Robot monitoring

Abstract

The invention relates to a method for monitoring at least one robot (10; 20), in which method the following steps are carried out at several points in time: providing (S30) ascertained current values of state parameters of the robot; predicting (S40) first expected values for these state parameters on the basis of these ascertained current values and a first numerical model; and comparing (S50) the predicted first expected values and ascertained current values, with a potential anomaly being identified for a state parameter if a deviation between the predicted first expected value and the ascertained current value for this state parameter satisfies an anomaly criterion, and with a present anomaly being identified for a state parameter if a potential anomaly has been identified and an additional verification criterion has been satisfied for this state parameter, and with an error response being triggered (S90) if a present anomaly has been identified. The invention also relates to a system and to a computer program (product).

Description

[0001] Description

[0002] Robot monitoring

[0003] The present invention relates to a method and system for monitoring a robot and a computer program or computer program product for carrying out the method.

[0004] The object of the present invention is to improve the monitoring of robots.

[0005] This object is achieved by a method having the features of claim 1. Claims 14 and 15 protect a system, computer program, or computer program product for implementing a method described herein. The subclaims relate to advantageous developments.

[0006] According to one embodiment of the present invention, for monitoring at least one robot, in a further development for, preferably parallel, monitoring of several robots, the following steps are carried out for the robots, each for several points in time:

[0007] - Providing determined current values ​​of state parameters of the robot;

[0008] - Predicting expected values ​​for these state parameters based on

[0009] - these determined current values ​​and

[0010] - a numerical model, which are referred to herein without loss of generality as first expected values ​​or first numerical model, in one embodiment, calculating the expected first values ​​using the first model, which in one embodiment maps the determined current values ​​to the first expected values; and

[0011] - Comparing the predicted first expected values ​​and the determined current values; wherein a potential anomaly for a state parameter is identified if a deviation between the predicted first expected value and the determined current value for this state parameter fulfills a, in particular predetermined, anomaly criterion; wherein an existing anomaly for a state parameter is identified if both a potential anomaly has been identified for this state parameter and an additional verification criterion is fulfilled; and wherein an error reaction is triggered if an existing anomaly has been identified.

[0012] In one embodiment, the robot comprises a robot arm with at least three, preferably at least six, (movement) axes or joints, preferably rotational axes or joints, and / or a stationary base and a mobile base having a chassis. It can, in particular, be an industrial robot and / or operate in an automated production facility. In one embodiment, the robot is at least temporarily operated during monitoring, in particular moved, preferably for the, preferably cyclic, execution of a robot or work program.

[0013] The state parameters depend in one embodiment on the state of the robot, in another embodiment on the state of the drives, in another embodiment on (drive) motors and / or gears, in another embodiment on the temperatures of the drives, and in another embodiment they specify these. The temperatures of the drives have proven to be easy to determine and, on the other hand, particularly predictive for predicting future performance, in particular impairment or failure, of the robot. The state parameters are therefore defined here, without loss of generality, as T(t) = [Ti(f), --- , T n (f)] Twhere t denotes a current time (point in time) and 7i denotes the state, preferably the temperature, of a drive i of the robot whose drives 1,...n are (to be) taken into account in the monitoring. The determination of current values ​​of state parameters of the robot comprises, in one embodiment, a metrological recording of the values ​​of the state parameters. The provision of determined current values ​​comprises, in one embodiment, the determination of the values ​​and / or the storage of the (determined) values ​​and / or the reception and / or reading of the (determined or stored) values, and can in particular consist of this.

[0014] One aspect of the present invention is based on the idea of ​​predicting expected values ​​for these state parameters based on determined current values ​​and a numerical model, instead of comparing a current state parameter value with a fixed limit value or the like, and comparing them with the determined current values. It has been shown that this allows errors or problems of the robot to be detected particularly well, in particular more reliably and / or earlier. The anomaly criterion can be met, in particular, precisely when the deviation exceeds a specified tolerance value.

[0015] By means of a two-stage method in which a potential anomaly is identified and additionally, preferably thereafter, particularly preferably only if a potential anomaly has been identified, it is checked or determined whether (also) an additional verification criterion is met and, preferably only in this case, a (more likely) existing anomaly is identified or determined and a corresponding error reaction is then triggered, the probability of a false-positive prognosis or unfounded triggering of the error reaction and thus an unnecessary disruption of the robot operation can be advantageously reduced in one embodiment.

[0016] In one embodiment, the method comprises the steps:

[0017] - Determining a second numerical model based on values ​​of the state parameters determined for the robot in a previous update period;

[0018] - predicting a second expected value for one or more of the state parameters based on the determined current values ​​and this second numerical model, in one embodiment calculating the expected second value(s) using this second model, which in one embodiment maps the determined current values ​​to the second expected value(s); and

[0019] - Comparing the predicted second expected value and the determined current value for the (respective) state parameter(s); whereby the additional verification criterion for the respective state parameter is not met if a deviation between the predicted second expected value and the determined current value for this state parameter fulfills a non-anomaly criterion. In particular, the non-anomaly criterion can be met precisely if this deviation does not exceed a specified tolerance value.

[0020] This is based on the following idea: the first model used to identify the potential anomaly may no longer be sufficiently accurate due to changing boundary conditions. By additionally verifying an identified potential anomaly with a second model, the probability of a false-positive prediction or unjustified triggering of the error reaction, and thus unnecessary disruption of robot operation due to changed boundary conditions, can be advantageously reduced.

[0021] In a further development, the second numerical model is determined if a potential anomaly for the corresponding state parameter has been identified or a potential anomaly for a state parameter identified using the first model triggers the determination of the corresponding second model.

[0022] This is based on the following idea: (Re)determining a model based on values ​​of the state parameters can be complex. By determining the second model if a potential anomaly has been identified, one implementation allows the second model to be determined as needed for verification, thus advantageously reducing the required effort.In one embodiment, the second numerical model is determined, in particular with the aid of machine learning, preferably with the aid of numerical optimization, on the basis of values ​​of the state parameters which have been determined for one or more (past) points in time in a previous update period for the robot, wherein the update period in one embodiment is at least one hour and / or at most four weeks, preferably at least one day and / or at most three weeks, and / or at least one hour and / or at most four weeks, preferably at least one day and / or at most three weeks, in the past and / or wherein determined values ​​of the state parameters are used which have previously been provided as determined current values ​​or compared with the corresponding, previously predicted first expected values.By selecting an appropriate update period, especially a sliding one, a more accurate model can be determined or used in one execution, and by using the values ​​already determined previously for comparison, the effort can be reduced.

[0023] In addition or alternatively to the non-anomalies criterion checked using the second model, in one embodiment, the additional verification criterion is not met if a determined current operating state of the robot meets an operating state criterion. The operating state criterion can be met, in particular, precisely when the determined current operating state of the robot belongs to a predefined class with one or more operating states, in particular one or more non-automatic operating states, and / or is not an automatic operating state and / or is not a normal or standard operating state.

[0024] This is based in particular on the following idea: numerical models, in particular the first and / or second model, are advantageously based on a (past or previous) normal or standard operating state, usually an automatic operating state. In other operating states, for example, manual teaching or the like, significantly different state parameter values ​​can occur without this resulting from a robot error. By verifying an identified potential anomaly, preferably in addition to the non-anomaly criterion, (also) based on a current operating state of the robot, the probability of a false-positive prognosis or unjustified triggering of the error reaction and thus an unnecessary disruption of robot operation due to changed boundary conditions can be advantageously (further) reduced.

[0025] In one embodiment, for one or more of the state parameters, in particular on the basis of the respective numerical model, at least one expected value, in particular the first expected value (based on the first numerical model) and / or the second expected value (based on the second numerical model), is predicted on the basis of the determined current values ​​of one or more, preferably all, of the other state parameters, in a further development on the basis of a, in one embodiment linear, regression (θi(f) = ({θi ( / ), ... , T n (f)}\7i(f)), for example 7i(f) = Sj\i(aj- 7j(f))).

[0026] This is based on the following idea: in a fault-free robot, various state parameters, especially states and temperatures of different drives, behave essentially similarly. Therefore, the values ​​of one state parameter can be predicted particularly well, especially quickly and / or reliably, based on the current values ​​of the other state parameters.

[0027] In one embodiment, in a further development, the first numerical model is (has been) determined on the basis of values ​​of the state parameters that were determined in a previous training period, wherein in one embodiment the start of this training period and preferably also the end of this training period lies (in time) before the start of the update period that is used for the determination of the second model.

[0028] This is based in particular on the following ideas: determining a model based on values ​​of the state parameters can be complex. Therefore, it can be expedient to determine the first model based on values ​​of the state parameters that were determined in a previous training period, in a further development, whereby this training period preferably lies before a robot operation to be monitored. However, boundary conditions can change so that such a (first) model determined based on a previous training period is no longer accurate. By then determining the second model for a more recent or (temporally) more recent update period, a more accurate second model can be used for verification in one embodiment, especially if necessary.

[0029] In one embodiment, the first model is, in particular, determined using machine learning, preferably using numerical optimization.

[0030] Additionally or alternatively, in one embodiment, the second model is determined using machine learning. This allows the (respective) model to be determined and / or operate particularly quickly and / or reliably in one embodiment, or to map the determined current values ​​to the first or second expected values, or to predict them.

[0031] In one embodiment, the first numerical model is (or has been) determined individually for the robot to be monitored, preferably using machine learning and / or based on values ​​of the state parameters determined for the robot during a previous training period. This allows the use of a (robot-)specific model that is particularly accurate and / or reliable.

[0032] In another embodiment, the first numerical model is (or has been) determined collectively for a group of robots, to which the robot (to be monitored) preferably also belongs, preferably with the aid of machine learning and / or based on values ​​of the state parameters that were determined in a previous training period for one or more robots in the group and / or robots of the same type. This advantageously allows a broader database to be used, allowing the model to be determined and / or operated particularly quickly and / or robustly.

[0033] In one embodiment, preferably in combination with a determination of the first model for a group of robots and / or a prediction of an expected value for at least one of the state parameters on the basis of the determined current values ​​of at least one other of the state parameters, statistical values, in particular an average value and / or a standard deviation and / or a minimum value and / or a maximum value or the like, are determined on the basis of the determined current values ​​of the state parameters, and the first expected values ​​for the state parameters are also predicted on the basis of these determined statistical values.

[0034] This is based in particular on the following idea: by considering statistical values, individual boundary conditions can be advantageously taken into account. In a simple and illustrative example, due to an increased ambient temperature, the temperatures of the drives of a robot to be monitored can be higher than predicted by a first model that was determined for a group of robots and / or in a previous training period with a lower ambient temperature. By also considering the average value of the current temperatures of the drives of the robot to be monitored and / or their maximum and / or minimum values, or similar, this can be taken into account and the probability of a false-positive prediction or unjustified triggering of the error reaction, and thus an unnecessary disruption to robot operation, can be advantageously reduced.

[0035] In one embodiment, a potential and / or an existing anomaly is identified individually for individual drives of the robot and, accordingly, in a further development, an existing anomaly is identified individually for individual drives of the robot if a potential anomaly has been identified for the respective drive and an additional verification criterion is met, and an error reaction is triggered for the respective drive if an existing anomaly has been identified for it, wherein this additional verification criterion is preferably not met if a deviation between a second expected value forecast for the respective drive and a current value determined for the respective drive fulfills a non-anomaly criterion and / or a determined current operating state of the robot fulfills an operating state criterion. In this way, a more specific error reaction can advantageously be triggered orthe drive in question or affected can be specifically addressed, for example examined, maintained and / or replaced.

[0036] In one embodiment, the error response comprises a notification of a monitoring entity, preferably automatic or human, in particular thus of monitoring personnel, preferably operating and / or maintenance personnel, in one embodiment by means of electronic communication, in a further embodiment by means of email or the like. This notification preferably contains a, preferably graphical, time series of the current and / or forecasted first and / or second expected values ​​determined for the various points in time and / or a specification of the drive(s) for which an existing anomaly has been identified.

[0037] In one embodiment, the current values ​​of the state parameters are stored in a data cloud. This can improve the data transfer and / or the processing of the data, in particular by means of the first and / or second numerical model. Alternatively or particularly preferably, in one embodiment the first and / or second numerical model are additionally stored in the data cloud. By storing it in the data cloud, the model can be determined and / or made available in a particularly advantageous manner, and by storing both the current values ​​and the model(s) processing them, efficiency and / or safety can be advantageously increased. According to one embodiment of the present invention, a system for monitoring at least oneof the robot, in a further development for, preferably in parallel, monitoring of several robots, in particular hardware and / or software, in particular program technology, for carrying out a method described here and / or has:.

[0038] - Means for providing determined current values ​​of state parameters of the robot;

[0039] - means for predicting first expected values ​​for these state parameters on the basis of these determined current values ​​and a first numerical model;

[0040] - means for comparing the forecasted first expected values ​​and determined current values;

[0041] - means for identifying a potential anomaly for a state parameter if a deviation between the predicted first expected value and the determined current value for that state parameter meets an anomaly criterion;

[0042] - means for identifying an existing anomaly for a state parameter if a potential anomaly has been identified for that state parameter and an additional verification criterion is met; and

[0043] - Means for triggering an error reaction if an existing anomaly has been identified.

[0044] In one embodiment, the system or its means comprises:

[0045] - means for determining a second numerical model, in particular using machine learning, on the basis of values ​​of the state parameters determined for the robot in a previous update period, in particular if a potential anomaly for a state parameter has been identified;

[0046] - Means for predicting a second expected value for this or at least one of the state parameters based on the determined current values ​​and this second numerical model; and - Means for comparing this predicted second expected value and the determined current value of this state parameter; wherein the additional verification criterion for this state parameter is not met if a deviation between the predicted second expected value and the determined current value for this state parameter satisfies a non-anomaly criterion.

[0047] In one embodiment, the system or its means comprises:

[0048] - means for determining the first numerical model, in particular individually for the robot or collectively for a group of robots; and / or

[0049] Means for determining statistical values ​​on the basis of the determined current values ​​of the state parameters, wherein the means for predicting first expected values ​​is configured to also predict the first expected values ​​for the state parameters on the basis of these determined statistical values.

[0050] A means within the meaning of the present invention can be designed in hardware and / or software, in particular at least one, in particular digital, processing unit, in particular a microprocessor unit (CPU), graphics card (GPU) or the like, preferably connected to a memory and / or bus system for data or signals, and / or one or more programs or program modules. The processing unit can be designed to execute instructions implemented as a program stored in a memory system, to detect input signals from a data bus, and / or to output output signals to a data bus. A memory system can have one or more, in particular different, storage media, in particular optical, magnetic, solid-state, and / or other non-volatile media. The program can be designed in such a way that it embodies the methods described here oris capable of executing, so that the processing unit can carry out the steps of such methods and thus in particular can monitor the robot. In one embodiment, a computer program product can have, in particular be, a storage medium, in particular a computer-readable and / or non-volatile one, for storing a program or instructions or with a program or instructions stored thereon. In one embodiment, execution of this program or these instructions by a system or a controller, in particular a computer or an arrangement of several computers, causes the system or the controller, in particular the computer(s), to carry out a method described here or one or more of its steps, or the program or the instructions are configured to do so.

[0051] In one embodiment, one or more, in particular all, steps of the method are fully or partially computer-implemented or one or more, in particular all, steps of the method are fully or partially automated, in particular by the system or its means.

[0052] In one embodiment, the system comprises the robot(s) and / or at least one data processing based at least partially on machine learning for determining the first model and / or the second model using machine learning.

[0053] Further advantages and features emerge from the subclaims and the exemplary embodiments. The following shows, partly schematically:

[0054] Fig. 1 : a system for monitoring robots according to an embodiment of the present invention; and

[0055] Fig. 2: a method for monitoring the robots according to an embodiment of the present invention.

[0056] Fig. 1 shows a system for monitoring a first robot 10 and a second robot 20 according to an embodiment of the present invention. For this purpose, their robot controllers 11 and 21, respectively, communicate with a data cloud 30, with which a computer 31 also communicates, which in turn can send emails to a terminal of an operator 4. Fig. 2 shows a method for monitoring the robots 10, 20 according to an embodiment of the present invention. The method described below is executed for both robots 10, 20.

[0057] In a step S10, temperatures of the drives (indicated in black in Fig. 1 ) of the respective robot 10 or 20 are first recorded for a training period for several points in time, for example every 5 or 10 minutes, every hour or the like, and sent to the data cloud 30 and stored there.

[0058] In a step S20, a first model is machine-learned for each robot, which estimates the temperature of the drives of the respective robot from the current temperatures of the other drives of the respective robot using linear regression, for example 7j(f) = Sj\i(aj-7](f)), i = 1 , ... , 6; wherein the coefficients aj are preferably determined using numerical optimization, for example, determined in such a way that the quadratic error between the drive temperatures predicted by linear regression and those actually determined is minimized.

[0059] In a step S30, current temperatures of the drives of the respective robot are recorded for a point in time and sent to the data cloud 30.

[0060] On the basis of these determined current values ​​of the first numerical model, the server 31 predicts first expected values ​​for the temperatures of the drives of the respective robot (Fig. 2: step S40).

[0061] In a step S50, it is checked whether the deviation between the respective predicted first expected value and the determined current value meets an anomaly criterion, for example, exceeds a predetermined tolerance value.

[0062] If this is not the case for any of the drives (S50: "N"), the method returns to step S30. However, if the anomaly criterion is met for at least one drive or the corresponding state parameter or its temperature (S50: "Y"), a potential anomaly is identified for this drive.

[0063] Therefore, for this robot or at least for this drive, based on the temperatures stored in the data cloud 30 for a previous update period, for example, the previous (working) day, the previous (working) week, or the like, a second model is machine-learned, analogous to step S20. This second model estimates the temperature of at least the corresponding drive from the current temperatures of the other drives of the respective robot using linear regression. With the help of this second model, a second expected value for the temperature of the respective drive is predicted based on the determined current values ​​(Fig. 2: steps S60, S65). Like the first model, the second model can predict the expected temperatures for all drives or only for the drive(s) for which a potential anomaly has been identified.

[0064] In a step S70, it is then checked whether the deviation between the respective predicted second expected value and the determined current value meets a non-anomaly criterion, for example, does not exceed a predetermined tolerance value.

[0065] If the non-anomaly criterion is met for the drive(s) for which a potential anomaly has been identified (S70: “Y”), the method returns to step S30.

[0066] If, however, the non-anomalies criterion is not met for at least one of the drives (S70: "N"), a check is carried out in step S80 to determine whether a determined current operating state of the robot meets an operating state criterion, for example, a manual teach mode (mode) or another operating state (mode) deviating from a normal automatic mode (mode). If the operating state criterion is met (S80: "Y"), the method also returns to step S30; otherwise (S80: "N"), an error response is triggered. For example, the server 31 sends an email notification to the terminal 32 of the personnel 4 (Fig. 2: step S90) in which the drive(s) for which an existing anomaly has been identified using the second model is / are identified.

[0067] In the present disclosure, "has an X" generally does not imply an exhaustive list, but is a shortened form of "has at least one X" and also includes "has two or more Xs" and "has Y in addition to X." Although exemplary embodiments have been explained in the foregoing description, it should be noted that numerous modifications are possible. Furthermore, it should be noted that the exemplary embodiments are merely examples and are not intended to limit the scope, applications, or construction in any way.Rather, the foregoing description provides the person skilled in the art with a guide for the implementation of at least one exemplary embodiment, whereby various changes, in particular with regard to the function and arrangement of the described components, can be made without departing from the scope of protection as it results from the claims and combinations of features equivalent to these.

[0068] List of reference symbols

[0069] 10 robots

[0070] 11 robot control 20 robots

[0071] 21 Robot control

[0072] 30 Data Cloud

[0073] 31 computers

[0074] 32 Terminal 4 operating personnel

Claims

Patent claims 1. A method for monitoring at least one robot (10; 20), wherein the following steps are carried out for several points in time: - Providing (S30) determined current values of state parameters of the robot; - predicting (S40) first expected values for these state parameters based on these determined current values and a first numerical model; and - Comparing (S50) the predicted first expected values and determined current values; wherein a potential anomaly is identified for a state parameter if a deviation between the predicted first expected value and the determined current value for this state parameter fulfills an anomaly criterion; wherein an existing anomaly is identified for a state parameter if a potential anomaly has been identified for this state parameter and an additional verification criterion is fulfilled; and wherein an error reaction is triggered (S90) if an existing anomaly has been identified.

2. Method according to claim 1, characterized by the steps: - determining (S60) a second numerical model, in particular using machine learning, on the basis of values of the state parameters determined for the robot in a previous update period; - predicting (S65) a second expected value for at least one of the state parameters based on the determined current values and this second numerical model; and - comparing (S70) this predicted second expected value and the determined current value of this state parameter; wherein the additional verification criterion for this state parameter is not met if there is a deviation between the predicted second expected value and determined current value for this state parameter fulfills a non-anomaly criterion.

3. Method according to the preceding claim, characterized in that the second numerical model is determined if a potential anomaly for this state parameter has been identified.

4. Method according to one of the preceding claims, characterized in that the additional verification criterion is not met if a determined current operating state of the robot meets an operating state criterion.

5. Method according to one of the preceding claims, characterized in that for at least one of the state parameters at least one expected value is predicted on the basis of the determined current values of at least one other of the state parameters.

6. Method according to one of the preceding claims, characterized in that the first numerical model has been determined, in particular with the aid of machine learning, on the basis of values of the state parameters which have been determined in a previous training period.

7. Method according to one of the preceding claims, characterized in that the first numerical model has been determined individually for the robot.

8. Method according to one of the preceding claims 1 to 6, characterized in that the first numerical model has been determined collectively for a group of robots.

9. Method according to one of the preceding claims, characterized in that statistical values are determined on the basis of the determined current values of the state parameters and the first expected Values for the state parameters are also predicted on the basis of these determined statistical values.

10. Method according to one of the preceding claims, characterized in that the state parameters depend on a state, in particular a temperature, of drives of the robot. 11 .Method according to one of the preceding claims, characterized in that an anomaly is identified individually for individual drives of the robot.

12. Method according to one of the preceding claims, characterized in that the error reaction comprises a notification of a monitoring instance.

13. Method according to one of the preceding claims, characterized in that the current values of the state parameters and / or the first numerical model are stored in a data cloud.

14. System for monitoring at least one robot (10; 20) which is designed to carry out a method according to one of the preceding claims and / or comprises: - Means for providing determined current values of state parameters of the robot; - means for predicting first expected values for these state parameters on the basis of these determined current values and a first numerical model; - means for comparing the forecasted first expected values and determined current values; - means for identifying a potential anomaly for a state parameter if a deviation between the predicted first expected value and the determined current value for that state parameter meets an anomaly criterion; - means for identifying an existing anomaly for a state parameter if a potential anomaly has been identified for that state parameter and an additional verification criterion is met; and - means for triggering an error reaction if an existing anomaly has been identified.

15. A computer program or computer program product, wherein the computer program or computer program product contains instructions, in particular stored on a computer-readable and / or non-volatile storage medium, which, when executed by one or more computers or a system according to claim 14, cause the computer(s) or the system to carry out a method according to one of claims 1 to 13.

Images