A method for industrial robot fault diagnosis based on similarity measurement

Abstract

The present invention relates to the field of robot fault diagnosis methods, and discloses an industrial robot fault diagnosis method based on similarity measurement, comprising the following steps: collecting original signals and preprocessing the original signals to obtain preprocessed signals, wherein the preprocessing is used to remove noise signals in the original signals; setting reference signals, and segmenting the reference signals and preprocessed signals in one-to-one correspondence; calculating the Manhattan distance values ​​between all segmented reference signals and preprocessed signals, and judging the robot working state through all Manhattan distance values. By segmenting the reference signal and the preprocessed signal, the reference signal and the preprocessed signal are divided, and the divided signal segments are calculated respectively, and the working state of the robot is judged through the Manhattan distance values ​​of the signal segments, that is, comparison is performed from multiple aspects, thereby increasing the accuracy of the calculation results and the accuracy of the judgment.

Description

Technical Field

[0001] The invention relates to the field of robot fault diagnosis methods, and in particular to an industrial robot fault diagnosis method based on similarity measurement. Background Art

[0002] There are currently few fault diagnosis methods for industrial robot applications, and most patents exist in universities, which are far from being put into practice. The vast majority of current academic research on fault diagnosis focuses on the research of "intelligent" diagnostic methods. Deep learning has a difficult-to-cross gap - poor interpretability, so it is difficult to apply in industry.

[0003] Industrial robot fault diagnosis is different from fault diagnosis of high-value-added equipment such as high-speed rail, airplanes, wind power, and nuclear power. If each axis of each robot is equipped with a high-sensitivity vibration sensor, it will cause high economic costs. At the same time, during the operation of the industrial robot, each axis is moving relative to the base coordinate system, which will also cause interference to the vibration data. Therefore, fault diagnosis by collecting robot operation data is the most economical option in the field of industrial robots.

[0004] The sampling frequency of the operating data obtained from the controller is low. The original real signal is a smooth curve, but under the influence of the low sampling rate, the smooth curve is distorted into a sawtooth curve. Filtering is performed on the low sampling rate data to facilitate the subsequent calculation of statistical features.

[0005] Industrial robots are adjusted in real time by PID, and the values ​​change in real time. That is, when the same machine runs the same program, the current data collected twice cannot be completely consistent, so the impact of this deviation needs to be removed.

[0006] There are many interference factors in industrial application sites, which may cause data collection distortion. To ensure the reliability of the data, it is necessary to first check whether the data meets the requirements of subsequent analysis. Summary of the invention

[0007] The present invention aims to solve the above problems and provides an industrial robot fault diagnosis method based on similarity measurement to solve the problems of low accuracy, large amount of calculation and low efficiency of existing diagnosis methods.

[0008] To achieve this object, the present invention adopts the following technical solutions:

[0009] A method for industrial robot fault diagnosis based on similarity measurement includes the following steps:

[0010] Collecting the original signal and preprocessing the original signal to obtain a preprocessed signal, the preprocessing is used to remove the noise signal in the original signal;

[0011] Set a reference signal, and segment the reference signal and the preprocessing signal in one-to-one correspondence;

[0012] Calculate the Manhattan distance values ​​between all segmented reference signals and preprocessed signals, and judge the robot working status through all Manhattan distance values.

[0013] Preferably, the preprocessing method is a wavelet denoising method, which can filter out noise signals in the original signal.

[0014] Preferably, the reference signal and the preprocessing signal are divided into 6 signal segments in a one-to-one correspondence;

[0015] The six signal segments of the reference signal are: forward acceleration segment-reference signal, forward uniform speed segment-reference signal, forward deceleration segment-reference signal, reverse acceleration segment-reference signal, reverse uniform speed segment-reference signal, and reverse deceleration segment-reference signal;

[0016] The preprocessed signal is divided into 6 signal segments: forward acceleration segment-preprocessed signal, forward uniform speed segment-preprocessed signal, forward deceleration segment-preprocessed signal, reverse acceleration segment-preprocessed signal, reverse uniform speed segment-preprocessed signal, and reverse deceleration segment-preprocessed signal.

[0017] Preferably, after the original signal is acquired, the original signal is additionally saved, and a correlation test is performed between the additionally saved original signal and the preprocessed signal;

[0018] A correlation test is performed on the six signal segments divided by the reference signal and the six signal segments divided by the preprocessed signal.

[0019] Preferably, the method used for correlation test is Pearson test.

[0020] Preferably, the six signal segments into which the reference signal is divided are matched one by one with the six signal segments into which the preprocessed signal is divided, and calculations are performed to obtain Manhattan distance values ​​of the six signal segments.

[0021] Preferably, the values ​​calculated by the Manhattan distance value are: the Manhattan distance value of the forward acceleration segment, the Manhattan distance value of the forward uniform speed segment, the Manhattan distance value of the forward deceleration segment, the Manhattan distance value of the reverse acceleration segment, the Manhattan distance value of the reverse uniform speed segment and the Manhattan distance value of the reverse deceleration segment.

[0022] Preferably, the robot working state determination includes the following steps:

[0023] (1) Set the threshold of the Manhattan distance value of the 6-segment signal;

[0024] (2) Calculate the cumulative Manhattan distance value of the Manhattan distance values ​​of the six signal segments;

[0025] (3) Set the threshold of the cumulative Manhattan distance value;

[0026] (4) The Manhattan distance values ​​of the six signal segments and one cumulative Manhattan distance value are compared with their corresponding thresholds to determine the working status of the robot.

[0027] The contribution of the present invention is that the present invention divides the reference signal and the preprocessing signal into segments, calculates each segment of the signal respectively, and judges the working state of the robot by the Manhattan distance value of each segment of the signal, that is, compares from multiple aspects, thereby increasing the accuracy of the calculation result and the accuracy of the judgment.

[0028] Manhattan distance is used to calculate and judge the robot state. Compared with Euclidean distance, the present invention has a smaller amount of calculation. Since the amount of calculation is greatly reduced, the running speed is also faster, and the working state of the robot can be judged more quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic flow chart of the robot fault diagnosis method of the present invention;

[0030] Figure 2 It is a schematic diagram of the correlation test method of the reference signal and the preprocessing signal before and after segmentation according to the present invention. DETAILED DESCRIPTION

[0031] The following examples are provided to further explain and supplement the present invention and do not constitute any limitation to the present invention.

[0032] like Figure 1 As shown, an industrial robot fault diagnosis method based on similarity measurement includes the following steps:

[0033] Collecting the original signal and preprocessing the original signal to obtain a preprocessed signal, the preprocessing is used to remove the noise signal in the original signal;

[0034] Set a reference signal, and segment the reference signal and the preprocessing signal in one-to-one correspondence;

[0035] Calculate the Manhattan distance values ​​between all segmented reference signals and preprocessed signals, and judge the robot working status through all Manhattan distance values.

[0036] The collection of original signals can be achieved through scripts. Specifically, when the robot is running, the robot's operating data signal will be fed back to the servo driver through the encoder. By writing a script externally, the operating data signal of the robot in the servo driver can be obtained. The present invention generally collects more than two groups of original signals at a time.

[0037] After collecting the original signal, it needs to be preprocessed. Since the original signal contains some noise signals, if these noise signals are not removed, it will greatly affect the subsequent calculation results of the Manhattan distance value. The noise signals in the original signal can be removed by preprocessing to facilitate subsequent segmentation and calculation.

[0038] A reference signal is set, where the reference signal is the signal when the robot works normally. The reference signal and the preprocessed signal are segmented and processed, and the Manhattan distance value between the segmented reference signal and the preprocessed signal is calculated. The present invention divides the reference signal and the preprocessed signal into segments, and calculates each segment of the divided signal separately. The working state of the robot is judged by the Manhattan distance value of each segment of the signal, that is, comparison is performed from multiple aspects, thereby increasing the accuracy of the calculation result and the accuracy of the judgment.

[0039] Manhattan distance is used to calculate and judge the robot state. Compared with Euclidean distance, the present invention has a smaller amount of calculation. Since the amount of calculation is greatly reduced, the running speed is also faster, and the working state of the robot can be judged more quickly.

[0040] There are two main methods for existing fault diagnosis. The first method, such as a method for identifying abnormal tool signals with patent number CN116079498A, selects the primary segment to be identified in the initial signal and the primary comparison segment in the standard template signal, and performs Manhattan calculation on all matching points in the two. Due to data differences and regional causes of faults (such as some faults are obvious in the acceleration segment but not in the uniform speed segment), the judgment results obtained by this method are not accurate, and the working status of the tool cannot be accurately judged, and there are many false alarms.

[0041] The second method, such as the rolling bearing fault diagnosis method based on wavelet decomposition and improved Manhattan distance with patent number CN111222495A, selects specific working conditions, analyzes and compares the specific working conditions to determine the state of the bearing. This method has certain limitations and can only determine a few specific working conditions. If other faults occur, it cannot be determined.

[0042] Both of the above methods have certain limitations. The present invention divides the reference signal and the preprocessing signal into segments, and calculates each segment of the signal respectively, and judges the working status of the robot by the Manhattan distance value of each segment of the signal, that is, comparing from multiple aspects, thereby increasing the accuracy of the calculation results and the accuracy of the judgment. On the other hand, the present invention is not limited to the diagnosis of several faults. As long as the robot has a fault abnormality, it can be discovered, and there is no problem of diagnostic limitation.

[0043] The preprocessing method used is the wavelet denoising method, which can filter out the noise signal in the original signal.

[0044] Wavelet denoising is a time-frequency localization analysis method with a fixed window area, variable window shape, and variable time window and frequency window. Wavelet denoising has a higher frequency resolution and a lower time resolution in the low-frequency part, and a higher time resolution and a lower frequency resolution in the high-frequency part. It is very suitable for detecting the components of mutation signals in normal signals. It can use a long time interval to obtain more detailed low-frequency signal information, and a short time interval to obtain high-frequency signal information. In actual engineering applications, the analyzed signal may contain many spikes or mutation parts, and the noise is not a stable white noise.

[0045] Wavelet denoising can decompose the various frequency components in the signal into non-overlapping frequency bands, providing an effective way for signal filtering, signal-noise separation and feature extraction.

[0046] The basic idea of ​​wavelet denoising is that after the signal is transformed by wavelet (using Mallat algorithm), the wavelet coefficients generated by the signal contain important information of the signal. After the signal is decomposed by wavelet, the wavelet coefficients are larger, and the wavelet coefficients of the noise are smaller, and the wavelet coefficients of the noise are smaller than the wavelet coefficients of the signal. By selecting a suitable threshold, the wavelet coefficients greater than the threshold are considered to be generated by the signal and should be retained, and those less than the threshold are considered to be generated by noise and set to zero to achieve the purpose of denoising.

[0047] From the perspective of signal science, wavelet denoising is a problem of signal filtering. Although wavelet denoising can be regarded as low-pass filtering to a large extent, it is superior to traditional low-pass filters in that it can successfully retain signal characteristics after denoising.

[0048] The reference signal and the pre-processed signal are divided into 6 signal segments one by one;

[0049] The six signal segments of the reference signal are: forward acceleration segment-reference signal, forward uniform speed segment-reference signal, forward deceleration segment-reference signal, reverse acceleration segment-reference signal, reverse uniform speed segment-reference signal, and reverse deceleration segment-reference signal;

[0050] The preprocessed signal is divided into 6 signal segments: forward acceleration segment-preprocessed signal, forward uniform speed segment-preprocessed signal, forward deceleration segment-preprocessed signal, reverse acceleration segment-preprocessed signal, reverse uniform speed segment-preprocessed signal, and reverse deceleration segment-preprocessed signal.

[0051] According to the six working conditions of the robot, the reference signal and the pre-processing signal can be divided into six signal segments respectively, namely forward acceleration segment, forward uniform speed segment, forward deceleration segment, reverse acceleration segment, reverse uniform speed segment and reverse deceleration segment (the lengths of each signal segment divided by the reference signal and the pre-processing signal are one-to-one corresponding). Through the above six working conditions, the robot's status is comprehensively analyzed and calculated to ensure that the robot's working status can be judged from multiple aspects to improve the accuracy of the judgment.

[0052] Specifically, the segmentation process is as follows:

[0053] Taking the preprocessed signal as an example, the preprocessed signal includes several data signal segments:

[0054]

[0055] Forward acceleration stage - pre-processing

[0056] If the values ​​of the five consecutive items starting from diff_v(i) are positive, and the values ​​of the six vcmd(i) items corresponding to the five diff_v(i) items are all greater than 0, it means that the segment starts from here as a positive acceleration; when the first five items of diff_v(j) are positive, and the values ​​of the first and last four items of diff_v(j) are non-positive, it is judged that the positive acceleration segment ends at diff_v(j), that is, the data signal segment corresponding to the segment from vcmd(i) to vcmd(j+1) is extracted as the forward acceleration segment-preprocessing signal.

[0057] The data signal segments contained in the forward acceleration segment-preprocessing signal are marked as: forward acceleration segment-1, forward acceleration segment-2, ..., forward acceleration segment-n.

[0058] Forward uniform speed section-pretreatment

[0059] If the values ​​starting from diff_v(i) are all 0 for five consecutive items, and the values ​​of the six vcmd(i) items corresponding to the five diff_v(i) items are all greater than 0, it means that the segment is in the positive uniform speed segment, that is, the data signal segment corresponding to the segment from vcmd(i) to vcmd(j+1) is extracted as the positive uniform speed segment-preprocessing signal.

[0060] The data signal segments contained in the forward uniform speed segment-preprocessing signal are marked as: forward uniform speed segment-1, forward uniform speed segment-2, ..., forward uniform speed segment-n.

[0061] Forward deceleration section-pretreatment

[0062] If the values ​​of the five consecutive items starting from diff_v(i) are negative, and the values ​​of the six vcmd (i) items corresponding to the five diff_v(i) items are all greater than 0, it means that the segment starts from here as a positive deceleration; when the first five items of diff_v(j) are negative, and the values ​​of the first and last four items of diff_v(j) are non-negative, it is judged that the positive deceleration segment-preprocessing ends at diff_v(j), that is, the data signal segment corresponding to the vcmd (i) to vcmd (j+1) segment is extracted as the positive deceleration-preprocessing signal.

[0063] The data signal segments included in the forward deceleration segment-preprocessing signal are marked as: forward deceleration segment-1, forward deceleration segment-2, ..., forward deceleration segment-n.

[0064] Similarly, the reverse acceleration segment-preprocessing signal, the reverse uniform speed segment-preprocessing signal and the reverse deceleration segment-preprocessing signal of the preprocessing signal can be obtained.

[0065] Using the above segmentation method, the forward acceleration segment-reference signal, the forward uniform speed segment-reference signal, the forward deceleration segment-reference signal, the reverse acceleration segment-reference signal, the reverse uniform speed segment-reference signal, and the reverse deceleration segment-reference signal can also be obtained after the reference signal is segmented.

[0066] like Figure 2 As shown, after the original signal is obtained, the original signal is additionally saved, and a correlation test is performed between the additionally saved original signal and the preprocessed signal;

[0067] A correlation test is performed on the six signal segments divided by the reference signal and the six signal segments divided by the preprocessed signal.

[0068] The method used for correlation test is Pearson test.

[0069] The purpose of correlation test is to ensure the reliability of the signal and avoid signal distortion. Specifically, in the present invention, each time a signal is processed, a correlation test is performed on the signal to ensure that the signal calculated subsequently will not be distorted. The method used for correlation test in the present invention is the Pearson test method.

[0070] To further explain, after the original signal is processed by wavelet denoising, it is necessary to perform a correlation test on the original signal and the preprocessed signal (after passing the test, the original signal and the preprocessed signal are segmented one by one) to avoid signal distortion. The correlation detection coefficient is 0.95, that is, when the correlation between the original signal and the preprocessed signal is not less than 0.95, it can be considered that the original signal and the preprocessed signal meet the consistency requirement. Otherwise, the consistency requirement is not met, the correlation test between the original signal and the preprocessed signal fails, and the group of original signals and preprocessed signals needs to be discarded.

[0071] Similarly, after the reference signal and the preprocessed signal are segmented, a correlation test is also required to avoid signal distortion. The correlation test system is also 0.95, that is, when the correlation between the original signal and the preprocessed signal is not less than 0.95, it can be considered that the original signal and the preprocessed signal meet the consistency requirement. Otherwise, the consistency requirement is not met and the group of reference signals, preprocessed signals, original signals and preprocessed signals are discarded.

[0072] The six signal segments divided by the reference signal and the six signal segments divided by the preprocessed signal are matched one by one and calculated to obtain the Manhattan distance values ​​of the six signal segments.

[0073] The calculation formula of the Manhattan distance value of the 6-segment signal is as follows:

[0074]

[0075] in, is the Manhattan distance between the 6 segments of the preprocessed signal and the 6 segments of the reference signal, m=1, 2, 3, 4, 5, 6, 1-6 represent 6 working conditions respectively. For example, when m=1, It represents the Manhattan distance between the forward acceleration segment-preprocessing and the forward acceleration segment-benchmark. To preprocess the data signal segments under various working conditions, is the data signal segment of each working condition of the reference signal corresponding to each working condition of the preprocessing, and n is the data signal segment corresponding to each working condition. For example, Forward acceleration segment-preprocessed forward acceleration segment-1, It is the forward acceleration segment-1 of the forward acceleration segment-baseline.

[0076] The values ​​calculated by the Manhattan distance value are: the Manhattan distance value of the forward acceleration segment, the Manhattan distance value of the forward uniform speed segment, the Manhattan distance value of the forward deceleration segment, the Manhattan distance value of the reverse acceleration segment, the Manhattan distance value of the reverse uniform speed segment and the Manhattan distance value of the reverse deceleration segment.

[0077] The Manhattan distance values ​​under 6 working conditions are obtained, and the robot status is judged according to different working conditions to increase the accuracy.

[0078] The robot working status judgment includes the following steps:

[0079] (1) Set the threshold of the Manhattan distance value of the 6-segment signal;

[0080] (2) Calculate the cumulative Manhattan distance value of the Manhattan distance values ​​of the six signal segments;

[0081] (3) Set the threshold of the cumulative Manhattan distance value;

[0082] (4) The Manhattan distance values ​​of the six signal segments and one cumulative Manhattan distance value are compared with their corresponding thresholds to determine the working status of the robot.

[0083] In order to further improve the accuracy of judgment, before judging the working state of the robot, it is also necessary to calculate the cumulative Manhattan distance value, a total of 7 Manhattan distance values, namely, the Manhattan distance value of the forward acceleration segment, the Manhattan distance value of the forward uniform speed segment, the Manhattan distance value of the forward deceleration segment, the Manhattan distance value of the reverse acceleration segment, the Manhattan distance value of the reverse uniform speed segment, the Manhattan distance value of the reverse deceleration segment and the cumulative Manhattan distance value;

[0084] After obtaining 7 Manhattan distance values, thresholds of the 7 Manhattan distance values ​​are set, and the 7 Manhattan distance values ​​are compared with their corresponding thresholds respectively. When 4 or more of the 7 Manhattan distance values ​​are greater than the corresponding thresholds, it indicates that the robot's working state is abnormal, otherwise it is normal.

[0085] Although the present invention is disclosed through the above embodiments, the protection scope of the present invention is not limited thereto. Without departing from the concept of the present invention, any deformation, replacement, etc. made to the above components shall fall within the scope of the claims of the present invention.

Claims

1. A method for industrial robot fault diagnosis based on similarity measurement, characterized in that: The steps include: Collecting the original signal and preprocessing the original signal to obtain a preprocessed signal, the preprocessing is used to remove the noise signal in the original signal; Set a reference signal, and segment the reference signal and the preprocessing signal in one-to-one correspondence; Calculate the Manhattan distance values ​​between all segmented reference signals and preprocessed signals, and judge the robot working status through all Manhattan distance values; Among them, the preprocessing method used is the wavelet denoising method, which can filter out the noise signal in the original signal; In addition, the reference signal and the pre-processed signal are divided into 6 signal segments corresponding to each other. The 6 signal segments of the reference signal are: forward acceleration segment-reference signal, forward uniform speed segment-reference signal, forward deceleration segment-reference signal, reverse acceleration segment-reference signal, reverse uniform speed segment-reference signal, and reverse deceleration segment-reference signal; the 6 signal segments of the pre-processed signal are: forward acceleration segment-pre-processed signal, forward uniform speed segment-pre-processed signal, forward deceleration segment-pre-processed signal, reverse acceleration segment-pre-processed signal, reverse uniform speed segment-pre-processed signal, and reverse deceleration segment-pre-processed signal. After the original signal is obtained, the original signal is additionally saved, and a correlation test is performed between the additionally saved original signal and the preprocessed signal; a correlation test is performed between the six signal segments divided by the reference signal and the six signal segments divided by the preprocessed signal, and the correlation test method adopted is the Pearson test method; The six signal segments divided by the reference signal and the six signal segments divided by the preprocessed signal are matched one by one and calculated to obtain the Manhattan distance values ​​of the six signal segments.

2. The industrial robot fault diagnosis method based on similarity measurement according to claim 1 is characterized in that: The values ​​calculated by the Manhattan distance value are: the Manhattan distance value of the forward acceleration segment, the Manhattan distance value of the forward uniform speed segment, the Manhattan distance value of the forward deceleration segment, the Manhattan distance value of the reverse acceleration segment, the Manhattan distance value of the reverse uniform speed segment and the Manhattan distance value of the reverse deceleration segment.

3. The industrial robot fault diagnosis method based on similarity measurement according to claim 2 is characterized in that: The robot working status judgment includes the following steps: (1) Set the threshold of the Manhattan distance value of the 6-segment signal; (2) Calculate the cumulative Manhattan distance value of the Manhattan distance values ​​of the six signal segments; (3) Set the threshold of the cumulative Manhattan distance value; (4) The Manhattan distance values ​​of the six signal segments and one cumulative Manhattan distance value are compared with their corresponding thresholds to determine the working status of the robot.

Images