Industrial robot anomaly detection method based on recursive feature elimination and auto-encoder

Abstract

The invention discloses an industrial robot anomaly detection method based on recursive feature elimination and an auto-encoder. The method comprises the steps of obtaining a first preset number of normal sample signals; constructing a one-dimensional convolution auto-encoder model for inputting the training set and the verification set of the normal sample signals into the one-dimensional convolution auto-encoder model for training and verification so as to obtain reconstruction signals of the motion trail of the robot; constructing a feature set 1 according to the extracted time domain features; a recursive feature elimination method is introduced for effective feature screening; according to the weights corresponding to the obtained effective features, the percentage of the weights of the effective features is obtained; subtracting the weight feature of the to-be-detected acceleration signal from the weight feature of the reconstructed signal to obtain a feature error of the to-be-detected acceleration signal; and judging the state of the industrial robot according to the characteristic error of the acceleration signal to be detected. According to the method, an operator does not need to master a large amount of professional knowledge of the industrial robot, and does not need to establish a complex dynamic model, detection can be directly carried out based on the original acceleration signal collected by the sensor, and the industrial robot anomaly detection process is simplified.

Description

Technical Field

[0001] The invention relates to an industrial robot anomaly detection method based on recursive feature elimination and autoencoder, and belongs to the field of industrial robot anomaly detection based on terminal vibration signals. Background Art

[0002] Abnormalities in the robot during operation may lead to a decline in product quality. Therefore, abnormal state detection of the robot is a necessary measure to ensure the service life of the robot and product quality. Timely detection and elimination of abnormal conditions are crucial to ensure the continuous operation of industrial robots. It helps managers determine maintenance timing based on changes in their operating status and formulate reasonable maintenance strategies, thereby improving the reliability and safety of the robot.

[0003] In the field of industrial robot anomaly detection, at present, the research methods of anomaly detection are mainly divided into three categories: deep learning-based methods, data distribution-based methods, and distance-based methods. Deep learning methods have high anomaly detection accuracy, but they require a large amount of data and are sensitive to parameters; data distribution-based methods are fast, but have low detection accuracy; distance-based methods are simple to calculate, but are too sensitive to the choice of distance metrics. In the comparison of several anomaly detection methods, the autoencoder reconstruction method in deep learning methods has significant advantages in automatic feature extraction, nonlinear relationship processing, and large-scale data adaptation. In view of the lack of sufficient industrial robot anomaly data in data-driven technology, Chen T et al. proposed an unsupervised anomaly detection method for industrial robots - sliding window convolutional variational autoencoder. This method realizes real-time anomaly detection in space and time by processing multivariate time series data. In view of the data scarcity faced in industrial robot anomaly detection, which limits the accuracy of data-driven anomaly detection, Zhong ZD et al. proposed a sliding window one-dimensional convolutional autoencoder to determine whether the operating status of the industrial robot is normal.

[0004] In robot anomaly detection based on reconstruction methods, the goal is to distinguish normal signals from abnormal signals. Therefore, accurately extracting the features of original signals and reconstructed signals is crucial to achieve anomaly detection accuracy. Summary of the invention

[0005] The present invention provides an industrial robot anomaly detection method based on recursive feature elimination and autoencoder, which can detect whether anomalies occur during the movement of the robot in the presence of abnormalities.

[0006] The technical solution of the present invention is:

[0007] According to a first aspect of the present invention, a method for detecting anomalies of an industrial robot based on recursive feature elimination and autoencoder is provided, and the method steps are as follows:

[0008] Step 1: According to the selected motion mode, the end acceleration signal of the industrial robot under normal motion is collected through the posture sensor installed at the end of the industrial robot, and the acceleration signal is segmented according to the motion cycle to obtain a first preset number of normal sample signals, and the training set and the verification set are divided;

[0009] Step 2: Construct a one-dimensional convolutional autoencoder model to input the training set and verification set of normal sample signals into the one-dimensional convolutional autoencoder model for training and verification, and obtain the reconstructed signal of the robot's motion trajectory;

[0010] Step 3: extracting M time domain features from the second preset number of normal sample signals; constructing feature set 1 based on the extracted M time domain features; introducing a recursive feature elimination method to screen effective features, and obtaining a third preset number of N effective features;

[0011] Step 4: According to the weights corresponding to the obtained effective features, the percentage of the effective feature weights is obtained;

[0012] Step 5: Calculate the weighted features of the acceleration signal to be detected based on the effective features of the acceleration signal to be detected and the percentage of the effective feature weights; calculate the weighted features of the reconstructed signal based on the effective features of the reconstructed signal obtained in step 2 and the percentage of the effective feature weights; subtract the weighted features of the acceleration signal to be detected from the weighted features of the reconstructed signal to obtain the characteristic error of the acceleration signal to be detected;

[0013] Step 6: According to the characteristic error of the acceleration signal to be detected, the state of the industrial robot is judged to realize abnormality detection of the industrial robot.

[0014] Furthermore, the one-dimensional convolutional autoencoder model is divided into two parts: an encoder and a decoder. The encoder includes a convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer; the decoder includes a transposed convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer.

[0015] Furthermore, the recursive feature elimination method is specifically as follows: feature set 1 is used as the initial input of the support vector machine model, and the support vector machine model is iterated MN times. In each iteration, the time domain features with the smallest weight are removed through the weight information provided by the support vector machine model. For the time domain features retained after the sth iteration, feature set s+1 is constructed as the input of the support vector machine model for the s+1th iteration until the required number of iterations is reached. The weight information provided by the support vector machine model under the MNth iteration is sorted, and the time domain features represented by the first N weight information are retained as valid features.

[0016] Furthermore, the percentage formula of the effective feature weight is: Among them, |ω j | is the absolute value of the weight of the jth effective feature, is the sum of the absolute values ​​of all valid feature weights.

[0017] Furthermore, the weighted features of the acceleration signal to be detected, the weighted features of the reconstructed signal, and the characteristic error of the acceleration signal to be detected are expressed as follows:

[0018]

[0019] Among them, W j represents the jth effective feature weight, f j is the jth effective feature of the acceleration signal to be detected, F j is the jth effective feature of the reconstructed signal; H F is the characteristic error of the acceleration signal to be detected, H o is the weight feature of the acceleration signal to be detected, H r is the weight feature of the reconstructed signal.

[0020] Furthermore, the state of the industrial robot is judged based on the characteristic error of the acceleration signal to be detected, and the specific expression is:

[0021]

[0022] Among them, UL is the upper boundary, H F is the characteristic error of the acceleration signal to be detected.

[0023] According to a second aspect of the present invention, an industrial robot anomaly detection device based on recursive feature elimination and autoencoder is provided, comprising a module for executing any one of the above-mentioned industrial robot anomaly detection methods based on recursive feature elimination and autoencoder.

[0024] According to a third aspect of the present invention, there is provided a terminal device, comprising a memory, a processor, and a program stored in the memory and executable by the processor, wherein when the processor executes the program, the industrial robot anomaly detection method based on recursive feature elimination and autoencoder as described in any one of the above is implemented.

[0025] The beneficial effects of the present invention are: the present invention does not require the operator to master a large amount of professional knowledge of industrial robots, nor does it require the establishment of a complex dynamic model. It can directly detect based on the original acceleration signal collected by the sensor, simplifying the process of abnormal detection of industrial robots; specifically: the present invention can reconstruct the sample signal through the designed one-dimensional convolutional autoencoder model, and perform effective feature screening according to the recursive feature elimination method, which can reduce the computational complexity and noise interference to the detection; further, according to the characteristic error and threshold of the acceleration signal to be detected, the state of the signal to be detected can be effectively distinguished, thereby improving the detection effect of abnormal conditions of industrial robots. Compared with other methods, the present invention not only provides a new perspective in theory, but also has significant advantages in practical applications, providing a more effective and economical solution for the health monitoring and abnormal detection of industrial robot joints, and has important research significance and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of the present invention;

[0027] Figure 2 It is an industrial robot sample signal acquisition test bench;

[0028] Figure 3 This is a one-dimensional convolutional autoencoder model structure diagram;

[0029] Figure 4 This is a classification result diagram of the acceleration signal to be detected of the industrial robot. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily without conflict.

[0031] Example 1: Figure 1-Figure 4 As shown, according to a first aspect of an embodiment of the present invention, a method for detecting anomalies of an industrial robot based on recursive feature elimination and autoencoder is provided, and the method steps are as follows:

[0032] Step 1: According to the selected motion mode, the end acceleration signal of the industrial robot under normal motion is collected through the posture sensor installed at the end of the industrial robot, and the acceleration signal is segmented according to the motion cycle to obtain a first preset number of normal sample signals, and the training set and the verification set are divided;

[0033] Step 2: Construct a one-dimensional convolutional autoencoder model to input the training set and verification set of normal sample signals into the one-dimensional convolutional autoencoder model for training and verification, and obtain the reconstructed signal of the robot's motion trajectory;

[0034] Step 3: extracting M time domain features from the second preset number of normal sample signals; constructing feature set 1 based on the extracted M time domain features; introducing a recursive feature elimination method to screen effective features, and obtaining a third preset number of N effective features;

[0035] Step 4: According to the weights corresponding to the obtained effective features, the percentage of the effective feature weights is obtained;

[0036] Step 5: Calculate the weighted features of the acceleration signal to be detected based on the effective features of the acceleration signal to be detected and the percentage of the effective feature weights; calculate the weighted features of the reconstructed signal based on the effective features of the reconstructed signal obtained in step 2 and the percentage of the effective feature weights; subtract the weighted features of the acceleration signal to be detected from the weighted features of the reconstructed signal to obtain the characteristic error of the acceleration signal to be detected;

[0037] Step 6: According to the characteristic error of the acceleration signal to be detected, the state of the industrial robot is judged to realize abnormality detection of the industrial robot.

[0038] Furthermore, the one-dimensional convolutional autoencoder model is divided into two parts: an encoder and a decoder. The encoder includes a convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer; the decoder includes a transposed convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer.

[0039] Furthermore, the recursive feature elimination method is specifically as follows: feature set 1 is used as the initial input of the support vector machine model, and the support vector machine model is iterated MN times. In each iteration, the time domain features with the smallest weight are removed through the weight information provided by the support vector machine model. For the time domain features retained after the sth iteration, feature set s+1 is constructed as the input of the support vector machine model for the s+1th iteration until the required number of iterations is reached. The weight information provided by the support vector machine model under the MNth iteration is sorted, and the time domain features represented by the first N weight information are retained as valid features.

[0040] Furthermore, the percentage formula of the effective feature weight is: Among them, |ω j | is the absolute value of the weight of the jth effective feature, is the sum of the absolute values ​​of all valid feature weights.

[0041] Furthermore, the weighted features of the acceleration signal to be detected, the weighted features of the reconstructed signal, and the characteristic error of the acceleration signal to be detected are expressed as follows:

[0042]

[0043] Among them, W j represents the jth effective feature weight, f j is the jth effective feature of the acceleration signal to be detected, F j is the jth effective feature of the reconstructed signal; H F is the characteristic error of the acceleration signal to be detected, H o is the weight feature of the acceleration signal to be detected, H r is the weight feature of the reconstructed signal.

[0044] Furthermore, the state of the industrial robot is judged based on the characteristic error of the acceleration signal to be detected, and the specific expression is:

[0045]

[0046] Among them, UL is the upper boundary, H F is the characteristic error of the acceleration signal to be detected.

[0047] According to the second aspect of an embodiment of the present invention, there is provided an industrial robot anomaly detection device based on recursive feature elimination and autoencoder, comprising a module for executing any one of the above-mentioned industrial robot anomaly detection methods based on recursive feature elimination and autoencoder. Specifically comprising: a first module for executing step 1: according to the selected motion mode, the acceleration signal of the end of the industrial robot under normal motion is collected by the posture sensor installed at the end of the industrial robot, and the acceleration signal is segmented according to the motion cycle to obtain a first preset number of normal sample signals, and a training set and a verification set are divided; a second module for executing step 2: constructing a one-dimensional convolutional autoencoder model, which is used to input the training set and verification set of the normal sample signal into the one-dimensional convolutional autoencoder model for training and verification, and obtain a reconstructed signal of the robot motion trajectory; a third module for executing step 3: extracting M time domain features from the second preset number of normal sample signals; constructing feature set 1 based on the extracted M time domain features; introducing the recursive feature elimination method for effective Feature screening, obtaining a third preset number N of valid features; a fourth module, used to execute step 4: obtaining the percentage of valid feature weights based on the weights corresponding to the obtained valid features; a fifth module, used to execute step 5: calculating the weighted features of the acceleration signal to be detected based on the valid features of the acceleration signal to be detected and the percentage of the valid feature weights; calculating the weighted features of the reconstructed signal based on the valid features of the reconstructed signal obtained in step 2 and the percentage of the valid feature weights; subtracting the weighted features of the acceleration signal to be detected from the weighted features of the reconstructed signal to obtain the feature error of the acceleration signal to be detected; a sixth module, used to execute step 6: judging the state of the industrial robot based on the feature error of the acceleration signal to be detected to realize abnormality detection of the industrial robot.

[0048] According to a third aspect of an embodiment of the present invention, a terminal device is provided, comprising a memory, a processor, and a program stored in the memory and executable by the processor, wherein when the processor executes the program, the industrial robot anomaly detection method based on recursive feature elimination and autoencoder as described in any one of the above is implemented.

[0049] Example 2: Reference Figure 1-Figure 4 As shown, an optional implementation of the present invention is described in detail below, and the specific steps of the method are as follows:

[0050] Step 1: According to the test path published in the industrial robot dynamic stability test method, select the robot's movement as a rectangular trajectory movement in the same plane; based on the selected rectangular trajectory movement mode, collect the terminal acceleration signal of the industrial robot under normal movement through the posture sensor (MTI-630) installed at the end of the industrial robot, and divide the acceleration signal according to the movement period to obtain a first preset number of normal sample signals, and divide them into training set and validation set, of which 90% is used as training set and 10% is used as validation set.

[0051] Specifically, the experiment was conducted on a QJR6-1 six-axis industrial robot, and an MTI-630 attitude sensor (i.e., the acceleration sensor in the figure) was installed at the end flange of the industrial robot. Figure 2 As shown. According to the test path selection in the dynamic stability test method of industrial robots, the movement of the robot is a rectangular trajectory in the same plane. The side length of the rectangle is 500mm, and the movement speed is 250mm / s. The acceleration signal generated by the industrial robot during movement is collected, and the sampling frequency of the attitude sensor is set to 400Hz. In order to avoid the influence of temperature on signal acquisition, the robot moves normally for 30 minutes before collecting signals. The signal in the y direction of each complete cycle is saved as a sample signal, with a total of 5120 points, and a total of 300 sample signals are collected for model training and verification. In the above, the dynamic stability test method of industrial robots is specifically: Chongqing Robotics Society Group Standard, Industrial Robot Dynamic Stability Test Method. [T / CRSS 0007-2023].ICS25.040.30.CCS J28.

[0052] Industrial robot anomalies include short-term anomalies and long-term anomalies. Short-term anomalies are manifested as deviations from normal signal patterns in a short period of time, such as transient shocks. The collected normal sample signals are used to construct an abnormal sample simulation data set. 100 sample signals are randomly selected from 300 sample signals as the basic samples for abnormal sample construction. An abnormal signal with a random shock amplitude is inserted at 4000-4100 points of each basic sample (the shock amplitude range is 2-3m / s 2 ), forming a short-term abnormal sample data set. Long-term abnormality is manifested as abnormal vibration that lasts for a long time. The eccentric motor is installed on the robot's arm, and the start of the eccentric motor is controlled by the robot IO. When the robot arm moves, the eccentric motor starts synchronously, and 100 groups of long-term abnormal signals are collected as the long-term abnormal sample data set.

[0053] Step 2: Construct a one-dimensional convolutional autoencoder model to input the training set and verification set of normal sample signals into the one-dimensional convolutional autoencoder model for training and verification, and obtain the reconstructed signal of the robot's rectangular motion trajectory;

[0054] The one-dimensional convolutional autoencoder model is divided into two parts: encoder and decoder, and its structure is as follows: Figure 3 As shown in the figure, the encoder includes a convolution layer, a ReLU activation function layer, a dropout layer, and a pooling layer; the convolution layer and the pooling layer encode the data to achieve downsampling of the sample data. The decoder includes a transposed convolution layer, a ReLU activation function layer, a dropout layer, and a pooling layer to achieve upsampling of the sample data. After constructing the one-dimensional convolutional neural network autoencoder model, the parameters of the model are determined. Among them, the convolution kernel size of the one-dimensional convolution layer is 7, the step size is 2, and the number of channels is 32; the convolution kernel size of the transposed convolution layer is 7, the step size is 2, and the number of channels is 16.

[0055] Specifically, when training the one-dimensional convolutional autoencoder model, the training set of normal sample signals is used as the input of the encoder, and the output of the encoder is H n =σ0(W0X n +b0), Among them, X n is the original sample signal input, H n is the hidden feature extracted from the encoder, is the output vector reconstructed by the encoder, W1, W2 are weights, b0, b1 are biases, σ0, σ1 are ReLU activation functions; the output of the pooling layer is j = max(y1, y2, y3, …, y n ), where n is the width of the pooling area, y n is the nth data in the pooling area.

[0056] The output of the encoder is used as the input of the decoder. The transposed convolution layer in the decoder is the reverse calculation of the convolution layer. The output of the i-th channel of the transposed convolution layer is: D i =ReLU(∑X*ω i +b i ), where ω i is the transposed convolution kernel, * indicates the transposed convolution calculation, ReLU is the activation layer, b i is the bias; the final decoder output is: Among them, j k The position of the maximum value recorded for the decoder pooling layer.

[0057] When the validation set of normal sample signals is input into the trained one-dimensional convolutional autoencoder model for verification, the mean square error loss MSE function is used. The MSE function measures the error by calculating the difference between the predicted value and the true value. The specific calculation formula is: Among them, C represents the number of samples, X i represents the true value of the normal sample signal, Represents the predicted value obtained by the one-dimensional convolutional autoencoder model. When the value of the Loss function is infinitely close to 0, it means that the model fits the sample signal in the validation set better, that is, the difference between the predicted value and the true value is smaller. At this time, the training is stopped and the final robot reconstruction signal is obtained.

[0058] Step 3: extracting M time domain features from a second preset number of normal sample signals; constructing a feature set 1 based on the extracted time domain features; introducing a recursive feature elimination method to screen effective features, and obtaining a third preset number of N effective features;

[0059] The recursive feature elimination method is specifically:

[0060] Feature set 1 is used as the initial input of the support vector machine model, and the support vector machine model is iterated MN times. In each iteration, the time domain features with the smallest weight are removed through the weight information provided by the support vector machine model. For the time domain features retained after the sth iteration, feature set s+1 is constructed as the input of the support vector machine model for the s+1th iteration, until the required number of iterations is reached, the weight information provided by the support vector machine model under the MNth iteration is sorted, and the time domain features represented by the first N weight information are retained as valid features. Exemplarily, feature set 1 is constructed by 10 time domain features, and the third preset number is 5. Then, the support vector machine model is iterated 5 times. After feature set 1 is input into the support vector machine model for the first iteration, feature set 2 constructed by 9 time domain features is obtained; after feature set 2 is input into the support vector machine model for the second iteration, feature set 3 constructed by 8 time domain features is obtained, and so on. After 5 iterations, feature set 6 constructed by 5 time domain features is finally obtained.

[0061] The M time domain features include mean, variance, standard deviation, maximum, minimum, root mean square, range, median, kurtosis, and skewness. After effective feature screening by recursive feature elimination method, mean, variance, standard deviation, range, and root mean square are used as effective features.

[0062] In the embodiment of the present invention, the first preset number and the second preset number are both 300; if the normal samples originally collected are sufficient, the first preset number and the second preset number can be increased as needed.

[0063] Step 4: According to the weights corresponding to the obtained effective features, the percentage of the effective feature weights is obtained;

[0064] The percentage of effective feature weight can better understand the proportion of each feature in the classification process. The percentage formula of effective feature weight is: Among them, |ω j | is the absolute value of the weight of the jth effective feature, is the sum of the absolute values ​​of all valid feature weights.

[0065] Exemplarily, the following Table 1 compares the weights of each effective feature of 5 iterations. The feature marked with 0 indicates that the feature is the feature that needs to be removed under the current number of iterations. Taking the Iter1 column as an example, after the first iteration, the time domain feature Min with the smallest weight is removed, and feature set 2 is constructed as the input of the second iteration. The same goes for the others. After 5 iterations, the percentage of effective feature weights is obtained based on the weights corresponding to the effective features obtained, as shown in Table 2.

[0066] Table 1 Feature weights during iteration

[0067] feature Iter1 Iter2 Iter3 Iter4 Iter5 Mean 0.0246 0.23909 0.2561 0.0228 0.0246 Variance 0.0315 0.3562 0.0378 0.0441 0.0427 Std 0.037 0.041 0.0436 0.4917 0.0483 Kurtosis <![CDATA[5×10 -5 ]]> 0 0 0 0 Skewnss <![CDATA[3×10 -5 ]]> <![CDATA[7×10 -5 ]]> 0.0001 0 0 Max <![CDATA[8×10 -5 ]]> 0.0001 0 0 0 Min 0 0 0 0 0 Range <![CDATA[5×10 -5 ]]> 0.0002 0.0004 <![CDATA[8×10 -5 ]]> <![CDATA[8×10 -5 ]]> Median 0.0001 0.0003 0.0002 <![CDATA[1×10 -5 ]]> 0 RMS 0.0245 0.0239 0.0256 0.2262 0.0244

[0068] Table 2 Percentage of effective feature weights

[0069] Feature Name Weight Percent Mean 17.38％ Variance 30.77％ Std 34.69％ Range 0.02％ RMS 17.14％

[0070] Step 5: Calculate the weighted features of the acceleration signal to be detected based on the effective features of the acceleration signal to be detected and the percentage of the effective feature weights; calculate the weighted features of the reconstructed signal based on the effective features of the reconstructed signal obtained in step 2 and the percentage of the effective feature weights; subtract the weighted features of the acceleration signal to be detected from the weighted features of the reconstructed signal to obtain the characteristic error of the acceleration signal to be detected;

[0071] The weight feature of the acceleration signal to be detected, the weight feature of the reconstructed signal, and the characteristic error of the acceleration signal to be detected are expressed as follows:

[0072]

[0073] Among them, W j represents the jth effective feature weight, f j is the jth effective feature of the acceleration signal to be detected, F j is the jth effective feature of the reconstructed signal; H F is the characteristic error of the acceleration signal to be detected, H o is the weight feature of the acceleration signal to be detected, H r is the weight feature of the reconstructed signal.

[0074] Step 6: Perform abnormal judgment on the acceleration signal to be detected of the industrial robot. The judgment expression is as follows:

[0075]

[0076] Among them, UL is the upper boundary, H F is the characteristic error of the acceleration signal to be detected.

[0077] That is, after the acceleration signal to be detected is judged, if the characteristic error of the acceleration signal to be detected is less than or equal to the upper boundary, the current state of the robot is considered to be normal; otherwise, the current state of the robot is considered to be abnormal.

[0078] The upper boundary is expressed as follows:

[0079] UL=Q3+1.5IQR

[0080] Among them, Q3 is the upper quartile of the weighted features of the fourth preset number of normal acceleration signals to be detected, and IQR is the quartile interval of the weighted features of the fourth preset number of normal acceleration signals to be detected.

[0081] 100 normal sample signals and 50 short-time abnormal samples are selected as acceleration signals to be detected, and UL=0.2 is obtained based on the upper quartile of the weighted features of the 100 normal acceleration signals to be detected and the quartile interval of the weighted features of the acceleration signals to be detected. Each acceleration signal to be detected is detected by the method of the present invention, and the detection results are as follows: Figure 4 As shown in (1), through Figure 4 (1) The anomaly detection effect of the method of the present invention on normal sample signals and short-time abnormal signals is shown to have an overall accuracy of 87.3%.

[0082] 100 normal sample signals and 50 long-term abnormal samples are selected as acceleration signals to be detected, and UL=0.2 is obtained based on the upper quartile of the weighted features of the 100 normal acceleration signals to be detected and the quartile interval of the weighted features of the acceleration signals to be detected. Each acceleration signal to be detected is detected by the method of the present invention, and the detection results are as follows: Figure 4 As shown in (2), through Figure 4 (2) The anomaly detection effect of the method of the present invention on normal sample signals and long-term abnormal signals can be seen, and the overall accuracy rate is 90%.

[0083] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. An industrial robot anomaly detection method based on recursive feature elimination and autoencoder, characterized in that: The method steps are as follows: Step 1: According to the selected motion mode, the end acceleration signal of the industrial robot under normal motion is collected through the posture sensor installed at the end of the industrial robot, and the acceleration signal is segmented according to the motion cycle to obtain a first preset number of normal sample signals, and the training set and the verification set are divided; Step 2: Construct a one-dimensional convolutional autoencoder model to input the training set and verification set of normal sample signals into the one-dimensional convolutional autoencoder model for training and verification, and obtain the reconstructed signal of the robot's motion trajectory; Step 3: extracting M time domain features from the second preset number of normal sample signals; constructing feature set 1 based on the extracted M time domain features; introducing a recursive feature elimination method to screen effective features, and obtaining a third preset number of N effective features; Step 4: According to the weights corresponding to the obtained effective features, the percentage of the effective feature weights is obtained; Step 5: Calculate the weighted features of the acceleration signal to be detected based on the effective features of the acceleration signal to be detected and the percentage of the effective feature weights; calculate the weighted features of the reconstructed signal based on the effective features of the reconstructed signal obtained in step 2 and the percentage of the effective feature weights; subtract the weighted features of the acceleration signal to be detected from the weighted features of the reconstructed signal to obtain the characteristic error of the acceleration signal to be detected; Step 6: According to the characteristic error of the acceleration signal to be detected, the state of the industrial robot is judged to realize abnormality detection of the industrial robot.

2. The industrial robot anomaly detection method based on recursive feature elimination and autoencoder according to claim 1, characterized in that: The one-dimensional convolutional autoencoder model is divided into two parts: an encoder and a decoder. The encoder includes a convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer; the decoder includes a transposed convolution layer, a ReLU activation function layer, a dropout layer and a pooling layer.

3. The industrial robot anomaly detection method based on recursive feature elimination and autoencoder according to claim 1, characterized in that: The recursive feature elimination method is specifically: Feature set 1 is used as the initial input of the support vector machine model, and the support vector machine model is iterated MN times. In each iteration, the time domain feature with the smallest weight is removed through the weight information provided by the support vector machine model. For the time domain features retained after the sth iteration, feature set s+1 is constructed as the input of the support vector machine model for the s+1th iteration until the required number of iterations is reached. The weight information provided by the support vector machine model under the MNth iteration is sorted, and the time domain features represented by the first N weight information are retained as valid features.

4. The industrial robot anomaly detection method based on recursive feature elimination and autoencoder according to claim 1, characterized in that: The percentage formula of the effective feature weight is: Among them, |ω j | is the absolute value of the weight of the jth effective feature, is the sum of the absolute values ​​of all valid feature weights.

5. The industrial robot anomaly detection method based on recursive feature elimination and autoencoder according to claim 1, characterized in that: The weighted features of the acceleration signal to be detected, the weighted features of the reconstructed signal, and the characteristic error of the acceleration signal to be detected are expressed as follows: Among them, W j represents the jth effective feature weight, f j is the jth effective feature of the acceleration signal to be detected, F j is the jth effective feature of the reconstructed signal; H F is the characteristic error of the acceleration signal to be detected, H o is the weight feature of the acceleration signal to be detected, H r is the weight feature of the reconstructed signal.

6. The industrial robot anomaly detection method based on recursive feature elimination and autoencoder according to claim 1, characterized in that: The state of the industrial robot is judged based on the characteristic error of the acceleration signal to be detected. The specific expression is: Among them, UL is the upper boundary, H F is the characteristic error of the acceleration signal to be detected.

7. An industrial robot anomaly detection device based on recursive feature elimination and autoencoder, characterized in that: The invention comprises a module for executing the industrial robot anomaly detection method based on recursive feature elimination and autoencoder as described in any one of claims 1 to 6.

8. A terminal device, characterized in that: It includes a memory, a processor, and a program stored in the memory and executable by the processor. When the processor executes the program, the industrial robot anomaly detection method based on recursive feature elimination and autoencoder as described in any one of claims 1 to 6 is implemented.

Images