Gear motor fault detection method and system based on vibration signal

Abstract

The present application relates to the field of fault detection, and more particularly to a gear motor fault detection method and system based on vibration signals, the method comprising: collecting vibration signal sequences, load parameter sequences, rotation speed sequences and fault labels; performing multi-modal decomposition on the vibration signal sequences to obtain sub-sequences of a plurality of vibration frequency bands, and calculating the contribution degree of a target frequency band; calculating the relevance of the target frequency band and the target fault based on the first cooperativity and the second cooperativity; obtaining the importance weight of the target frequency band in the target fault to obtain a weighted sub-sequence, and taking the sum of all weighted sub-sequences as the reconstructed vibration signal sequence of any gear motor; inputting the obtained three-dimensional matrix of the gear motor to be detected into the trained fault detection model, and outputting the predicted value of the gear motor to be detected in the target fault. Through the technical scheme of the present application, the accuracy of the gear motor fault detection result can be improved.

Description

Technical Field

[0001] This invention relates to the field of fault detection, and more particularly to a method and system for fault detection of gear motors based on vibration signals. Background Technology

[0002] A geared motor is an integrated power actuator that combines a power motor and a gear transmission mechanism. It is a core power component in equipment such as substation inspection robots, photovoltaic support drive systems, and precision machine tools. During long-term operation, the operating status of the geared motor directly affects the reliability and safety of the entire equipment. Therefore, timely and accurate fault detection of the geared motor is of great importance. Different mechanical faults cause changes in the characteristics of vibration signals, and these vibration signals can be collected in real time by non-contact sensors without affecting the normal operation of the equipment, thus enabling geared motor fault detection.

[0003] A Chinese patent application with publication number CN118242331A discloses a method for detecting the mechanical condition of a hydraulic motor, comprising: acquiring vibration data; decomposing the vibration data to obtain component signals; performing frequency domain transformation on the component signals to obtain several frequency bandwidths of the component signals; obtaining the uniformity characteristics of each component signal; obtaining the optimal decomposition scale of the vibration data; obtaining the main frequency of the vibration data, local data segments of the vibration data, and the main frequency of the local data segments based on the optimal decomposition scale; obtaining the importance of the main frequency of the local data segments and the reliability of the local data segments; obtaining filtered vibration data based on the importance and reliability, and making anomaly judgments on the mechanical condition of the hydraulic motor.

[0004] However, vibration signals are cross-coupled with rotational speed parameters and load parameters. Detection methods that rely solely on vibration signals or do not consider the correlation between parameters cannot effectively separate fault characteristics in different frequency bands, resulting in low accuracy of fault detection results. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a gear motor fault detection method based on vibration signals, comprising: acquiring a vibration signal sequence, a load parameter sequence, and a speed sequence of any gear motor over a preset time period; taking any fault type as a target fault; obtaining a fault label for the target fault of any gear motor over the preset time period; performing multimodal decomposition on the vibration signal sequence to obtain subsequences of several vibration frequency bands; taking any vibration frequency band as a target frequency band; calculating the contribution of the target frequency band; calculating the first synergy between the subsequence of the target frequency band and the load parameter sequence; similarly obtaining the second synergy between the subsequence of the target frequency band and the speed sequence; and calculating the target frequency band based on the first and second synergies. The correlation between the frequency band and the target fault is determined. The product of the contribution and the correlation is used as the importance weight of the target frequency band in the target fault. The subsequences of the target frequency band are weighted using the importance weight to obtain weighted subsequences. The weighted subsequences of any gear motor in each vibration frequency band are obtained by traversing the frequency band. The sum of all weighted subsequences is used as the reconstructed vibration signal sequence of any gear motor. The reconstructed vibration signal sequence, load parameter sequence, and speed sequence are constructed into a three-dimensional matrix. The fault detection model is trained using the three-dimensional matrix. The obtained three-dimensional matrix of the gear motor to be detected is input into the trained fault detection model, and the predicted value of the gear motor to be detected in the target fault is output, thus completing the fault detection.

[0007] Preferably, the step of obtaining the fault label of any gear motor in a preset time period includes: in response to any gear motor having a target fault in the preset time period, the fault label value is 1; in response to any gear motor not having a target fault in the preset time period, the fault label value is 0.

[0008] Preferably, the calculation of the contribution of the target frequency band includes: obtaining all extreme points in the subsequence of the target frequency band, the extreme points including maxima and minima; calculating the time interval between any two adjacent extreme points; constructing all time intervals into a time interval sequence and dividing it into interval intervals; statistically analyzing the probability distribution of each interval interval; using the entropy value of the time interval sequence as the extreme interval distribution entropy feature; traversing to obtain the extreme interval distribution entropy feature of each gear motor in the target frequency band; discretizing all extreme interval distribution entropy features in the target frequency band using equidistant interval processing; and calculating the mutual information value between the target frequency band and the target fault; using the normalized mutual information value as the contribution of the target frequency band.

[0009] Preferably, the calculation of the first synergy between the subsequence of the target frequency band and the load parameter sequence includes: calculating the cross-correlation function between the subsequence of the target frequency band and the load parameter sequence, and taking the absolute value of the maximum value in the cross-correlation function as the first synergy between the subsequence of the target frequency band and the load parameter sequence.

[0010] Preferably, the calculation of the correlation between the target frequency band and the target fault includes: dividing the gear motors into a fault group and a non-fault group according to the values ​​of all gear motors at the target fault; calculating the mean of all first synergies in the fault group as the first mean, calculating the mean of all first synergies in the non-fault group as the second mean, and calculating the first absolute difference between the first mean and the second mean; calculating the mean of all second synergies in the fault group as the third mean, calculating the mean of all second synergies in the non-fault group as the fourth mean, and calculating the second absolute difference between the third mean and the fourth mean; obtaining the first standard deviation of all first synergies in the fault group, obtaining the second standard deviation of all second synergies in the fault group; calculating the sum of the first mean and the third mean as the first sum, calculating the sum of the first absolute difference and the second absolute difference as the second sum, using the product of the first sum and the second sum as the numerator, using the sum of the first standard deviation, the second standard deviation, and the preset adjustment coefficient as the denominator, and using the normalized result of the ratio of the numerator to the denominator as the correlation between the target frequency band and the target fault.

[0011] Preferably, the step of weighting the subsequences of the target frequency band using importance weights to obtain weighted subsequences includes: multiplying the importance weights by each element in the subsequences of the target frequency band to obtain weighted subsequences.

[0012] Preferably, the fault detection model is an LSTM model. The input of the LSTM model is a three-dimensional matrix of any gear motor in the history, and the output is the predicted value of the fault label of any gear motor in the history. The label is the true value of the fault label of any gear motor in the history. The loss function of the LSTM model is the cross-entropy loss function.

[0013] Secondly, the present invention also provides a gear motor fault detection system based on vibration signals, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned gear motor fault detection method based on vibration signals is implemented.

[0014] The present invention has the following effects:

[0015] Compared with existing technologies, this invention has significant advantages. First, it effectively solves the technical problem of difficulty in separating fault features due to the cross-coupling between vibration signals and speed and load parameters. By introducing a dual evaluation mechanism of contribution and correlation, it not only quantifies the sensitivity of each vibration frequency band to the fault itself, but also deeply explores the changing law of the dynamic coupling relationship between vibration signals and operating parameters under fault conditions, thereby accurately identifying key frequency bands that contain rich fault information and can effectively distinguish operating interference. Second, the signal reconstruction process based on importance weights is essentially an adaptive filtering enhancement operation, which significantly suppresses noise components unrelated to the fault and fluctuations in normal operating conditions, greatly improving the signal-to-noise ratio and feature discrimination power of the reconstructed signal. Finally, combined with the ability of the LSTM model to mine deep correlations of multi-source parameter time series, this invention can achieve high-precision fault detection in complex and ever-changing operating environments, improving the robustness and generalization ability of the detection system. Attached Figure Description

[0016] Figure 1 This is a flowchart of a gear motor fault detection method based on vibration signals according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Reference Figure 1 The gear motor fault detection method based on vibration signals includes steps S1-S5, as detailed below:

[0020] S1: Collect the vibration signal sequence, load parameter sequence and speed sequence of any gear motor in a preset time period, take any fault type as the target fault, and obtain the fault label of the target fault of any gear motor in the preset time period.

[0021] In one embodiment, multiple gear motors with different operating states (including normal operating conditions and various fault conditions) from the past are selected, and a unified preset time period is set. Based on this, the vibration signal sequence, load parameter sequence, and speed sequence of each historical gear motor within the preset time period are collected. The vibration signal sequence is composed of acceleration signals collected by vibration sensors, which is used to reflect the vibration response characteristics of the gear motor during operation. The load parameter sequence and speed sequence together reflect the operating conditions of the gear motor, providing a condition benchmark for subsequent fault characteristic analysis.

[0022] Take any fault type as the target fault, obtain the fault label of the target fault of any gear motor in a preset time period. In response to the occurrence of the target fault in any gear motor in the preset time period, the fault label value is 1; in response to the absence of the target fault in any gear motor in the preset time period, the fault label value is 0.

[0023] S2: Perform multimodal decomposition on the vibration signal sequence to obtain subsequences of several vibration frequency bands, take any vibration frequency band as the target frequency band, and calculate the contribution of the target frequency band.

[0024] In one embodiment, a multimodal decomposition method is used to process the vibration signal sequence of any gear motor during a preset time period. The preset number of decomposition layers (exemplarily 6 layers) is used. Utilizing the variational optimization framework of multimodal decomposition, the center frequency and bandwidth of each mode are adaptively determined by constructing and solving a constrained variational model. This decomposes the original complex non-stationary vibration signal sequence into several independent subsequences of vibration frequency bands. Each subsequence represents a vibration component within a specific frequency band range, achieving effective separation of the vibration signal in the frequency domain and avoiding mutual interference between components of different frequency bands.

[0025] It should be noted that due to the complexity of the gear motor's operating environment, the vibration signal sequence not only contains fault characteristics but is also heavily influenced by fluctuations in the load parameter sequence and speed sequence. This results in severe cross-coupling between the vibration signal sequence and the load parameter and speed sequences, making it difficult for traditional methods to isolate operating condition interference. Furthermore, the fault characterization capabilities of different vibration frequency bands vary significantly for different target faults; some frequency bands may only reflect noise or normal operating condition changes rather than the essence of the fault. Therefore, step S2 identifies key frequency bands that are both sensitive to the target fault and can effectively distinguish operating conditions by quantifying the contribution of each target frequency band to the target fault and its correlation with operating parameters. This highlights effective features and suppresses coupling interference when generating the reconstructed vibration signal sequence, resolving the feature ambiguity problem caused by signal aliasing.

[0026] Taking any vibration frequency band as the target frequency band, the extreme value interval distribution entropy feature is extracted, including: traversing all sampling points in the subsequence of the target frequency band, identifying all local extreme points in the subsequence by calculating the second derivative, covering both maxima and minima; then calculating the time interval between any two adjacent extreme points, constructing all time intervals into a time interval sequence to reflect the periodic variation law of the vibration waveform; next, dividing the time interval sequence into intervals, counting the frequency of extreme value intervals in each interval, and then calculating the probability distribution of each interval; finally, using the Shannon entropy formula to calculate the entropy value of the time interval sequence as the extreme value interval distribution entropy feature.

[0027] It needs to be explained that under normal operating conditions, the vibration signal sequence of the gear motor exhibits a relatively stable periodic fluctuation pattern. The time intervals between adjacent extreme points are relatively concentrated and highly regular, resulting in a low extreme interval distribution entropy characteristic. However, when a target fault occurs in the gear motor, the impact, wear, or meshing abnormalities caused by the fault disrupt the original vibration periodicity, leading to a random shift in the timing of extreme points. The time intervals between adjacent extreme points fluctuate more widely and become more discrete, thus significantly increasing the extreme interval distribution entropy characteristic value. Since Shannon entropy is essentially a measure of the uncertainty or complexity of a sequence, the change in the extreme interval distribution entropy characteristic can intuitively reflect the degree to which the subsequence of the target frequency band is significantly affected by the target fault in the time domain structure. The higher the value of the extreme interval distribution entropy characteristic, the richer the fault-sensitive information contained in the target frequency band, and the stronger the characterization ability of the target fault.

[0028] To quantify the direct characterization capability of the subsequences of the target frequency band for the target fault, it is necessary to calculate the contribution of the target frequency band. This includes: obtaining the extreme value interval distribution entropy feature corresponding to each gear motor; discretizing the extreme value interval distribution entropy feature based on the value range of all gear motors using equidistant interval processing, where the number of intervals can be preset according to the data scale (e.g., 8 intervals), with each interval corresponding to a discrete category, thereby transforming the continuous extreme value interval distribution entropy feature into a discrete feature to meet the requirement of variable discreteness for mutual information calculation; subsequently, combining the discrete extreme value interval distribution entropy feature corresponding to the target frequency band with the fault label corresponding to the target fault, calculating the mutual information value between the target frequency band and the target fault. The mutual information value essentially measures the statistical dependence between the target frequency band and the target fault, i.e., the amount of effective information about the fault state contained in the feature; furthermore, to eliminate the dimensional differences of the mutual information value between different frequency bands and facilitate subsequent fusion with correlation, the mutual information value is normalized by maximum and minimum values, and the normalized mutual information value is used as the contribution of the target frequency band.

[0029] The larger the contribution value, the stronger the correlation between the extreme interval distribution entropy characteristics of the target frequency band and the target fault, the more prominent the sensitivity and discrimination ability of the target frequency band to the target fault in the time domain structure, and the more significant the direct contribution to identifying the target fault.

[0030] S3: Calculate the first coherence between the subsequence of the target frequency band and the load parameter sequence. Similarly, obtain the second coherence between the subsequence of the target frequency band and the rotational speed sequence. Calculate the correlation between the target frequency band and the target fault based on the first and second coherence.

[0031] In one embodiment, the cross-correlation function between the subsequence of the target frequency band and the load parameter sequence is calculated, and the absolute value of the maximum value in the cross-correlation function is taken as the first coherence between the subsequence of the target frequency band and the load parameter sequence. Similarly, the second coherence between the subsequence of the target frequency band and the rotational speed sequence is obtained.

[0032] It should be explained that vibration signal sequences, speed sequences, and load parameter sequences often have inherent time lags. If only linear correlation analysis is used, it is difficult to accurately capture the dynamic coupling relationship between vibration signal sequences and speed sequences, and between vibration signal sequences and load parameter sequences. Therefore, step S3 uses cross-correlation analysis to characterize the degree of correlation between the sub-sequences of the target frequency band and the operating parameters.

[0033] The larger the absolute value of the synergy, the tighter the coupling relationship between the vibration component of the target frequency band and the corresponding operating parameters. By quantifying the first synergy between the subsequence of the target frequency band and the load parameter sequence, and the second synergy between the subsequence of the target frequency band and the speed sequence during the operation of different gear motors, the dynamic correlation characteristics between the vibration characteristics and operating parameters of the gear motor under different operating conditions can be effectively reflected. This ensures that the target fault extraction can adapt to the time shift caused by changes in operating conditions, avoids the distortion of correlation features caused by ignoring time lag, and thus improves the robustness of the fault detection model to operating condition coupling interference.

[0034] Specifically, gear motors are divided into faulty and non-faulty groups based on the value of the target fault for all gear motors.

[0035] Calculate the mean of all first synergies in the fault group as the first mean, calculate the mean of all first synergies in the non-fault group as the second mean, and calculate the first absolute difference between the first mean and the second mean. Calculate the mean of all second synergies in the fault group as the third mean, calculate the mean of all second synergies in the non-fault group as the fourth mean, and calculate the second absolute difference between the third mean and the fourth mean.

[0036] Obtain the first standard deviation of all first coherences in the fault group, and obtain the second standard deviation of all second coherences in the fault group.

[0037] The sum of the first mean and the third mean is calculated as the first sum, and the sum of the first absolute difference and the second absolute difference is calculated as the second sum. The product of the first sum and the second sum is used as the numerator, and the sum of the first standard deviation, the second standard deviation, and the preset adjustment coefficient is used as the denominator. The normalized result of the ratio of the numerator to the denominator is used as the correlation between the target frequency band and the target fault. For example, the preset adjustment coefficient is 0.001, which is used to prevent the denominator from being 0.

[0038] It should be explained that the first sum directly reflects the overall strength of the coupling relationship between the target frequency band and the operating parameters under the target fault. The larger the first sum, the more significant the coordinated response of the vibration component with the load and speed when the fault occurs. The second sum quantifies the offset of the correlation characteristics before and after the introduction of the fault state. The larger the second sum, the stronger the sensitivity of the target frequency band to the fault and the more prominent its distinguishing ability. The denominator characterizes the dispersion and stability of the correlation characteristics under the target fault. The smaller the standard deviation, the more consistent and less volatile the coupling relationship between the target frequency band and the operating parameters is under the target fault.

[0039] S4: The product of contribution and correlation is used as the importance weight of the target frequency band in the target fault. The subsequence of the target frequency band is weighted using the importance weight to obtain the weighted subsequence. The weighted subsequence of any gear motor in each vibration frequency band is obtained by traversing the frequency band. The sum of all weighted subsequences is used as the reconstructed vibration signal sequence of any gear motor.

[0040] In one embodiment, based on the calculated contribution and correlation, the product of the two is used as the importance weight of the target frequency band under the target fault. The importance weight comprehensively quantifies the frequency band's ability to directly characterize the fault and its distinguishing reliability under operating condition coupling. Subsequently, the subsequence of the target frequency band is weighted using the importance weight. Specifically, the importance weight is multiplied by each data element in the subsequence one by one to obtain a weighted subsequence containing fault feature enhancement information.

[0041] By traversing all vibration frequency bands obtained after multimodal decomposition of any gear motor, the weighted subsequence corresponding to each vibration frequency band is obtained, and all weighted subsequences are summed to synthesize the reconstructed vibration signal sequence of any gear motor.

[0042] Through the weighted reconstruction process, the noise components and fault-independent operating condition coupling components that were originally mixed in the original signal are suppressed due to their low importance weight, while the frequency band components that are sensitive to the target fault and have high discrimination are retained and enhanced due to their high weight. The final reconstructed vibration signal sequence has a higher signal-to-noise ratio and feature discrimination power, and can be used together with the load parameter sequence and speed sequence to construct a three-dimensional matrix, which can be used as input to the fault detection model to achieve high-precision fault prediction.

[0043] S5: Construct a three-dimensional matrix from the reconstructed vibration signal sequence, load parameter sequence, and speed sequence. Use the three-dimensional matrix to train a fault detection model. Input the obtained three-dimensional matrix of the gear motor to be tested into the trained fault detection model and output the predicted value of the gear motor to be tested at the target fault, thus completing the fault detection.

[0044] In one embodiment, the fault detection model is an LSTM (Long Short Term Memory) model. The input layer receives a three-dimensional matrix of any gear motor in the history, constructed from the reconstructed vibration signal sequence, load parameter sequence, and speed sequence, aiming to fully explore the deep correlation features of multi-source parameters in the time dimension. After processing by the internal network, the model generates a predicted value of the fault label of any gear motor in the history at the output layer. This predicted value represents the model's estimate of the probability of the target fault existing under the current working condition and is compared with the true value of the fault label of any gear motor in the history as a supervision signal.

[0045] During the model training phase, the cross-entropy loss function is used to quantify the difference between the predicted and true values. The backpropagation algorithm is used to minimize the value of this loss function to iteratively optimize the weight parameters inside the model. This allows the model to gradually learn the nonlinear mapping relationship between the fault evolution law and the state label implicit in the three-dimensional matrix. Finally, the trained fault detection model can accurately output the predicted value of the gear motor under the target fault based on the input three-dimensional matrix of the gear motor under test. This achieves fault state discrimination based on binary classification logic, ensuring that the detection system still has high accuracy and strong generalization ability when facing complex working conditions.

[0046] The gear motor fault detection system based on vibration signals includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the gear motor fault detection method based on vibration signals according to the first aspect of the present invention.

[0047] The vibration signal-based gear motor fault detection system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0048] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for fault detection of gear motors based on vibration signals, characterized in that, include: Collect the vibration signal sequence, load parameter sequence and speed sequence of any gear motor in a preset time period, take any fault type as the target fault, and obtain the fault label of the target fault of any gear motor in the preset time period. Multimodal decomposition of the vibration signal sequence yields several subsequences of vibration frequency bands. Any vibration frequency band is taken as the target frequency band, and the contribution of the target frequency band is calculated, including: Obtain all extreme points in the subsequence of the target frequency band, including maximum and minimum values. Calculate the time interval between any two adjacent extreme points, construct all time intervals into a time interval sequence, divide the intervals into intervals, and statistically analyze the probability distribution of each interval. Use the entropy value of the time interval sequence as the entropy feature of the extreme interval distribution. The extreme value interval distribution entropy characteristics of each gear motor in the target frequency band are obtained by traversal. All extreme value interval distribution entropy characteristics in the target frequency band are discretized by equal interval processing, and the mutual information value between the target frequency band and the target fault is calculated. The normalized mutual information value is used as the contribution of the target frequency band. The first coherence between the subsequence of the target frequency band and the load parameter sequence is calculated. Similarly, the second coherence between the subsequence of the target frequency band and the rotational speed sequence is obtained. Based on the first and second coherence, the correlation between the target frequency band and the target fault is calculated. The product of contribution and correlation is used as the importance weight of the target frequency band in the target fault. The subsequence of the target frequency band is weighted using the importance weight to obtain the weighted subsequence. The weighted subsequence of any gear motor in each vibration frequency band is obtained by traversing the frequency band. The sum of all weighted subsequences is used as the reconstructed vibration signal sequence of any gear motor. The reconstructed vibration signal sequence, load parameter sequence, and speed sequence are used to construct a three-dimensional matrix. The three-dimensional matrix is ​​then used to train a fault detection model. The obtained three-dimensional matrix of the gear motor to be tested is input into the trained fault detection model, and the predicted value of the gear motor to be tested at the target fault is output, thus completing the fault detection.

2. The gear motor fault detection method based on vibration signals according to claim 1, characterized in that, The fault label for obtaining the target fault of any gear motor within a preset time period includes: If a target fault occurs in any gear motor within a preset time period, the fault label value is 1; if no target fault occurs in any gear motor within a preset time period, the fault label value is 0.

3. The gear motor fault detection method based on vibration signals according to claim 1, characterized in that, The first synergy between the subsequence of the calculated target frequency band and the load parameter sequence includes: Calculate the cross-correlation function between the subsequence of the target frequency band and the load parameter sequence, and take the absolute value of the maximum value in the cross-correlation function as the first coherence between the subsequence of the target frequency band and the load parameter sequence.

4. The gear motor fault detection method based on vibration signals according to claim 1, characterized in that, The correlation between the calculated target frequency band and the target fault includes: The gear motors are divided into faulty and non-faulty groups according to the values ​​of all gear motors at the target fault. Calculate the mean of all first synergies in the fault group as the first mean, calculate the mean of all first synergies in the non-fault group as the second mean, and calculate the first absolute difference between the first mean and the second mean. Calculate the mean of all second synergies in the fault group as the third mean, calculate the mean of all second synergies in the non-fault group as the fourth mean, and calculate the second absolute difference between the third mean and the fourth mean; Obtain the first standard deviation of all first coherences in the fault group, and obtain the second standard deviation of all second coherences in the fault group; The sum of the first mean and the third mean is calculated as the first sum, and the sum of the first absolute difference and the second absolute difference is calculated as the second sum. The product of the first sum and the second sum is used as the numerator, and the sum of the first standard deviation, the second standard deviation and the preset adjustment coefficient is used as the denominator. The result after normalizing the ratio of the numerator to the denominator is used as the correlation between the target frequency band and the target fault.

5. The gear motor fault detection method based on vibration signals according to claim 1, characterized in that, The step of weighting the subsequences of the target frequency band using importance weights to obtain weighted subsequences includes: The importance weight is multiplied by each element in the subsequence of the target frequency band to obtain the weighted subsequence.

6. The gear motor fault detection method based on vibration signals according to claim 1, characterized in that, The fault detection model is an LSTM model. The input of the LSTM model is a three-dimensional matrix of any gear motor in the history, and the output is the predicted value of the fault label of any gear motor in the history. The label is the true value of the fault label of any gear motor in the history. The loss function of the LSTM model is the cross-entropy loss function.

7. A gear motor fault detection system based on vibration signals, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the gear motor fault detection method based on vibration signals according to any one of claims 1-6.

Images