Method for resistance welding and computer with processor device

Abstract

The invention relates to a method for resistance welding for joining at least two workpieces, wherein a welding process comprises supplying a welding current to the workpieces at a welding point via electrodes attached to at least one welding tongs and exerting a mechanical electrode force on the workpieces at the welding point via the welding tongs by applying pressure to the workpieces, so that a connection is created between the workpieces by the pressure and by welding heat due to an electrical material resistance experienced by the welding current.The invention provides that from at least a part of real-time data of at least one welding process, as a result of at least one prediction model based on at least one machine learning algorithm, a probability for an occurrence of at least one weld spatter, in which molten weld metal splashes out of the weld, for at least one future weld and at least one main influencing cause for the occurrence of the weld spatter for the at least one future weld is output.

Description

The invention relates to a method for resistance welding, in particular resistance spot welding, and to a computer having a processor device for carrying out the method.For joining at least two workpieces, a method based on resistance welding may be used. Resistance welding is defined in that a welding current is supplied to at least two electrically conductive workpieces to be connected. The welding current generates welding heat due to an electric resistance of the workpieces. In addition to the welding current, a mechanical electrode force is applied to the workpieces. The workpieces are pressed against one another by the mechanical electrode force. The mechanical electrode force and the welding heat, preferably at the same location and / or in the same region of the workpieces, generate a connection (joint), in particular a cohesive connection, between the at least two workpieces.One method based on resistance welding is resistance spot welding. In resistance spot welding, the welding current is supplied to the workpieces at a welding location via electrodes attached to at least one welding gun. In addition to supplying the welding current, the electrodes exert a mechanical electrode force at the welding location. The supporting surface of the electrodes on the workpieces is preferably circular.For the following description, a coordinate system is selected such that at least two workpieces contacting in the z-direction, in particular directly contacting, are to be connected to one another via resistance spot welding, then the center points of the two support surfaces are preferably located on an axis parallel to or on the z-direction and / or the mechanical electrode force is preferably exerted in the z-direction. A construction as used in resistance spot welding is shown by way of example in DE 10 2020 204 671 A1 Fig. 1.During resistance welding, at least one weld spatter may be produced during the welding process. A weld gun is defined by molten weld metal spraying out of the weld site. In other words, a welding spray is to be understood as meaning at least one drop of molten material which escapes from the at least one workpiece during a welding process. The material is the material of at least one of the workpieces to be connected or a mixture of the materials of at least two workpieces to be connected.EP 3 412 397 A1 describes a method for predicting a probability of occurrence of a spatter in resistance welding by means of a classification model.DE 10 2020 204 671 A1 describes a method for resistance welding, wherein an evaluation of welding data is carried out by means of a regression model, wherein in the course of this evaluation a probability with which a spatter occurs in a next welding process and also a spatter time at which the spatter occurs in the next welding process are determined.The invention is based on the object of avoiding a spatter in a future welding process.The object is achieved by the subject matters of the independent claims. Advantageous further developments of the invention are described by the dependent patent claims, the following description and the figures.The invention is based on the finding that, when detecting the occurrence of at least one spatter, the at least one spatter cannot be avoided only during the welding process. The detection of a spatter during the welding process can be carried out, for example, on the basis of a falling dynamic material resistance. In order to avoid at least one spatter, a prediction model is advantageous which predicts a probability for the occurrence of at least one spatter in at least one future welding process from a predeterminable part of the real-time data influencing the welding process, for example.Real-time data is understood to mean data relevant and / or characteristic for the welding process. In particular, the real-time data comprise process parameters, in particular automatically adaptable process parameters, which directly influence the process of resistance welding and can be automatically adapted, such as a welding current intensity, a voltage applied to the electrodes and / or an applied mechanical electrode force. Additionally or alternatively, the real-time data comprises component information describing the workpieces to be joined via resistance welding, for example material, welding location, thickness of the workpieces at the welding location, time interval to the previous welding process and / or the component to be produced by the joining. Additionally or alternatively, the real-time data includes information about the plant, such as information about and / or wear (useful life and / or frequency of use) of the forceps and / or electrodes used. In particular, real-time data firstly comprise data which are collected in real time during the welding process, and secondly real-time data can comprise data which are already predefined and known in advance.The real-time data on which the model result (result) of the at least one prediction model is based is referred to as input data. The input data is thus a real or spurious subset of the real-time data. In other words, the difference between real-time data and input data is that the prediction does not have to be based on all determined real-time data of all welding processes.The method according to the invention comprises outputting from the input data of at least one future and / or past welding process as a model result of at least one prediction model a probability for an occurrence of at least one spatter and at least one main cause of influence for the occurrence of the spatter for the at least one future welding location. The at least one prediction model is based in each case on at least one machine learning algorithm.The method has the advantage that spatters can be predicted, if necessary, already before the welding process and can thereby preferably be prevented. Preventing at least one spatter enables preventing the problems which may be associated with at least one spatter, in particular a disturbance of the assembly process and / or contamination of at least one installation component, which may lead to increased maintenance times and / or maintenance costs. In particular, by avoiding at least one spatter, manual post-processing can be dispensed with, which can be time-consuming and / or expensive.The model result of the at least one prediction model is preferably based on input data comprising real-time data from N past welding processes. N may be a natural number greater than or equal to 0. Preferably, N is between 5 and 20. In particular, the last N welding processes can be available as "lag variables" to the at least one prediction model.This results in a particularly reliable prediction of at least one weld spatter, since the real-time data of the current welding process is more similar to the real-time data of the last N welding processes than the real-time data of welding processes which were carried out before the last N welding processes. By considering the last N welding processes, in which spatter occurred, there is an advantageous focusing on the causes that have led to the last N spatters.The last N welding processes are defined as follows for the following description: If the last N welding processes are numbered from 1 to N and the current welding process is numbered from N+1, no welding process takes place between that numbered i and that numbered i+1 (i ranges from 1 to N). The last N welding processes in which spatter occurred are defined as follows: If the last N welding processes in which spatter occurred are numbered from 1 to N and the current welding process is numbered N+1, no welding process in which spatter occurred is performed between that numbered i and that numbered i+1, but a welding process in which no spatter occurred is performed between the welding process numbered i and the welding process numbered i+1 (i ranges from 1 to N).In addition, the real-time data can comprise data of at least one future welding process. Accordingly, this real-time data is real-time data already known before a welding process. This relates in particular to settings for the installation for resistance welding and / or information about the workpieces for the welding process, for which at least the probability of at least one spatter occurring is to be determined.Real-time data of already completed welding processes are preferably used together with the spatter information as to whether at least one spatter has occurred in the corresponding welding process ("ground truth" data) for training at least one prediction model based on a machine learning algorithm. The real-time data used for training is referred to as training data. Using the training data can comprise using the training data, for example, to define at least one parameter of at least one prediction model. This can be done in a manner known per se from the prior art. The machine learning algorithm can be a process based on at least one of the following methods known from the prior art: random forest (RF), eXtreme gradient boosting (XGBoost), feedforward neural network (FNN) and / or long short-term memory (LSTM).The use of real-time data as training data differs from the use of real-time data, in particular of completed welding processes, as input data for the at least one prediction model in that the input data is the data on the basis of which the at least one prediction model determines the model result. In contrast to the input data, the training data is that data which preferably influence the value of predefinable parameters of the at least one prediction model itself in a training phase. A parameter can be, for example, a weighting, i.e. how strongly a predeterminable part of the input data influences the model result.The model result comprises the probability of the occurrence of at least one spatter for at least one future welding location. In particular, it can be indicated whether the probability for a spatter for at least one future welding process exceeds a predefinable threshold value.Furthermore, the model result comprises at least one main cause of influence for the occurrence of the spatter for the at least one future welding location. This means that different causes are determined, on the basis of which the at least one prediction model comes to the output probability for the occurrence of the at least one spatter. The influence factor of the various causes can be determined, that is to say it can be determined which of the causes contributes most percentage points to the probability. For example, that at least one cause whose influencing factor is the greatest and / or exceeds a predefinable value and / or a predefinable proportion can be considered as the main cause of influence. In particular, that at least one cause and / or that at least one variable which contributes most to the probability of the occurrence of at least one spatter can be regarded as the at least one main cause of influence. Alternatively, the at least one main influence cause can be the at least one cause for which it is detected that, in the event of a change in the input data influenced by the at least one main influence cause, the output probability for the occurrence of at least one spatter can decrease most in at least one future welding process.The invention also comprises developments, by means of which additional advantages result.A development comprises that the model result comprises at least one action recommendation for a reduction in the probability of the spatter and / or at least one automatic adaptation of a process parameter of the at least one future welding process for the at least one future welding location for the reduction in the probability of the spatter in the at least one future welding process. This means that it is output which at least one action must take place in order to reduce the probability of the occurrence of a spatter at the at least one future welding location. The recommendation for action can comprise that certain settings should be made during the welding process, for example at at least one installation and / or installation component which is preferably involved in the welding process, and / or that other properties should be given to the workpieces, for example by adapting their thickness and / or material and / or shape and / or provided welding point. In the case of parameters relating to the welding process itself, which can be automatically regulated (in particular process parameters), for example a current intensity of the welding current used in resistance welding and / or the mechanical electrode force exerted, the setting can take place automatically.This results in the advantage that it is output which changes to a reduction in the probability of the occurrence of at least one spatter in the at least one future welding process are to be carried out. This enables a reduction in the probability of the occurrence of at least one spatter, in particular because what has to be done and no interpretation of data has to take place. If data are to be interpreted, measures may be taken which are not as effective as the at least one proposed and / or executed measure. The future welding process refers to a welding process that has not yet started upon output of the model result. At the welding point of the future welding process, therefore, there has not yet been any connection of the at least two workpieces.One development comprises using an ensemble of different prediction models based on different machine learning algorithms and combining the model results of the different prediction models into a single overall result, which replaces the model result. This means that at least two prediction models are used, which independently of one another each output a model result. The at least two different prediction models are based on different machine learning algorithms or machine learning processes. Possible algorithms or processes are, for example, RF, XGBoost, FNN and / or LSTM. The prediction models are preferably trained based on the training data and the associated spatter information according to methods known in the prior art. Based on the input data, each prediction model outputs a model result, respectively. After obtaining the model results, the model results are combined into the single overall result. In the case of a determination of a total result, this total result can replace the otherwise written model result.This can be effected, for example, by displaying the model result of the model which outputs the highest probability for at least one spatter in a future welding process. Additionally or alternatively, each prediction model may be assigned a model weighting, preferably depending on its reliability during prediction. The model weighting can be multiplied by the probability output by the prediction model. The model result which outputs the highest probability when considering the model weighting can be output as the overall result. Additionally or alternatively, at least one main cause of influence can be determined for each prediction model and the probability of the occurrence of at least one spatter in at least one future welding process during correction and / or removal of the at least one main cause of influence. The model result of the prediction model whose probability for at least one spatter of at least one future welding process is the lowest during correction and / or removal of the at least one main cause of influence can be output as the overall result.This yields the advantage of a more reliable overall result by combining different prediction models based on different machine learning algorithms.A development comprises that the real-time data and / or the input data comprise at least one of the following variables:• a variation over time of the welding current intensity: the welding current intensity can influence the welding heat, for example. The welding current intensity is in particular an automatically adaptable process parameter. An automatically adaptable process parameter can be defined in that a change in the value that this process parameter assumes preferably leads to a reduction, preferably a minimization, of the probability for at least one spatter. Automatic adaptation may mean that no action by a person is necessary, the adaptation by which a reduction in the probability of the occurrence of a spatter is achieved, i.e. takes place mechanically. If the model result yields an automatically adaptable process parameter as the main cause of influence, the value of this automatically adaptable process parameter is preferably automatically adapted.• A time profile of the electrical material resistance: the electrical material resistance can be determined from the welding current intensity and the electrical voltage applied to the electrodes. The electrical material resistance is, in particular, an automatically adaptable process parameter via the welding current intensity and the voltage applied to the electrodes, in that the welding current intensity and the voltage applied to the electrodes are adapted as a function of one another. The electrical material resistance may depend on the material of the workpieces. On the basis of the electrical material resistance, the presence of at least one spatter can be detected in the current welding process.• At least one reference resistance curve: by way of a comparison of the electrical material resistance with the reference resistance curve, which indicates the time profile of the electrical material resistance in a welding process without the occurrence of a spatter, in particular in an optimally running welding process, it is possible to infer the occurrence of a spatter in a future welding process. In particular in the case of deviations of the temporal profile of the electrical material resistance from the reference resistance curve, it is possible to infer the occurrence of a spatter in a future welding process.• a model, in particular a design and / or a design, of the welding gun: in the case of welding guns there may be different models, in particular different designs. For example, X-tongs, in which the electrodes are brought together via a swivel joint, such as in scissors, and which can apply a lower mechanical electrode force, are distinguished from C-tongs, which have a rectilinear movement of the electrodes and can only have a limited ejection. The choice of the type of welding gun can influence the welding process, in particular the generation of spatter, in particular via the respective advantages and disadvantages of the respective type. The model of the welding gun can influence and / or restrict, for example, the welding current and / or the mechanical electrode force generated and / or the voltage applied to the electrodes. Additionally or alternatively, the design of the welding gun can influence the welding process, in particular the formation of spatter. The design of a welding gun can include, for example, its stiffness. In particular, the design of the welding tongs comprises the construction of the welding tongs, i.e. for example their material and / or their dimensions. As a recommendation for action, for example, the use of another welding gun, in particular the use of a different design and / or design of the welding gun, can be provided.• a generated mechanical electrode force: the generated mechanical electrode force is the force with which the electrodes press on the workpieces, in particular the force with which the electrodes press the workpieces together at the welding location and / or press them against one another. The mechanical electrode force generated is in particular an automatically adaptable process parameter.• A number of welding processes already carried out with the electrodes: The quality of the electrodes decreases in particular with the number of welding processes carried out. The decrease in the quality of the electrodes with an increasing number of welding processes occurs in particular due to wear and / or contamination of the electrodes. The quality of the electrodes influences the electrical material resistance in particular via their electrical conductivity, which decreases in particular with an increasing number of welding processes. If the electrodes are determined as the main cause of influence, an exchange and / or a revision of these electrodes can be output as recommendation for action. In particular, a particularly advantageous point in time for an exchange and / or a reworking of the electrodes can be determined.• A material of the electrodes: The electrodes can be manufactured, for example, from copper alloys. Different materials, in particular different copper alloys, may have different properties. In particular, a mechanical strength and / or the electrical conductivity of the electrodes can be influenced by the selection of the material of the electrodes. The material of the electrodes can thus influence the welding process, in particular the formation of spatter, and / or the quality of the electrodes after a predeterminable number of welding processes. In particular, depending on the material of the workpieces, a different material may be expedient for the electrodes, in particular for avoiding spatter.• Geometry of the electrodes: The geometry of the electrodes can influence the area on which the mechanical electrode force is exerted and / or the transmission of the welding current to the workpieces.• Thicknesses of the various workpieces to be connected at the at least one welding location: In particular, in the event of repeated occurrence of spatters at a thickness in a specific interval and / or in the case of a material of the workpieces to be connected, a change in the thickness and / or of the material of the component can be provided in order to avoid spatters.• Materials of the various workpieces to be joined at the at least one welding location.• Information about the component to be produced: This can include, for example, the distance of the welding points. In particular, if the time interval between two welding processes is too short, the electrodes may not be sufficiently cooled, which may lead to a greater wear of the electrodes. By knowing the component, instead of waiting between two welding processes to increase the time interval between the welding processes, for example, the spatial interval and / or the order of the welding processes can be adjusted, so that a time interval between at least two welding processes can be increased without accompanying an increase in costs due to an increase in the time for welding the component. In particular, this can be advantageous in the case of components which have an unnecessarily large time interval for cooling the electrodes between two welding processes which differ from the welding processes with the excessively short time interval. By adapting the sequence of the welding processes, the time intervals between the welding processes can be adapted. Advantageously, it can be determined whether at least one spatter occurs in certain components with an increased probability in contrast to other components. Accordingly, an advantageous adaptation of the at least one component can be effected.In particular, by including the model and / or the type of welding gun, the material and / or the geometry of the electrodes, at least one property of the workpieces and / or component information in the real-time data, the probability for at least one weld spatter can be reduced not only during a future welding process, but also for a plurality of future welding processes, since this can be used to adapt to the specific requirements of the welding process. For example, it can be determined that a specific welding process requires other requirements, for example a higher mechanical electrode force and / or a higher welding current intensity, and the system components such as, for example, the welding tongs and / or the electrodes are adapted thereto and / or an adaptation of the material and / or the welding point can take place.In particular, when observing the real-time data of completed welding processes, it is possible to infer future welding processes. This means that if at least one spatter has occurred in the case of identical real-time data, for example in the case of the same component, electrodes and / or welding tongs, when welding processes have been completed, the probability of at least one spatter occurring in the case of unchanged real-time data is higher than if no spatter was detected in the case of the completed welding processes.This results in the advantage that more accurate action recommendations and / or more accurate adaptations of process parameters can be determined by incorporating different real-time data. This thus results in the advantage of a more accurate model result.A development comprises that at least one of the following variables is available for the prediction model by feature engineering, a transformation of raw data into attributes suitable for machine learning, in particular for training the prediction model:• Time interval between successive welding points,• the value of the electrical material resistance at at least one extreme of the time profile of the electrical material resistance of a welding process and the at least one associated time,• at least one ascent and / or descent rate of the time profile of the electrical material resistance of a welding process,• at least one difference value between the time profile of the material resistance and the reference resistance curve, in particular the difference between the time profile of the material resistance and the reference resistance curve at at least one extreme of the time profile of the material resistance and the reference resistance curve, respectively.In particular, the time interval between successive welding points has proven to be an important indicator for predicting at least one spatter.On the basis of the temporal profile of the electrical material resistance, it is possible to detect during the welding process whether a spatter will occur and / or has occurred in the current welding process. In particular, the value and / or the time of the at least one extreme and / or at least one rate of rise and / or fall of the temporal electrical material resistance can disclose the occurrence of at least one weld spatter. In particular, at least one weld spatter can be inferred in at least one future welding process via the temporal profile of the electrical material resistance. Preferably, a spatter occurring in previous welding processes can be predicted by observing the time profile of the material resistance.A development comprises that, for subsequent checking of the welding process, at least one image of the welding point is taken after the welding process and evaluation data is generated as the image result, which describes whether at least one spatter has occurred. This means that after at least one welding process, indications for at least one spatter are sought on the image. This can be effected, for example, by machine, image-based object recognition.This is because a drop in the time profile of the electrical material resistance can not only indicate at least one spatter, but can also have other causes. By the additional verification by means of image data, it is possible to check after the welding process whether the temporal profile of the electrical material resistance provides the correct spatter information for the welding process.By means of the subsequent check as to whether at least one spatter was present in a welding process, the at least one prediction model can be adapted, in particular by means of the real-time data from the welding process. In other words, for adapting the at least one prediction model, the image result from the examination of the image and the real-time data or the input data which were available to the at least one prediction model during the welding process are used as training data for the at least one machine learning algorithm.The real-time data of at least one welding process and the weld spatter information can be used as training data for adapting the at least one prediction model. The spatter information can be obtained via the image result and / or via the time profile of the electrical material resistance, in particular by searching for a drop in the resistance curve.The image result and the evaluation of the material resistance over time are preferably combined in order to ensure a reliable, subsequent detection of a spatter in a welding process. The real-time data and the spatter information thus serve as training data for the at least one machine learning algorithm of the at least one prediction model.This results in the advantage for the continuous adaptation, preferably improvement, of the at least one prediction model. In particular, an improvement of the at least one prediction model results from an increase in the training data, which preferably comprise as many different real-time data as possible.For use cases or application situations which can arise in the method and which are not explicitly described here, provision can be made for an error message and / or a request for inputting a user feedback to be output and / or for a default setting and / or a predetermined initial state to be set according to the method.As a further solution, the invention comprises a computer with a processor device. The computer is configured to predict a probability that at least one spatter occurs in a future welding process from input data comprising real-time data of at least one welding process by means of at least one prediction model, which is based in each case on at least one machine learning algorithm. In addition, the computer is configured to ascertain a main cause of influence by means of the at least one prediction model.The computer can have a data processing device or a processor device which is configured to carry out an embodiment of the method according to the invention. For this purpose, the processor device can have at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (field programmable gate array) and / or at least one DSP (digital signal processor). In particular, a CPU (central processing unit), a GPU (graphic processing unit) or an NPU (neural processing unit) can be used as the microprocessor in each case. Furthermore, the processor device can have program code which is configured to carry out the embodiment of the method according to the invention when executed by the processor device. The program code can be stored in a data memory of the processor device. The processor device can be based on at least one circuit board and / or on at least one SoC (system on chip), for example.The invention also includes developments of the computer according to the invention, which have features as have already been described in connection with the developments of the method according to the invention. For this reason, the corresponding developments of the computer according to the invention are not described again here.As a further solution, the invention also comprises a computer-readable storage medium comprising program code which, when executed by a computer or a computer cluster, causes the latter to execute an embodiment of the method according to the invention. The storage medium may be provided at least partially as a non-volatile data memory (e.g. as a flash memory and / or as an SSD-solid state drive) and / or at least partially as a volatile data memory (e.g. as a RAM-random access memory). The storage medium can be arranged in the computer or computer network. However, the storage medium can also be operated, for example, as a so-called store server and / or cloud server on the Internet. The computer or computer network can provide a processor circuit with, for example, at least one microprocessor. The program code can be provided as binary code and / or as assembler code and / or as source code of a programming language (e.g. C) and / or as a program script (e.g. Python).The invention also includes the combinations of the features of the described embodiments. The invention therefore also comprises implementations which each have a combination of the features of a plurality of the described embodiments, provided that the embodiments have not been described as mutually exclusive.Exemplary embodiments of the invention are described below. The following shows: FIG. 1 shows a flow chart of an embodiment of the method according to the invention.The exemplary embodiments explained below are preferred embodiments of the invention. In the exemplary embodiments, the described components of the embodiments each represent individual features of the invention that are to be considered independently of one another and that also develop the invention independently of one another. Therefore, the disclosure is intended to include combinations of the features of the embodiments other than those illustrated. Furthermore, the described embodiments can also be supplemented by further features of the invention that have already been described.In the figures, identical reference numerals designate functionally identical elements.FIG. 1 shows a flow chart of an embodiment of the method according to the invention. By means of the method, at least one device for resistance welding and / or at least one setting of the device for resistance welding can be adapted in order to avoid spatter as far as possible. The device for resistance welding enables in particular the connection of a plurality of workpieces, in particular in vehicle construction. The first step S 1 of the method comprises acquiring real-time data. In particular, the real-time data comprise data from the feature engineering, wherein these data are data which are extracted from raw data and are converted into data for training the machine learning algorithm.The real-time data can relate to at least one already completed welding process, the current welding process and / or at least one future welding process.At least one predeterminable part of the determined real-time data is provided as input data for at least one prediction model. The at least one prediction model is based in each case on a machine learning algorithm, for example on RF, XGBoost, FNN and / or LSTM, and in a step S 2 called spatter prediction, determines at least the probability for the occurrence of at least one spatter in at least one future welding process. Different prediction models are preferably based on different machine learning algorithms and can each determine a model result of their own comprising at least the probability of the occurrence of at least one spatter during at least one future welding process. In the case of only one prediction model, the model result corresponds to the final output total result. In the case of a plurality of prediction models, a model result can be output in each case for each prediction model and / or the model results can be combined, such that a single output overall result can be output.In addition to the probability of the occurrence of at least one spatter during at least one future welding process, a step S 3 called a case-based cause examination comprises the determination of at least one main cause of influence for the occurrence of at least one spatter. The at least one main cause of influence is, for example, the at least one cause, so that in the event of a change in the input data influenced by the at least one main cause of influence, the probability that at least one spatter occurs can decrease most in the case of at least one future welding process.The model result and / or the overall result preferably comprises at least one recommendation for action and / or at least one automatic adaptation of at least one process parameter, which are determined in a step S 4 called a case-based recommendation for action. A recommendation for action can comprise the display of a request, so that when the at least one instruction given in the request, for example an exchange and / or an adaptation of specific components and / or specific process parameters, is carried out, the probability of at least one spatter occurring in at least one future welding process can be reduced. In the case of automatic adaptation, the request can be executed automatically.The recommendation for action and / or the automatic adaptation is preferably based on the at least one main cause of influence determined. In particular, it is provided that the at least one prediction model determines which at least one change must be made in order to change the input data via the at least one main cause of influence in such a way that the probability of at least one spatter occurring during at least one future welding process is as low as possible.By means of a final step S 5, which carries the designation validation and feedback, an adaptation of the at least one prediction model can be provided. This can comprise that after a welding process at least one image of the welding location is recorded and for example by object recognition at least one spatter is sought. As a result, it is possible to validate weld spatter information determined from the temporal profile of the electrical material resistance about whether at least one weld spatter was present during the welding process under consideration. This is because even if no spatter has occurred, the change over time in the electrical material resistance can mimic the time profile of the electrical material resistance that is exhibited when at least one spatter occurs. The evaluation of the image can be regarded as a validation of the weld spatter information from the temporal profile of the electrical material resistance.This spatter information can be used in addition to the input data and / or real-time data of the welding process used in the observed completed welding process in order to provide training data for the at least one prediction model based on machine learning. In other words, it is known from the temporal profile of the electrical material resistance and its validation by the image result (the image result describes whether, for example, by object recognition on the image, the occurrence of at least one spatter was detected in the welding process under consideration) whether it was a welding process in which at least one spatter occurred or whether there was no spatter present. Together with real-time data determined during and / or for the welding process, training data can be provided for an adaptation, preferably an improvement, of the at least one prediction model.This at least one modified prediction model can in turn be used as at least one prediction model, by means of which, for example, in the steps of weld spatter prediction (S 3) and / or case-based cause examination (S 4), the probability for the occurrence of at least one weld spatter in at least one future welding process and / or the at least one main cause of influence is determined (S 6).A particularly advantageous exemplary embodiment is given below:By means of resistance welding, workpieces can be connected to one another in a materially integral manner. For example, in the course of automated body shell construction, different workpieces, for example metal sheets, are welded to one another by means of resistance welding using robot-guided welding tongs. Spatter may be generated during the welding process. In this case, in particular, molten weld metal sprays out of the welding location and can adhere to the workpiece, to the electrodes and / or to system components.In particular, the material properties, process parameters, but also process fluctuations such as surface dirt, gripper imperfections or the varying gap sizes between the joining partners can influence the formation of spatter.The current detection of spatters (on the basis of the falling dynamic resistance; the dynamic resistance is the time profile of the electrical material resistance) makes it possible to detect at least one spatter during and / or after the welding process.At present, spatter is detected only during and / or after the process and cannot be avoided in good time as a result. Furthermore, the case-dependent causes for the generation of a spatter are not known, so that it is not possible to adequately deal with. Spatter can then interfere with the process and / or contaminate plant components. This leads to increased maintenance times and / or maintenance costs. Sharp-edged material adhesions on at least one workpiece can damage at least one cable harness during the assembly. Therefore, spatters must be remachined manually, which is time consuming and / or expensive.In order to avoid spatter, these must be recognized in good time in the process and the underlying cause must be determined. A data-based prediction (prediction model) of spatters analyzes automatically, online and / or in real-time IoT process data and / or real-time data and / or input data and predicts a probability of occurrence of a spatter for the at least one future and / or next welding process and / or welding location based on previous and / or current and / or future real-time data (welding point data and / or data of the welding process and / or process settings for the at least one future welding location and / or a future welding process, the material information, component information and installation information (e.g. type of tongs) for the future welding process and / or the future welding location). Additionally or alternatively, the at least one main influencing factor is output. As a result, the welding process can be adapted before a weld spatter is produced. A validation and feedback pipeline can continuously improve the prediction model.The detection of spatters can be effected on a rule-based basis via the falling dynamic resistance in the resistance curve. The rule-based recognition of spatters makes it possible to subdivide the data set into two classes: welding processes and / or welding points with spatters and welding processes and / or welding points without spatters.With the aid of a classified data set comprising input data which comprise real-time data of selected welding processes and the information as to whether at least one spatter was present in the selected welding process assigned to the input data, different machine learning algorithms (RF, XGBoost, FNN) are trained which can predict a probability for the occurrence of a spatter for the future welding point on the basis of information on previous spatters (input data in the form of lag variables) and the process parameters of the next welding point. An LSTM is trained to discover time encoded information in the data. Additionally or alternatively, an ensemble of trained models may be used for prediction.The data and / or real-time data are process data (for example current and voltage curves and the resistance curves calculated therefrom), for example from the Bosch 6000 SPS, process settings (reference resistance curves), tong data (for example tong model, motor current, force values, arm bending), electrode data (for example number of already welded welds, electrode material, geometry of the electrode caps, electrode path), material characteristics (material thickness combination MDK), and / or vehicle data (component information). Additionally or alternatively, further information is incorporated into the model by feature engineering.The trained model receives as input data the described data set comprising at least a part of the real-time data in an automated, online and real-time manner and outputs a percentage probability for the occurrence of a spatter for the at least one future welding process and / or the at least one future welding location.Feature Engineering:The features produced include, among other things, the time interval between successive welding points, meaningful points in the dynamic resistance curve (e.g. resistance maxima and / or minima and / or associated points in time in the resistance welding curve and / or resistance descent and / or rates of rise).Validation Pipeline:Linking the described input and output data of the model to image data of welded body components from the series enables validation of the prediction of weld spatters and optimization and / or adaptation of the prediction model. The validation pipeline enables a continuous improvement and / or adaptation of the prediction model by means of a feedback loop.Generation of Action Recommendations Based on the Analysis of the Most Relevant Features:By means of case-based root cause examinations and correlation analyses, decisive influencing factors for the formation of spatters are determined. This makes it possible to communicate case-based action recommendations to the maintenance person and / or user, in order to thus avoid the generation of at least one spatter. For example, a spatter is predicted from at least one prediction model and the major impact factor, whatever the model predicted a spatter, is electrode failure. Therefore, the recommendation would be for the maintenance person and / or user to service the welding gun. If the main influencing factor is a process parameter, the process can be automatically adapted.Overall, the examples show how prediction and automatic validation of spatter can be provided in resistance spot welding by machine learning.References included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedDE 10 2020 204 671 A1 [0004, 0007]EP 3 412 397 A1

[0006]

Claims

Method for resistance welding for connecting at least two workpieces, wherein a welding process comprises supplying a welding current to the workpieces at a welding point via electrodes attached to at least one welding gun and applying a mechanical electrode force to the workpieces via the welding gun by pressure on the workpieces at the welding point, such that a connection between the workpieces is generated between the workpieces by the pressure and by a welding heat due to an electrical material resistance experienced by the welding current, characterized in that a probability for an occurrence of at least one welding spatter, in which molten weld material sprays out of the welding point, is produced from at least a part of real-time data of at least one welding process as a model result of at least one prediction model, for at least one future welding location and at least one main cause of influence for the occurrence of the spatter is output for the at least one future welding location.Method according to claim 1, wherein the model result additionally comprises: • at least one action recommendation for a reduction of the probability of the spatter and / or • at least one automatic adaptation of a process parameter of the at least one future welding process for the at least one future welding location for reduction of the probability of the spatter in the at least one future welding process.Method according to one of the preceding claims, wherein an ensemble of different prediction models based on different machine learning algorithms is used and the model results of the different prediction models are combined into a single overall result.Method according to one of the preceding claims, wherein the real-time data comprise at least one of the following variables: • a time profile of a welding current intensity, • a time profile of the electrical material resistance, • at least one reference resistance curve, • model of the welding gun, • a design and / or a design of the welding gun, • a generated mechanical electrode force, • a number of welding processes already carried out with the electrodes, • a material of the electrodes, • a geometry of the electrodes, • thicknesses of the various workpieces to be connected at the at least one welding location, • materials of the various workpieces to be connected at the at least one welding location, • information about the component to be produced.Method according to one of the preceding claims, wherein at least one of the following variables is available for the prediction model by feature engineering: • time interval between successive welding points, • the value of the electrical material resistance at at least one extreme of the time profile of the electrical material resistance of a welding process and the at least one associated point in time, • at least one rate of rise and / or fall of the time profile of the electrical material resistance of a welding process, • at least one difference value between the time profile of the material resistance and the reference resistance curve.Method according to one of the preceding claims, wherein, for subsequent checking of the welding process, at least one image of the welding point is made after the welding process and evaluation data are generated as the image result, which data describe whether at least one weld spatter has occurred.Method according to Claim 6, wherein, in order to adapt the at least one prediction model, the image result from the examination of the image and the real-time data which were available to the at least one prediction model during the welding process are used as training data for the at least one machine learning algorithm.Computer having a processor device which is set up to carry out the method according to one of the preceding claims.

Images