Servo motor fault determination method and device and servo motor fault detection system

Abstract

The invention provides a fault determination method and device of a servo motor and a fault detection system of the servo motor. The method comprises the following steps: acquiring feature data; fusing the plurality of feature data to obtain comprehensive feature data; a fault analysis model is constructed, the fault analysis model is obtained through training of multiple sets of training data, and each set of training data in the multiple sets of training data comprises historical comprehensive feature data obtained in a historical time period and historical analysis results corresponding to the historical comprehensive feature data, the historical analysis result is used for representing whether the servo motor has a fault or not in a historical time period; and inputting the comprehensive feature data into a fault analysis model to obtain an analysis result corresponding to the comprehensive feature data. According to the scheme, the problem of poor fault detection accuracy of the servo motor in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the technical field of servo motor detection. Specifically, it relates to a method and device for determining faults of a servo motor, a computer program product, and a servo motor fault detection system. Background Art

[0002] Servo motors are key components widely used in modern automation systems, and are widely used in fields such as robotics, numerical control machine tools, automated production lines, precision machining equipment, and aerospace. The main feature of a servo motor is its ability to achieve precise control of speed, position, and acceleration, which makes it an indispensable core component in application scenarios that require high precision and high response speed. To ensure the efficient operation of a servo motor, a closed-loop control system is usually adopted, which can continuously adjust the input of the motor according to real-time feedback to achieve an ideal motion state.

[0003] However, with the wide application of servo motors in various high-precision equipment, their failure rate and health status monitoring issues have gradually attracted the attention of researchers and engineers. Traditional servo motor health monitoring methods usually rely on the monitoring of simple physical quantities such as temperature, current, vibration, etc. These methods can often only provide a basic judgment of the health status, and the accuracy of fault detection is relatively poor. Summary of the Invention

[0004] The main objective of the present application is to provide a method and device for determining faults of a servo motor, a computer program product, and a servo motor fault detection system, so as to at least solve the problem of relatively poor accuracy of fault detection of servo motors in the prior art.

[0005] To achieve the above objective, according to one aspect of the present application, a method for determining faults of a servo motor is provided, including: obtaining characteristic data, where the characteristic data is data of key parameters during the operation of the servo motor, and the key parameters include at least one or more of temperature, current value, position, speed, and vibration signal. There are multiple pieces of the characteristic data, and the characteristic data corresponds to time windows one by one; fusing multiple pieces of the characteristic data to obtain comprehensive characteristic data, where the fusion method includes at least weighted average; constructing a fault analysis model, where the fault analysis model is trained using multiple sets of training data, and each set of training data in the multiple sets of training data includes historical comprehensive characteristic data obtained within a historical time period and a historical analysis result corresponding to the historical comprehensive characteristic data, where the historical analysis result is used to represent whether the servo motor has a fault during the historical time period; inputting the comprehensive characteristic data into the fault analysis model to obtain an analysis result corresponding to the comprehensive characteristic data.

[0006] Optionally, obtain feature data, including: performing time-domain feature extraction on the key parameters to obtain first sub-feature data; performing frequency-domain feature extraction on the key parameters to obtain second sub-feature data; performing time-frequency domain feature extraction on the key parameters to obtain third sub-feature data.

[0007] Optionally, performing time-domain feature extraction on the key parameters to obtain first sub-feature data includes: calculating the mean of the key parameters according to a first formula, where the first formula is:

[0008]

[0009] μ represents the mean, N represents the total number of samples within the time window, and x i represents the data at the i-th sampling point; calculating the variance of the key parameters according to a second formula, where the second formula is:

[0010]

[0011] σ 2 represents the variance; calculating the peak value of the key parameters according to a third formula, where the third formula is:

[0012] x peak = max(x i ), i = 1, 2, …, N,

[0013] x peak represents the peak value; calculating the kurtosis of the key parameters according to a fourth formula, where the fourth formula is:

[0014]

[0015] Kurtosis represents the kurtosis.

[0016] Optionally, performing frequency-domain feature extraction on the key parameters to obtain second sub-feature data includes: calculating the frequency component of the key parameters according to a fifth formula, where the fifth formula is:

[0017]

[0018] X(k) represents the frequency component; calculating the main frequency of the key parameters according to a sixth formula, where the sixth formula is:

[0019]

[0020] f peak represents the main frequency, |X(f k )| represents the spectral amplitude, and f kf represents frequency; according to the seventh formula, calculate the total spectral energy of the key parameter, where the seventh formula is:

[0021]

[0022] E represents the total spectral energy; according to the eighth formula, calculate the spectral center rate of the key parameter, where the eighth formula is:

[0023]

[0024] f c represents the spectral center rate; according to the ninth formula, calculate the spectral bandwidth of the key parameter, where the ninth formula is:

[0025] B = fhigh - flow,

[0026] B represents the spectral bandwidth, f high represents the highest frequency at which the spectral energy exceeds the energy threshold, f low represents the lowest frequency at which the spectral energy exceeds the energy threshold.

[0027] Optionally, perform time - frequency domain feature extraction on the key parameter to obtain the third sub - feature data, including: according to the tenth formula, calculate the frequency distribution of the key parameter, where the tenth formula is:

[0028]

[0029] W(a,b) represents the frequency distribution, a represents the frequency resolution, b represents the time resolution, ψ(t) represents the mother wavelet function; according to the eleventh formula, calculate the band energy of the key parameter, where the eleventh formula is:

[0030]

[0031] E band represents the band energy; according to the twelfth formula, calculate the energy center time of the key parameter, where the twelfth formula is:

[0032]

[0033] t c represents the energy center time.

[0034] Optionally, the feature data includes first feature data, second feature data, and third feature data. Fusing multiple pieces of the feature data to obtain comprehensive feature data includes: performing a linear transformation on the first feature data to obtain a first matrix; performing a linear transformation on the second feature data to obtain a second matrix; performing a linear transformation on the third feature data to obtain a third matrix; and fusing the first matrix, the second matrix, and the third matrix to obtain the comprehensive feature data.

[0035] Optionally, fusing the first matrix, the second matrix, and the third matrix to obtain the comprehensive feature data includes: calculating the correlation between different time windows according to the thirteenth formula, where the thirteenth formula is:

[0036]

[0037] A ij represents the correlation, Q i represents the first matrix corresponding to the i-th piece of the feature data, K j represents the second matrix corresponding to the j-th feature data, and T represents the time window; calculating the attention weight according to the fourteenth formula, where the fourteenth formula is:

[0038]

[0039] α ij represents the attention weight; calculating the weighted feature vector according to the fifteenth formula, where the fifteenth formula is:

[0040]

[0041] Z i represents the weighted feature vector, V j represents the third matrix corresponding to the j-th piece of the feature data; generating the comprehensive feature data according to the sixteenth formula, where the sixteenth formula is:

[0042]

[0043] Z fused represents the comprehensive feature data.

[0044] According to another aspect of the present application, there is provided a fault determination device for a servo motor, including: an acquisition unit configured to acquire feature data, where the feature data is data of key parameters during the operation of the servo motor, and where the key parameters include at least one or more of temperature, current value, position, speed, and vibration signal, there are multiple pieces of the feature data, and the feature data corresponds to time windows one by one; a fusion unit configured to fuse multiple pieces of the feature data to obtain comprehensive feature data, where the fusion method includes at least weighted average; a construction unit configured to construct a fault analysis model, where the fault analysis model is trained using multiple sets of training data, and each set of training data in the multiple sets of training data includes historical comprehensive feature data acquired within a historical time period and a historical analysis result corresponding to the historical comprehensive feature data, where the historical analysis result is used to represent whether the servo motor has a fault during the historical time period; and an analysis unit configured to input the comprehensive feature data into the fault analysis model to obtain an analysis result corresponding to the comprehensive feature data.

[0045] According to yet another aspect of the present application, there is provided a computer program product including a computer program, and when the computer program is executed by a processor, the steps of any one of the fault determination methods for the servo motor are implemented.

[0046] According to still another aspect of the present application, there is provided a servo motor fault detection system, including: one or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include those for executing any one of the fault determination methods for the servo motor.

[0047] Applying the technical solution of the present application, there are multiple pieces of feature data, which are sampled from different time windows, so as to reflect the operating state of the servo motor in different time periods, and thus multi-dimensional feature data can be extracted, providing a rich information source for subsequent feature fusion and state evaluation. Furthermore, multi-data fusion is performed to fuse the features of multiple time windows and multiple modalities into a comprehensive feature representation. The fused features can better reflect the key information of the motor operation, improving the accuracy of subsequent evaluation. Then, the state of the servo motor is evaluated according to the neural network model, and the operating state of the servo motor is accurately judged through machine learning, with high accuracy. Description of the Drawings

[0048] The schematic drawings forming a part of this application are used to provide a further understanding of this application. The illustrative embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0049] Figure 1 The block diagram of the hardware structure of a mobile terminal for implementing a method for determining the fault of a servo motor provided in an embodiment of the present application is shown;

[0050] Figure 2 The schematic flow chart of a method for determining the fault of a servo motor provided in an embodiment of the present application is shown;

[0051] Figure 3 The schematic diagram of the overall framework of the present solution is shown;

[0052] Figure 4 The schematic diagram of multi-time window feature extraction is shown;

[0053] Figure 5 The schematic diagram of the attention mechanism is shown;

[0054] Figure 6 The block diagram of the structure of a device for determining the fault of a servo motor provided in an embodiment of the present application is shown.

[0055] Among them, the above-mentioned drawings include the following reference numerals:

[0056] 102, processor; 104, memory; 106, transmission device; 108, input / output device. Detailed implementation manners

[0057] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0058] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0059] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so as to implement the embodiments of the present application described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0060] Specifically, it is difficult for existing solutions to accurately predict the fault type and occurrence time of the motor. Moreover, in a complex industrial environment, the operating state of the motor is often affected by multiple factors, and simple monitoring methods are difficult to cope with complex fault modes, resulting in the motor not being maintained or replaced in a timely manner at an early stage, thus affecting the stability of the entire system.

[0061] In recent years, with the development of deep learning and intelligent algorithms, more and more research has begun to attempt to improve the accuracy of motor health monitoring by using the operation data of servo motors and through multi-dimensional feature extraction and learning. The existing related technologies mainly focus on the following aspects:

[0062] Health monitoring methods based on traditional feature extraction: These methods mainly extract features in the time domain, frequency domain and time-frequency domain from the operation data of the motor (such as temperature, current, vibration, etc.), and combine traditional machine learning algorithms (such as support vector machines, decision trees, etc.) to judge the health state.

[0063] Motor health monitoring based on deep learning: With the progress of deep learning technology, researchers have begun to use deep learning models such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) to automatically learn features and detect anomalies from the operation data of the motor. These methods significantly improve the accuracy and real-time performance of motor health monitoring by automatically extracting high-level features and classifying faults.

[0064] Methods based on multi-modal data fusion: Some studies have proposed to improve the accuracy of motor health monitoring by fusing data from multiple sensors (such as temperature, current, position, etc.) and using data fusion technology. These methods can capture more potential fault information by comprehensively analyzing different types of data.

[0065] In the prior art, most of the servo motor health state prediction and diagnosis methods rely on simple physical quantity monitoring and rule-based analysis methods. Traditional methods mainly perform single-dimensional time-domain or frequency-domain analysis on signals such as temperature, current, and position. Although they can provide a basic health state assessment, it is difficult to accurately predict the time and type of faults when facing complex fault modes and variable working conditions. In recent years, deep learning technology has been introduced into the field of motor health monitoring, attempting to improve the prediction accuracy through automatic feature extraction. However, such methods usually require a large amount of high-quality data for training, consume a high amount of computing resources, and have limitations in feature fusion and time series analysis, making it difficult to effectively utilize the operating data in different time periods. In addition, the multi-modal data fusion technology is not yet mature, and most methods cannot distinguish the importance of different data sources for health state prediction, resulting in insufficient reliability and adaptability of the prediction results.

[0066] As introduced in the background art, the accuracy of fault detection of servo motors in the prior art is poor. To solve the above problems, embodiments of the present application provide a method, apparatus, computer program product, and servo motor fault detection system for determining faults of a servo motor.

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0068] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal for a method of determining faults of a servo motor according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1 only one is shown in Figure 1 a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than Figure 1 shown, or have a different configuration from

[0069] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above-mentioned method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the mobile terminal through a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (Radio Frequency, abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0070] In this embodiment, a method for determining a fault of a servo motor running on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0071] Figure 2 It is a schematic flowchart of a method for determining a fault of a servo motor according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0072] Step S201, obtain characteristic data, where the above-mentioned characteristic data is data of key parameters during the operation of the servo motor, and where the above-mentioned key parameters include at least one or more of temperature, current value, position, speed, and vibration signal. There are multiple pieces of the above-mentioned characteristic data, and the above-mentioned characteristic data corresponds one-to-one with a time window;

[0073] Specifically, during the operation of the servo motor, various types of sensors are installed on the device to collect key operation data. These data reflect the operation status of the servo motor under different working conditions and provide basic support for the prediction and diagnosis of the health status. The sampling frequency is set according to the real-time requirements of the motor operation, such as 1 kHz. The collected data includes the following categories:

[0074] Temperature data. The temperature sensor is installed near the outer shell, bearing or winding of the servo motor to monitor the change of the thermal state during motor operation. These data can reflect the change of the internal thermal load of the motor and timely capture the fault risks caused by abnormal temperature rise, such as problems like overload and poor heat dissipation.

[0075] Current data. The current sensor is installed in the power supply circuit of the motor to measure the change of the actual current during motor operation. The current data can reflect information such as load change, overload, and motor faults (such as winding short circuit or open circuit), and is an important parameter for fault monitoring.

[0076] Position and speed data. By installing an encoder or a position sensor on the shaft of the servo motor, the rotational position and speed change of the motor can be obtained in real time. The position and speed data can reflect the motion state and control accuracy of the motor, and judge whether there are faults such as abnormal motion and positioning inaccuracy by comparing with the preset trajectory.

[0077] Vibration data. In some application scenarios, the vibration of the motor can be monitored by installing an accelerometer or a vibration sensor. Abnormal vibration is usually related to structural looseness, bearing wear or imbalance of dynamic balance.

[0078] Other data. Depending on the specific application requirements, other key parameters can also be collected, such as torque, voltage, power, etc., to further enhance the accuracy of the health status assessment.

[0079] Specifically, as Figure 3 shown, the solution of this application includes multi-time window feature extraction. By sampling the data of the past 1 second, 2 seconds and 3 seconds and respectively extracting the time domain, frequency domain and time-frequency domain features, a multi-dimensional feature representation is formed to comprehensively reflect the operation status of the motor at different time periods. This method can capture the dynamic behavior of the servo motor at different time scales and provides a rich information source for subsequent feature fusion and health status prediction. Compared with the traditional method, multi-time window feature extraction significantly improves the ability to capture complex fault patterns, thereby enhancing the accuracy of health status prediction.

[0080] Step S202, fuse the multiple above-mentioned feature data to obtain comprehensive feature data, where the fusion method includes at least weighted average;

[0081] Specifically, the feature data obtained from different time windows are fused to form a comprehensive feature dataset. The fusion process is not just simple data stacking. Instead, techniques such as weighted averaging are adopted, and different weights are given according to the contribution of the data in each time window to the prediction of the motor health status, so as to more accurately reflect the operating state of the motor and potential health problems. Through this weighted fusion method, the more relevant time windows and feature parameters can be highlighted, improving the pertinence and accuracy of subsequent analysis.

[0082] Specifically, as Figure 3 shown, the solution of this application includes feature fusion based on the attention mechanism. The attention mechanism is introduced into the diagnosis of the servo motor health status, and a dynamic weighting strategy is designed. According to the importance of different time windows and feature domains, the weights are automatically adjusted, and the features of multiple time windows and multi-modal are fused into a comprehensive feature representation. Through the attention mechanism, the importance adaptive learning of features is realized, and the time series features of the motor are extracted with little increase in computational complexity. The fused features can better reflect the key information of the motor operation, significantly improving the robustness and accuracy of the health status prediction.

[0083] Step S203: Construct a fault analysis model. The fault analysis model is obtained by training with multiple sets of training data. Each set of the multiple sets of training data includes historical comprehensive feature data obtained within a historical time period and the corresponding historical analysis results of the historical comprehensive feature data, where the historical analysis results are used to characterize whether the servo motor has a fault during the historical time period.

[0084] Specifically, a fault analysis model is constructed, which is obtained by training through a machine learning algorithm. During the training process, multiple sets of historical data are used. Each set of data contains the comprehensive feature data collected within a specific historical time period and the analysis results (i.e., whether there is a fault) corresponding to that time period. These historical data are used to train the model so that it can learn the association pattern between the operating state of the servo motor and faults. In this way, the fault analysis model can predict whether the motor may be in a fault state or face a fault risk based on the current comprehensive feature data, providing support for real-time monitoring and prediction.

[0085] Step S204: Input the comprehensive feature data into the fault analysis model to obtain the analysis result corresponding to the comprehensive feature data.

[0086] Specifically, the comprehensive feature data obtained from step S202 is used as input and passed to the fault analysis model constructed in step S203 for processing. The model analyzes the input data according to the patterns and rules it has learned, and finally outputs an analysis result, which is usually a numerical value or a classification label, indicating the current health status or fault prediction result of the servo motor. Through this process, the motor status can be monitored in real time, and early warnings can be issued in a timely manner when abnormalities occur, providing a decision-making basis for preventive maintenance and fault troubleshooting.

[0087] Specifically, as Figure 3 shown, the solution of this application includes health status prediction and diagnosis. The fused comprehensive features are input into a neural network for health status prediction and diagnosis. The neural network model determines the operating status of the servo motor based on the input features and outputs a decision on whether maintenance or replacement is required. Through the optimized design of feature extraction and fusion, the health status prediction of this application has high accuracy and reliability. The model can accurately predict the fault trend of the motor, reduce the risk of unexpected downtime, and provide strong support for the intelligent maintenance and optimized management of industrial equipment.

[0088] Through this embodiment, there are multiple pieces of feature data, which are sampled from different time windows. This can reflect the operating status of the servo motor at different time periods, and thus multi-dimensional feature data can be extracted, providing a rich information source for subsequent feature fusion and status evaluation. Then, multi-data fusion is performed to fuse the features of multiple time windows and multiple modalities into a comprehensive feature representation. The fused features can better reflect the key information of the motor operation, improve the accuracy of subsequent evaluation, and then perform the status evaluation of the servo motor according to the neural network model. By means of machine learning, the operating status of the servo motor can be accurately judged, with high accuracy.

[0089] Specifically, the solution of this application proposes an efficient and accurate method for predicting and diagnosing the health status of a servo motor by fusing multi-time window feature extraction and an attention mechanism. It can comprehensively capture the operating status of the motor and accurately predict its health status. The multi-time window feature extraction method fully explores the dynamic characteristics of the servo motor operation data at different time scales, and the introduction of the attention mechanism ensures the intelligence and flexibility of the feature fusion process, making the generated comprehensive features obtain not only the information of the time series but also without excessive increase in computational complexity. Combined with the accurate prediction of the health status by the neural network, this method not only significantly improves the accuracy and robustness of fault prediction but also reduces the risk caused by fault downtime. Overall, the invention provides a motor health monitoring and diagnosis solution applicable to complex industrial environments, providing reliable support for the intelligent maintenance and optimized management of industrial equipment.

[0090] In the specific implementation process, the acquisition of feature data can be achieved through the following steps: extracting the time-domain features of the above key parameters to obtain the first sub-feature data; extracting the frequency-domain features of the above key parameters to obtain the second sub-feature data; extracting the time-frequency domain features of the above key parameters to obtain the third sub-feature data.

[0091] In this solution, through time-domain, frequency-domain, and time-frequency domain feature extraction, it is possible to comprehensively analyze multiple dimensions of the servo motor operation data. It can not only capture the static and dynamic characteristics of the motor but also reveal the hidden spectral information and time-varying frequency characteristics in the signal. This multi-dimensional feature extraction method significantly improves the accuracy and reliability of fault prediction, can detect potential fault signs earlier, and provides strong support for preventive maintenance. At the same time, due to the real-time nature of time-domain feature extraction and the sensitivity of frequency-domain and time-frequency domain feature extraction to complex faults, this method can meet the high requirements for motor health status monitoring in industrial scenarios, reduce the risk of production interruption caused by faults, and further improve the operation efficiency and safety of the equipment.

[0092] Specifically, time-domain feature extraction refers to directly performing statistical analysis on the time-series signal to extract features reflecting the signal strength, change trend, and volatility. These features may include the mean, variance, peak value, kurtosis, etc. of the signal. Through time-domain feature extraction, static and dynamic indicators reflecting the operation status of the servo motor can be obtained, such as the temperature of the motor, the change trend of the current, etc. These information are crucial for judging whether there are abnormal instantaneous behaviors in the motor.

[0093] Time-domain feature extraction can directly and quickly obtain the statistical information of key parameters from the original signal, capture the instantaneous changes and volatility of the motor operation status, and has high sensitivity for identifying sudden faults (such as overheating, overload).

[0094] Specifically, frequency-domain feature extraction is to transform the time-series signal into the frequency domain and analyze the energy distribution, main frequency components, and frequency characteristics of the signal through methods such as the fast Fourier transform (FFT). Frequency-domain feature extraction focuses on the distribution of different frequency components in the signal and can reveal possible periodic or quasi-periodic abnormalities during the motor operation, such as bearing faults, gear meshing abnormalities, etc. These abnormalities are often not easily detected in the time domain.

[0095] Frequency-domain feature extraction can discover the spectral information hidden in the signal, capture potential problems in the internal structure of the motor, and improve the detection ability for complex and hidden faults. Especially for detecting abnormalities in low-frequency or high-frequency signals, it is very crucial for predicting and diagnosing the long-term health trend of the servo motor.

[0096] Specifically, time-frequency domain feature extraction combines time-domain and frequency-domain analysis. Usually, methods such as wavelet transform are adopted, which can observe the frequency distribution changes of signals at different times and provide time-frequency diagrams or time-frequency spectra of signals. This method is especially suitable for analyzing non-stationary signals, that is, the frequency characteristics of signals that change over time. Time-frequency domain feature extraction can reveal the operating characteristics of servo motors under dynamic conditions, including the transient components and frequency change trends of signals.

[0097] Time-frequency domain feature extraction can capture the dynamic changes in the operating state of servo motors, identify transient events and the changes of frequency over time, which is very effective for diagnosing the types of faults that occur with the changes in operating conditions, and can improve the adaptability and prediction accuracy of the prediction model under non-stable conditions.

[0098] Specifically, as Figure 4 shown, the above scheme of this application samples the motor data in multiple time windows of the past 1 second, 2 seconds, and 3 seconds, extracts time-domain, frequency-domain, and time-frequency domain features, and comprehensively captures the operating characteristics of servo motors. Based on the attention mechanism, these features are weighted and fused, which can automatically learn the relative importance of different time periods and feature domains, and generate high-quality comprehensive feature representations. The fused features are input into an optimized neural network for health state prediction, which can accurately judge the operating state of the servo motor and output suggestions on whether maintenance or replacement is required. Through the innovative feature fusion strategy and lightweight model design, this method not only improves the accuracy and real-time performance of prediction, but also reduces the demand for computing resources, and can better adapt to the complex and changeable operating environment in industrial scenarios.

[0099] Specifically, multi-time window feature extraction refers to dividing the collected time series signals into multiple time windows, and respectively taking the data windows of the past 1 second, 2 seconds, and 3 seconds. Feature extraction is performed on the signals within each time window, specifically including:

[0100] Time-domain feature extraction: Calculate indicators such as mean, variance, peak value, and kurtosis to characterize the statistical characteristics of signals in the time domain;

[0101] Frequency-domain feature extraction: Extract the spectral characteristics of signals through the fast Fourier transform (FFT), including the main frequency, spectral energy distribution, etc.;

[0102] Time-frequency domain feature extraction: Use methods such as wavelet transform to obtain the frequency distribution of signals at different time points and extract refined dynamic characteristic indicators.

[0103] In some embodiments, time-domain feature extraction is performed on the above key parameters to obtain the first sub-feature data, which can be specifically implemented through the following steps: According to the first formula, calculate the mean of the above key parameters, where the above first formula is:

[0104]

[0105] Let μ represent the above-mentioned mean value, N represent the total number of samples within the above-mentioned time window, and x i represent the data at the i-th sampling point; according to the second formula, calculate the variance of the above-mentioned key parameter, where the above-mentioned second formula is:

[0106]

[0107] σ 2 represent the above-mentioned variance; according to the third formula, calculate the peak value of the above-mentioned key parameter, where the above-mentioned third formula is:

[0108] x peak = max(x i ), i = 1, 2, …, N,

[0109] x peak represent the above-mentioned peak value; according to the fourth formula, calculate the kurtosis of the above-mentioned key parameter, where the above-mentioned fourth formula is:

[0110]

[0111] Let Kurtosis represent the above-mentioned kurtosis.

[0112] In this solution, the extraction of time-domain features can comprehensively evaluate the statistical characteristics of motor operation data by calculating the mean value, variance, peak value, and kurtosis of key parameters, including the stability, volatility, maximum intensity, and peak degree of signal distribution of the motor. These time-domain features can help the system quickly identify whether the motor is in a normal or abnormal working state, and provide direct and effective information for predicting possible motor faults, such as overload, wear, vibration, etc.

[0113] Specifically, the mean value is the average of all sampling point data of a signal or data set within a time window, which reflects the central tendency of the signal or the basic level of the motor operation state. The calculation of the mean value can quickly understand whether the motor operation state is stable within a certain period of time, and is an important indicator for monitoring the basic working condition of the motor. Abnormal mean value changes may indicate unstable or abnormal operation trends of the motor in terms of temperature, current, position, etc.

[0114] Specifically, the variance measures the degree of deviation between the signal value and the mean value, reflecting the volatility and data dispersion degree of the signal. The magnitude of the variance can reflect the stability of the motor parameters. A higher variance may indicate that the motor working state is unstable, with problems such as overload and poor heat dissipation. The variance value can help the system determine whether the motor fluctuates within the normal working range.

[0115] Specifically, the peak value is the maximum value of all the sampled data points within the time window, which can reveal the maximum intensity of the signal or the maximum state of the motor operation. The peak value is an important basis for judging whether the motor has instantaneous overload or abnormal high temperature and vibration. By monitoring the peak value, it is possible to promptly detect abnormal high values that may be caused by instantaneous large loads or internal faults in the motor, which helps to implement preventive maintenance measures in advance.

[0116] Specifically, kurtosis describes the peakedness of the signal distribution, that is, the thickness of the tails and the sharpness of the center of the signal distribution. The calculation of kurtosis helps to identify the non-normality of the signal distribution, that is, whether there are abnormal large or small peak values in the signal, which helps to early detect potential problems such as friction and vibration inside the motor. In the motor operation data, an abnormally high kurtosis value may indicate internal structure problems such as bearing wear, and this information is particularly important in time-domain analysis because they may indicate impending failures.

[0117] In the specific implementation process, frequency-domain feature extraction is performed on the above key parameters to obtain the second sub-feature data, which can be achieved through the following steps: According to the fifth formula, calculate the frequency components of the above key parameters, where the above fifth formula is:

[0118]

[0119] X(k) represents the above frequency components; according to the sixth formula, calculate the main frequency of the above key parameters, where the above sixth formula is:

[0120]

[0121] f peak represents the above main frequency, |X(f k )| represents the spectral amplitude, and f k represents the frequency; according to the seventh formula, calculate the total spectral energy of the above key parameters, where the above seventh formula is:

[0122]

[0123] E represents the above total spectral energy; according to the eighth formula, calculate the spectral center rate of the above key parameters, where the above eighth formula is:

[0124]

[0125] f c represents the above spectral center rate; according to the ninth formula, calculate the spectral bandwidth of the above key parameters, where the above ninth formula is:

[0126] B = fhigh - flow,

[0127] B represents the above-mentioned spectral bandwidth, and f high represents the highest frequency at which the spectral energy exceeds the energy threshold, and f low represents the lowest frequency at which the spectral energy exceeds the energy threshold.

[0128] In this scheme, frequency-domain feature extraction can reveal potential fault information related to frequency during motor operation by performing spectral analysis on the signal, including the main vibration frequency, energy distribution, concentration, and range of frequency distribution. These features are crucial for detecting and diagnosing faults in common components such as bearings and gears in servo motors. Compared with time-domain features, frequency-domain features can capture the operating characteristics of the motor from different perspectives, and combining the two can provide a more comprehensive health status assessment.

[0129] Specifically, NNN is the number of FFT points. The calculated total spectral energy, spectral center frequency, and spectral bandwidth are used as feature inputs. In actual use, other data can be added as features for the frequency-domain description.

[0130] Specifically, the fifth formula uses the basic principle of the fast Fourier transform (FFT) to convert the time-series signal x(n) into the frequency domain, obtaining the frequency components X(k) at different frequencies k. Converting the signal to the frequency domain can reveal the hidden periodic and harmonic components in the motor operation data, helping analysts identify vibrations or anomalies at specific frequencies, which is very useful for detecting faults in components such as bearings and gears.

[0131] Specifically, the main frequency is the frequency point with the largest amplitude in the signal spectrum, reflecting the main frequency components of the signal. Identifying the main frequency helps determine the main vibration frequency during motor operation, which can be compared with the known natural frequencies of the motor or components to judge whether there are faults such as imbalance, misalignment, or damaged bearings.

[0132] Specifically, the total spectral energy reflects the overall energy distribution of the signal in the frequency domain. The magnitude of the total spectral energy can provide overall information about the signal intensity, and abnormal energy distribution may indicate abnormal vibrations or noises inside the motor, helping to reveal fault signs at an early stage.

[0133] Specifically, the spectral center frequency is the weighted average frequency of the signal energy, reflecting the central position of the signal energy distribution. The spectral center frequency can help analyze the concentration of frequency components in the signal. If the spectral center frequency changes significantly, it may mean that the operating mode or load condition of the motor has changed, which may in turn affect its health status.

[0134] Specifically, the spectral bandwidth is the frequency range where the signal energy exceeds a specific threshold, reflecting the bandwidth characteristics of the signal. Technical effect: The widening or narrowing of the spectral bandwidth can reveal the distribution range of the frequency components in the signal. A wide frequency band may mean the existence of abnormal vibrations of multiple frequencies, while a narrow frequency band may indicate a fault mode at a specific frequency (such as the gear meshing frequency).

[0135] In some embodiments, time-frequency domain feature extraction is performed on the above key parameters to obtain third sub-feature data, which can be specifically implemented through the following steps: According to the tenth formula, calculate the frequency distribution of the above key parameters, where the above tenth formula is:

[0136]

[0137] W(a, b) represents the above frequency distribution, a represents the frequency resolution, b represents the time resolution, and ψ(t) represents the mother wavelet function; According to the eleventh formula, calculate the band energy of the above key parameters, where the above eleventh formula is:

[0138]

[0139] E band represents the above band energy; According to the twelfth formula, calculate the energy center time of the above key parameters, where the above twelfth formula is:

[0140]

[0141] t c represents the above energy center time.

[0142] In this solution, time-frequency domain feature extraction decomposes the signal into a time-frequency diagram representation through wavelet transform, so that the frequency distribution characteristics of the signal in different time periods can be analyzed. This analysis method is particularly effective for detecting transient events in non-stationary signals and faults of specific components. The calculation of band energy and energy center time can provide detailed information on the energy distribution of the signal at specific frequency bands and time points, helping to identify potential fault signs early, so as to take preventive maintenance measures, reduce equipment failure downtime, and improve the stability and efficiency of equipment operation.

[0143] Specifically, a is the scale parameter (controlling the frequency resolution), b is the time parameter (controlling the time resolution), and the calculated band energy and energy center time are used as feature inputs. In actual use, other data can be added to the time-frequency domain description as features.

[0144] Specifically, the tenth formula is based on the wavelet transform principle and converts the time-domain signal x(t) into a time-frequency domain representation, thereby revealing the distribution of the signal at different time points and different frequencies. The wavelet transform provides a powerful time-frequency analysis tool that can identify changes in frequency components within different time periods of the signal, which is particularly effective for detecting transient events (such as momentary overloads, vibration mutations) in non-stationary signals. It can help analysts locate problems in time and understand the specific manifestations of problems in the frequency spectrum, providing more detailed clues for fault diagnosis.

[0145] Specifically, the band energy reflects the total energy of the signal within a specific frequency range and can be used to identify which frequency band in the signal has concentrated or abnormal energy. Calculating the band energy helps identify abnormal energy distributions in which frequency bands, which is crucial for detecting faults in specific components (such as bearings, gears) of a servo motor. For example, abnormally high band energy may indicate bearing wear or poor gear meshing because these faults usually generate abnormal vibrations or noises at specific frequencies.

[0146] Specifically, the energy center time is the time average weighted by the signal energy and reveals the concentrated position of the signal energy on the time axis. The energy center time can help analysts locate the time points where the signal energy is concentrated, which is particularly important when detecting non-stationarity or transient events during the operation of a servo motor. If the energy center time changes significantly, it may mean that the motor has experienced different loads or internal faults (such as sudden vibrations or overheating) at specific time points.

[0147] In the specific implementation process, the above-mentioned feature data includes first feature data, second feature data, and third feature data. Fusing multiple pieces of the above-mentioned feature data to obtain comprehensive feature data can be achieved through the following steps: performing a linear transformation on the above-mentioned first feature data to obtain a first matrix; performing a linear transformation on the above-mentioned second feature data to obtain a second matrix; performing a linear transformation on the above-mentioned third feature data to obtain a third matrix; fusing the above-mentioned first matrix, the above-mentioned second matrix, and the above-mentioned third matrix to obtain the above-mentioned comprehensive feature data.

[0148] In this solution, fusing the feature data in the time domain, frequency domain, and time-frequency domain can generate comprehensive feature data containing multi-faceted information about the operating state of the motor. This kind of fusion makes full use of the advantages of different features, overcomes the limitations of single-feature representation, and improves the accuracy and comprehensiveness of health state prediction.

[0149] Specifically, such as Figure 5As shown, the input feature vectors, namely the feature vectors X1, X2, and X3 extracted from multiple time windows, are input into the attention mechanism. These vectors represent the features of the past 1 second, 2 seconds, and 3 seconds respectively, contain the information extracted from the time domain, frequency domain, and time-frequency domain, and reflect the operating states of the servo motor in different time periods.

[0150] Specifically, for each input feature vector X 1 , X 2 , X 3 , a linear transformation is performed to map them to the query matrix Q, the key matrix K, and the value matrix V respectively. The formulas are as follows:

[0151] Q = X i W Q , K = X i W K , V = X i W V .

[0152] Among them, W Q , W K and W V are weight matrices that need to be learned through training.

[0153] Specifically, the time-domain feature data is converted into a matrix form suitable for subsequent fusion operations through a linear mapping (such as multiplying by a weight matrix). Such a matrix can be more conveniently combined with other feature data. Through linear transformation, the representation form of the time-domain features can be adjusted to match the dimensions and formats of other feature data, facilitating unified fusion processing. This process also allows weighting of the importance of the time-domain features to ensure that the time-domain features are appropriately reflected in the final comprehensive feature data.

[0154] Specifically, by mapping the frequency-domain features to matrix form, data can be prepared for the fusion process. The linear transformation of the frequency-domain features can adjust the weights of the features so that they occupy appropriate positions in the comprehensive feature data. Since the frequency-domain features are very crucial for detecting specific types of motor faults (such as bearing faults), this processing helps to improve the recognition rate and prediction accuracy of such faults.

[0155] Specifically, it is converted into a matrix form suitable for fusion through linear transformation. The time-frequency domain features can capture the frequency characteristics of the signal at different time periods and are particularly useful for the analysis of non-stationary signals and transient events. Converting the time-frequency domain features into matrix form can ensure that these dynamically changing features are effectively considered in the fusion process, thereby improving the prediction ability of the motor health state under complex working conditions.

[0156] Specifically, the fusion process can adopt weighted average, splicing, or more complex deep learning techniques (such as the attention mechanism) to integrate the features extracted from the three matrices, generating a comprehensive feature data that contains time-domain, frequency-domain, and time-frequency-domain information. By fusing different feature data, the comprehensive feature data can comprehensively reflect the operating state of the servo motor in multiple dimensions, including not only stable operating parameters (such as mean and variance), but also the frequency characteristics of the signal and the characteristics that change over time. This multi-dimensional feature representation significantly improves the accuracy and robustness of fault prediction, enabling earlier and more accurate identification of potential problems in the motor, reducing the uncertainty of fault diagnosis, and being of great significance for preventive maintenance and fault warning.

[0157] In some embodiments, the above-mentioned comprehensive feature data is obtained by fusing the above-mentioned first matrix, the above-mentioned second matrix, and the above-mentioned third matrix, and can be specifically implemented through the following steps: Calculate the correlation between different above-mentioned time windows according to the thirteenth formula, where the above-mentioned thirteenth formula is:

[0158]

[0159] A ij represents the above-mentioned correlation, Q i represents the above-mentioned first matrix corresponding to the i-th above-mentioned feature data, K j represents the above-mentioned second matrix corresponding to the j-th feature data, T represents the above-mentioned time window; Calculate the attention weight according to the fourteenth formula, where the above-mentioned fourteenth formula is:

[0160]

[0161] α ij represents the above-mentioned attention weight; Calculate the weighted feature vector according to the fifteenth formula, where the above-mentioned fifteenth formula is:

[0162]

[0163] Z i represents the above-mentioned weighted feature vector, V j represents the above-mentioned third matrix corresponding to the j-th above-mentioned feature data; Generate the above-mentioned comprehensive feature data according to the sixteenth formula, where the above-mentioned sixteenth formula is:

[0164]

[0165] Z fused represents the above-mentioned comprehensive feature data.

[0166] In this solution, the fusion process of the attention mechanism enables the model to automatically learn the importance of different time windows and feature domains. Through weighted summation and integration, comprehensive feature data that can fully reflect the health status of the servo motor is generated. This data fusion method not only improves the prediction accuracy but also ensures the adaptability of the model to complex working conditions and non-stationary signals.

[0167] Specifically, the calculation method of the attention weight is to calculate the attention weight using the query matrix Q and the key matrix K. The dot product scoring mechanism is used to measure the correlation between different time windows. The formula is as shown in the thirteenth formula, where d k is the dimension of the key matrix K, which is used for scaling to avoid overly large attention weights. Then, the Softmax function is used to normalize the scoring result. The formula for calculating the attention weight is as shown in the fourteenth formula. Using the attention α ij weight to weight the value matrix V to generate the weighted feature vector Z for each time window i , and the formula is as shown in the fifteenth formula. Fuse the weighted feature vectors of all time windows (such as a summation operation) to generate the final comprehensive feature Z fused is represented by the formula as shown in the sixteenth formula. The comprehensive feature Z fused , as the output of the attention mechanism, contains the key information of different time windows and provides input for the subsequent prediction of the servo motor health status.

[0168] The comprehensive feature Z fused is the feature vector fused through the attention mechanism and represents the key operating information of the servo motor within different time windows. Pass Z fused as the input and pass it to the neural network for health scoring. The neural network uses a simple fully connected structure. The length of the input layer is the dimension of Z fused . The hidden layer consists of multiple fully connected layers, and the number of neurons in each layer decreases sequentially. The output is a single numerical value representing the health score S (ranging from 0 - 100). The neurons in all layers use the Sigmoid activation function.

[0169] The inventors of this application found in the research on the prediction and diagnosis method of the servo motor health status that traditional monitoring methods usually rely on simple time-domain or frequency-domain feature extraction. Although they can meet the requirements of certain basic scenarios, when faced with the complex and changeable working conditions of the servo motor, the prediction accuracy and robustness often fail to meet the actual requirements. Existing deep learning methods, although having high feature automatic extraction capabilities, are limited in their application in the actual industrial environment due to their dependence on a large amount of high-quality data during the training process. These difficulties lead to instability and inaccuracy in the fault prediction and health status assessment of the servo motor, restricting the efficient maintenance of industrial equipment and the extension of service life.

[0170] Based on the above analysis, the inventor deeply mines the operation data of the servo motor through a method combining multi-time window feature extraction and attention mechanism, and proposes a technical solution for extracting time domain, frequency domain, and time-frequency domain features from three time windows of the past 1 second, 2 seconds, and 3 seconds. On this basis, an attention mechanism model is designed, which can dynamically weight these features according to the actual data of the motor operation to generate a comprehensive representation, avoiding the deficiencies of manual feature selection and the uneven importance of different features. During the research process, the inventor verified the performance of the attention mechanism by combining the actual motor operation data, and improved the processing ability of multi-modal comprehensive features by optimizing the deep neural network structure, significantly enhancing the accuracy of health state prediction. Finally, through multiple verifications in the simulation platform and the actual motor operation environment, it is proved that the solution of this application can balance computational efficiency and accuracy and is applicable to the online monitoring and maintenance of servo motors in industrial scenarios.

[0171] In summary, the solution of this application realizes a multi-dimensional accurate description of the motor operation state by performing multi-time window sampling on the servo motor operation data and extracting time domain, frequency domain, and time-frequency domain features. At the same time, the attention mechanism is used to dynamically weight and fuse the importance of different time windows and feature domains, overcoming the deficiencies of existing methods in feature extraction and data fusion. On this basis, the comprehensive features are predicted through an optimized neural network, improving the accuracy and reliability of health state assessment. In addition, the solution of this application reduces the dependence on the amount of training data and computing resources, and can meet the real-time requirements in industrial scenarios while adapting to complex working condition changes.

[0172] The embodiment of this application also provides a fault determination device for a servo motor. It should be noted that the fault determination device for a servo motor in the embodiment of this application can be used to execute the fault determination method for a servo motor provided in the embodiment of this application. The device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0173] The following introduces the fault determination device for a servo motor provided in the embodiment of this application.

[0174] Figure 6 is a structural block diagram of a fault determination device for a servo motor according to an embodiment of this application. As Figure 6 shown, the device includes:

[0175] An acquisition unit 10 for acquiring feature data, where the feature data is data of key parameters during the operation of a servo motor, and the key parameters at least include one or more of temperature, current value, position, speed, and vibration signal. There are multiple pieces of the feature data, and the feature data corresponds to time windows one by one;

[0176] A fusion unit 20 for fusing multiple pieces of the feature data to obtain comprehensive feature data, where the fusion method at least includes weighted average;

[0177] A construction unit 30 for constructing a fault analysis model, where the fault analysis model is trained using multiple sets of training data, and each set of the multiple sets of training data includes historical comprehensive feature data obtained within a historical time period and the corresponding historical analysis result of the historical comprehensive feature data, where the historical analysis result is used to represent whether the servo motor has a fault during the historical time period;

[0178] An analysis unit 40 for inputting the comprehensive feature data into the fault analysis model to obtain the analysis result corresponding to the comprehensive feature data.

[0179] Through this embodiment, there are multiple pieces of feature data, which are sampled from different time windows, so as to reflect the operating state of the servo motor in different time periods, and thus multi-dimensional feature data can be extracted, providing a rich information source for subsequent feature fusion and state evaluation. Furthermore, multi-data fusion is performed to fuse the features of multiple time windows and multiple modalities into a comprehensive feature representation. The fused features can better reflect the key information of the motor operation, improving the accuracy of subsequent evaluation. Then, the state of the servo motor is evaluated according to the neural network model, and the operating state of the servo motor is accurately judged through machine learning, with high accuracy.

[0180] In the specific implementation process, the acquisition unit includes a first extraction module, a second extraction module, and a third extraction module. The first extraction module is used for extracting time-domain features of the key parameters to obtain first sub-feature data; the second extraction module is used for extracting frequency-domain features of the key parameters to obtain second sub-feature data; the third extraction module is used for extracting time-frequency domain features of the key parameters to obtain third sub-feature data.

[0181] In this solution, through time-domain, frequency-domain, and time-frequency-domain feature extraction, multiple dimensions of the servo motor's operating data can be comprehensively analyzed. It can not only capture the static and dynamic characteristics of the motor but also reveal the hidden spectral information and time-varying frequency characteristics in the signal. This multi-dimensional feature extraction method significantly improves the accuracy and reliability of fault prediction, can detect potential fault signs earlier, and provides strong support for preventive maintenance. At the same time, due to the real-time nature of time-domain feature extraction and the sensitivity of frequency-domain and time-frequency-domain feature extraction to complex faults, this method can meet the high requirements for motor health status monitoring in industrial scenarios, reduce the risk of production interruption caused by faults, and further improve the operating efficiency and safety of the equipment.

[0182] In some embodiments, the first extraction module includes a first calculation sub-module, a second calculation sub-module, a third calculation sub-module, and a fourth calculation sub-module. The first calculation sub-module is used to calculate the mean of the above key parameters according to the first formula, where the first formula is:

[0183]

[0184] μ represents the above mean, N represents the total number of samples within the above time window, and x i represents the data at the i-th sampling point; the second calculation sub-module is used to calculate the variance of the above key parameters according to the second formula, where the second formula is:

[0185]

[0186] σ 2 represents the above variance; the third calculation sub-module is used to calculate the peak value of the above key parameters according to the third formula, where the third formula is:

[0187] x peak =max(x i ), i = 1, 2, …, N,

[0188] x peak represents the above peak value; the fourth calculation sub-module is used to calculate the kurtosis of the above key parameters according to the fourth formula, where the fourth formula is:

[0189]

[0190] Kurtosis represents the above kurtosis.

[0191] In this solution, time-domain feature extraction can comprehensively evaluate the statistical characteristics of motor operation data by calculating the mean, variance, peak value, and kurtosis of key parameters, including the stability, volatility, maximum intensity, and peak degree of signal distribution of the motor. These time-domain features can help the system quickly identify whether the motor is in a normal or abnormal working state, providing direct and effective information for predicting possible motor faults, such as overload, wear, vibration, etc.

[0192] In the specific implementation process, the second extraction module includes a fifth calculation sub-module, a sixth calculation sub-module, a seventh calculation sub-module, an eighth calculation sub-module, and a ninth calculation sub-module. The fifth calculation sub-module is used to calculate the frequency components of the above key parameters according to the fifth formula, where the above fifth formula is:

[0193]

[0194] X(k) represents the above frequency component; the sixth calculation sub-module is used to calculate the main frequency of the above key parameters according to the sixth formula, where the above sixth formula is:

[0195]

[0196] f peak represents the above main frequency, |X(f k )| represents the spectral amplitude, and f k represents the frequency; the seventh calculation sub-module is used to calculate the total spectral energy of the above key parameters according to the seventh formula, where the above seventh formula is:

[0197]

[0198] E represents the above total spectral energy; the eighth calculation sub-module is used to calculate the spectral center rate of the above key parameters according to the eighth formula, where the above eighth formula is:

[0199]

[0200] f c represents the above spectral center rate; the ninth calculation sub-module is used to calculate the spectral bandwidth of the above key parameters according to the ninth formula, where the above ninth formula is:

[0201] B = f high - f low ,

[0202] B represents the above spectral bandwidth, f high represents the highest frequency at which the spectral energy exceeds the energy threshold, and f low represents the lowest frequency at which the spectral energy exceeds the energy threshold.

[0203] In this solution, frequency-domain feature extraction can reveal potential fault information related to frequency during motor operation by performing spectral analysis on the signal, including the main vibration frequency, energy distribution, concentration, and range of frequency distribution. These features are crucial for detecting and diagnosing faults in common components such as bearings and gears in servo motors. Compared with time-domain features, frequency-domain features can capture the operating characteristics of the motor from different perspectives, and combining the two can provide a more comprehensive health status assessment.

[0204] In some embodiments, the third extraction module includes a tenth calculation sub-module, an eleventh calculation sub-module, and a twelfth calculation sub-module. The tenth calculation sub-module is used to calculate the frequency distribution of the above key parameters according to the tenth formula, where the tenth formula is:

[0205]

[0206] W(a, b) represents the above frequency distribution, a represents the frequency resolution, b represents the time resolution, and ψ(t) represents the mother wavelet function; the eleventh calculation sub-module is used to calculate the band energy of the above key parameters according to the eleventh formula, where the eleventh formula is:

[0207]

[0208] E band represents the above band energy; the twelfth calculation sub-module is used to calculate the energy center time of the above key parameters according to the twelfth formula, where the twelfth formula is:

[0209]

[0210] t c represents the above energy center time.

[0211] In this solution, time-frequency domain feature extraction decomposes the signal into a time-frequency diagram representation through wavelet transform, so that the frequency distribution characteristics of the signal in different time periods can be analyzed. This analysis method is particularly effective for detecting transient events and faults of specific components in non-stationary signals. The calculation of band energy and energy center time can provide detailed information on the energy distribution of the signal at specific frequency bands and time points, helping to identify potential fault signs early, so as to take preventive maintenance measures, reduce equipment failure downtime, and improve the stability and efficiency of equipment operation.

[0212] In the specific implementation process, the above-mentioned feature data includes first feature data, second feature data, and third feature data, and the fusion unit includes a first transformation module, a second transformation module, a third transformation module, and a fusion module. The first transformation module is used to perform a linear transformation on the above-mentioned first feature data to obtain a first matrix; the second transformation module is used to perform a linear transformation on the above-mentioned second feature data to obtain a second matrix; the third transformation module is used to perform a linear transformation on the above-mentioned third feature data to obtain a third matrix; the fusion module is used to fuse according to the above-mentioned first matrix, the above-mentioned second matrix, and the above-mentioned third matrix to obtain the above-mentioned comprehensive feature data.

[0213] In this solution, the feature data in the time domain, frequency domain, and time-frequency domain are fused, and a comprehensive feature data containing multi-faceted information about the motor operating state can be generated. This kind of fusion makes full use of the advantages of different features, overcomes the limitations of single-feature representation, and improves the accuracy and comprehensiveness of health state prediction.

[0214] In some embodiments, the fusion module includes a thirteenth calculation sub-module, a fourteenth calculation sub-module, a fifteenth calculation sub-module, and a sixteenth calculation sub-module. The thirteenth calculation sub-module is used to calculate the correlation between different above-mentioned time windows according to the thirteenth formula, where the above-mentioned thirteenth formula is:

[0215]

[0216] A ij represents the above-mentioned correlation, Q i represents the above-mentioned first matrix corresponding to the i-th above-mentioned feature data, K j represents the above-mentioned second matrix corresponding to the j-th feature data, and T represents the above-mentioned time window; the fourteenth calculation sub-module is used to calculate the attention weight according to the fourteenth formula, where the above-mentioned fourteenth formula is:

[0217]

[0218] α ij represents the above-mentioned attention weight; the fifteenth calculation sub-module is used to calculate the weighted feature vector according to the fifteenth formula, where the above-mentioned fifteenth formula is:

[0219]

[0220] Z i represents the above-mentioned weighted feature vector, V j represents the above-mentioned third matrix corresponding to the j-th above-mentioned feature data; the sixteenth calculation sub-module is used to generate the above-mentioned comprehensive feature data according to the sixteenth formula, where the above-mentioned sixteenth formula is:

[0221]

[0222] Z fused Represents the above comprehensive feature data.

[0223] In this solution, the fusion process of the attention mechanism enables the model to automatically learn the importance of different time windows and feature domains. Through weighted summation and synthesis, comprehensive feature data that can comprehensively reflect the health status of the servo motor is generated. This data fusion method not only improves the accuracy of prediction but also ensures the adaptability of the model to complex working conditions and non-stationary signals.

[0224] The above-mentioned fault determination device of the servo motor includes a processor and a memory. The above-mentioned acquisition unit, fusion unit, construction unit, analysis unit, etc. are all stored in the memory as program units, and the corresponding functions are implemented by the processor executing the above program units stored in the memory. The above modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0225] The processor contains a kernel, and the kernel retrieves the corresponding program units from the memory. One or more kernels can be set, and by adjusting the kernel parameters, the problem of poor accuracy in fault detection of the servo motor in the prior art can be solved.

[0226] The memory may include non-permanent memory in the computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory includes at least one storage chip.

[0227] An embodiment of the present invention provides a computer-readable storage medium. The above computer-readable storage medium includes a stored program. Wherein, when the above program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned fault determination method of the servo motor.

[0228] An embodiment of the present invention provides a processor. The above processor is used to run a program. Wherein, when the above program runs, it executes the above-mentioned fault determination method of the servo motor.

[0229] An embodiment of the present invention provides a device. The device includes a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements at least the steps of the fault determination method of the servo motor. The device herein can be a server, a PC, a PAD, a mobile phone, etc.

[0230] A computer program product includes a non-volatile computer-readable storage medium. The above non-volatile computer-readable storage medium stores a computer program. When the above computer program is executed by a processor, it implements the steps of the above-mentioned fault determination method of the servo motor in each embodiment of the present application.

[0231] The present application also provides a servo motor fault detection system, including one or more processors, a memory, and one or more programs, wherein the above one or more programs are stored in the above memory and are configured to be executed by the above one or more processors, and the above one or more programs include methods for performing the fault determination of any of the above servo motors.

[0232] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order than here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0233] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0234] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions. These computer program instructions can be provided to the processors of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0235] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means realizes the functions in the process Figure 1one or more processes and / or blocks Figure 1 the functions specified in one or more blocks.

[0236] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one or more processes and / or blocks Figure 1 or more processes and / or the functions specified in one or more blocks.

[0237] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0238] The memory may include non-permanent memory in the computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0239] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for information storage. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0240] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0241] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various modifications and changes can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for determining a servo motor fault, characterized in that: include: Acquire characteristic data, wherein the characteristic data is data of key parameters in the operation process of the servo motor, wherein the key parameters include at least one or more of temperature, current value, position, speed, and vibration signal, and there are multiple characteristic data, and the characteristic data and the time window correspond one to one; Fusion of the plurality of feature data to obtain comprehensive feature data, wherein the fusion method at least includes weighted averaging; Constructing a fault analysis model, wherein the fault analysis model is obtained by training using multiple sets of training data, each set of training data in the multiple sets of training data includes historical comprehensive feature data acquired in a historical time period and historical analysis results corresponding to the historical comprehensive feature data, wherein the historical analysis results are used to characterize whether the servo motor has a fault in the historical time period; The comprehensive feature data is input into the fault analysis model to obtain analysis results corresponding to the comprehensive feature data.

2. The method according to claim 1, characterized in that Get feature data, including: Performing time domain feature extraction on the key parameter to obtain first sub-feature data; Performing frequency domain feature extraction on the key parameters to obtain second sub-feature data; Perform time-frequency domain feature extraction on the key parameters to obtain third sub-feature data.

3. The method according to claim 2, characterized in that Performing time domain feature extraction on the key parameters to obtain first sub-feature data includes: According to the first formula, the mean value of the key parameter is calculated, wherein the first formula is: μ represents the mean value, N represents the total number of samples in the time window, and x i Represents the data of the i-th sampling point; According to the second formula, the variance of the key parameter is calculated, wherein the second formula is: σ 2 represents the variance; According to the third formula, the peak value of the key parameter is calculated, wherein the third formula is: x peak =max(x i ),i=1,2,…,N, x peak represents said peak value; According to the fourth formula, the kurtosis of the key parameter is calculated, wherein the fourth formula is: Kurtosis represents the kurtosis.

4. The method according to claim 2, characterized in that: Perform frequency domain feature extraction on the key parameters to obtain second sub-feature data, including: According to the fifth formula, the frequency component of the key parameter is calculated, wherein the fifth formula is: X(k) represents the frequency component; According to the sixth formula, the main frequency of the key parameter is calculated, wherein the sixth formula is: f peak represents the main frequency, |X(f k )| represents the spectrum amplitude, f k Indicates frequency; According to the seventh formula, the total spectrum energy of the key parameter is calculated, wherein the seventh formula is: E represents the total energy of the spectrum; According to the eighth formula, the spectrum center rate of the key parameter is calculated, wherein the eighth formula is: f c represents the center rate of the spectrum; According to the ninth formula, the spectrum bandwidth of the key parameter is calculated, wherein the ninth formula is: B=fhigh-flow, B represents the spectrum bandwidth, f high Indicates the highest frequency where the spectrum energy exceeds the energy threshold, f low Indicates the lowest frequency at which the spectrum energy exceeds the energy threshold.

5. The method according to claim 2, characterized in that: Performing time-frequency domain feature extraction on the key parameters to obtain third sub-feature data includes: According to the tenth formula, the frequency distribution of the key parameter is calculated, wherein the tenth formula is: W(a,b) represents the frequency distribution, a represents the frequency resolution, b represents the time resolution, ψ(t) represents the mother wavelet function; According to the eleventh formula, the frequency band energy of the key parameter is calculated, wherein the eleventh formula is: E band represents the energy of the frequency band; According to the twelfth formula, the energy center time of the key parameter is calculated, wherein the twelfth formula is: t c Indicates the energy center time.

6. The method according to claim 2, characterized in that The feature data includes first feature data, second feature data and third feature data. A plurality of the feature data are fused to obtain comprehensive feature data, including: Performing a linear transformation on the first characteristic data to obtain a first matrix; Performing a linear transformation on the second characteristic data to obtain a second matrix; Performing a linear transformation on the third characteristic data to obtain a third matrix; The comprehensive feature data is obtained by fusing the first matrix, the second matrix and the third matrix.

7. The method according to claim 6, characterized in that The comprehensive feature data is obtained by fusing the first matrix, the second matrix and the third matrix, including: According to the thirteenth formula, the correlation between different time windows is calculated, wherein the thirteenth formula is: A ij represents the correlation, Q i represents the first matrix corresponding to the i-th feature data, K j represents the second matrix corresponding to the j-th feature data, and T represents the time window; According to the fourteenth formula, the attention weight is calculated, wherein the fourteenth formula is: α ij represents the attention weight; According to the fifteenth formula, the weighted eigenvector is calculated, wherein the fifteenth formula is: Z i represents the weighted feature vector, V j Represents the third matrix corresponding to the j-th feature data; According to the sixteenth formula, the comprehensive feature data is generated, wherein the sixteenth formula is: Z fused Represents the comprehensive feature data.

8. A servo motor fault determination device, characterized in that: include: An acquisition unit, used for acquiring characteristic data, wherein the characteristic data is data of key parameters in the operation process of the servo motor, wherein the key parameters at least include one or more of temperature, current value, position, speed, and vibration signal, and there are multiple characteristic data, and the characteristic data and the time window correspond one to one; A fusion unit, used for fusing the plurality of feature data to obtain comprehensive feature data, wherein the fusion method at least includes weighted averaging; A construction unit, configured to construct a fault analysis model, wherein the fault analysis model is obtained by training using multiple sets of training data, each set of training data in the multiple sets of training data includes historical comprehensive feature data acquired in a historical time period, and historical analysis results corresponding to the historical comprehensive feature data, wherein the historical analysis results are used to characterize whether the servo motor has a fault in the historical time period; The analysis unit is used to input the comprehensive feature data into the fault analysis model to obtain the analysis result corresponding to the comprehensive feature data.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the servo motor fault determination method according to any one of claims 1 to 7 are implemented.

10. A servo motor fault detection system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for executing a fault determination method for a servo motor as described in any one of claims 1 to 7.

Images