 1997-09-02
Filing date: 1998-09-02
Publication date: 1999-03-25
Also published as: AU9013498A; SE9703157L; SE9703157D0

Abstract

Method of bending on press brakes or similar whereby the desired result, e.g. the desired final bending angle, is input into a computer with a connected neural network that has 'learned' the relations between the parameters of bending procedures, after which the computer, with the help of the neural network, controls the movement of the pressing tool to obtain the desired result.

Description

METHOD FOR A BENDING PROCEDURE

Technical area

The invention concerns a computer-controlled method for obtaining the desired result, for example, a product with one or more bent angles, during a procedure for bending on press brakes or similar.

The prior art

Bending is a method of forming shapes where the moment of bending produces the main deformation of a material. There are two main categories of bending; press brake forming and edge folding. During press brake forming, a force is used that, via the downward movement of the pressing beam and the shape of the tool, creates a bending moment in the sheet metal plate where the loading finally exceeds the yielding point, whereby the sheet begins to deform. During edge folding, the sheet is clamped between two tong-like tool parts. That part of the sheet that is to be bent extends out from the tong-like tool parts and rests against a foldable tool part, e.g. one able to rotate. When this folds, rotating around its axis, the sheet bends to a certain angle.

The invention primarily concerns shaping sheet metal by press brakes, but it can even be used for other materials and other procedures.

As the material characteristics and the thickness of the material vary from sheet to sheet, it is usual that the press brake machinery has to be reset during the process of working. This is time-consuming and difficult, which is why an objective of the present invention is to solve this problem.

An exact, final angle of bending is difficult to calculate since there occurs a certain spring- back after bending, i.e. an elastic material has the ability to return to its original shape when it has been deformed. This elastic deformation or spring-back can be predicted and taken into account when calculating how great a distance a pressing beam or a pressing table needs to move in order that the desired final angle be obtained. A common method of achieving a certain angle of bending takes place by bending test samples, which is expensive and time-consuming. Even mathematical formulae and tables are used to calculate an angle of bending. There are at least two reasons why these methods are not reliable. Firstly, the parameters specified for the material, e.g. the yield limit and thickness of the material, are seldom the actual values. Secondly, the same material can have different spring-back in different directions of bending depending on the tension remaining and the texture caused by the rolling process.

The only certain way of determining the angle obtained is by measuring every finished product. Mechanical measurements are expensive and difficult to perform. There is thus a need for a simpler, cheaper and time-saving method of determining the exact final angle during bending procedures with press brakes.

Brief description of the invention

The invention thus describes a method of obtaining a desired result, e.g. a product with one or more exactly bent angles, during a procedure for bending on press brakes or similar. According to the invention, use is thus made of a computer with a neural network (a software) that has "learned" the bending procedure where different parameters have been varied, after which the neural network can then process the information from the current bending procedure and thereafter calculate output data, e.g. for controlling the movement of the frame of the machinery.

Other characteristics of the invention are given in the claims that follow.

Description of one embodiment of the invention

One embodiment of the invention is described below with the help of the drawing showing an example of the embodiment. Figure 1 shows a press brake. A neural network is a computer program that resembles the human brain's way of working. The program is fed with data for a number of parameters and output data measured from this. By successively "learning" the program by inputting further known input data and output data, the program is tuned to "recognise" relationships between different parameters. Neural networks are themselves known, which is why no further description is given here.

In cases where, for example, a press brake is used to shape sheet metal, one can, for example, measure the force and displacement of the pressing beam or pressing table, the deformation in me frame, the final angle of the bent sheet, and other parameters, and input these into the program. The neural network can then "learn" to "recognise" deflections of the frame or the final angle of the sheet corresponding with a certain used force and displacement. When using the neural network with press brakes, for example, one can input the desired angle of bending, after which the computer can determine how far the pressing beam/pressing table is to be pressed for this angle to be achieved.

In figure 1 , 1 designates die material that is to be bent. 2 designates the die or pressing table and 3 designates a stamp, knife or pressing beam. Parameters that can be used are, for example, the thickness of material t, the angle of bending α, the pressing depth a, the die width d, the radius of the die u, the radius of the knife r , etc.

In order to "learn", the neural network needs many sample test bendings where the different parameters and the obtained results are input. According to experiments performed, one can in this way obtain a final angle with a 1 % deviation after approximately 200 test measurements.

From experiments, one has concluded that neural networks can be designed so that using only power and distance of travel as variable input data, d ey are able to predict the pressing depth with good accuracy when the material quality, sheet thickness and me desired final angle vary. Nevertheless, for the method to be useful in practice, the neural network must be able to handle variations in, for example, the width of the die. The invention refers especially to neural networks connected in parallel with the control system of the bending arrangement. The neural network can provide the control system with necessary information so that the exact final angle can be obtained. This parallel system must be so general that a new learning process for each new tool is not necessary. A number of tool parameters plus, among other things, the nominal value of the thickness of the sheet metal is the only information that needs to be added to the neural network. The neural network can naturally also be integrated into me control system.

In experiments, a model has been designed that did not take account of me actual quality of the material or the actual thickness of the material, but that instead only varied the geometric parameters within predetermined limits. Such parameters include die width, knife radius, nominal thickness of material and me set point of the angle. The distance die knife/pressing beam/pressing table needs to travel to obtain a certain bending angle depends on these geometric parameters, but also on the plastic and elastic deformation characteristics of the material. The neural network learns by testing the relation between all of these parameters. One can then expand the experiment by varying the friction and die radius.

The example given above was based on predicting the pressing depth so that the desired final angle of the material that is to be bent is obtained. The method according to me invention can also be used to predict die original propagation length (length of propagation) to obtain a final end point irrespective of the number of bends. This can be achieved wid a special neural network, e.g. a neural network that complements the neural network specified above and d at has certain parameters in common.

The invention has been described with the help of an example of an embodiment showing the bending of material on a press brake. It is obvious for a person skilled in the art that the invention is not limited to mis example of an embodiment but that other bending procedures, materials and parameters can be used within me scope of that specified in d e following claims.

Claims

1. Method of obtaining a desired result, for example, a product with one or more bent angles, during a procedure for bending on press brakes and similar characterised i n t h a t information about the desired result plus parameters originating from the work piece and the said press brake are input into a computer with a neural network that has learned de relations between the parameters of the said procedure and diat can process the information from several said procedures, after which the computer can, with the help of the neural network, calculate output data for controlling the movement of the pressing tool, for example.

2. Method according to claim lcharacterised in that at least one parameter from the said press brake and at least one parameter from me work piece are used during the learning process of the neural network.

3. Method according to claim 2characterised in that me parameters for the said press brake diat are used during the learning process of the neural network are selected from among the following: the pressing depth of the pressing tool, the deformation of the frame of the machinery, me pressing force, the die widtii, the die radius, the die angle, de knife radius, me roll diameter.

4. Method according to claim 2characterised in that me parameters for the work pieces that are used during me learning process of the neural network are selected from among the following: nominal thickness of material, actual thickness of material, the coefficient of friction, the quality of me material, e.g. the plastic and elastic deformation, the final angle of the bent material, the propagation length of the material and the widdi of the material in die direction of bending.

5. Method according to any of the previous claims characterised in that the parameters from the said neural network are used together with additional parameters to teach a further neural network to calculate the length of propagation for a material that is to be bent in a desired way.