Reducer maintenance method, maintenance device and storage medium

Abstract

The present invention discloses a method for maintaining a reducer, a maintenance device for a reducer, and a storage medium. The method comprises: obtaining a target current threshold of the target reducer of the robot and a target change relationship of the target reducer, wherein the target current threshold is a critical current value for distinguishing whether the target reducer has failed, and the target change relationship represents the change pattern of the operating current of the target reducer with the use time; determining the target time for maintenance of the target reducer according to the target change relationship and the target current threshold; and outputting maintenance prompt information corresponding to the target reducer according to the target time. The present invention aims to improve the timeliness and accuracy of equipment maintenance of the reducer in the robot to ensure the continuous and stable operation of the production line.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment, and in particular to a maintenance method, a maintenance device and a storage medium for a reducer. Background Art

[0002] Industrial robots are multi-jointed manipulators or multi-degree-of-freedom devices widely used in industry. They possess a degree of autonomy and rely on their own power and control capabilities to perform various industrial processing and manufacturing functions. They are widely used in industries such as vehicle manufacturing, electronics, and logistics. Industrial robots typically have multiple mechanical joints, each equipped with an independent reducer. These reducers control the operation of the robot to perform various actions.

[0003] When the robot is running for a long time, the bearings corresponding to the reducer will wear and produce iron powder. Currently, the iron powder concentration in the space where the bearings are located in the reducer is generally tested regularly, and manual analysis is performed to determine whether the reducer needs to be maintained and replaced. However, this method easily makes the maintenance time of the reducer inconsistent with the actual needs of the reducer, causing sudden production line shutdowns. Summary of the Invention

[0004] The main purpose of the present invention is to provide a reducer maintenance method, a reducer maintenance device and a storage medium, aiming to improve the timeliness and accuracy of equipment maintenance of the reducer in the robot to ensure the continuous and stable operation of the production line.

[0005] To achieve the above object, the present invention provides a method for maintaining a reducer, which is applied to a robot. The method for maintaining a reducer comprises the following steps:

[0006] Obtaining a target current threshold of a target reducer of the robot and a target change relationship of the target reducer, wherein the target current threshold is a current critical value used to distinguish whether the target reducer has failed, and the target change relationship represents a change pattern of the operating current of the target reducer with usage time;

[0007] Determining a target time for maintenance of the target reducer according to the target change relationship and the target current threshold;

[0008] Output maintenance prompt information corresponding to the target reducer according to the target time.

[0009] Optionally, the step of determining the target time for maintenance of the target reducer according to the target change relationship and the target current threshold includes: determining a predicted time for failure of the target reducer according to the target change relationship and the target current threshold;

[0010] The maintenance time and / or replacement time of the target reducer is determined according to the predicted time, and the target time includes the maintenance time and / or the replacement time.

[0011] Optionally, the step of obtaining the target current threshold of the target reducer includes:

[0012] Acquiring a first working condition parameter of the robot;

[0013] Inputting the first operating condition parameter into a prediction model, where the prediction model is a machine learning model for predicting an iron powder concentration threshold in a space corresponding to a bearing where the target reducer is located or the target current threshold when the target reducer fails;

[0014] The target current threshold is determined according to a result output by the prediction model.

[0015] Optionally, the first operating condition parameter includes at least one of the following parameters:

[0016] The oil adding time of the robot;

[0017] Load parameters of the robot;

[0018] The motion characteristic parameters of the robot include motion amplitude, motion speed, motion mileage, work done during motion and / or motion kinetic energy.

[0019] Optionally, before the step of inputting the first operating condition parameter into the prediction model, the method further includes:

[0020] Get multiple sample data;

[0021] Using the plurality of sample data to train a preset machine learning model to obtain the prediction model;

[0022] Among them, the sample data includes multiple features, and the multiple features include the attribute information of the corresponding target robot and the second operating condition parameters of the reducer in the target robot, the current value when the reducer fails, and the iron powder concentration in the space corresponding to the bearing when the reducer fails.

[0023] Optionally, the step of using the plurality of sample data to train a preset machine learning model to obtain the prediction model includes:

[0024] Filter the plurality of sample data according to the correlation coefficient to obtain a training sample set;

[0025] Using the training sample set to train a preset machine learning model to obtain the prediction model;

[0026] The correlation coefficient indicates whether the current value in the corresponding sample data and the corresponding iron powder concentration are positively correlated.

[0027] Optionally, the preset machine learning model includes a preset decision tree model, and the step of using the training sample set to train the preset machine learning model to obtain the prediction model includes:

[0028] Dividing the training sample set using the attribute information as an initial splitting point of the preset decision tree model to obtain at least two sample sets;

[0029] Determining feature importances corresponding to different target features in each of the sample sets, where the target features are features other than the features to be predicted by the prediction model among the multiple features;

[0030] Determining a target splitting point corresponding to each sample set according to the feature importance;

[0031] Divide the corresponding sample set according to the target splitting point to obtain a new sample set, and return to the step of determining the feature importance corresponding to different target features in each of the sample sets until a preset condition for completing the training of the preset decision tree model is met;

[0032] The prediction model splitting point is determined according to the preset decision tree model constructed with the initial splitting point and its corresponding target splitting point.

[0033] Optionally, the step of obtaining the target change relationship of the target reducer includes:

[0034] obtaining first data;

[0035] Training a preset regression model based on the first data to obtain a target regression model;

[0036] Determining the target regression model as the target change relationship;

[0037] The first data includes a first usage time of the target reducer detected before the current moment and a corresponding first operating current.

[0038] Optionally, after the step of training a preset regression model according to the first data to obtain a target regression model, the method further includes:

[0039] Obtaining the current change rate of the target reducer;

[0040] When the current change rate is greater than a preset threshold, correcting the target regression model according to the second data, and determining the corrected target regression model as the target change relationship;

[0041] When the current change rate is less than or equal to the preset threshold, performing the step of determining that the target regression model is the target change relationship;

[0042] The second data includes a plurality of second usage durations of reducers other than the target reducer and their corresponding second operating currents.

[0043] Optionally, the preset regression model includes more than one polynomial linear regression model, and different polynomial linear regression models have different degrees. The step of training the preset regression model according to the first data to obtain the target regression model includes:

[0044] Training the more than one polynomial linear regression models according to the first data to obtain more than one candidate change relationship;

[0045] Determine an evaluation value corresponding to each of the candidate change relationships, wherein the evaluation value represents the difference between a predicted value and an actual value of the corresponding polynomial linear regression model;

[0046] One of the more than one candidate change relationships is determined as the target change relationship according to the evaluation value, and the difference between the predicted value and the actual value corresponding to the target change relationship is the smallest among all the candidate change relationships.

[0047] Optionally, after the step of training the more than one polynomial linear regression models according to the first data to obtain more than one candidate change relationship, the step further includes:

[0048] When the operating current of the reducer corresponding to each of the alternative change relationships has an increasing trend with the use time, the step of determining the evaluation value corresponding to each of the alternative change relationships is performed.

[0049] Optionally, after the step of training a preset regression model according to the first data to obtain a target regression model, the method further includes:

[0050] Obtain the current total usage time and current operating current of the target reducer;

[0051] Substituting the total usage time into a historical regression model to obtain a first current value, and substituting the total usage time into a target regression model to obtain a second current value; the historical regression model is a regression model generated before the current moment that characterizes the variation pattern of the operating current of the target reducer with usage time;

[0052] Determining a first difference between the current operating current and the first current value, and determining a second difference between the current operating current and the second current value;

[0053] When the first difference is less than the second difference, determining that the historical regression model is the target change relationship;

[0054] When the first difference is greater than or equal to the second difference, the step of determining that the target regression model is the target change relationship is performed.

[0055] In addition, in order to achieve the above-mentioned purpose, the present application also proposes a reducer maintenance device, which includes: a memory, a processor, and a reducer maintenance program stored on the memory and runnable on the processor. When the reducer maintenance program is executed by the processor, the steps of the reducer maintenance method described in any one of the above items are implemented.

[0056] In addition, in order to achieve the above-mentioned purpose, the present application also proposes a storage medium, on which a maintenance program for a reducer is stored. When the maintenance program for the reducer is executed by a processor, the steps of the maintenance method for the reducer as described in any of the above items are implemented.

[0057] The present invention proposes a method for maintaining a reducer. The method obtains a target current threshold for distinguishing whether a target reducer of a robot has failed, and a target change relationship that characterizes the change law of the current of the target reducer with the usage time. The target time for maintenance of the target reducer is determined based on the target change relationship and the preset current threshold. The target current threshold is used as a judgment condition for failure of the target reducer. Combined with the relationship between current and time, it can achieve accurate characterization of the remaining life of the reducer in the robot. Therefore, the target time for maintenance of the target reducer is determined in combination with the target current threshold and the target change relationship. The obtained target time can accurately reflect the maintenance requirements corresponding to the current remaining life of the target reducer. Maintenance prompt information is output according to the target time, so that equipment maintenance personnel can know in advance the time required for maintenance of the target reducer based on the prompt of the target time in the maintenance prompt information and make relevant maintenance preparations in advance, and perform equipment maintenance on the target reducer in time at the accurate maintenance time, thereby effectively improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, avoiding sudden shutdown of the production line, and ensuring continuous and stable operation of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a schematic diagram of the hardware structure involved in the operation of an embodiment of a maintenance device for a reducer of the present invention;

[0059] Figure 2 1. A schematic flow chart of an embodiment of a method for maintaining a reducer according to the present invention;

[0060] Figure 3 for Figure 2 Detailed flow chart of step S20;

[0061] Figure 4 Schematic diagram of a prediction model for determining reducer failure by combining a target change relationship and a target current threshold value, involved in an embodiment of a reducer maintenance method of the present invention;

[0062] Figure 5 1. A schematic flow chart of another embodiment of a method for maintaining a reducer according to the present invention;

[0063] Figure 6 Schematic diagram of a process of another embodiment of a method for maintaining a reducer of the present invention;

[0064] Figure 7 The figure is a flow chart of another embodiment of the method for maintaining a reducer of the present invention.

[0065] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0066] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0067] The main solution of an embodiment of the present invention is: obtaining the target current threshold of the target reducer of the robot and the target change relationship of the target reducer, the target current threshold is the current critical value used to distinguish whether the target reducer has failed, and the target change relationship characterizes the change pattern of the operating current of the target reducer with the usage time; determining the target time for maintenance of the target reducer based on the target change relationship and the target current threshold; and outputting the maintenance prompt information corresponding to the target reducer based on the target time.

[0068] In the existing technology, the iron powder concentration in the bearing space of the reducer is generally regularly tested, and manual analysis is performed to determine whether the reducer needs to be maintained and replaced. However, this method easily makes the maintenance time of the reducer inconsistent with the actual needs of the reducer, causing sudden shutdown of the production line.

[0069] The present invention provides the above-mentioned solution, aiming to improve the timeliness and accuracy of equipment maintenance of the reducer in the robot, so as to ensure the continuous and stable operation of the production line.

[0070] This embodiment of the present invention provides a speed reducer maintenance device 1 that can be used to monitor the life of a speed reducer in a robot 2. In this embodiment, robot 2 is used in a vehicle manufacturing line. In other embodiments, robot 2 can also be used in a logistics line or an electronic equipment manufacturing line.

[0071] Reference Figure 1The reducer maintenance device 1 is connected to the robot 2. The reducer maintenance device 1 can be built into the robot 2 or independently provided outside the robot 2 and communicatively connected to the robot 2. The reducer maintenance device 1 can be used to obtain the current operating parameters of the robot 2 to which it is connected.

[0072] In the embodiment of the present invention, referring to Figure 1 The reducer maintenance device 1 includes a processor 1001 (e.g., a CPU), a memory 1002, a timer 1003, and the like. The various components of the control device are connected via a communication bus. The memory 1002 can be a high-speed RAM memory or a non-volatile memory such as a disk drive. Alternatively, the memory 1002 can be a storage device independent of the processor 1001.

[0073] Those skilled in the art will understand that Figure 1 The device structure shown in the figure does not constitute a limitation of the device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0074] like Figure 1 As shown, the memory 1002 as a storage medium may include a maintenance program for the reducer. Figure 1 In the device shown, the processor 1001 can be used to call the reducer maintenance program stored in the memory 1002 and execute the relevant steps of the reducer maintenance method in the following embodiment.

[0075] An embodiment of the present invention further provides a method for maintaining a reducer, which is applied to the above-mentioned maintenance device for the reducer.

[0076] Reference Figure 2 , an embodiment of a maintenance method for a reducer of the present application is proposed. In this embodiment, the maintenance method for the reducer includes:

[0077] Step S10, obtaining a target current threshold of a target reducer of the robot and a target change relationship of the target reducer, wherein the target current threshold is a current critical value for distinguishing whether the target reducer has failed, and the target change relationship represents a change pattern of the operating current of the target reducer with use time;

[0078] Each reducer installed on the robot can be used as the target reducer. Alternatively, the reducer pre-set on the robot for life monitoring can be used as the target reducer.

[0079] The target current threshold may be a fixed value predetermined based on experience, or a value obtained by analyzing data detected during actual operation of the target reducer before the current moment and / or data detected during operation of reducers other than the target reducer.

[0080] Specifically, the target reducer can be pre-set with the required accuracy range for the corresponding robot motion. If the reducer fails, the corresponding robot motion accuracy is outside this accuracy range, indicating that the reducer cannot meet the required accuracy range for robot operation. If the reducer is not failed, the corresponding robot motion accuracy is within this accuracy range.

[0081] The target change relationship may include a functional relationship between the operating current and the usage time, a curve of the operating current changing with the usage time, a mapping relationship between the operating current and the usage time, etc. The usage time specifically starts from the time when the target reducer is first powered on.

[0082] Different types of target reducers may correspond to different target current thresholds and / or target change relationships. For the same type of target reducer, different target current thresholds and / or target change relationships may correspond to different operating conditions of the target reducer.

[0083] Specifically, step S10 is executed in real time or at preset intervals during the use of the target reducer on the robot, or step S10 can be executed before the target reducer is installed on the robot.

[0084] Step S20, determining a target time for maintenance of the target reducer according to the target change relationship and the target current threshold;

[0085] The target time may specifically include a target speed reducer replacement time, a target speed reducer maintenance time, and / or a target speed reducer inspection time.

[0086] Specifically, the target time here can be determined based on the predicted time of failure of the target reducer determined by the target change relationship and the target current threshold. Alternatively, a reference current threshold corresponding to the target time can be determined based on the target current threshold, and the target time here can be determined based on the target change relationship and the reference current threshold. Alternatively, the remaining life of the target reducer can be determined based on the target change relationship and the target current threshold, and the target time here can be determined based on the remaining life, and so on.

[0087] Step S30: Outputting maintenance prompt information corresponding to the target reducer according to the target time.

[0088] Specifically, the maintenance reminder information may be output at the target time or before the target time, or the maintenance reminder information including the target time may be output at the current moment.

[0089] A maintenance method for a reducer is proposed in an embodiment of the present invention. The method obtains a target current threshold for distinguishing whether a target reducer of a robot has failed, and a target change relationship that characterizes the change law of the current of the target reducer with the usage time. The target time for maintenance of the target reducer is determined based on the target change relationship and the preset current threshold. The target current threshold is used as a judgment condition for the failure of the target reducer. Combined with the relationship between current and time, it can achieve accurate characterization of the remaining life of the reducer in the robot. Therefore, the target time for maintenance of the target reducer is determined in combination with the target current threshold and the target change relationship. The obtained target time can accurately reflect the maintenance requirements corresponding to the current remaining life of the target reducer. Maintenance prompt information is output according to the target time, so that equipment maintenance personnel can know in advance the time required for maintenance of the target reducer based on the prompt of the target time in the maintenance prompt information and make relevant maintenance preparations in advance, and perform equipment maintenance on the target reducer in time at the accurate maintenance time, thereby effectively improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, avoiding sudden shutdown of the production line, and ensuring continuous and stable operation of the production line.

[0090] Furthermore, in the above embodiment, referring to Figure 3 The step of determining the target time for maintenance of the target reducer according to the target change relationship and the target current threshold comprises:

[0091] Step S21, determining the predicted time of failure of the target reducer according to the target change relationship and the target current threshold;

[0092] When the target change relationship is a change curve, such as Figure 4 As shown, the change curve can be drawn in the preset coordinate system with the horizontal axis representing the usage time and the vertical axis representing the operating current. The current critical line is drawn in the preset coordinate system according to the target current threshold, and the horizontal coordinate corresponding to the intersection of the change curve and the current critical line is used as the predicted time.

[0093] When the target change relationship is a functional relationship, the predicted time of target reducer failure can be calculated by substituting the target current threshold into the functional relationship.

[0094] Step S22: determining the maintenance time and / or replacement time of the target reducer according to the predicted time, wherein the target time includes the maintenance time and / or the replacement time.

[0095] Specifically, the remaining life of the target reducer can be determined based on the time difference between the predicted time and the current time, and the maintenance time and / or replacement time here can be determined based on the remaining life. The remaining life is specifically the estimated time period that the target reducer is allowed to be used from the current time to the time when the target reducer fails. In this embodiment, the calculation result obtained by subtracting the current time from the predicted time is used as the remaining life. In other embodiments, the calculation result obtained by subtracting the current time from the predicted time can also be determined first, and the value obtained after correcting the calculation result by a preset correction value is used as the remaining life. Specifically, the maintenance time may include a maintenance cycle, and the maintenance cycle is specifically the target time period required for the interval between two adjacent maintenance operations corresponding to the target reducer. The remaining life can be positively correlated with the maintenance cycle, that is, the longer the remaining life, the longer the maintenance cycle can be, and the shorter the remaining life, the shorter the maintenance cycle can be. Based on this, it is beneficial to extend the actual life of the target reducer. In addition, the moment when the remaining life reaches the preset minimum life can also be used as the replacement time.

[0096] In addition, the predicted time can be directly used as the replacement time, and the time obtained by subtracting the preset time from the predicted time can be used as the maintenance time.

[0097] In this embodiment, the failure time of the target reducer is first predicted by the target current threshold and the target change relationship, and then the maintenance time and / or replacement time of the target reducer is determined according to the predicted failure time, thereby ensuring that the equipment maintenance personnel can accurately know the maintenance time and / or replacement time of the target reducer based on the output maintenance prompt information, thereby effectively improving the timeliness and accuracy of the maintenance and replacement of the reducer in the robot, further avoiding sudden shutdown of the production line, and ensuring the continuous and stable operation of the production line.

[0098] In other embodiments, the predicted maintenance time of the target reducer can also be determined based on the target current threshold. Specifically, the result obtained by subtracting the preset amplitude from the target current threshold is used as the reference current threshold. The reference current threshold is a current critical value used to distinguish whether the target reducer needs maintenance. When the target change relationship is a change curve, the change curve can be drawn in a preset coordinate system with the horizontal axis representing the usage time and the vertical axis representing the operating current. The current critical line is drawn in the preset coordinate system based on the reference current threshold, and the horizontal coordinate corresponding to the intersection of the change curve and the current critical line is used as the predicted maintenance time. When the target change relationship is a functional relationship, the predicted maintenance time of the target reducer can be calculated by substituting the reference current threshold into the functional relationship. Based on this, the timeliness and accuracy of the maintenance of the reducer in the robot can also be effectively improved, further avoiding sudden shutdowns of the production line and ensuring the continuous and stable operation of the production line.

[0099] Furthermore, based on the above embodiment, another embodiment of the maintenance method of the reducer of the present application is proposed. In this embodiment, referring to Figure 5, the step of obtaining the target current threshold of the target reducer includes:

[0100] Step S11, obtaining a first working condition parameter of the robot;

[0101] The first operating condition parameter is specifically a characteristic parameter of the robot's operating requirements for the reducer after the current or target reducer is powered on for the first time.

[0102] In this embodiment, the first operating condition parameter includes at least one of the following parameters:

[0103] The oil adding time of the robot;

[0104] Load parameters of the robot;

[0105] The motion characteristic parameters of the robot include motion amplitude, motion speed, motion mileage, work done during motion and / or motion kinetic energy.

[0106] Among them, the oil adding time is specifically the time when the target reducer of the robot is added with oil during the time period between the first power-on of the target reducer and the current moment, and specifically can be the time of the most recent oil adding from the current moment. The load parameter is specifically the load-bearing capacity of the target reducer at its location on the robot. The motion amplitude is specifically the amplitude of rotation or the amplitude of translation required for the target reducer at its location on the robot. The motion speed is specifically the rotation speed (specifically the maximum speed or average speed) or the speed of translation required (such as the maximum speed or average speed) required for the target reducer at its location on the robot. The motion mileage is specifically the angular displacement or translation displacement required for the target reducer at its location on the robot. The work done during the motion process and the motion kinetic energy are specifically obtained by accumulating the moment of inertia of the robot during the motion process within the target duration.

[0107] In other embodiments, in addition to the above parameters, the first operating condition parameter may also include other operating condition parameters, such as the number of commutations of the target reducer at the position on the robot, the length of time the target reducer stays in the target area at the position on the robot, etc.

[0108] Step S12: inputting the first operating condition parameter into a prediction model, wherein the prediction model is a machine learning model for predicting an iron powder concentration threshold value of a bearing space corresponding to failure of the target reducer or the target current threshold value;

[0109] In this embodiment, the prediction model is a GBDT model. In other embodiments, the preset model may also be other types of machine learning models, such as one or more of an XGBoost model, a LightGBM model, a KNN model, etc.

[0110] In addition to the first operating condition parameters, the attribute information of the robot where the target reducer is located can also be obtained, and the first operating condition parameters and the attribute information are input into the prediction model.

[0111] Step S13: determining the target current threshold according to the result output by the prediction model.

[0112] In one implementation, when the prediction model is used to predict the iron powder concentration threshold, a correspondence between the iron powder concentration threshold and the target current threshold may be pre-established. The correspondence may be a calculation relationship, a mapping relationship, etc. Based on this correspondence, the target current threshold corresponding to the result output by the prediction model may be determined.

[0113] In another implementation, the prediction model is used to predict the target current threshold, and the result output by the prediction model can be directly used as the target current threshold.

[0114] In this embodiment, the target current threshold when the target reducer fails is predicted in combination with the actual working condition of the robot, which is conducive to ensuring that the obtained target current threshold can match the actual working condition of the robot, so as to improve the accuracy of the obtained target current threshold, thereby further improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, avoiding sudden shutdown of the production line, and ensuring the continuous and stable operation of the production line.

[0115] Furthermore, based on any of the above embodiments, another embodiment of the maintenance method of the reducer of the present application is proposed. In this embodiment, referring to Figure 6 , before step S12, further comprising:

[0116] Step S01, obtaining a plurality of sample data, wherein the sample data includes a plurality of features, wherein the plurality of features include attribute information of a corresponding target robot, a second operating condition parameter of a reducer in the target robot, a current value when the reducer fails, and an iron powder concentration in a space corresponding to a bearing when the reducer fails;

[0117] The target robot may include multiple robots other than the robot where the target reducer is located, and may further include the robot where the target reducer is located. The sample data may be detected and obtained during the operation of the corresponding target robot.

[0118] The amount of sample data can be as large as possible and can be increased as the location of the target reducer and the service life of the robot increase.

[0119] The attribute information specifically includes information such as model, online time, service life, number of failures, number of maintenance times and / or load.

[0120] The parameters included in the second operating condition parameters may be the same as the above-mentioned first operating condition parameters, or may further include more other types of operating condition parameters in addition to the parameters of the same type as the above-mentioned first operating condition parameters. Specifically, in the present embodiment, the second operating condition parameters include the robot's motion amplitude, motion mileage, motion speed, number of commutations, and other parameters (such as added value peak). Specifically, the motion amplitude may include the robot's rotation amplitude on multiple axes. The motion mileage may include the robot's rotation mileage on multiple axes. The motion speed may include the robot's maximum speed on multiple axes. The second operating condition parameters are specifically detected at a set time interval after the reducer of the target robot is first powered on and before it fails.

[0121] The current value here specifically refers to the peak operating current value reached when the reducer fails. The specific current value may include the peak current value of the robot on multiple axes.

[0122] It should be noted that different axes in the multi-axis here correspond to different degrees of freedom of the robot.

[0123] Step S02: sampling the plurality of sample data to train a preset machine learning model to obtain the prediction model.

[0124] The preset machine learning model is a machine learning model with unknown model parameters.

[0125] Specifically, a training sample set and a test sample set can be extracted from multiple sample data. The training sample set is then learned according to preset rules to determine the model parameters of the preset machine learning model and obtain a prediction model. Different preset rules can be used for different robots and different target reducers.

[0126] In this embodiment, the prediction machine learning model is a decision tree model. In other embodiments, the prediction machine learning model may be a neural network model. In this embodiment, the prediction machine learning model is a GBDT model.

[0127] The GBDT model is specifically composed of an iterative decision tree algorithm and consists of multiple decision trees. The conclusions of all trees are accumulated as the prediction value, and one of the trees can be understood as a weak learner. Specifically, the mean iron powder concentration of all sample data in the training sample set can be determined. The weak learner in the preset GBDT model is initialized based on the mean iron powder concentration. The training sample set is learned by the initialized weak learner. The best fitting value of each decision node in each weak learner in the GBDT model is determined based on the negative gradient fitting method of the decision tree. The strong learner formed by combining all decision trees is used as the prediction model, and the loss function corresponding to the prediction model is minimized.

[0128] In this embodiment, the prediction model corresponding to the target current threshold is obtained by training in the above manner, which is conducive to flexible processing of various types of data involved in the operation of the robot reducer, so as to comprehensively consider the influence of various factors such as working conditions and robot properties on the relationship between the current value and iron powder concentration when the target reducer fails, thereby further improving the accuracy of the target current threshold determined based on the prediction model, so as to further improve the timeliness and accuracy of equipment maintenance of the reducer in the robot, avoid sudden shutdown of the production line, and ensure the continuous and stable operation of the production line.

[0129] In other embodiments, the sample data may also include the second operating parameters of the reducer in the corresponding target robot, the current value when the reducer fails, and the iron powder concentration in the space corresponding to the bearing when the reducer fails, but does not include the above-mentioned attribute information.

[0130] Furthermore, in this embodiment, step S02 includes: filtering the multiple sample data according to the correlation coefficient to obtain a training sample set; using the training sample set to train a preset machine learning model to obtain the prediction model; wherein the correlation coefficient represents whether the current value in the corresponding sample data and its corresponding iron powder concentration are positively correlated.

[0131] In this embodiment, the correlation coefficient is specifically the Pearson correlation coefficient. In other embodiments, the correlation coefficient may also be other types of coefficients, such as the Spearman correlation coefficient.

[0132] Specifically, in this embodiment, the sample data that satisfies the positive correlation between current value and specific powder concentration among all sample data can be determined based on the correlation coefficient as the training sample set. In other embodiments, the sample data that satisfies the positive correlation between current value and specific powder concentration among all sample data can be determined based on the correlation coefficient as positive sample data, and the sample data that does not satisfy the positive correlation between current value and specific powder concentration among all sample data can be determined based on the correlation coefficient as negative sample data. The set of sample data with positive sample data labels and negative sample data labels is used as the training sample set. Features are extracted from the training sample set based on the sample data labels to obtain target features.

[0133] In this embodiment, the above method is conducive to ensuring that the output results of the prediction model conform to the positive correlation between the iron powder concentration and the operating current, and ensuring the accuracy of the critical state characterization of whether the target reducer fails based on the output results of the prediction model. It is conducive to further improving the accuracy of the determined maintenance time, thereby further improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, and ensuring the continuous and stable operation of the production line.

[0134] Furthermore, in this embodiment, the preset machine learning model includes a preset decision tree model. In this embodiment, the preset decision tree model is a preset regression tree model (such as a GBDT model). In other embodiments, the preset decision tree model may also be a preset classification tree model. The step of extracting the features in the training sample set as the target features includes: dividing the training sample set with the attribute information as the initial splitting point of the preset decision tree model to obtain at least two sample sets; determining the feature importance corresponding to different target features in each of the sample sets, wherein the target features are features other than the features to be predicted by the prediction model in the multiple features; determining the target splitting point corresponding to each of the sample sets according to the feature importance; dividing the corresponding sample sets according to the target splitting point to obtain a new sample set, and returning to execute the step of determining the feature importance corresponding to different target features in each of the sample sets until the preset condition for the end of the training of the preset decision tree model is reached; and determining the prediction model splitting point according to the preset decision tree model constructed with the initial splitting point and its corresponding target splitting point.

[0135] Specifically, in this embodiment, the attribute information includes the working years of the robot. Specifically, whether the working years of the robot are less than a preset number of years can be used as the initial splitting point of the decision tree.

[0136] When the prediction model is used to predict the iron powder concentration threshold, the feature to be predicted is the iron powder concentration among multiple features; when the prediction model is used to predict the target current threshold, the feature to be predicted is the current value among multiple features.

[0137] Each feature importance represents the importance of the corresponding target feature relative to the feature being predicted by the prediction model. Feature importance is used to evaluate the difference between the predicted and actual values ​​in a sample set obtained by partitioning the corresponding sample set using different thresholds for each target feature. The greater the feature importance, the smaller the difference. The smallest difference serves as the most reliable basis for branching in the decision tree. A feature importance is calculated for each target feature. Feature importance can be specifically characterized by the mean squared error growth or Gini index growth corresponding to each target feature.

[0138] The target feature with the highest feature importance among all target features is used as the target splitting point for the corresponding sample set. After determining the target splitting point, the corresponding sample set is further divided according to the target splitting point to obtain a new sample set. The new sample set can then re-determine a new splitting point based on the corresponding feature importance, and the corresponding sample set is divided based on the new splitting point until a preset condition is met. The preset condition can be that the number of splits of the training sample set reaches a preset number. Alternatively, the preset condition can be that the loss function corresponding to the decision tree model constructed using the initial splitting point and all its corresponding target splitting points is minimized.

[0139] Specifically, the preset decision tree model constructed by the initial splitting point and its corresponding target splitting point can be directly used as the prediction model. After obtaining the preset decision tree model constructed by the initial splitting point and its corresponding target splitting point, the residual between the predicted value and the corresponding actual value of the prediction decision tree model can be determined, and new sample data can be obtained based on the residual and another decision tree model with a determined splitting point can be determined by analogy with the training method of the above-mentioned preset decision tree model. This cycle is repeated until the loss function of the final model formed by the combination of all the decision tree models with determined splitting points is minimized, and the final model with the minimum loss function is used as the prediction model. Different attribute information (such as working years and robot type, etc.) can also be used as the initial splitting points of different preset decision tree models. After all the splitting points of each preset decision tree model are determined in the above-mentioned manner based on the feature importance, the preset decision tree models of all the determined splitting points are combined to form a prediction model.

[0140] In this embodiment, the importance of different features is analyzed in combination with attribute information such as the working years of the robot, and the determined feature importance is used as the splitting point in the decision tree model, so that the prediction model constructed by the decision tree model is more in line with the actual operation of the robot where the target reducer is located, so as to effectively improve the accuracy of the target current threshold determined based on the prediction model, thereby further improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, and ensuring the continuous and stable operation of the production line.

[0141] Furthermore, based on any of the above embodiments, another embodiment of the maintenance method of the reducer of the present application is proposed. In this embodiment, referring to Figure 7 The step of obtaining the target change relationship of the target reducer includes:

[0142] Step S14, obtaining first data; wherein, the first data includes a first usage duration of the target reducer detected before the current moment and a corresponding first operating current.

[0143] There may be multiple first usage durations, each corresponding to a first operating current. Specifically, during the period from the first power-on of the target reducer to the current moment, the first usage duration and its corresponding first operating current are detected each time the target reducer is powered on for a set duration or each time it is powered off. The number of first usage durations and their corresponding first operating currents in the first data is positively correlated with the usage duration of the target reducer.

[0144] Step S15: training a preset regression model based on the first data to obtain a target regression model;

[0145] In this embodiment, the preset regression model is a linear regression model. In other embodiments, the preset regression model may also be a nonlinear regression model.

[0146] The preset regression model may include one or more models. When the preset regression model includes more than one model, different models may be trained based on the first data to obtain more than one candidate model, and one of the more than one candidate models may be selected as the target regression model.

[0147] Furthermore, in this embodiment, after obtaining the first data, preprocessing is performed, and the preprocessed data is used to train a preset regression model to obtain a target regression model. The preprocessing may specifically include processing the operating current in the data for outliers, normalization, daily maximum value calculation, and / or modeling data acquisition.

[0148] Step S16, determining that the target regression model is the target change relationship;

[0149] In this embodiment, a target regression model is obtained by performing regression analysis on the current collected during the actual use of the target reducer and the usage time, and serves as the relationship between the operating current of the target reducer and the usage time. Based on this, the regression analysis is helpful in predicting the trend of rapid changes in the current of the target reducer in the future. Based on the target regression model, the accuracy of the target time determined subsequently can be further improved, thereby further improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, and ensuring the continuous and stable operation of the production line.

[0150] Furthermore, in this embodiment, after step S15, it also includes: obtaining the current change rate of the target reducer; when the current change rate is greater than a preset threshold, correcting the target regression model according to the second data, and determining the corrected target regression model as the target change relationship; when the current change rate is less than or equal to the preset threshold, executing the step of determining the target regression model as the target change relationship; the second data includes multiple second usage times and their corresponding second operating currents of other reducers other than the target reducer within the set time before failure.

[0151] Specifically, there are multiple other reducers, and each reducer has at least one second usage duration and its corresponding second operating current. Specifically, the multiple second usage durations and their corresponding second operating currents can be obtained by obtaining operating data recorded by multiple other reducers before the current moment.

[0152] The current change rate specifically represents the rate of change of the target reducer's operating current. Specifically, the third current value of the target reducer in its current power-on state, the fourth current value of the target reducer in its previous power-on state, and the duration between the power-on times in the current power-on state and the previous power-on time can be obtained. The current difference between the third and fourth current values ​​can be determined, and the ratio of the current difference to the duration can be determined as the current change rate.

[0153] The preset threshold is specifically a critical value used to distinguish whether the target reducer currently has a failure risk.

[0154] When the current change rate is greater than the preset threshold, the target regression model used to predict the target current threshold of the target reducer is corrected based on the data of current change with usage time collected when other reducers are about to fail. In this way, by shortening the interval period of input model data and taking the experience of failure of other reducers as a reference, it is ensured that the obtained target regression model can more accurately predict the failure time of the target reducer, thereby improving the accuracy and timeliness of the maintenance of the target reducer.

[0155] In other embodiments, the current change rate may not be determined, and the first data and the second data may be used to directly train a preset regression model to obtain a target regression model.

[0156] Furthermore, in this embodiment, the preset regression model includes more than one polynomial linear regression model, and different polynomial linear regression models have different degrees. Step S15 includes:

[0157] Step S151: training the more than one polynomial linear regression model according to the first data to obtain more than one candidate change relationship;

[0158] Specifically, the polynomial linear regression model is a polynomial linear regression model with usage time as the independent variable, operating current as the dependent variable, and unknown coefficients corresponding to the independent variables. The order of the independent variables in different terms in the polynomial linear regression model is different.

[0159] The first data is used to train more than one polynomial linear regression model, target values ​​of coefficients in the polynomial linear regression model are determined, and the polynomial linear regression model with the target values ​​of the coefficients is used as an alternative change relationship.

[0160] Step S152, determining an evaluation value corresponding to each candidate change relationship, wherein the evaluation value represents the difference between the predicted value and the actual value of the corresponding polynomial linear regression model;

[0161] Specifically, multiple data can be extracted from the first data as test samples, and the usage time of each data in the test sample can be substituted into the alternative change relationship to obtain the predicted current corresponding to each data, and the evaluation value corresponding to each alternative change relationship is calculated based on all the predicted currents corresponding to each alternative change relationship and the operating current in the corresponding data.

[0162] Evaluation values ​​may include mean absolute error, mean square error, root mean square error, coefficient of determination, or mean absolute percentage error.

[0163] Step S153 : determining, based on the evaluation value, one of the more than one candidate change relationships as the target change relationship, wherein the difference between the predicted value and the actual value corresponding to the target change relationship is the smallest among all the candidate change relationships.

[0164] Specifically, the alternative change relationship with the smallest mean square error can be used as the target change relationship; or, the alternative change relationship with the determination coefficient closest to 1 can be used as the target change relationship, and so on.

[0165] In this embodiment, the first data is used to train more than one polynomial linear regression model of different degrees to obtain more than one alternative change relationship, and a relationship with the smallest difference between the predicted value and the target value is selected from the more than one alternative change relationship as the target change relationship. This can effectively improve the accuracy of the law of change of the operating current of the target reducer represented by the target change relationship with the usage time, thereby effectively improving the accuracy of the subsequently determined target time and the output maintenance prompt information, thereby further improving the timeliness and accuracy of equipment maintenance of the reducer in the robot, and ensuring the continuous and stable operation of the production line.

[0166] Furthermore, in this embodiment, after the step of training the more than one polynomial linear regression models according to the first data to obtain more than one alternative change relationship, it also includes: when the operating current of the reducer corresponding to each of the alternative change relationships has an increasing trend with the use time, executing the step of determining the evaluation value corresponding to each alternative change relationship.

[0167] Among them, when there is an alternative change relationship whose operating current of the reducer corresponding to the use time does not have an increasing trend, the first data can be re-preprocessed, and the time range covered by the extracted data can be narrowed during the preprocessing process. More than one polynomial linear regression model can be re-trained based on the re-preprocessed data until more than one alternative change relationship that meets the target condition is obtained, and the target condition is that the operating current of the reducer corresponding to each relationship in the more than one alternative change relationship has an increasing trend with the use time.

[0168] In this embodiment, after obtaining more than one alternative change relationship, the target change relationship will be further selected only after ensuring that the operating current of the reducer corresponding to each alternative change relationship has an increasing trend with the use time, thereby effectively avoiding the overfitting of the polynomial linear regression model, so as to truly and effectively reflect the relationship between the actual operating current of the target reducer and the use time, so as to further improve the accuracy of the subsequently determined target time and the output maintenance prompt information, thereby further improving the timeliness and accuracy of the equipment maintenance of the reducer in the robot, and ensuring the continuous and stable operation of the production line.

[0169] Furthermore, in this embodiment, step S15 also includes: obtaining the current total usage time and current operating current of the target reducer; substituting the total usage time into the historical regression model to obtain a first current value, and substituting the total usage time into the target regression model to obtain a second current value; the historical regression model is a regression model generated before the current moment that characterizes the change law of the operating current of the target reducer with the usage time; determining a first difference between the current operating current and the first current value, and determining a second difference between the current operating current and the second current value; when the first difference is less than the second difference, determining the historical regression model as the target change relationship; when the first difference is greater than or equal to the second difference, executing the step of determining the target regression model as the target change relationship.

[0170] After the target regression model is corrected based on the second data, the target total usage time may be substituted into the corrected target regression model to obtain the second current value.

[0171] In this embodiment, the current operating data of the target reducer is used to compare the established target regression model and the historical regression model, and the regression model with the smallest deviation between the predicted value and the actual value is used to predict the target time, which is conducive to the output of maintenance prompt information being more in line with the actual maintenance needs of the target reducer, so as to effectively ensure the timeliness and accuracy of equipment maintenance of the reducer in the robot and ensure the continuous and stable operation of the production line.

[0172] In other embodiments, multiple data may be extracted from the first data as test samples, and the usage time of each data in the test samples may be substituted into the historical regression model and the target regression model to obtain the corresponding predicted current. The second evaluation values ​​corresponding to the historical regression model and the target regression model are calculated based on all the predicted currents corresponding to the historical regression model and the target regression model and the operating currents in the corresponding data. The second evaluation value may include mean absolute error, mean square error, root mean square error, determination coefficient or mean absolute percentage error, etc. Specifically, the historical regression model and the regression model with the smallest mean square error in the target regression model may be used as the target change relationship; or, the historical regression model and the regression model with the determination coefficient closest to 1 in the target regression model may be used as the target change relationship, and so on.

[0173] In addition, an embodiment of the present invention further proposes a storage medium on which a maintenance program for a reducer is stored. When the maintenance program for the reducer is executed by a processor, the relevant steps of any embodiment of the above reducer maintenance method are implemented.

[0174] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0175] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0176] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, reducer maintenance device, or network equipment, etc.) to execute the methods described in each embodiment of the present invention.

[0177] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for maintaining a reducer, applied to a robot, characterized in that: The maintenance method of the reducer comprises the following steps: Obtaining a target current threshold of a target reducer of the robot and a target change relationship of the target reducer, wherein the target current threshold is a current critical value used to distinguish whether the target reducer has failed, and the target change relationship represents a change pattern of the operating current of the target reducer with usage time; Determining a target time for maintenance of the target reducer according to the target change relationship and the target current threshold; Outputting maintenance prompt information corresponding to the target reducer according to the target time; Acquire a first operating condition parameter of the robot, where the first operating condition parameter is a characteristic parameter of the robot's operating requirements for the target reducer after the target reducer is powered on for the first time; Inputting the first operating condition parameter into a prediction model, where the prediction model is a machine learning model for predicting an iron powder concentration threshold value in a space where a bearing is located when the target reducer fails; Based on the corresponding relationship between the iron powder concentration threshold and the target current threshold, the target current threshold corresponding to the result output by the prediction model is determined.

2. The maintenance method for a reducer according to claim 1, wherein: The step of determining the target time for maintenance of the target reducer according to the target change relationship and the target current threshold comprises: Determining a predicted time of failure of the target reducer according to the target change relationship and the target current threshold; The maintenance time and / or replacement time of the target reducer is determined according to the predicted time, and the target time includes the maintenance time and / or the replacement time.

3. The maintenance method of the reducer according to claim 1, characterized in that: The first operating condition parameter includes at least one of the following parameters: The oil adding time of the robot; Load parameters of the robot; The motion characteristic parameters of the robot include motion amplitude, motion speed, motion mileage, work done during motion and / or motion kinetic energy.

4. The maintenance method for a reducer according to claim 1, wherein: Before the step of inputting the first operating condition parameter into the prediction model, the method further includes: Get multiple sample data; Using the plurality of sample data to train a preset machine learning model to obtain the prediction model; Among them, the sample data includes multiple features, and the multiple features include the attribute information of the corresponding target robot and the second operating condition parameters of the reducer in the target robot, the current value when the reducer fails, and the iron powder concentration in the space corresponding to the bearing when the reducer fails.

5. The method for maintaining a reducer according to claim 4, wherein: The step of using the plurality of sample data to train a preset machine learning model to obtain the prediction model comprises: Filter the plurality of sample data according to the correlation coefficient to obtain a training sample set; Using the training sample set to train a preset machine learning model to obtain the prediction model; The correlation coefficient indicates whether the current value in the corresponding sample data and the corresponding iron powder concentration are positively correlated.

6. The method for maintaining a reducer according to claim 5, wherein: The preset machine learning model includes a preset decision tree model, and the step of using the training sample set to train the preset machine learning model to obtain the prediction model includes: Dividing the training sample set using the attribute information as an initial splitting point of the preset decision tree model to obtain at least two sample sets; Determining feature importances corresponding to different target features in each of the sample sets, where the target features are features other than the features to be predicted by the prediction model among the multiple features; Determining a target splitting point corresponding to each sample set according to the feature importance; Divide the corresponding sample set according to the target splitting point to obtain a new sample set, and return to the step of determining the feature importance corresponding to different target features in each of the sample sets until a preset condition for completing the training of the preset decision tree model is met; The prediction model splitting point is determined according to the preset decision tree model constructed with the initial splitting point and its corresponding target splitting point.

7. The method for maintaining a reducer according to any one of claims 1 to 6, characterized in that: The step of obtaining the target change relationship of the target reducer includes: obtaining first data; Training a preset regression model based on the first data to obtain a target regression model; Determining the target regression model as the target change relationship; The first data includes a first usage time of the target reducer detected before the current moment and a corresponding first operating current.

8. The method for maintaining a reducer according to claim 7, wherein: After the step of training a preset regression model according to the first data to obtain a target regression model, the method further includes: Obtaining the current change rate of the target reducer; When the current change rate is greater than a preset threshold, correcting the target regression model according to the second data, and determining the corrected target regression model as the target change relationship; When the current change rate is less than or equal to the preset threshold, performing the step of determining that the target regression model is the target change relationship; The second data includes a plurality of second usage times of other reducers other than the target reducer within a set time period before failure and their corresponding second operating currents.

9. The method for maintaining a reducer according to claim 7, wherein: The preset regression model includes more than one polynomial linear regression model, and different polynomial linear regression models have different degrees. The step of training the preset regression model according to the first data to obtain the target regression model includes: Training the more than one polynomial linear regression models according to the first data to obtain more than one candidate change relationship; Determine an evaluation value corresponding to each of the candidate change relationships, wherein the evaluation value represents the difference between a predicted value and an actual value of the corresponding polynomial linear regression model; One of the more than one candidate change relationships is determined as the target change relationship according to the evaluation value, and the difference between the predicted value and the actual value corresponding to the target change relationship is the smallest among all the candidate change relationships.

10. The method for maintaining a reducer according to claim 9, wherein: After the step of training the more than one polynomial linear regression models according to the first data to obtain more than one candidate change relationship, the method further includes: When the operating current of the reducer corresponding to each of the alternative change relationships has an increasing trend with the use time, the step of determining the evaluation value corresponding to each of the alternative change relationships is performed.

11. The method for maintaining a reducer according to claim 7, wherein: After the step of training a preset regression model according to the first data to obtain a target regression model, the method further includes: Obtain the current total usage time and current operating current of the target reducer; Substituting the total usage time into a historical regression model to obtain a first current value, and substituting the total usage time into a target regression model to obtain a second current value; the historical regression model is a regression model generated before the current moment that characterizes the variation pattern of the operating current of the target reducer with usage time; Determining a first difference between the current operating current and the first current value, and determining a second difference between the current operating current and the second current value; When the first difference is less than the second difference, determining that the historical regression model is the target change relationship; When the first difference is greater than or equal to the second difference, the step of determining that the target regression model is the target change relationship is performed.

12. A maintenance device for a reducer, characterized in that: The reducer maintenance device includes: a memory, a processor, and a reducer maintenance program stored in the memory and executable on the processor. When the reducer maintenance program is executed by the processor, the steps of the reducer maintenance method according to any one of claims 1 to 11 are implemented.

13. A storage medium, characterized in that: The storage medium stores a maintenance program for the reducer, and when the maintenance program for the reducer is executed by the processor, the steps of the maintenance method for the reducer according to any one of claims 1 to 11 are implemented.

Images