Method and control unit for resistance welding

Abstract

The invention relates to a method for resistance welding, wherein welding data describing the welding process are respectively determined in the course of the welding process, wherein the welding data are evaluated with the aid of a regression model, wherein in the course of the evaluation the probability of weld spatter occurring in a subsequent welding process is determined and furthermore the spattering time at which weld spatter occurs in the subsequent welding process is determined, wherein when the determined probability reaches a threshold value, an adaptation of at least one of the predefined welding parameters is performed, wherein in the course of the adaptation at least one relationship between the probability and / or the spattering time as dependent variables on the one hand and at least one of the welding parameters as an independent variable on the other hand is determined with the aid of the regression model, wherein at least one of the predefined welding parameters is selected and adapted with the aid of the at least one determined relationship.

Description

Technical Field

[0001] The invention relates to a method for resistance welding as well as a control unit, a welding device and a computer program for carrying out the method. Background Art

[0002] By means of welding processes, such as resistance welding, workpieces can be connected to one another in a material-locking manner. For example, in the process of automated body-in-white production, different workpieces, such as sheet metal, are welded to one another by means of resistance welding using robot-guided welding tongs.

[0003] During resistance welding, the two welding electrodes of the welding tongs are first pressed against the welding point of the workpieces by means of an electrode drive during a so-called force build-up phase until a predetermined electrode force is reached. The actual welding process then takes place, during which the welding electrodes are energized with welding current for the duration of the welding time. This results in resistance heating of the two workpieces to be welded between the welding electrodes until the desired welding temperature is reached.

[0004] In EP 3 412 397 A1, the probability of the occurrence of weld spatter is determined with the aid of a classification model. Summary of the Invention

[0005] Against this background, a method for resistance welding, as well as a control unit, a welding device, and a computer program for carrying out the method are proposed, having the features of the independent claims. Advantageous embodiments are the subject matter of the dependent claims and the following description.

[0006] The present invention uses the following measures to predict the probability and time point of the occurrence of welding spatter for the next welding process with the help of regression analysis of welding data of the welding process that has been performed, and determine at least one relationship between the probability or time point and the welding parameters, and adapt the welding parameters based on the relationship so as to prevent the predicted welding spatter.

[0007] Within the scope of the present invention, a welding process is performed in which a welding electrode is pressed against a welding point of a workpiece according to predetermined welding parameters and energized with welding current. Thus, a welding point is welded in each of the welding processes.

[0008] Welding parameters represent, in particular, predetermined setpoint values ​​or their temporal profile, according to which the corresponding welding process is to be performed. Welding parameters may, in particular, be electrical and / or mechanical parameters related to the movement or current flow of the welding electrode. For example, the welding parameters for the welding process to be performed can be predetermined depending on the workpiece to be welded. Welding parameters can, in particular, be stored in the form of so-called welding programs, which are expediently executed by a corresponding welding control unit. Typical welding parameters are welding current and electrode force, each as a temporal profile.

[0009] Furthermore, welding data describing the welding process are determined during the course of each of these welding processes. These welding data can, for example, be detected during the course of the respective welding process using measurement technology and / or derived from detected measured values. For example, these can be actual values ​​that are compared with predefined setpoint values ​​during the control of the welding process. For example, electrical actual or measured values ​​can be determined as welding data, such as measured values ​​for the welding current and welding voltage, or mechanical or electromechanical measured values ​​of the welding electrode or electrode drive, such as the torque of the electrode drive and the electrode force or the position or path of the welding electrode.

[0010] Within the scope of the present method, the welding data is evaluated using a regression model. During this evaluation, the probability of weld spatter occurring in the next welding process is determined, as well as the time at which weld spatter occurs or is likely to occur in the next welding process. For example, an implicit probability can be determined, for example, in the form of a binary output, particularly such that the occurrence of weld spatter in the next welding process is determined to be likely or unlikely. Furthermore, an explicit probability, according to which weld spatter will occur, can be advantageously determined, particularly as a real number between 0 and 1 or as a percentage value between 0% and 100%.

[0011] Regression models or regression analysis are statistical analysis methods used to model relationships between dependent and independent variables and, for example, to quantitatively describe these relationships or predict the values ​​of the dependent variables. Therefore, within the scope of this method, the collected welding data is analyzed using regression analysis to predict future behavior during the next welding process to be performed. In this context, weld spatter is understood to be droplets of molten metal that break apart due to the extreme intensity of heat and forces applied at or near the weld point.

[0012] The regression model can be trained, in particular, using training data, for example, archived data collected during the course of previously performed welding processes. In particular, the archived data or the correspondingly performed welding processes may have features or patterns that indicate weld spatter. For example, the regression model can include internal and external independent variables. Such internal independent variables can be, for example, one or more weights determined during a training phase based on the training data. Such external independent variables can be, for example, welding parameters x determined from the archived data or from performed welds. After this training phase, the weights are, in particular, permanently predefined. When the regression model is applied to current welding data, new welding parameters x can be supplied to the model. The model outputs can be y(x) and / or t(x), where y is the spatter probability and t is the spatter time.

[0013] The regression model can be trained, in particular, to learn the corresponding relationship between measured values, characteristics, and signals from previous welds and the occurrence of weld spatter at subsequent weld points. The regression model can advantageously memorize these characteristics from previous weld points or weld data and fully utilize them for the next weld point or subsequent weld data. If the regression model identifies, based on the current weld data, that weld spatter is likely in the next weld process, the model outputs, for example, a corresponding probability and, in particular, an estimated time of occurrence.

[0014] When the determined probability reaches a threshold value (particularly user-definable), at least one of the predefined welding parameters is adapted using the regression model. During this adaptation, the regression model is used to determine the relationship between the probability and / or spatter time as dependent variables, on the one hand, and the welding parameter as an independent variable, on the other. Based on this, it can then be determined which welding parameter should be changed and in what manner to avoid weld spatter.

[0015] In particular, an optimization method can be implemented for this purpose. Here, the probability and / or the spatter time can be used as variables to be optimized, i.e., the output of the model (e.g., y(x), t(x)) can be converted into a target function for the optimization method. Advantageously, one or more corresponding welding parameters are adapted so that the probability is as low as possible and, for example, falls below a threshold value, and / or so that the spatter time is shifted as far back as possible and, in particular, after the end of the next welding process.

[0016] Preferably, a measure of the influence of at least one of the predetermined welding parameters on the probability and / or the time of spatter can also be determined based on the at least one determined relationship. This can be used to determine which welding parameter should be changed and in what manner in order to most reliably prevent the occurrence of weld spatter. For example, the relationship can be analyzed functionally to determine a gradient, for example. This gradient then provides a measure of the influence of the welding parameter on the occurrence of weld spatter or the time of spatter.

[0017] Therefore, one or more welding parameters for the next welding process to be performed are adapted based on the evaluation of the welding data in such a way that the predicted weld spatter is minimized in this next welding process.

[0018] Weld spatter can negatively impact weld quality and lead to contamination of metal surfaces, particularly workpiece surfaces and surfaces of the welding equipment itself, such as the electrode cap. Therefore, it is important to suppress weld spatter and its negative effects. This method can reduce the occurrence of weld spatter by adaptively and automatically adjusting welding parameters. Manual adjustments can be conveniently eliminated. Furthermore, so-called Q-stops—line stops due to quality issues—caused by excessive spatter can be avoided.

[0019] Conventionally, welding programs that frequently produce weld spatter are often manually re-parameterized, often with a lower current at specific points during the welding process. Because spatter is material expelled from the weld nugget, welding equipment with UIR control ensures that the welding time is extended to reform the weld nugget and ensure point quality. However, the welding duration of the spatter point is prolonged, which slows down the production line. In contrast, according to the present invention, through statistical analysis of welding data, welding parameters can be automatically adapted to reduce or minimize weld spatter in future welding processes. This improves the quality and efficiency of the welding process.

[0020] It can be provided that one or more welding parameters are modified to varying degrees depending on the predicted spatter time, so that no weld spatter occurs at that time. Depending on the specific spatter time, the modification can be more or less severe. If weld spatter is predicted at a relatively early spatter time, it is expedient to modify the welding parameters relatively strongly. Conversely, if weld spatter is predicted at a relatively late time, a less drastic adaptation of the welding parameters is sufficient to prevent spatter.

[0021] According to a particularly preferred embodiment, at least one of the predefined welding parameters is adapted with the aid of a gradient method. The gradient method, also known in English as the "gradient descent method", is a numerical method for solving optimization problems. In this case, one generally proceeds from a starting point in a descending direction until no numerical improvement can be achieved. Therefore, in particular, a gradient is determined as a measure of the influence of the welding parameters on the occurrence of weld spatter or the time of spatter. With the gradient method, in particular, all inputs of the regression model that contribute to this gradient are optimized. For example, if the current welding parameter x is fed to a regression model and the model outputs a probability y(x) as output, this output y(x) can be converted into a target function for the gradient method. For example, this target function can be selected as Q(x) = y 2 (x). The signed gradient can be determined as the derivative of dQ with respect to dx: Q'(x) = 2y(x)*y'(x). Thus, for example, the gradient method can be applied to find the set of welding parameters x* corresponding to the minimum value on the hypersurface Q.

[0022] The set of optimal welding parameters x* can be predetermined for the next welding process and returned to the control device as a set of welding parameters for the upcoming subsequent welding.

[0023] According to a particularly advantageous embodiment, the regression model is implemented as an artificial neural network, in particular a recursive artificial neural network. Such neural networks or artificial neural networks (referred to as "artificial neural networks," ANNs) typically consist of nodes (so-called (artificial) neurons) arranged in multiple layers and directed edges connecting the nodes / neurons. The output of a node is, in particular, a function of the weighted inputs to the node (inputs and outputs are specified by the direction of the directed edges) and an activation function. More specifically, the output is a function of the weighted sum of the inputs, with the weights being assigned to the edges. In neural networks based on the so-called "feedforward" principle, neuron outputs are typically directed in only one processing direction and cannot be fed back via edges. Therefore, such "feedforward" neural networks typically have only one output layer.

[0024] In particular, recursive or feedback neural networks can also be used within the scope of the present method, in which the neuron outputs can also be directed in different processing directions. Thus, the outputs of neurons can be fed back via specific recursive edges. Thus, in a recursive neural network, neurons of one layer can also be connected to neurons of the same layer or a previous layer.

[0025] Machine learning can particularly advantageously be achieved using neural networks. Neural networks can be trained using training data to automatically and autonomously analyze and, if necessary, adapt welding parameters. As is known, neural networks can be trained by adapting their parameters or weights so that when the neural network is applied to the training data, the error with respect to a given error function is minimized. Welding data is preferably used as input values, which can be fed into the input neurons of the neural network. In particular, the neural network can be trained so that it provides probabilities and spatter times as outputs at the output neurons.

[0026] The welding data preferably includes a plurality of different physical variables. By using a regression model, the user does not have to select which data should be included in the analysis. Instead, all available data can be fed in, and the model itself "recognizes" which parameters have an influence on the result.

[0027] The method is therefore particularly advantageous in that it automatically identifies which physical variables lead to the most effective improvement and within which range these physical variables should be changed. Within the scope of the method, an evaluation and decision is expediently performed automatically and autonomously, in particular by means of a neural network or a regression model formed by a neural network, as to which of the welding data allows for the most accurate possible prediction of the next welding process and, furthermore, which of the welding parameters should be changed in what manner in order to prevent the predicted weld spatter as effectively as possible.

[0028] Particularly preferably, the welding parameters and / or welding data each include the electrode force and / or the position of the welding electrode and / or the welding current and / or the welding voltage and / or the welding time and / or the time profile of the electrode force and / or the time profile of the position of the welding electrode and / or the time profile of the welding current intensity and / or the time profile of the welding voltage. Therefore, the welding parameters to be adapted or the welding data to be evaluated each advantageously involve mechanical and electrical variables or their time profiles.

[0029] This method is particularly advantageously suitable for body-in-white production, and in particular for automated welding processes in body-in-white production, preferably during motor vehicle production. In particular, sheet metal parts are welded together to produce the body of a motor vehicle. During the production process of a single body, up to several thousand weld points (for example, approximately 5,000 weld points for a mid-size vehicle) can be processed automatically. This method allows each weld point to be welded with the highest possible quality.

[0030] A control unit (processing unit) according to the present invention, for example a welding control unit of a welding system, is configured, in particular in terms of programming, to carry out the method according to the present invention. The control unit or welding control unit can be designed, for example, as a programmable logic controller (SPS), numerical control (NC), or computerized numerical control (CNC).

[0031] The welding device according to the present invention for resistance welding comprises, in particular, welding tongs with welding electrodes and an electrode drive for moving the welding electrodes. A robot can also be provided to operate the welding tongs, for example in the case of pneumatic welding tongs. Furthermore, the welding device includes a preferred embodiment of the control unit according to the present invention.

[0032] It is also advantageous to implement the method according to the invention in the form of a computer program or computer program product having program code for executing all method steps, since this results in particularly low costs, especially when the implemented controller is also used for other tasks and is therefore already available. Suitable data carriers for providing the computer program are, in particular, magnetic, optical, and electronic storage media, such as hard disks, flash memories, EEPROMs, DVDs, etc. The program can also be downloaded via a computer network (Internet, intranet, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Further advantages and configurations of the invention are apparent from the description and the accompanying drawings.

[0034] It goes without saying that the features mentioned above and those yet to be explained below can be used not only in the respectively specified combination but also in other combinations or alone without departing from the scope of the present invention.

[0035] The invention is schematically illustrated in the drawings on the basis of exemplary embodiments and is described in more detail below with reference to the drawings.

[0036] Figure 1 A preferred embodiment of the welding device according to the invention is schematically shown, which is provided for carrying out a preferred embodiment of the method according to the invention.

[0037] Figure 2 A preferred embodiment of the method according to the invention is schematically shown as a block diagram. DETAILED DESCRIPTION

[0038] exist Figure 1 In FIG. 1 , a welding apparatus for resistance welding is schematically shown and referenced as 100 .

[0039] Welding device 100 can be used to connect workpieces 120 to one another in a materially bonded manner by resistance welding. In particular, workpieces 120 are welded to one another during body-in-white production, particularly the manufacture of motor vehicle bodies. For example, two aluminum sheets 121 and 122 are welded to one another as workpieces.

[0040] The welding device 100 has a welding tongs 110 with two welding electrodes 111 and 112. An electrode drive 130 is provided to move the welding electrodes 111, 112. Figure 1 In FIG, the welding tongs 110 are shown as servo welding tongs, for example, with an electrode drive 130 designed as a servo motor. It is also conceivable that the electrode drive 130 can be designed as an electric motor, a hydraulic motor or a pneumatic motor, for example.

[0041] During the resistance welding process, during the so-called force buildup phase, the welding electrodes 111 and 112 are pressed against the sheets 121 and 122 at the welding point 125 with the aid of the electrode drive 130 using an electrode force. Subsequently, during the actual welding process, the welding electrodes 111 and 112 are energized with a welding current for the duration of the welding time, thereby causing resistance heating of the sheets 121 and 122 at the welding point 125 and liquefaction of the surfaces of the workpieces 121, 122.

[0042] The welding device 100 also includes a control unit (welding control) 140, which can be configured, for example, as a programmable logic controller (SPS). The control unit 140 is configured to control the electrode drive 130 and the welding tongs 110, as indicated by reference numerals 151 and 152, and thereby regulate the welding process. To this end, a corresponding control program or welding program 141 is executed in the control unit 140, which controls the electrode drive 130 and the welding tongs 110 according to predefined welding parameters. These welding parameters can include, for example, the temporal profile of the electrode force and the welding current.

[0043] Furthermore, the control unit 140 is configured to evaluate the collected measurement data or welding data of the previous welding process to determine whether welding spatter is likely to occur in the next welding process and to adapt the welding parameters if necessary in order to prevent such welding spatter. For this purpose, the control unit 140 is configured in particular in terms of programming to execute a preferred embodiment of the method according to the invention, which is described in Figure 2 It is shown schematically as a block diagram in FIG and is explained below.

[0044] For analyzing the welding data and adapting the welding parameters, an artificial neural network, in particular a recurrent neural network, is provided, which expediently implements a regression method or forms a regression model. In step 201, the neural network is first trained, in particular using appropriate training data, for example, archived data from previously performed welding processes.

[0045] After the neural network has been trained, a welding process is performed in normal operation according to predefined welding parameters in step 202. Furthermore, welding data related to the welding process is collected. Welding parameters and welding data may include, for example, setpoint values ​​or actual values ​​of physical variables such as electrode force, welding electrode position, welding current, welding voltage, welding time, and the temporal profiles of electrode force, welding electrode position, welding current intensity, and welding voltage.

[0046] In step 203 , the neural network performs an evaluation of the collected welding data using a regression model, in particular in order to be able to assess the occurrence of weld spatter in the next welding process.

[0047] In step 204 , as a result of the analysis process, the neural network provides a probability that weld spatter will occur in the next welding process and also provides a point in time at which weld spatter is likely to occur in the next welding process.

[0048] In step 205, a check is performed to determine whether the determined probability lies above a limit value or threshold value, which can be predefined, in particular by the user. For example, the threshold value can be predefined as 10% or, for example, as 25%. If the threshold value is not reached, the probability of occurrence is too low, and therefore no welding parameter adaptation is performed. In this case, the next welding process is performed according to step 202 using the currently predefined welding parameters.

[0049] If, on the other hand, the threshold value is reached or exceeded, the welding parameters are adapted in steps 206 and 207, in particular using a gradient method or a "gradient descent" method, to prevent the predicted occurrence of weld spatter. During the gradient method, the welding parameters or corresponding physical variables are selected autonomously and automatically to determine how to adapt them in order to avoid weld spatter. For example, during the gradient method, the welding current and welding voltage, and their temporal profiles, can be selected as welding parameters for adaptation.

[0050] In step 206, welding parameters are selected according to their influence on the result of the adaptation. For example, a measure of the influence of the parameter on the result, such as the gradient, can be determined for each of all welding parameters.

[0051] In step 207, a new parameter set is determined for which the probability of weld spatter occurring is below a threshold value. For this purpose, the corresponding welding parameters are optimized in the course of the gradient method in such a way that the probability can be reduced to a value below the threshold value.

[0052] Then, in step 208 , the next welding process is performed with altered parameters in order to prevent the predicted occurrence of weld spatter.

[0053] Within the scope of the present invention, the spatter probability and spatter time for the next weld are predicted based on a regression model. If the probability of spatter occurrence for the next weld estimated by the regression model exceeds a limit value defined (e.g., by the user), the welding parameters for the next weld are automatically adapted, in particular using a gradient method, preferably without limiting the weld quality. The regression model is particularly advantageously based on a recurrent neural network and inherently identifies which input variables are to be considered in its output or target value.

[0054] In particular, the model can be fed with all recorded welding data from the welding control system, for example, a time series of reference and / or actual curves, the welding parameters used for welding, and derived data such as the time intervals between individual welds. The outputs of the model can advantageously be spatter probability and spatter time, the correlation of which with the input variables is determined. Using this correlation, the welding parameters are automatically adapted, in particular using a gradient method, to minimize the probability of spatter occurrence.

Claims

1. A method for resistance welding, wherein: A welding process is performed, during the course of which welding electrodes (111, 112) are pressed against a welding point (125) of a workpiece (121, 122) according to predetermined welding parameters and energized with a welding current, wherein welding data describing the welding process are determined in each case during the course of the welding process, wherein an analysis of the welding data is performed with the aid of a regression model, wherein in the course of the analysis the probability of weld spatter occurring in the next welding process is determined and also the spattering time point at which weld spatter occurs in the next welding process is determined, wherein when the determined probability reaches a threshold value, at least one of the predetermined welding parameters is adapted, wherein in the course of the adaptation at least one relationship between the probability and / or the spattering time point as dependent variables on the one hand and at least one of the welding parameters as an independent variable on the other hand is determined with the aid of the regression model, wherein at least one of the predetermined welding parameters is selected and adapted with the aid of the at least one determined relationship.

2. The method according to claim 1, wherein A measure of the influence of at least one of the predefined welding parameters on the probability and / or the spattering time is determined using the at least one determined relationship.

3. The method according to claim 1 or 2, wherein: At least one of the predefined welding parameters is adapted by means of a gradient method. The method according to claim 1 , wherein the regression model is implemented as an artificial neural network. The method of claim 4 , wherein the regression model is implemented as a recurrent artificial neural network.

6. The method according to claim 1 or 2, wherein the welding parameters respectively include one or more of the following physical parameters: -Electrode force; - the position of the welding electrodes (111, 112); - welding current; - welding voltage; - welding time; -Time variation curve of electrode force, - a temporal variation of the position of the welding electrodes (111, 112); - the time profile of the welding current intensity; and -Temporal variation of the welding voltage.

7. A control unit (140) configured to execute the method according to any one of claims 1 to 6.

8. A welding device (100) having a control unit (140) according to claim 7.

9. A computer program product comprising a computer program which, when executed on a control unit (140), causes the control unit (140) to perform the method according to any one of claims 1 to 6.

10. A machine-readable storage medium having a computer program stored thereon, which, when executed on a control unit (140), causes the control unit (140) to perform the method according to any one of claims 1 to 6.

Images