Robotic process

Abstract

A method for performing a process according to the invention, in particular by means of at least one robot (10), comprises the following steps: performing (S10) a run of the process with a process control, in particular a process control of the robot; detecting (S10) values of a first process variable (x) for the performance; and detecting (S10) an evaluation (E) of the performed process run; an evaluation learning step comprising the following multiple repetitions: changing (S20), by means of an optimizer (4), at least one parameter of the process control, in particular a regulator (31), to a changed process control based on values of a quality criterion of the performed process run; performing (S20) a run of the process with the changed process control; detecting (S20) values of the first process variable for the performance; and detecting an evaluation (E) of the performed process run; wherein a first quality factor model (51) of the process is machine learned based on the detected evaluations and values of the first process variable, the first quality factor model determining a quality factor (E’) of the process based on the first process variable (x); and an process control optimization step comprising the following multiple repetitions: changing (S40), by means of an optimizer, the process control to a changed process control based on values of a quality criterion of the performed process run; performing (S40) a run of the process with the changed process control; and detecting (S40) values of the first process variable for the performance, wherein values of the quality criterion for at least one performed process run with the changed process control are determined based on the quality factor (E’) determined by the machine learned first quality factor model from the detected values of the first process variable for the process run.

Description

TECHNICAL FIELD

[0001] The invention relates to a method, a system and a computer program product for executing a process, in particular with the aid of at least one robot. BACKGROUND

[0002] It is known from in-house practice to optimize a robot process, for example to adjust parameters, by means of an optimizer which optimizes a quality criterion, for example minimizes a cost function, etc.

[0003] However, the optimization, in particular the convergence and the finding of a (local) optimum, depends to a large extent on the quality criterion, but it is often difficult in practice to formulate suitable, in particular sensible, quality criteria, for example in terms of robot-assisted snap-on or snap-in locking: Thus, it is possible to judge very simply whether a robot-assisted plug-in or snap-in locking process has been successful, but it is not easy to read this from the force or joint angle curve of the robot. SUMMARY

[0004] It is an object of the invention to improve the execution of a process, in particular a robot-assisted process.

[0005] One embodiment according to the invention comprises a method for executing a process, in one embodiment with the aid of, in particular by means of, at least one robot, which method comprises the following steps:

[0006] - executing a run of the process with a process control, in one embodiment with a robot process control;

[0007] - detecting (one- or multi-dimensional) values of a one- or multi-dimensional first process variable for the execution, in particular during the execution of the process; and

[0008] - detecting an evaluation of the executed process run.

[0009] According to one embodiment of the invention, after the execution of the process with an initial process control, which in one embodiment is a default, in particular standard, process control or has already been pre-optimized, in particular empirically, and for this or in this the values of the first process variable have been detected and (immediately after the process run) an evaluation of the executed (original) process run has been detected (in one embodiment by a person), the following evaluation learning step is repeated a plurality of times:

[0010] - based on the value of the quality criterion of the executed process run, the process control, in particular the hitherto or previous process control, in one embodiment one or more parameters of the process control, in one extension one or more parameters of the regulator, is changed by the optimizer, in particular an optimization method, preferably a numerical optimization method or measure, in particular an algorithm, to a changed process control;

[0011] - the (further) run of the process is executed with the changed process control;

[0012] - the value of the first process variable for the execution, in particular during the execution, is detected; and

[0013] - an evaluation of the executed process run is detected, in one embodiment similar to the evaluation for the initial process control or during the initial process control;

[0014] - the (further) run of the process is executed with the changed process control;

[0015] In particular,

[0016] - based on the value of the quality criterion of the executed original process run, the initial process control is changed by the optimizer to a changed process control,

[0017] - the run of the process is re-executed with the changed process control,

[0018] - the value of the first process variable for the re-execution, in particular during the re-execution, is detected, and

[0019] - an evaluation of the executed process run is detected;

[0020] and multiple times

[0021] - based on the value of the quality criterion of the hitherto executed process run, the process control, in particular the already or multiple times changed or previous process control, is further changed by the optimizer to a changed process control,

[0022] - the next run of the process is executed with the changed process control,

[0023] - the value of the first process variable for the next execution, in particular during the next execution, is detected, and

[0024] - an evaluation of the executed process run is detected;

[0025] wherein based on the detected values of the evaluation and the first process variable, a first quality factor model of the process is machine learned, which determines a quality factor of the process based on (values of) the first process variable.

[0026] In an embodiment, the idea is based on using at least one machine learning method for evaluating a process run. Thereby, in an embodiment, also complex and / or varying processes can advantageously be optimized, in particular when the success of the process is hardly directly measurable. As mentioned in the outset, for example in the context of robot- assisted snap-in or click-in processes, a human, image and / or audio processing can easily judge whether a robot- assisted plugging or (click-in) click-in process was successful, in particular by a human's visual judgment of the result of the engagement, by image processing, in particular image recognition, of the result of the engagement, or by audio processing of (click-in) click-in noises, whereas these cannot be easily read from the robot's force or joint angle curves.

[0027] Accordingly, according to an embodiment of the present invention, the method comprises, after the evaluation learning phases or evaluation learning steps, further process control optimization steps which are repeated a plurality of times:

[0028] - based on the values of the quality criterion of the performed process runs, changing the process control to a changed process control by an optimizer;

[0029] - performing a run of the process with the changed process control; and

[0030] - detecting values of the first process variable for the performance, in particular during the performance; wherein the values of the quality criterion of one or more of the process runs which have been performed with the changed process control are determined based on (respectively) a quality factor which is determined by a first quality factor model which is machine learned from the detected values of the first process variable for (or during) the respective process run.

[0031] Hence, in an embodiment, the optimization of the process by an optimizer is combined with a judgment of the process (success) by a machine learning method. In an embodiment, the quality factor of the process is the success of the process or other factor or component of the quality criterion. Accordingly, in an embodiment, the evaluation of the process run comprises a value (rating level) of one or more quality factors or the detection of the evaluation of the performed process run comprises a detection of a value of a quality factor of the performed process run. In an embodiment, the quality criterion is a (minimized) cost function of the optimizer.

[0032] In an embodiment, the value of the quality criterion for one or more process runs performed in the evaluation learning step with changed process control is determined (based on the detected evaluation of the process run, respectively) in the same way as in the one or more process control optimization steps, wherein (in an embodiment only) the detected evaluation is used instead of the quality factor determined by the quality factor model.

[0033] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability is increased and / or the required time and / or the number of runs is reduced, and / or the found Optimums or the practicability of the optimized process control, in particular robot control, is improved.

[0034] In an embodiment, the method comprises the following steps:

[0035] - detecting (one-dimensional or multi-dimensional) values of an additional or second one- or multi-dimensional process variable for the execution of the process run, in particular during the execution of the process run, in the evaluation learning step and / or in the process control optimization step.

[0036] In an extended embodiment, the value of the quality criterion for one or more process runs performed in the evaluation learning step with changed process control is determined (based on the detected values of the second process variable for the process run, respectively) additionally or depends additionally on the detected values of the second process variable for the process run, in particular in the process run, wherein the detected evaluation of the process run is independent of the values of the second process variable.

[0037] Additionally or alternatively, in an extended embodiment, the value of the quality criterion for one or more process runs performed in the process control optimization step with changed process control is determined (based on the detected values of the second process variable for the process run, respectively) additionally or depends additionally on the detected values of the second process variable for the process run, in particular in the process run, wherein the quality factor determined by the first quality factor model is independent of the values of the second process variable.

[0038] Thus, in an embodiment, in addition to the detected evaluation or the quality factor determined by the quality factor model, a further or second one- or multi-dimensional variable, e.g. a process duration, etc., is considered in determining the value of the quality criterion or in the optimization process.

[0039] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability and / or the required time and / or the number of runs can be increased and / or the found optimization or the practicability of the optimized process control, in particular the robot controller, can be improved.

[0040] In an embodiment, for one or more process runs performed with the changed process control in the process control optimization step, an evaluation of the process run is detected (respectively); in an embodiment, the evaluation of the process run is detected in the same way as in the one or more evaluation learning steps and, in an embodiment, is compared with the quality factor determined by the machine-learned first quality factor model based on the values of the first process variable, in particular detected in the process run, for this process run.

[0041] In an extended embodiment, the first quality factor model is further machine-learned based on the values of the first process variable and the detected evaluation.

[0042] In other words, the machine learning is also further machine-learned during the (further) optimization of the process control in the one or more process control optimization phases or steps. This can be particularly expedient or advantageous in the case where the process control is significantly changed in the optimization phase, for example additional movements are established in the joining process, so that the first process variable, for example the joint coordinates or the force curve, is changed accordingly significantly.

[0043] Additionally or alternatively, in an extended embodiment, a notification is issued, in an embodiment an optical and / or acoustic notification, in an embodiment on the robot, for example by a light fixed to the robot or the like, when, in particular as long as, a tolerance parameter depending on the deviation between the detected evaluation and the determined quality factor is outside a predetermined tolerance range.

[0044] Thereby, in the one or more process control optimization phases or steps, during the (further) optimization of the process control, further machine learning can advantageously be continued as long as necessary.

[0045] Additionally or alternatively, in an extended embodiment, the detection of an evaluation of a further process run performed with the changed process control in the process control optimization step is performed depending on the result of the comparison or is dependent on the result of the comparison.

[0046] In an embodiment, the first quality factor model is to further machine learn, if, in particular as long as, the quality factor determined by the first quality factor model is too far off from the detected evaluation deviation or a tolerance parameter depending on the deviation is outside a predetermined tolerance range. Additionally or alternatively, when the quality factor determined by the first quality factor model is sufficiently in agreement with the detected evaluation or a tolerance parameter depending on the deviation is within a predetermined tolerance range, then the interval of further comparisons, in an embodiment a time interval, can be increased or the further comparisons can be omitted.

[0047] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability can be increased and / or the required time and / or the number of runs can be reduced and / or the found optimization or the practicability of the optimized process control, in particular robot control, can be improved.

[0048] In an embodiment, based on the evaluation of the first process variable detected for the evaluation learning step or in the evaluation learning step and the value, at least one further quality factor model of the process is machine learned, which determines the quality factor of the process based on (values of) the first process variable, wherein the first quality factor model and the further quality factor model are different.

[0049] In an extended approach, in at least one process control optimization step, a value of a quality criterion of a process run performed with a changed process control is determined based on the quality factor determined by the further quality factor model after machine learning from the value of the first process variable detected for the process run, in particular in the process run. Thereby, in an embodiment, two or more different machine learning approaches or machine learned quality factor models are bundled together, in an embodiment combined together (used).

[0050] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability can be increased and / or the required time and / or the number of runs can be reduced and / or the found optimization or the practicability of the optimized process control, in particular robot control, can be improved.

[0051] Additionally or alternatively, in an extension, a notification is issued, in an embodiment an optical and / or acoustic notification, in an embodiment on the robot, e.g. by a light fixed to the robot, when a tolerance parameter depending on a deviation between the quality factor determined by the first quality factor model and by the further quality factor model from a value of the first process variable detected in the process control optimization step for or in a process run is outside a predetermined tolerance range. In an embodiment, a signal (by the notification) is issued that operator intervention is required.

[0052] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability is increased and / or the required time and / or the number of runs is reduced, and / or the found optimal or optimized process control, in particular robot control, is improved in its practicability.

[0053] In an embodiment, a confidence interval of the first quality factor model is determined, and a notification is issued, in an embodiment an optical and / or acoustic notification, in an embodiment on the robot, e.g. by a light fixed to the robot, when the confidence interval exceeds a boundary value, in particular a predetermined boundary value. In an embodiment, a signal (by the notification) is issued that operator intervention is required.

[0054] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability is increased and / or the required time and / or the number of runs is reduced, and / or the found optimal or optimized process control, in particular robot control, is improved in its practicability.

[0055] In an embodiment, at least one process run performed with the changed process control in the evaluation learning step is evaluated based on an electronic transmission signal, in particular without direct view on the process. The electronic transmission signal can in particular have one or more taken images of the process.

[0056] Thereby, in an embodiment, the safety of the operator can be increased, and / or the operator can evaluate simultaneously multiple decentralized processes, in particular of multiple decentralized robots, or optimize these processes.

[0057] In one embodiment, the first process variable and / or the second process variable has actual data and / or target data specific to the robot, which in one embodiment is detected by sensors on the robot side and / or outside the robot, in particular coordinates of a pose of at least one reference object fixed to the robot and / or a first-order and / or a higher-order time derivative thereof, forces on at least one reference object fixed to the robot and / or at least one drive variable of at least one robot drive, and / or visual data and / or audio data and / or time data, in particular duration.

[0058] Thus, in one embodiment, the first process variable and / or the second process variable comprises a robot trace, or a time curve of joint coordinates, and / or coordinates of an end effector pose in the work space and / or a time derivative thereof, and / or a time curve of external forces, in particular contact forces, acting on the robot, and / or a time curve of drive forces and / or currents and / or voltages of drives of the robot. For a more compact description, antiparallel force couples or torques are also collectively referred to as forces herein.

[0059] Additionally or alternatively, in one embodiment, the first process variable and / or the second process variable comprises visual data, in particular camera or image data, and / or audio signals, or is data determined, in particular determined, based on images, in particular camera images, and / or audio signals.

[0060] Additionally or alternatively, in one embodiment, the first process variable and / or the second process variable comprises a duration of the respective process run and / or one or more portions thereof.

[0061] Such process variables are particularly suitable for optimizing the process, on the one hand, and are particularly well considered or utilized by the quality factor model of the machine learning, on the other hand.

[0062] In one embodiment, the evaluation is detected by manual input. For example, an operator can indicate or evaluate, in one embodiment evaluate the success of, the quality factor of the process run, respectively, during and / or after the process run.

[0063] Thereby, in one embodiment, the practicability of the optimized or optimized process control, in particular robot control, found can be improved.

[0064] Additionally or alternatively, in one embodiment, the evaluation is detected automatically, in one embodiment sensor-assisted. For example, image recognition or image processing and / or audio recognition or audio processing can indicate or evaluate, in one embodiment evaluate the success of, the quality factor of the process run, respectively, during and / or after the process run.

[0065] Thereby, in an embodiment, the optimization, in particular the convergence, can be improved, in particular the stability is increased and / or the personnel outlay, the required time and / or the number of runs is reduced.

[0066] Additionally or alternatively, in an embodiment, the evaluation is bivalent, in particular "good" / "bad", "0" / "1", "ordered" / "unordered", etc.

[0067] Thereby, in an embodiment, the optimization, in particular the stability, can be increased.

[0068] In another embodiment, the evaluation is trivalent or more-valent, for example includes a score comprising a three-valued or multi-valued scale, a classification of three or more different quality levels, etc.

[0069] Thereby, in an embodiment, the utility of the found optimization or the optimized process control, in particular the robot control, can be improved.

[0070] In an embodiment, the first quality factor model has a neural network, a random forest model, a decision tree model, a k-nearest neighbor model, a logistic regression model or a linear model, in particular a generalized linear model.

[0071] Additionally or alternatively, in an embodiment, the further quality factor model has a neural network, a random forest model, a decision tree model, a k-nearest neighbor model, a logistic regression model or a linear model, in particular a generalized linear model, wherein, in an embodiment, the first quality factor model and the further quality factor model are different.

[0072] These machine learning methods are particularly suitable for determining a quality factor of a process, in particular a robot-assisted process.

[0073] In an embodiment, a different subsequent process is executed depending on the detected evaluation of the executed process in the evaluation learning step. Additionally or alternatively, in an embodiment, a different subsequent process is executed depending on the determined quality factor of the executed process in the process control optimization step. For example, in a (sufficiently) successful joining process, the component can be supplied to the normal subsequent process run, and in a (not sufficiently) successful joining process, the component can be sorted out or fed into a reprocessing.

[0074] Thereby, in an embodiment, the process, in particular the robot-assisted process, can be further improved.

[0075] According to an embodiment of the application, a system for carrying out a process is proposed, in one embodiment by means of, in particular by at least one robot, in particular hardware- and / or software-technically, in particular programmatically designed for carrying out the method described here, and / or comprising:

[0076] - means for carrying out a run of the process by process control, in particular by process control of a robot;

[0077] - means for detecting a value of a first process variable for the execution, in particular during the execution;

[0078] - means for detecting an evaluation of the executed process run;

[0079] - means for repeatedly executing the evaluation learning step a plurality of times:

[0080] - changing, by an optimizer, the process control, in particular at least one parameter of the process control, in particular of a regulator, to a changed process control based on the value of the quality criterion of the executed process run;

[0081] - carrying out a run of the process with the changed process control;

[0082] - detecting a value of a first process variable for the execution, in particular during the execution; and

[0083] - detecting an evaluation of the executed process run;

[0084] wherein a first quality factor model of the process is machine-learned based on the detected evaluation and the value of the first process variable, the first quality factor model determining a quality factor of the process based on the first process variable;

[0085] and

[0086] - means for repeatedly executing the process control optimization step a plurality of times:

[0087] - changing, by an optimizer, the process control to a changed process control based on the value of the quality criterion of the executed process run;

[0088] - carrying out a run of the process with the changed process control; and

[0089] - detecting a value of a first process variable for the execution, in particular during the execution; wherein a value of a quality criterion for at least one process run carried out with the changed process control is determined based on a quality factor determined by the first quality factor model machine-learned based on the detected value of the first process variable for the process run or in the process run.

[0090] In an embodiment, the system or the device thereof comprises:

[0091] - a device for determining a value of a quality criterion for at least one process run performed with the changed process control in the evaluation learning step based on the detected evaluation of the process run; and / or

[0092] - a device for detecting a value of a second process variable for the execution of a process run in the evaluation learning step and / or in the process control optimization step, wherein the value of the quality criterion for at least one process run performed with the changed process control in the evaluation learning step additionally depends on the detected value of the second process variable for the process run, the detected evaluation of the process run being independent of the value of the second process variable; and / or wherein the value of the quality criterion for at least one process run performed with the changed process control in the process control optimization step additionally depends on the detected value of the second process variable for the process run, the quality factor determined by the first quality factor model being independent of the value of the second process variable; and / or

[0093] - a device for detecting an evaluation of at least one process run performed with the changed process control in the process control optimization step and, in an embodiment, comparing it with a quality factor determined by the first quality factor model based on the detected value of the first process variable for the process run by machine learning, and for further machine learning of the first quality factor model based on the value of the first process variable and the detected evaluation and / or for issuing a notification when a tolerance parameter depending on a deviation between the detected evaluation and the determined quality factor is outside a predetermined tolerance range and / or for detecting an evaluation of a further process run performed with the changed process control in the process control optimization step depending on the result of the comparison; and / or

[0094] - a device for machine learning of at least one further quality factor model of the process based on the detected evaluation and the value of the first process variable in the evaluation learning step, in particular in the evaluation learning step, wherein the first quality factor model and the further quality factor model are different, and for determining a value of a quality criterion for a process run performed with the changed process control in at least one process control optimization step based on a quality factor determined by the machine-learned further quality factor model from the detected value of the first process variable for the process run, and / or for issuing a notification when a tolerance parameter depending on a deviation between the quality factors determined by the first quality factor model and by the further quality factor model based on the detected value of the first process variable for the process run or in the process run in the process control optimization step is outside a predetermined tolerance range; and / or

[0095] - means for determining a confidence interval for the first quality factor model and for issuing a notification if the confidence interval exceeds a boundary value; and / or

[0096] - means for evaluating at least one process run performed with the changed process in the evaluation learning step based on the electronic transmission signal, in particular without direct sight of the process; and / or

[0097] - means for detecting the evaluation by manual input and / or automatically, in particular sensor-aided;

[0098] and / or

[0099] - means for performing a different subsequent process depending on the detected evaluation of the process performed in the evaluation learning step and / or depending on the determined quality factor of the process performed in the process control optimization step.

[0100] The means in the sense of the present application can be constituted in hardware technology and / or software technology, in particular with a processing unit, in particular a digital processing unit, in particular a microprocessor unit (CPU), and / or one or more programs or program modules, preferably in data connection or signal connection with a storage system and / or a bus system. The CPU can be designed for this purpose to execute instructions of a program implemented as stored in the storage system, to take in input signals from a data bus and / or to send output signals to the data bus. The CPU can in particular have, in particular a graphics card (GPU) and / or at least one neural computing chip. The storage system can have one or more, in particular different, storage media, in particular optical, magnetic, solid and / or other non-volatile media. The program can be provided in such a way that it embodies or executes the method described herein, so that the CPU can execute the steps of the method and thereby in particular can execute the process or its optimization. In an embodiment, the computer program product can have, in particular be, in particular non-volatile, a storage medium for storing the program or a storage medium on which the program is stored, wherein the execution of the program causes the system, in particular a computer, to execute the method described herein or one or more steps of the method.

[0101] In an embodiment, one or more steps, in particular all steps, of the method are executed fully or partially automatically, in particular by the system or the means thereof. BRIEF DESCRIPTION OF DRAWINGS

[0102] Further advantages and features are given by the dependent claims and the embodiments. For this purpose, some are shown schematically:

[0103] Figure 1for a method for performing a process according to an embodiment of the application; and

[0104] Figure 2 for a system for performing the method according to an embodiment of the application. DETAILED DESCRIPTION

[0105] Figure 2 A system for performing a method according to an embodiment of the application is shown.

[0106] The system comprises a robot 10 with joints (drives) 11 which is to insert a clip 20 with its end effector 12 onto a component 21 conveyed on a conveyor belt 70.

[0107] For this, it is controlled by a robot controller 30 with regulators 31, where in this text the regulation is also referred to as control, and the regulators 31 can also have individual joint regulators for individual joints (drives) 11.

[0108] In a step S10, the insertion process is first performed using default values of the parameters of the regulators 31. The first process variable(s) (values) include the robot trajectory x, for example axis angles and / or axis speeds, and / or forces or torques in the drives and / or axes and / or on the end effector 12, and are transmitted to the machine learning method 50, in this embodiment an artificial neural network 51.

[0109] In addition, an operator (not shown) evaluates the process run, for example its success, by means of the input device 6. In another embodiment, it can also specify or input a value for another quality factor, respectively. This evaluation E is detected and likewise notified to the machine learning method 50 or artificial neural network 51 and to the optimizer 4.

[0110] The second process variable(s) (values) in the form of the process duration y are likewise notified to the optimizer 4.

[0111] Next, the parameters of the regulators are changed by the optimizer 4 a plurality of times, and as long as the interruption criterion is not met (step S30: "N"), for example the artificial neural network 51 has not been trained sufficiently or the quality factor model realized thereby has not been machine learned sufficiently, the process is reperformed with the process control thus changed, i.e. with the regulators thus changed (step S20). Figure 1

[0112] Here, in each of these repeatedly evaluated learning steps S20, the respective values x, y of the first or second process variable and the evaluation E are notified to the machine learning method 50 or artificial neural network 51 (x, E) or to the optimizer 4 (y, E), respectively, by the operator.

[0113] ​The optimizer 4 determines a value of the quality criterion, for example as a weighted sum of the individual process durations y and the evaluations E, and changes the parameters of the regulator, the quality factor model or the artificial neural network 51 based on the (value of the) quality criterion, based on which the evaluations E and the values of the first process variables x are trained or machine-learned.

[0114] If the interruption criterion is met (S30, "Y"), the system or method continues with the process control optimization.

[0115] Here, the parameters of the regulator are further changed by the optimizer 4, and the process is re-executed with the thus changed process control, i.e. the thus changed regulator parameters (step S40), until the interruption criterion is met (step S50: "Y"), for example the value of the quality criterion is within a predetermined range.

[0116] Here, the value of the quality criterion is determined in a similar manner as a weighted sum, wherein now the quality factors E' determined by the machine-learned quality factor model 51 based on the values of the first process variables x are used instead of the evaluations E.

[0117] If the interruption criterion is met (step S50, "Y"), the process can be continued with the thus optimized process control, in particular the robot control (S80), wherein here too further optimization and / or machine learning can take place.

[0118] If the process control optimization has not yet ended (S50: "N"), the quality of the machine-learned quality factor model is checked at regular intervals, for example the confidence intervals or the bias of the different types of second quality factor models used for parallel machine learning are checked or further detected evaluations E, in one embodiment further detected evaluations E with increasing and / or bias-dependent time intervals. If the confidence intervals or the bias are too large (S60: "Y"), the operator is requested or signaled to intervene (step S70), otherwise (S60: "N") the process control optimization is continued.

[0119] Depending on the detected evaluations E or the determined quality factors E', the controller 30 instructs the conveyor belt 70 to supply the component 21 after the joining process to the normal subsequent process or to reprocess it.

[0120] Although exemplary embodiments are set forth in the foregoing description, it should be understood that a wide variety of modifications can be made.

[0121] Thus, in this embodiment, the neural network 51 only learns based on sensor data x specific to the robot. In a variant, other data, for example process durations y, etc., can additionally or alternatively also be taken into account.

[0122] It should also be noted that these example embodiments are merely examples and should not form any limitation on the scope of protection, application and construction. On the contrary, the person skilled in the art, by virtue of the foregoing description, is able to derive the teaching to implement a conversion of at least one example embodiment, in which various changes can be implemented, in particular with regard to the functions and arrangements of the components, without departing from the scope of protection of the invention, for example as can be obtained according to the claims and equivalent combinations of features thereof.

[0123] List of reference signs

[0124] 4 optimizer

[0125] 6 input device

[0126] 10 robot

[0127] 11 joint (driver)

[0128] 12 end effector

[0129] 20 catch

[0130] 21 component

[0131] 30 controller

[0132] 31 regulator

[0133] 50 machine learning method

[0134] 51 artificial neural network (quality factor model of machine learning)

[0135] 70 conveyor belt

[0136] E evaluation

[0137] E' quality factor

[0138] x robot trajectory (first process variable)

[0139] y process duration (second process variable).

Claims

1. A method for executing a process, the method comprising the steps of: - executing (S10) a run of the process with a process control; - detecting (S10) a value of a first process variable (x) for this execution; and - detecting (S10) an evaluation (E) of the executed process run; the method comprising the following repeatedly performed evaluation learning step: - changing (S20), by an optimizer (4), at least one parameter of the process control to a changed process control based on a value of a quality criterion for the executed process run; - executing (S20) a run of the process with the changed process control; - detecting (S20) a value of the first process variable for this execution; and - detecting an evaluation (E) of the executed process run; wherein a first quality factor model (51) of the process is machine learned based on the detected evaluation and the value of the first process variable, the first quality factor model determining a quality factor (E’) of the process based on the first process variable (x); and the method comprising the following repeatedly performed process control optimization step: - changing (S40), by the optimizer, the process control to a changed process control based on a value of a quality criterion for the executed process run; - executing (S40) a run of the process with the changed process control; and - detecting (S40) a value of the first process variable for this execution; wherein the value of the quality criterion for at least one executed process run with the changed process control is determined based on the quality factor (E’) determined by the machine learned first quality factor model from the detected value of the first process variable for this process run.

2. The method of claim 1, wherein, The method is performed with the aid of at least one robot (10), the process control is a process control of the robot, and the at least one parameter of the process control is at least one parameter of a regulator (31).

3. The method of claim 1, wherein, The value of the quality criterion for at least one executed process run with the changed process control in the evaluation learning step is determined based on the detected evaluation of the process run.

4. The method of claim 1, wherein, comprising the steps of: - detecting a value of a second process variable (y) for the execution of a process run in the evaluation learning step and / or the process control optimization step; wherein - the value of the quality criterion for at least one executed process run with the changed process control in the evaluation learning step additionally depends on the detected value of the second process variable for this process run, the detected evaluation of this process run being independent of the value of the second process variable; and / or - the value of the quality criterion for at least one executed process run with the changed process control in the process control optimization step additionally depends on the detected value of the second process variable for this process run, the quality factor determined by the first quality factor model being independent of the value of the second process variable.

5. The method of claim 1, wherein, For at least one executed process run with the changed process control in the process control optimization step, an evaluation of this process run is detected, wherein, - the first quality factor model is based on a further machine learning of values of the first process variable and of the detected evaluations; and / or - a notification is issued (S70) when a tolerance parameter depending on a deviation between the detected evaluations and the determined quality factors is outside a predetermined tolerance range; and / or - the detection of the evaluations of the process runs performed in the process control optimization steps with the changed process control depends on the result of the comparison.

6. The method of claim 5, wherein, For at least one process run performed in the process control optimization steps with the changed process control, a comparison is made of the quality factors determined by the first quality factor model learned by machine learning based on the values of the first process variable detected for the process run.

7. The method according to any one of claims 1 to 6, characterized in that, Based on the evaluations and the values of the first process variable detected for the evaluation learning step, at least one further quality factor model of the process is learned by machine learning, which further quality factor model determines a quality factor of the process based on the first process variable, wherein the first quality factor model and the further quality factor model are different, wherein - in at least one process control optimization step, a value of a quality criterion for a process run performed with the changed process control is determined based on the quality factors determined by the further quality factor model learned by machine learning from the values of the first process variable detected for the process run; and / or - a notification is issued (S70) when a tolerance parameter depending on a deviation between the quality factors determined by the first quality factor model and by the further quality factor model based on the values of the first process variable detected for a process run in one of the process control optimization steps is outside a predetermined tolerance range.

8. The method according to any one of claims 1 to 6, characterized in that, A confidence interval of the first quality factor model is determined, and a notification is issued (S70) when the confidence interval exceeds a boundary value.

9. The method according to any one of claims 1 to 6, characterized in that, At least one process run performed in the evaluation learning step with the changed process control is evaluated based on an electronic transmission of a signal.

10. The method of claim 9, wherein, At least one process run performed in the evaluation learning step with the changed process control is evaluated without direct sight of the process.

11. The method according to any one of claims 1 to 6, characterized in that, The first process variable and / or the second process variable have actual data and / or target data specific to the robot.

12. The method of claim 11, wherein, The actual data and / or target data are detected by a robot-side and / or robot-external sensor, and the actual data and / or target data are coordinates of a pose of at least one reference object fixed to the robot and / or a time derivative thereof, at least one force on at least one reference object fixed to the robot and / or at least one drive parameter of at least one robot drive, and / or visual and / or audio and / or temporal data.

13. The method according to any one of claims 1 to 6, characterized in that, The at least one evaluation is detected by manual input and / or is detected automatically, and / or is two-valued or multi-valued.

14. The method of claim 13, wherein, The at least one evaluation is detected sensor-assisted.

15. The method of any one of claims 1 to 6, wherein, The first quality factor model or the further quality factor model has an artificial neural network (51), a random forest model, a decision tree model, a k-nearest neighbor model, a logistic regression model, or a linear model.

16. The method of claim 15, wherein, The linear model is a generalized linear model.

17. The method of any one of claims 1 to 6, wherein, Depending on the detected evaluation of the executed process in the evaluation learning step, a different subsequent process is executed, and / or depending on the determined quality factor of the executed process in the process control optimization step, a different subsequent process is executed.

18. A system for executing a process, the system being designed to execute the method according to any one of claims 1 to 17, and / or having: - means for executing a run of a process by process control; - means for detecting a value of a first process variable (x) for the execution; and - means for detecting an evaluation (E) of the executed process run; - means for repeatedly executing the following evaluation learning step a plurality of times: - changing, by an optimizer (4), at least one parameter of the process control to a changed process control based on the value of the quality criterion of the executed process run; - executing a run of the process with the changed process control; - detecting a value of the first process variable for the execution; and - detecting an evaluation (E) of the executed process run; wherein a first quality factor model (51) of the process is machine learned based on the detected evaluation and the value of the first process variable, the first quality factor model determining a quality factor of the process based on the first process variable; - means for repeatedly executing the following process control optimization step a plurality of times: - changing, by an optimizer, the process control to a changed process control based on the value of the quality criterion of the executed process run; - executing a run of the process with the changed process control; and - detecting a value of the first process variable for the execution; wherein the value of the quality criterion for at least one executed process run with the changed process control is determined based on the quality factor (E’) determined by the machine learned first quality factor model from the detected value of the first process variable for the process run.

19. A computer program product having program code stored on a computer readable medium for executing the method according to any one of claims 1 to 17. ​

Images