Robot speed reducer service life prediction method, device, equipment and medium

Abstract

The invention discloses a robot speed reducer service life prediction method and device, equipment and a medium, relates to the technical field of service life prediction, and is used for solving the problem that the existing robot speed reducer service life prediction accuracy is relatively low. The method comprises the following steps: carrying out data acquisition on a full life cycle of a to-be-predicted robot speed reducer to obtain target full-cycle data; wherein the target full-cycle data comprises multiple groups of speed data, multiple groups of angle data and multiple groups of current data of the speed reducer; inputting the target full-cycle data into a trained life prediction model to obtain a life prediction result corresponding to the to-be-predicted robot speed reducer; wherein the trained life prediction model is obtained based on a plurality of preset regression models and a plurality of preset classification models, so that the problem that the precision of a single model is insufficient when the data size is small is effectively solved while the life prediction accuracy of the robot speed reducer is improved.

Description

Technical Field

[0001] The present application relates to the technical field of life prediction, and provides a method, device, equipment and medium for predicting the life of a robot reducer. Background Art

[0002] As we all know, robot reducers will experience hardware wear and aging during long-term use, which greatly affects the use of the robot, making the service life of the reducer an important evaluation indicator for the robot.

[0003] At present, the robot reducer life is usually estimated by using manufacturer data and empirical formulas, and then the subsequent maintenance and care are carried out based on this approximate reducer life. However, this life prediction method is based on the empirical formula for preliminary calculation, and cannot be customized for each reducer. It is not dynamic data, so it has two shortcomings: ① It is easy to cause the robot to shut down before the expected life due to accelerated degradation of the reducer; ② If the reducer is still in a healthy state when it reaches the expected life, then replacing the reducer too early will cause a waste of use time.

[0004] Therefore, how to improve the accuracy of robot reducer life prediction has become a problem that needs to be solved urgently. Summary of the invention

[0005] The present application provides a method, device, equipment and medium for predicting the life of a robot reducer, which are used to solve the problem of low accuracy in the life prediction of existing robot reducers.

[0006] On the one hand, a method for predicting the life of a robot reducer is provided, the method comprising:

[0007] Collect data for the entire life cycle of the robot reducer to be predicted, and obtain target full-cycle data; wherein the target full-cycle data includes multiple sets of speed data, multiple sets of angle data, and multiple sets of current data of the reducer;

[0008] The target full-cycle data is input into a trained life prediction model to obtain a life prediction result corresponding to the robot reducer to be predicted; wherein the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

[0009] Optionally, before inputting the target full-cycle data into a trained life prediction model to obtain a life prediction result corresponding to the reducer of the robot to be predicted, the method further includes:

[0010] Collect data on the entire life cycle of multiple robot reducers to obtain multiple sets of full-cycle data and multiple sets of actual life values; one set of full-cycle data corresponds to one set of actual life values;

[0011] Performing outlier processing, data integration, and data standardization on the multiple sets of full-cycle data to obtain multiple sets of processed full-cycle data;

[0012] Normalizing the multiple groups of actual lifespan values ​​to obtain multiple groups of processed lifespan values;

[0013] The multiple groups of processed life span values ​​are classified to obtain multiple groups of classified life span values.

[0014] Optionally, after performing outlier processing, data integration, and data standardization on the multiple sets of full-cycle data to obtain multiple sets of processed full-cycle data, the method further includes:

[0015] For any group of processed full-cycle data, feature extraction is performed on the processed full-cycle data in the time domain, frequency domain and time-frequency domain to obtain multiple groups of time-domain feature quantities, multiple groups of frequency-domain feature quantities and multiple groups of time-frequency domain feature quantities; wherein the time-domain feature quantities include mean value, maximum value, minimum value, median value, peak-to-peak value, standard deviation, root mean square, peak factor, impulse factor, margin factor, waveform factor, kurtosis and skewness; the frequency-domain feature quantities include centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation; and the time-frequency domain feature quantities include mean value, standard deviation, skewness and kurtosis of spectral kurtosis;

[0016] Feature quantity dimensionality reduction is performed on the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities and the multiple groups of time-frequency domain feature quantities to obtain the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities and the multiple groups of time-frequency domain feature quantities after dimensionality reduction corresponding to any group of processed full-cycle data.

[0017] Optionally, the step of performing feature dimension reduction on the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities to obtain the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities corresponding to any group of processed full-cycle data after dimension reduction includes:

[0018] If it is determined that the correlation value between any two of the multiple groups of time-domain feature quantities, the multiple groups of frequency-domain feature quantities, and the multiple groups of time-frequency-domain feature quantities is greater than a first preset correlation value, the first group of feature quantities is deleted; wherein the first group of feature quantities is any one of the two groups of feature quantities;

[0019] If it is determined that the correlation value between any one of the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities and a group of actual life values ​​corresponding to any one of the processed full-cycle data is less than a second preset correlation value, the second group of feature quantities is deleted; wherein the second group of feature quantities is any one of the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities.

[0020] Optionally, before inputting the target full-cycle data into a trained life prediction model to obtain a life prediction result corresponding to the reducer of the robot to be predicted, the method further includes:

[0021] Using multiple groups of actual life values, multiple groups of time domain feature quantities after dimension reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, the multiple preset regression models are trained to obtain multiple first candidate life prediction models; wherein the preset regression models include linear regression models, polynomial regression models, Gaussian process regression models, and neural network models;

[0022] Using multiple groups of classified life values, multiple groups of time domain feature quantities after dimensionality reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, the multiple preset classification models are trained to obtain multiple second candidate life prediction models; wherein the preset classification models include a decision tree model, a random forest model, a support vector machine model, and a neural network model;

[0023] Determining a plurality of initial life prediction models from the plurality of first candidate life prediction models and the plurality of second candidate life prediction models;

[0024] The multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

[0025] Optionally, the step of determining a plurality of initial life prediction models from the plurality of first candidate life prediction models and the plurality of second candidate life prediction models comprises:

[0026] The accuracy of the plurality of first candidate life prediction models is calculated by using the mean absolute error method to determine the accuracy corresponding to each of the plurality of first candidate life prediction models;

[0027] Using a preset accuracy calculation method, the accuracy of the plurality of second candidate life prediction models is calculated to determine the accuracy corresponding to each of the plurality of second candidate life prediction models;

[0028] A plurality of initial life prediction models are determined according to the accuracy rates corresponding to the plurality of second candidate life prediction models and the accuracy rates corresponding to the plurality of second candidate life prediction models.

[0029] Optionally, the step of performing weighted fusion on the multiple initial life prediction models to obtain a life prediction fusion model includes:

[0030] Determining weights corresponding to the multiple initial life prediction models according to the accuracy rates corresponding to the multiple initial life prediction models;

[0031] According to the life prediction values ​​and weights corresponding to each of the initial life prediction models, the multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

[0032] On the one hand, a robot reducer life prediction device is provided, the device comprising:

[0033] A data acquisition unit is used to collect data on the full life cycle of the robot reducer to be predicted, and obtain target full-cycle data; wherein the target full-cycle data includes multiple groups of speed data, multiple groups of angle data, and multiple groups of current data of the reducer;

[0034] The life prediction unit is used to input the target full-cycle data into a trained life prediction model to obtain a life prediction result corresponding to the robot reducer to be predicted; wherein the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

[0035] On the one hand, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein any one of the above methods is implemented when the processor executes the computer program.

[0036] In one aspect, a computer storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, any of the above methods is implemented.

[0037] Compared with the prior art, the beneficial effects of this application are:

[0038] In the embodiment of the present application, when predicting the life of the robot reducer, first, data collection can be performed on the whole life cycle of the robot reducer to be predicted to obtain the target full-cycle data; wherein, the target full-cycle data includes multiple groups of speed data, multiple groups of angle data and multiple groups of current data of the reducer; then, the target full-cycle data can be input into the trained life prediction model to obtain the corresponding life prediction result of the robot reducer to be predicted; wherein, the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models. Therefore, in the embodiment of the present application, since the life prediction is performed based on the full-cycle data of the robot reducer to be predicted, compared with the traditional life prediction method, the present application can not only monitor and predict the life of the reducer in real time and dynamically, improve the accuracy and flexibility of life estimation, effectively avoid the individual difference problem of different reducers, but also do not need to add additional sensors, and only rely on the joint speed, angle and current data that the robot can collect to make predictions, and have easy deployment in actual scenarios. In addition, since the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models, it can effectively solve the problem of insufficient accuracy of a single model when the amount of data is small. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0040] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;

[0041] Figure 2 A schematic diagram of a process flow of a robot reducer life prediction method provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of a robot reducer life prediction device provided in an embodiment of the present application.

[0043] Markings in the figure: 10 - robot reducer life prediction equipment, 101 - processor, 102 - memory, 103 - I / O interface, 104 - database, 30 - robot reducer life prediction device, 301 - data acquisition unit, 302 - life prediction unit, 303 - data preprocessing unit, 304 - feature extraction and dimensionality reduction unit, 305 - model fusion unit. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the embodiment of the present application will be clearly and completely described below in conjunction with the drawings in the embodiment of the present application. Obviously, the described embodiment is only a part of the embodiment of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other arbitrarily. In addition, although the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in an order different from that here.

[0045] As we all know, robot reducers will experience hardware wear and aging during long-term use, which greatly affects the use of the robot, making the service life of the reducer an important evaluation indicator for the robot.

[0046] At present, the robot reducer life is usually estimated by using manufacturer data and empirical formulas, and then the subsequent maintenance and care are carried out based on this approximate reducer life. However, this life prediction method is based on the empirical formula for preliminary calculation, and cannot be customized for each reducer. It is not dynamic data, so it has two shortcomings: ① It is easy to cause the robot to shut down before the expected life due to accelerated degradation of the reducer; ② If the reducer is still in a healthy state when it reaches the expected life, then replacing the reducer too early will cause a waste of use time.

[0047] Based on this, the embodiment of the present application provides a method for predicting the life of a robot reducer, in which, first, data collection can be performed on the entire life cycle of the robot reducer to be predicted to obtain target full-cycle data; wherein, the target full-cycle data includes multiple groups of speed data, multiple groups of angle data and multiple groups of current data of the reducer; then, the target full-cycle data can be input into the trained life prediction model to obtain the corresponding life prediction result of the robot reducer to be predicted; wherein, the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models. Therefore, in the embodiment of the present application, since the life prediction is performed based on the full-cycle data of the robot reducer to be predicted, compared with the traditional life prediction method, the present application can not only monitor and predict the life of the reducer in real time and dynamically, improve the accuracy and flexibility of life estimation, effectively avoid the individual difference problem of different reducers, but also do not need to add additional sensors, and can only be predicted based on the joint speed, angle and current data that the robot can collect, and it has easy deployment in the actual scene. In addition, since the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models, it can effectively solve the problem of insufficient accuracy of a single model when the data volume is small.

[0048] After introducing the design ideas of the embodiments of the present application, the following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios introduced below are only used to illustrate the embodiments of the present application and are not limited. In the specific implementation process, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0049] like Figure 1 FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application. The application scenario may include a robot reducer life prediction device 10 .

[0050] Among them, the robot reducer life prediction device 10 can be used to predict the life of the robot reducer, for example, it can be a personal computer (PC), a server and a laptop. The robot reducer life prediction device 10 may include one or more processors 101, a memory 102, an I / O interface 103 and a database 104. Specifically, the processor 101 can be a central processing unit (CPU), or a digital processing unit, etc. The memory 102 can be a volatile memory (volatile memory), such as a random-access memory (RAM); the memory 102 can also be a non-volatile memory (non-volatile memory), such as a read-only memory, a flash memory (flash memory), a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 102 can be any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to this. The memory 102 can be a combination of the above memories. The memory 102 may store some program instructions of the robot reducer life prediction method provided in the embodiment of the present application, and these program instructions can be used to implement the steps of the robot reducer life prediction method provided in the embodiment of the present application when executed by the processor 101, so as to solve the problem of low accuracy of the existing robot reducer life prediction. The database 104 may be used to store the target full-cycle data involved in the scheme provided in the embodiment of the present application, the trained life prediction model, multiple preset regression models, multiple preset classification models, multiple groups of full-cycle data, multiple groups of actual life values, multiple groups of processed full-cycle data, multiple groups of processed life values, multiple time domain feature quantities after dimensionality reduction, multiple frequency domain feature quantities, and multiple time-frequency domain feature quantities and other data.

[0051] In the embodiment of the present application, the robot reducer life prediction device 10 can obtain the target full-cycle data through the I / O interface 103, and then the processor 101 of the robot reducer life prediction device 10 will improve the accuracy of the robot reducer life prediction according to the program instructions of the robot reducer life prediction method provided by the embodiment of the present application in the memory 102. In addition, the target full-cycle data, the trained life prediction model, multiple preset regression models, multiple preset classification models, multiple groups of full-cycle data, multiple groups of actual life values, multiple groups of processed full-cycle data, multiple groups of processed life values, multiple time domain feature quantities after dimensionality reduction, multiple frequency domain feature quantities, and multiple time-frequency domain feature quantities and other data can also be stored in the database 104.

[0052] Of course, the method provided in the embodiment of the present application is not limited to Figure 1 The application scenarios shown can also be used in other possible application scenarios, and the embodiments of the present application are not limited thereto. Figure 1 The functions that can be realized by each device in the application scenario shown will be described in the subsequent method embodiments, and will not be described in detail here. Below, the method of the embodiment of the present application will be introduced in conjunction with the accompanying drawings.

[0053] like Figure 2 As shown, it is a schematic diagram of a process flow of a method for predicting the life of a robot reducer provided in an embodiment of the present application. The method can be Figure 1 The robot reducer life prediction device 10 is used to perform the method. Specifically, the process of the method is introduced as follows.

[0054] Step 201: Collect data on the entire life cycle of the robot reducer to be predicted to obtain target full-cycle data.

[0055] In the embodiment of the present application, the target full-cycle data includes multiple sets of speed data, multiple sets of angle data, and multiple sets of current data of the reducer.

[0056] Specifically, encoders, Hall sensors and current sensors may be used to collect data on the entire life cycle of the robot reducer to be predicted, so as to obtain target full-cycle data generated by the robot reducer to be predicted during its entire life cycle.

[0057] Step 202: Input the target full-cycle data into the trained life prediction model to obtain the life prediction result corresponding to the robot reducer to be predicted.

[0058] In an embodiment of the present application, the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

[0059] Specifically, in order to improve the accuracy and flexibility of life estimation, and effectively solve the problem of insufficient accuracy of a single model when the amount of data is small, in the application embodiment, after obtaining the target full-cycle data, the target full-cycle data can be directly input into the trained life prediction model to obtain the life prediction result corresponding to the robot reducer to be predicted, thereby, it can be determined whether the reducer life condition can predict the life result before the life limit is reached.

[0060] In one possible implementation, in order to obtain a life prediction model with high accuracy in life estimation, in an embodiment of the present application, before inputting the target full-cycle data into the trained life prediction model and obtaining the life prediction result corresponding to the robot reducer to be predicted, the robot reducer can also be subjected to an "accelerated life test" to obtain relevant operating data of the reducer's full life cycle for model training, that is, a new reducer can be assembled on the robot, and data can be collected for the full life cycle of multiple robot reducers to obtain multiple sets of full-cycle data and multiple sets of actual life values ​​(the final life of the reducer) collected during the accelerated life test; wherein, a set of full-cycle data corresponds to a set of actual life values.

[0061] Specifically, when performing an accelerated life test, after setting the operation cycle, a fixed action can be performed at a fixed speed to collect data. The fixed speed can be 250mm / s, 300mm / s, 350mm / s, and 400mm / s, and the fixed action can be the rotation of the axis at angles of ±10°, ±15°, ±20°, and ±25° from the robot's zero position. In order to prevent single errors, in the embodiment of the present application, the robot can repeat the fixed action for 4 operation cycles.

[0062] Collect the speed, angle and current time series data of the above operation cycle, as well as the operation life data of the reducer corresponding to the corresponding collection time. Then, collect data of a reducer from brand new to the life limit according to this method, and you can get the data of the whole life cycle of a reducer. Repeat the above actions to collect the full cycle data of the whole life cycle of multiple reducers, and get multiple groups of reducer data to be processed.

[0063] In order to further improve the estimation accuracy of the life prediction model, in an embodiment of the present application, the collected full-cycle data and actual life value of the reducer can also be preprocessed.

[0064] (1) Multiple sets of full-cycle data can be processed for outliers, integrated, and standardized to obtain multiple sets of processed full-cycle data.

[0065] Specifically, when performing "outlier processing" on multiple groups of full-cycle data, all data with abnormalities in the full-cycle data can be deleted, for example, the collected data is incomplete, the collected data has large fluctuations, and the collected data has abnormal values.

[0066] When performing "data integration" on multiple sets of full-cycle data, the speed, angle and current data files collected within the same sampling period can be integrated one by one into one file, and then the four operating cycle data within the sampling period can be averaged to obtain a file containing the speed, angle and current data of a single operating cycle.

[0067] When performing "data standardization" on multiple sets of full-cycle data, the z-score method can be used to standardize the data. For any set of full-cycle data (including a set of speed data, a set of angle data, and a set of current data), the following formula (1) can be used to standardize the data:

[0068]

[0069] Among them, x ′ is the full-cycle data after standardization, x is the full-cycle data before standardization, μ is the mean value of the full-cycle data before standardization, and σ is the standard deviation of the full-cycle data before standardization.

[0070] Furthermore, the above formula (1) can be used to calculate each point in any set of full-cycle data (including a set of speed data, a set of angle data and a set of current data) to obtain new data after standardization of any set of full-cycle data.

[0071] (2) normalizing the multiple groups of actual life values ​​to obtain multiple groups of processed life values; and classifying the multiple groups of processed life values ​​to obtain multiple groups of classified life values.

[0072] Specifically, the following formula (2) can be used to normalize multiple groups of actual life values, that is, the following formula (2) can be used to calculate each point in each group of actual life values:

[0073]

[0074] Among them, y ′ is the lifespan value after normalization, y is the actual lifespan value before normalization, and y max and min is the maximum and minimum value of each group of actual life values.

[0075] Furthermore, through this normalization method, the minimum value of the reducer life can be changed to 0, and the maximum value can be changed to 1. This effectively reduces the situation where the life span is different due to individual differences in reducers.

[0076] Then, the multiple groups of processed life values ​​can be classified, that is, a set proportion of the life values ​​in each group of processed life values ​​can be set to 1 (indicating imminent damage), and the rest can be set to 0 (indicating a healthy state), wherein the set proportion can be any value, for example, it can be 90%, 92%, 95%, and 99%.

[0077] In one possible implementation, in order to improve the efficiency of model training and simplify the training and prediction process of the machine learning model, in an embodiment of the present application, the data set for model training may also be subjected to dimensionality reduction processing.

[0078] Specifically, first, for any group of processed full-cycle data, feature extraction can be performed on the time domain, frequency domain, and time-frequency domain to obtain multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities; wherein the time domain feature quantities include mean value, maximum value, minimum value, median value, peak-to-peak value, standard deviation, root mean square, peak factor, pulse factor, margin factor, waveform factor, kurtosis, and skewness; the frequency domain feature quantities include centroid frequency, mean square frequency, root mean square frequency, frequency variance, and frequency standard deviation; and the time-frequency domain feature quantities include mean value, standard deviation, skewness, and kurtosis of spectral kurtosis.

[0079] Then, the above 23 types of feature quantities can be used to reduce the dimension of feature quantities, that is, the 23 types of feature quantities, including multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, can be subjected to dimension reduction to obtain multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities corresponding to any group of processed full-cycle data after dimension reduction.

[0080] In a possible implementation, a “correlation analysis method” may be used to reduce the dimension of feature quantities to obtain relevant feature quantities that can be used to build a model.

[0081] Specifically, first, the one-to-one correlation between each characteristic quantity (23 characteristic quantities) and the actual life value, a total of 24 sets of data, can be calculated, wherein the correlation value can be calculated using the following formula (3):

[0082]

[0083] Among them, cov(X,Y) is the correlation value between the first set of data and the second set of data, n is the length of any set of data, X μ is the mean value of the first set of data, Y μ is the mean value of the second set of data, Y i is the i-th point of the first set of data, Y i is the i-th point of the second set of data.

[0084] Then, the correlation between each feature quantity can be determined, and when the correlation is greater than a set value, one of the correlations can be deleted. That is, if it is determined that the correlation value between any two sets of feature quantities among multiple sets of time domain feature quantities, multiple sets of frequency domain feature quantities, and multiple sets of time-frequency domain feature quantities is greater than a first preset correlation value, the first set of feature quantities is deleted; wherein the first set of feature quantities is any set of feature quantities among any two sets of feature quantities. Specifically, the first preset correlation value can be set to any value greater than 0.9 and less than or equal to 1, for example, it can be a value such as 1, 0.95, and 0.91.

[0085] Next, the correlation between each feature quantity and the life data can be determined, and when the correlation is less than the set value, the feature quantity can be deleted. That is, if it is determined that the correlation value between any one of the multiple groups of time-domain feature quantities, multiple groups of frequency-domain feature quantities, and multiple groups of time-frequency-domain feature quantities and a group of actual life values ​​corresponding to any group of processed full-cycle data is less than the second preset correlation value, the second group of feature quantities is deleted; wherein the second group of feature quantities is any one of the multiple groups of time-domain feature quantities, multiple groups of frequency-domain feature quantities, and multiple groups of time-frequency-domain feature quantities. Specifically, the second preset correlation value can be set to any value greater than 0 and less than or equal to 0.5, for example, it can be 0.45, 0.3, and 0.1.

[0086] In one possible implementation, before inputting the target full-cycle data into the trained life prediction model and obtaining the life prediction result corresponding to the robot reducer to be predicted, the life prediction fusion model can also be acquired. In addition, in order to improve the accuracy of the robot reducer life prediction and to solve the problem of insufficient accuracy of a single model when the data volume is small, in an embodiment of the present application, the above-mentioned feature quantities and corresponding life data can be used to build relevant models for machine learning. This construction can use both regression models and classification models.

[0087] Specifically, first, multiple groups of actual life values, multiple groups of time domain feature quantities after dimensionality reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities can be used to train multiple preset regression models to obtain multiple first candidate life prediction models; among which, the preset regression models include linear regression models, polynomial regression models, Gaussian process regression models, and neural network models.

[0088] Then, multiple groups of classified life values, multiple groups of time domain feature quantities after dimensionality reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities can be used to train multiple preset classification models to obtain multiple second candidate life prediction models; wherein the preset classification models include decision tree models, random forest models, support vector machine models, and neural network models.

[0089] Next, multiple initial life prediction models can be determined directly from the multiple first candidate life prediction models and the multiple second candidate life prediction models.

[0090] Finally, the life prediction fusion model can be obtained by weighted fusion of multiple initial life prediction models.

[0091] In a possible implementation, when multiple initial life prediction models are determined from multiple first candidate life prediction models and multiple second candidate life prediction models, model selection can be performed by arranging the verification accuracy of the above-mentioned candidate life prediction models in descending order.

[0092] Specifically, first, the mean absolute error (MAE) method can be used to calculate the accuracy of multiple first candidate life prediction models (regression models) to determine the accuracy of each of the multiple first candidate life prediction models. The mean absolute error MAE can be calculated using the following formula (4):

[0093]

[0094] Where n is the number of data points of each group of actual life values, is the actual lifespan value, m i is the predicted life span value.

[0095] Then, a preset accuracy calculation method can be used to calculate the accuracy of multiple second candidate life prediction models to determine the accuracy corresponding to each of the multiple second candidate life prediction models. The preset accuracy calculation method can be expressed by the following formula (5):

[0096]

[0097] Among them, F is the number of data with a lifespan value of 0 in each group of processed lifespan values, P is the number of data with a lifespan value of 1 after processing, TF is the number of data with a correctly predicted lifespan of 0, and TP is the number of data with a correctly predicted lifespan of 1.

[0098] Finally, multiple initial life prediction models can be determined according to the accuracy rates corresponding to each of the multiple second candidate life prediction models and the accuracy rates corresponding to each of the multiple second candidate life prediction models. That is, the accuracy rates corresponding to each of the multiple second candidate life prediction models and the multiple second candidate life prediction models can be arranged in descending order according to the size of the accuracy rates, and then a model selection quantity value A can be set to select the number of models whose accuracy is arranged within the model selection quantity value A, wherein the model selection quantity value A can be any integer greater than 0, for example, it can be 1, 2, 3, and 4.

[0099] In one possible implementation, when performing weighted fusion on multiple initial life prediction models to obtain a life prediction fusion model, in order to improve the robustness, generalization ability and classification accuracy of the life prediction fusion model, in an embodiment of the present application, a "voting method" can be used to fuse the initial life prediction models to obtain the final life prediction fusion model.

[0100] Specifically, first, the weights corresponding to the multiple initial life prediction models can be determined according to the accuracy rates corresponding to the multiple initial life prediction models. The weight of any initial life prediction model B can be calculated using the following formula (6):

[0101]

[0102] Then, according to the life prediction values ​​and weights corresponding to each of the initial life prediction models, multiple initial life prediction models can be weighted fused to obtain a life prediction fusion model. That is, the life prediction values ​​of each initial life prediction model can be multiplied by the weight of the corresponding initial life prediction model to obtain the output result (life prediction value) of the life prediction fusion model. Among them, if the output result is greater than or equal to 0.5, the final output result is 1, which is judged to be a damaged reducer; if the output result is less than 0.5, the final output result is 0, which is judged to be a healthy reducer.

[0103] In summary, since the life prediction is based on the full-cycle data of the robot reducer to be predicted, compared with the traditional life prediction method, this application can not only monitor and predict the life of the reducer in real time and dynamically, improve the accuracy and flexibility of life estimation, and effectively avoid the individual differences of different reducers, but also does not need to add additional sensors, and can only be predicted based on the joint speed, angle and current data that can be collected by the robot, which is easy to deploy in actual scenarios. In addition, since the trained life prediction model is based on multiple preset regression models and multiple preset classification models, it can effectively solve the problem of insufficient accuracy of a single model when the amount of data is small. And when predicting an unknown reducer, the accuracy of the result is as high as 98.68%, which can accurately predict the life of the reducer before the life reaches the limit, so as to provide early warning.

[0104] Based on the same inventive concept, the embodiment of the present application provides a robot reducer life prediction device 30, such as Figure 3 As shown, the robot reducer life prediction device 30 includes:

[0105] The data acquisition unit 301 is used to collect data on the full life cycle of the robot reducer to be predicted, and obtain target full-cycle data; wherein the target full-cycle data includes multiple groups of speed data, multiple groups of angle data and multiple groups of current data of the reducer;

[0106] The life prediction unit 302 is used to input the target full-cycle data into the trained life prediction model to obtain the life prediction result corresponding to the robot reducer to be predicted; wherein the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

[0107] Optionally, the robot reducer life prediction device 30 further includes a data preprocessing unit 303, which is used to:

[0108] Collect data on the entire life cycle of multiple robot reducers to obtain multiple sets of full-cycle data and multiple sets of actual life values; one set of full-cycle data corresponds to one set of actual life values;

[0109] Perform outlier processing, data integration and data standardization on multiple sets of full-cycle data to obtain multiple sets of processed full-cycle data;

[0110] Performing normalization processing on multiple groups of actual life span values ​​to obtain multiple groups of processed life span values;

[0111] The life span values ​​after multiple groups of processing are classified to obtain the life span values ​​after multiple groups of classification.

[0112] Optionally, the robot reducer life prediction device 30 further includes a feature extraction and dimension reduction unit 304, which is used to:

[0113] For any group of processed full-cycle data, feature extraction is performed on any group of processed full-cycle data in the time domain, frequency domain and time-frequency domain to obtain multiple groups of time-domain feature quantities, multiple groups of frequency-domain feature quantities and multiple groups of time-frequency domain feature quantities; wherein the time-domain feature quantities include mean value, maximum value, minimum value, median value, peak-to-peak value, standard deviation, root mean square, peak factor, impulse factor, margin factor, waveform factor, kurtosis and skewness, the frequency-domain feature quantities include centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation, and the time-frequency domain feature quantities include mean value, standard deviation, skewness and kurtosis of spectral kurtosis;

[0114] Feature dimension reduction is performed on multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities to obtain multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities after dimension reduction corresponding to any group of processed full-cycle data.

[0115] Optionally, the feature extraction and dimension reduction unit 304 is further used for:

[0116] If it is determined that the correlation value between any two groups of feature quantities among the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities is greater than the first preset correlation value, the first group of feature quantities is deleted; wherein the first group of feature quantities is any group of feature quantities among any two groups of feature quantities;

[0117] If it is determined that the correlation value between any group of feature quantities among multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities and a group of actual life values ​​corresponding to any group of processed full-cycle data is less than a second preset correlation value, the second group of feature quantities is deleted; wherein the second group of feature quantities is any group of feature quantities among multiple groups of time domain feature quantities, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities.

[0118] Optionally, the robot reducer life prediction device 30 further includes a model fusion unit 305, which is used to:

[0119] Using multiple groups of actual life values, multiple groups of time domain feature quantities after dimension reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, multiple preset regression models are trained to obtain multiple first candidate life prediction models; wherein the preset regression models include linear regression models, polynomial regression models, Gaussian process regression models, and neural network models;

[0120] Using multiple groups of classified life values, multiple groups of time domain feature quantities after dimensionality reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, multiple preset classification models are trained to obtain multiple second candidate life prediction models; wherein the preset classification models include a decision tree model, a random forest model, a support vector machine model, and a neural network model;

[0121] Determining a plurality of initial life prediction models from a plurality of first candidate life prediction models and a plurality of second candidate life prediction models;

[0122] Multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

[0123] Optionally, the model fusion unit 305 is further used for:

[0124] The accuracy of the plurality of first candidate life prediction models is calculated by using the mean absolute error method to determine the accuracy corresponding to each of the plurality of first candidate life prediction models;

[0125] Using a preset accuracy calculation method, the accuracy of the plurality of second candidate life prediction models is calculated to determine the accuracy corresponding to each of the plurality of second candidate life prediction models;

[0126] A plurality of initial life prediction models are determined according to the respective corresponding accuracy rates of the plurality of second candidate life prediction models and the respective corresponding accuracy rates of the plurality of second candidate life prediction models.

[0127] Optionally, the model fusion unit 305 is further used for:

[0128] Determining weights corresponding to the multiple initial life prediction models according to the accuracy rates corresponding to the multiple initial life prediction models;

[0129] According to the life prediction values ​​and weights corresponding to each initial life prediction model, multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

[0130] The robot reducer life prediction device 30 can be used to perform Figure 2 The method implemented in the embodiment shown in FIG. 1 is a method implemented in the embodiment shown in FIG. 1 . Therefore, the functions that can be realized by each functional module of the robot reducer life prediction device 30 can be referred to. Figure 2 The description of the illustrated embodiment is not repeated in detail.

[0131] In some possible implementations, various aspects of the method provided in the present application may also be implemented in the form of a program part, which includes a program code. When the program part is run on a computer device, the program code is used to enable the computer device to execute the steps of the method according to various exemplary embodiments of the present application described above in this specification. For example, the computer device may execute the following steps: Figure 2 The method performed in the illustrated embodiment.

[0132] It can be understood by those skilled in the art that all or part of the steps of the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above method embodiment are executed; and the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk. Alternatively, if the above integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent part, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present invention can be essentially or in other words, the part that contributes to the prior art can be embodied in the form of a software part, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. And the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disk or optical disk.

[0133] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0134] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A robot reducer life prediction method, characterized in that: The method comprises: Collect data for the entire life cycle of the robot reducer to be predicted, and obtain target full-cycle data; wherein the target full-cycle data includes multiple sets of speed data, multiple sets of angle data, and multiple sets of current data of the reducer; The target full-cycle data is input into a trained life prediction model to obtain a life prediction result corresponding to the robot reducer to be predicted; wherein the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

2. The method according to claim 1, characterized in that Before inputting the target full-cycle data into the trained life prediction model to obtain the life prediction result corresponding to the reducer of the robot to be predicted, the method further includes: Collect data on the entire life cycle of multiple robot reducers to obtain multiple sets of full-cycle data and multiple sets of actual life values; one set of full-cycle data corresponds to one set of actual life values; Performing outlier processing, data integration, and data standardization on the multiple sets of full-cycle data to obtain multiple sets of processed full-cycle data; Normalizing the multiple groups of actual lifespan values ​​to obtain multiple groups of processed lifespan values; The multiple groups of processed life span values ​​are classified to obtain multiple groups of classified life span values.

3. The method according to claim 2, characterized in that After performing outlier processing, data integration, and data standardization on the multiple sets of full-cycle data to obtain multiple sets of processed full-cycle data, the method further includes: For any group of processed full-cycle data, feature extraction is performed on the processed full-cycle data in the time domain, frequency domain and time-frequency domain to obtain multiple groups of time-domain feature quantities, multiple groups of frequency-domain feature quantities and multiple groups of time-frequency domain feature quantities; wherein the time-domain feature quantities include mean value, maximum value, minimum value, median value, peak-to-peak value, standard deviation, root mean square, peak factor, impulse factor, margin factor, waveform factor, kurtosis and skewness; the frequency-domain feature quantities include centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation; and the time-frequency domain feature quantities include mean value, standard deviation, skewness and kurtosis of spectral kurtosis; Feature quantity dimensionality reduction is performed on the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities and the multiple groups of time-frequency domain feature quantities to obtain the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities and the multiple groups of time-frequency domain feature quantities after dimensionality reduction corresponding to any group of processed full-cycle data.

4. The method according to claim 3, characterized in that The step of performing feature dimension reduction on the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities to obtain the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities corresponding to any group of processed full-cycle data after dimension reduction comprises: If it is determined that the correlation value between any two of the multiple groups of time-domain feature quantities, the multiple groups of frequency-domain feature quantities, and the multiple groups of time-frequency-domain feature quantities is greater than a first preset correlation value, the first group of feature quantities is deleted; wherein the first group of feature quantities is any one of the two groups of feature quantities; If it is determined that the correlation value between any one of the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities and a group of actual life values ​​corresponding to any one of the processed full-cycle data is less than a second preset correlation value, the second group of feature quantities is deleted; wherein the second group of feature quantities is any one of the multiple groups of time domain feature quantities, the multiple groups of frequency domain feature quantities, and the multiple groups of time-frequency domain feature quantities.

5. The method according to claim 3, characterized in that Before inputting the target full-cycle data into the trained life prediction model to obtain the life prediction result corresponding to the reducer of the robot to be predicted, the method further includes: Using multiple groups of actual life values, multiple groups of time domain feature quantities after dimension reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, the multiple preset regression models are trained to obtain multiple first candidate life prediction models; wherein the preset regression models include linear regression models, polynomial regression models, Gaussian process regression models, and neural network models; Using multiple groups of classified life values, multiple groups of time domain feature quantities after dimensionality reduction, multiple groups of frequency domain feature quantities, and multiple groups of time-frequency domain feature quantities, the multiple preset classification models are trained to obtain multiple second candidate life prediction models; wherein the preset classification models include a decision tree model, a random forest model, a support vector machine model, and a neural network model; Determining a plurality of initial life prediction models from the plurality of first candidate life prediction models and the plurality of second candidate life prediction models; The multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

6. The method according to claim 5, characterized in that The step of determining a plurality of initial life prediction models from the plurality of first candidate life prediction models and the plurality of second candidate life prediction models comprises: The accuracy of the plurality of first candidate life prediction models is calculated by using the mean absolute error method to determine the accuracy corresponding to each of the plurality of first candidate life prediction models; Using a preset accuracy calculation method, the accuracy of the plurality of second candidate life prediction models is calculated to determine the accuracy corresponding to each of the plurality of second candidate life prediction models; A plurality of initial life prediction models are determined according to the accuracy rates corresponding to the plurality of second candidate life prediction models and the accuracy rates corresponding to the plurality of second candidate life prediction models.

7. The method according to claim 5, characterized in that The step of performing weighted fusion on the multiple initial life prediction models to obtain a life prediction fusion model comprises: Determining weights corresponding to the multiple initial life prediction models according to the accuracy rates corresponding to the multiple initial life prediction models; According to the life prediction values ​​and weights corresponding to each of the initial life prediction models, the multiple initial life prediction models are weightedly fused to obtain a life prediction fusion model.

8. A robot reducer life prediction device, characterized in that: The device comprises: A data acquisition unit is used to collect data on the full life cycle of the robot reducer to be predicted, and obtain target full-cycle data; wherein the target full-cycle data includes multiple groups of speed data, multiple groups of angle data, and multiple groups of current data of the reducer; The life prediction unit is used to input the target full-cycle data into a trained life prediction model to obtain a life prediction result corresponding to the robot reducer to be predicted; wherein the trained life prediction model is obtained based on multiple preset regression models and multiple preset classification models.

9. An electronic device, characterized in that: The device comprises: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory, and execute any method according to claims 1-7 according to the obtained program instructions.

10. A storage medium, characterized in that: The storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute any one of the methods of claims 1-7.

Images