Motor operation fault prediction method and system

Abstract

The invention provides a motor operation fault prediction method and system. The method comprises the following steps: collecting vibration signals and temperature data in operation in real time; frequency domain features, statistical features and time sequence features are extracted from the vibration signals and the temperature data to form a feature set; obtaining a training set, wherein the training set comprises a plurality of positive samples and a plurality of negative samples; training a fault prediction model based on the training set, wherein the fault prediction model comprises a support vector machine SVM sub-model and a long short-term memory network LSTM sub-model; and predicting a motor operation fault based on the trained fault prediction model. The problems that motor bearing fault predictive maintenance in the prior art is hysteretic and inaccurate, and an existing method cannot fully combine the advantages of frequency domain characteristics and time sequence analysis are solved.

Description

Technical Field

[0001] The present invention relates to the field of motor fault prediction, and more specifically, to a motor operation fault prediction method and system. Background Art

[0002] In the existing industrial control system, the motor bearing is a key transmission component, and its health is directly related to the continuous and stable operation of the production line. Once the motor bearing fails, such as wear, poor lubrication or foreign matter intrusion, it will often cause the motor operation to be obstructed or even stuck, directly causing production stagnation. This failure not only affects production efficiency, but may also cause chain damage to other precision components inside the motor, greatly increasing maintenance costs. In addition, bearing failures can also cause abnormal noise, increased vibration and increased operating temperature, further aggravating the instability of the equipment, reducing product quality, and posing a potential threat to production safety.

[0003] The traditional method of maintaining motor bearings is highly dependent on manual regular inspections and the experience and judgment of maintenance personnel. This method is not only time-consuming and labor-intensive, but also has significant lags and inaccuracies. Regular inspections may not capture the early signs of bearing failures, and by the time the problem appears, it has often had an impact on production. In addition, experience-based judgments are limited by individual differences, and it is difficult to ensure consistency and accuracy, which can easily lead to missed faults or misjudgments. Although some studies have attempted to apply technology for predictive maintenance, most methods have failed to fully combine the advantages of frequency domain features and time series analysis, resulting in prediction accuracy and timeliness that need to be improved. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of lag and inaccuracy in the predictive maintenance of motor bearing faults in the prior art and the failure of the existing motor fault prediction method to fully combine the advantages of frequency domain characteristics and time series analysis, and to provide a motor operation fault prediction method and system.

[0005] According to a first aspect of the present invention, a method for predicting motor operation faults is provided, comprising:

[0006] Step S1, based on a high-sensitivity vibration sensor and a temperature sensor, real-time collection of vibration signals and temperature data during motor operation;

[0007] Step S2, preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing;

[0008] Step S3, extracting frequency domain features closely related to the fault from the vibration signal, and extracting statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set;

[0009] Step S4, obtaining a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a motor without faults, each of the negative samples includes a feature set of a motor with faults, and the fault type is marked;

[0010] Step S5, training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model;

[0011] Step S6, predicting motor operation faults based on the trained fault prediction model.

[0012] Based on the above technical solution, the present invention can also make the following improvements.

[0013] Optionally, the vibration sensor is a piezoelectric acceleration sensor, which is installed on the outside of the bearing shell of the motor to collect vibration signals during the operation of the motor; the temperature sensor is a thermocouple temperature sensor with a high response speed, which is close to the motor housing or the internal hot spot area to monitor the motor temperature in real time;

[0014] The vibration signal sensed by the vibration sensor and the motor temperature data sensed by the temperature sensor are collected based on the data acquisition card, and the collected vibration signal and temperature data are transmitted to a data processing center.

[0015] Optionally, the step S2, preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing, includes:

[0016] A digital filter is used to low-pass filter the vibration signal to remove high-frequency noise in the vibration signal and retain low-frequency vibration components related to motor failure; a moving average or median filter is used to smooth temperature fluctuations;

[0017] Eliminate the abnormal part of the vibration signal that exceeds the peak value of the normal vibration range, and standardize the vibration signal using a Z-score standardization method;

[0018] Abnormal temperature data that exceeds the temperature range during normal operation of the motor is eliminated.

[0019] Optionally, step S3, extracting frequency domain features closely related to the fault from the vibration signal, includes:

[0020] Performing FFT fast Fourier transform on the preprocessed vibration signal to convert the time domain signal into a frequency domain signal;

[0021] identifying a specific fault frequency and its harmonic components closely related to the bearing fault from the frequency domain signal;

[0022] The amplitude and phase of the specific fault frequency are calculated, and the ratio of each harmonic to the fundamental frequency and the relative amplitude between each harmonic are calculated to form a frequency domain feature.

[0023] Optionally, in step S3, extracting statistical features and time series features from the temperature data includes:

[0024] Calculate the mean and variance of the temperature data within a set time period, wherein the mean and variance of the temperature data constitute statistical features;

[0025] Based on the time series analysis method, the trend line of the temperature data changing over time and the rate of change of the temperature are extracted, and the trend line of the temperature data changing over time and the rate of change of the temperature constitute the time series characteristics.

[0026] Optionally, in step S5, a fault prediction model is trained based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model, including:

[0027] Training the support vector machine SVM sub-model based on the training set, and calculating the first loss of the support vector machine SVM sub-model according to the fault prediction result and the actual fault label of each training sample after passing through the support vector machine SVM sub-model;

[0028] Optimizing the model parameters of the support vector machine SVM sub-model based on the first loss until the first loss is less than a first preset threshold, and obtaining the trained support vector machine SVM sub-model;

[0029] Obtaining samples in the training set that are identified as having a failure risk by the trained support vector machine SVM sub-model;

[0030] Training the LSTM sub-model based on samples with fault risks in the training set, and calculating a second loss of the LSTM sub-model based on a fault prediction result and an actual fault label of each sample with fault risk after passing through the LSTM sub-model;

[0031] Based on the second loss, the model parameters of the long short-term memory network LSTM sub-model are optimized until the second loss is less than a second preset threshold, and the trained long short-term memory network LSTM sub-model is obtained.

[0032] Optionally, the step S6, predicting the motor operation fault based on the trained fault prediction model, includes:

[0033] Based on high-sensitivity vibration sensors and temperature sensors, the vibration signals and temperature data of the motor during operation in the current set time period are collected;

[0034] Preprocessing the vibration signal and the temperature data, and extracting frequency domain features in the vibration signal and statistical features and time series features in the temperature data to form a current motor feature set;

[0035] Inputting the current motor feature set into the trained support vector machine SVM sub-model to obtain a first prediction result output by the support vector machine SVM sub-model;

[0036] If the first prediction result is no fault, then the current running state of the motor is no fault;

[0037] If the first prediction result is a fault, the current motor feature set is input into the trained long short-term memory network LSTM sub-model to obtain a second prediction result output by the long short-term memory network LSTM sub-model, and the second prediction result is the current operating state of the motor.

[0038] According to a second aspect of the present invention, there is provided a motor operation fault prediction system, comprising:

[0039] The acquisition module is used to collect vibration signals and temperature data of the motor in real time based on high-sensitivity vibration sensors and temperature sensors;

[0040] A preprocessing module, used for preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing;

[0041] An extraction module, used to extract frequency domain features closely related to the fault from the vibration signal, and to extract statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set;

[0042] An acquisition module is used to acquire a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a motor without faults, each of the negative samples includes a feature set of a motor with faults, and the fault type is marked;

[0043] A training module, used for training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model;

[0044] The prediction module is used to predict the motor operation fault based on the trained fault prediction model.

[0045] The present invention provides a motor operation fault prediction method and system. Compared with the prior art, the technical solution of the present application has the following beneficial technical effects:

[0046] (1) High-sensitivity sensors are used to capture motor vibration and temperature data in real time. Data cleaning and standardization significantly improve the quality of the data source, laying the foundation for subsequent analysis;

[0047] (2) Multi-dimensional feature extraction combines frequency domain features, statistical features, and time series features to deeply analyze the bearing status, capture subtle fault signs, enrich information input, strengthen model construction, and improve prediction accuracy;

[0048] (3) The SVM and LSTM prediction models complement each other, combining fast classification with long-term trend capture to accurately predict faults. Supervised learning and cross-validation optimization models ensure reliable predictions. They are deployed in industrial control systems to achieve real-time risk assessment and automatic alarms, shorten fault response, reduce unplanned downtime, and ensure production continuity. The machine learning feedback mechanism is combined with customized maintenance strategies to form a closed-loop optimization system. The maintenance strategy is dynamically adjusted according to the equipment status to ensure timely and effective maintenance, further improving the pertinence and efficiency of predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a motor operation fault prediction method provided by the present invention;

[0050] Figure 2 A schematic diagram of the structure of a motor operation fault prediction system provided by the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not subject to the constraints of the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0052] See also Figure 1 , a flow chart of a method for predicting motor operation faults is provided, the method comprising:

[0053] Step S1, based on a high-sensitivity vibration sensor and a temperature sensor, real-time collection of vibration signals and temperature data during operation.

[0054] It is understandable that by using high-sensitivity vibration sensors and temperature sensors, the vibration and temperature data of the motor during operation are collected in real time.

[0055] Specifically, a piezoelectric acceleration sensor (vibration sensor) is installed on the outside of the motor's bearing housing to capture tiny vibration changes during motor operation. A thermocouple temperature sensor with high response speed is installed close to the motor housing or internal hot spot area to monitor the motor temperature in real time.

[0056] A data acquisition card or embedded system is used to simultaneously collect the vibration signal sensed by the vibration sensor and the temperature data sensed by the temperature sensor to ensure that the vibration signal and temperature data can be collected synchronously. The data acquisition card or embedded system transmits the sensor data (including vibration data and temperature data) to the data processing center in real time through communication protocols such as RS-485, CAN bus or Ethernet.

[0057] The present invention uses a piezoelectric acceleration sensor installed on the outside of the motor's bearing housing. This design greatly improves the ability to capture tiny vibration changes when the motor is running. With its high sensitivity and wide-band response characteristics, the piezoelectric sensor can accurately sense and convert it into an electrical signal. Even tiny vibration anomalies can be captured in time, providing a reliable basis for early fault warning.

[0058] Equipped with a thermocouple temperature sensor with high response speed, and placed close to the motor casing or internal hot spot area, it ensures real-time monitoring of the motor temperature. Thermocouple sensors can accurately reflect the temperature changes during motor operation due to their wide temperature measurement range, high accuracy and fast response. Especially under high load or abnormal operating conditions, they can quickly capture temperature anomalies and effectively prevent motor damage caused by overheating.

[0059] The use of data acquisition cards or embedded systems enables the synchronous collection of vibration and temperature data. This design ensures the consistency and integrity of the data and avoids misjudgment or missed judgment due to data asynchrony. At the same time, the efficient data collection and processing mechanism also provides a solid foundation for subsequent data analysis and fault diagnosis.

[0060] Sensor data is transmitted to the data processing center in real time through modern communication protocols such as RS-485, CAN bus or Ethernet. These communication protocols have the advantages of fast transmission speed, strong anti-interference ability and long transmission distance, ensuring the real-time and reliability of data. The data processing center can receive and analyze these data instantly and respond quickly to changes in the motor's operating status, providing strong support for production management and equipment maintenance.

[0061] Step S2, preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing.

[0062] It is understandable that for the collected vibration signals and temperature data, data cleaning algorithms are used to remove noise and outliers, and data consistency is ensured through standardization processing to provide a high-quality data source for subsequent analysis.

[0063] When cleaning the data, digital filtering techniques such as Butterworth low-pass filter are applied to the vibration data to remove high-frequency noise in the vibration signal and retain the low-frequency vibration components related to motor failure. For temperature data, simple moving average or median filtering is used to smooth short-term temperature fluctuations.

[0064] Statistical tests are implemented to identify outliers in vibration signals and temperature data. For vibration data, special attention is paid to peak values ​​that are far beyond the normal vibration range. Outliers are removed or marked to prevent them from interfering with subsequent analysis. Z-score standardization is used when standardizing data, and the mean of vibration data is adjusted to 0 and the standard deviation is adjusted to 1 to eliminate amplitude differences caused by hardware differences between different sensors or motors.

[0065] The temperature data is relativized according to the temperature range during normal operation of the motor, and the average temperature during normal operation is subtracted from the temperature to obtain the deviation relative to the average temperature.

[0066] The cleaned and standardized vibration and temperature data are integrated into a unified format and stored in a high-performance data storage system so that subsequent data analysis, model training, real-time monitoring and other applications can access them efficiently.

[0067] Among them, in the data cleaning stage, digital filtering technology such as Butterworth low-pass filter is applied to vibration data. This measure effectively removes high-frequency noise and retains low-frequency vibration components that are closely related to motor failure. This processing not only improves the signal-to-noise ratio of vibration data, but also reduces the interference of noise on fault feature extraction, making subsequent analysis more accurate and reliable. At the same time, for temperature data, simple moving average or median filtering is used to smooth short-term temperature fluctuations. This method effectively reduces the random fluctuations of temperature data, makes the temperature trend clearer, and helps to detect temperature anomalies in a timely manner.

[0068] Secondly, statistical tests are implemented to identify outliers in vibration and temperature data, with special attention paid to peak values ​​in vibration data that far exceed the normal range. This step avoids their adverse effects on subsequent analysis by eliminating or marking outliers, ensuring the purity of the data set and the accuracy of the analysis results. The processing of outliers is a crucial part of data preprocessing, which is directly related to the performance and reliability of the subsequent fault diagnosis model.

[0069] Furthermore, in the data standardization stage, the Z-score standardization method is used to process the vibration data, adjusting the mean to 0 and the standard deviation to 1. This measure eliminates the amplitude differences between different sensors or motors due to hardware differences, making the vibration data comparable between different motors or sensors, which facilitates subsequent data analysis and model training.

[0070] At the same time, the temperature data is relativized, that is, the average temperature during normal operation is subtracted from the temperature to obtain the deviation relative to the average temperature. This processing makes the temperature data more intuitively reflect the changes in the motor's operating status, which facilitates fault warning and diagnosis.

[0071] Finally, the cleaned and standardized vibration signals and temperature data are integrated into a unified format and stored in a high-performance data storage system. This measure ensures the consistency and integrity of the data and provides efficient data access capabilities for subsequent data analysis, model training, and real-time monitoring applications. The high-performance data storage system can also support the rapid reading and writing of large-scale data and efficient management, further improving the efficiency of data processing and analysis.

[0072] Step S3, extracting frequency domain features closely related to the fault from the vibration signal, and extracting statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set.

[0073] Exemplarily, step S3, extracting frequency domain features closely related to the fault from the vibration signal, includes: performing FFT fast Fourier transform on the preprocessed vibration signal to convert the time domain signal into a frequency domain signal; identifying specific fault frequencies and their harmonic components closely related to the bearing fault from the frequency domain signal; calculating the amplitude and phase of the specific fault frequency, as well as calculating the ratio of each harmonic to the fundamental frequency and the relative amplitude between each harmonic to form frequency domain features.

[0074] Specifically, when extracting features from vibration signals and temperature data, the preprocessed vibration signal is subjected to a fast Fourier transform (FFT) when extracting vibration signal features to convert the time domain signal into a frequency domain signal. This step is the basis for extracting frequency domain features and can reveal the distribution characteristics of the signal in frequency.

[0075] The converted frequency domain signal is analyzed to identify specific fault frequencies that are closely related to bearing faults. These frequencies are usually related to the geometric dimensions, rotational speed and fault type of the bearing. The amplitude, phase and other parameters of these specific fault frequencies are calculated as important frequency domain features. After identifying the specific fault frequencies, the harmonic components of these frequencies are further analyzed. The presence and intensity of the harmonic components can reflect the severity and complexity of the fault. The ratio of harmonics to the fundamental frequency, the relative amplitude between harmonics and other features are calculated to more accurately describe the fault state of the bearing. The amplitude and phase of the specific fault frequency, the ratio of the harmonic to the fundamental frequency, and the relative amplitude between the harmonics constitute the frequency domain characteristics of the vibration signal.

[0076] The present invention converts the time domain signal into a frequency domain signal by performing FFT (Fast Fourier Transform) on the preprocessed vibration signal. This step lays a solid foundation for extracting frequency domain features. The FFT transform not only reveals the distribution characteristics of the signal in frequency, but also makes those periodic components that are difficult to detect in the time domain become clearly discernible in the frequency domain. This conversion process greatly enriches the analysis dimension of the vibration signal, allowing us to have a deeper understanding of the intrinsic structure and characteristics of the signal.

[0077] Secondly, analyzing the FFT transformation results and identifying the specific fault frequencies that are closely related to the bearing fault is a key step in feature extraction. These specific fault frequencies are usually closely related to the bearing's geometric dimensions, rotational speed and fault type, and are a direct reflection of the bearing fault. By accurately identifying these fault frequencies, the fault source can be quickly located, providing a strong basis for subsequent fault diagnosis.

[0078] Furthermore, the amplitude, phase and other parameters of these specific fault frequencies are calculated as important frequency domain features. These features not only directly reflect the intensity and directionality of the fault, but also provide rich information for subsequent fault mode identification. By comparing the frequency domain features under different fault states, accurate classification and identification of faults can be achieved.

[0079] In addition, after identifying the specific fault frequency, it is also crucial to further analyze the harmonic components of these frequencies. The existence and intensity of harmonic components can often reflect the severity and complexity of the fault. By calculating the ratio of harmonics to fundamental frequency, the relative amplitude between harmonics and other characteristics, the fault state of the bearing can be described more finely, providing strong support for early warning and in-depth analysis of faults.

[0080] Exemplarily, in step S3, statistical features and time series features are extracted from the temperature data, including: calculating the mean and variance of the temperature data within a set time period, the mean and variance of the temperature data constitute statistical features; based on a time series analysis method, extracting the trend line of the temperature data over time and the rate of change of temperature, the trend line of the temperature data over time and the rate of change of temperature constitute time series features.

[0081] Specifically, when extracting temperature data features, the mean, variance and other statistical features of the temperature data within the collection period are calculated. These features can reflect the overall level and fluctuation of the temperature. The mean reflects the average operating temperature of the bearing. The variance reveals the severity of the temperature change, which may be related to factors such as the lubrication status of the bearing and load changes.

[0082] Based on the time series analysis method, the trend line of the temperature data changing over time is extracted. The trend line can reveal the long-term trend of temperature changing over time and help predict future temperature changes. The rate of change of temperature over time, that is, the rate at which the temperature changes over time, is calculated. The magnitude and positive and negative of the rate of change can reflect the stability of the bearing's operating state and whether there is a sudden temperature rise or drop. The trend line of the temperature data changing over time and the rate of change of temperature over time constitute the time series characteristics of the temperature data.

[0083] All features extracted from vibration signals and temperature data are integrated to form a comprehensive feature set. Feature selection is performed based on the importance, relevance, and needs of subsequent analysis to remove redundant or irrelevant features, reduce the dimension of the feature set, and improve the efficiency and accuracy of subsequent analysis.

[0084] By calculating the mean, variance and other statistical characteristics of temperature data, we can obtain a quantitative description of the overall level and fluctuation of temperature data. The mean, as the average operating temperature of the bearing, directly reflects the basic thermal state of the bearing during operation; and the variance reveals the severity of temperature changes. This information is of great significance for evaluating factors such as the lubrication status and load changes of the bearing. These statistical characteristics provide basic quantitative indicators for subsequent fault diagnosis and prediction.

[0085] Secondly, the temperature data trend line extracted based on the time series analysis method further reveals the long-term trend of temperature changes over time. The existence of the trend line not only helps to understand the regularity and periodicity of temperature changes, but also provides an important basis for predicting future temperature changes. This is of great significance for timely detection of temperature anomalies and prevention of motor failures caused by overheating.

[0086] Furthermore, calculating the temperature change rate, that is, the rate at which the temperature changes over time, provides direct information about the stability of the bearing's operating state. The magnitude and sign of the change rate can sensitively reflect whether there is a sudden temperature rise or drop in the bearing. These phenomena are often early signs of bearing failure or abnormal operating conditions. By monitoring the change rate, potential failure risks can be discovered and responded to in a timely manner.

[0087] Finally, all the features extracted from the vibration signal and temperature data are integrated to form a comprehensive feature set, and feature selection is performed based on the importance, relevance, and needs of subsequent analysis. This process not only removes redundant or irrelevant features and reduces the dimension of the feature set, but also improves the efficiency and accuracy of subsequent analysis. The comprehensive feature set provides strong support for building an efficient and accurate fault diagnosis and prediction model, enabling the model to capture information on the motor's operating status more comprehensively and improve the accuracy of fault diagnosis and prediction precision.

[0088] Step S4, obtaining a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a non-faulty motor, each of the negative samples includes a feature set of a faulty motor, and the fault type is marked.

[0089] It can be understood that after step S4 extracts the feature sets from the vibration signal and the temperature data respectively, a data set is organized, wherein the data set includes positive samples and negative samples. The so-called positive samples refer to the feature set data of the motor running in a normal state, and the so-called negative samples refer to the feature set data of the motor in a fault state. When the motor is in a fault state, the fault type of the motor is marked.

[0090] Collect historical fault data and normal operation data from data sources such as system or device logs and sensors to ensure that the data covers a variety of operating conditions and fault types to improve the generalization ability of the model. Identify and process missing values, outliers, etc. to ensure the integrity and accuracy of the data, smooth or interpolate time series data to fill in gaps or missing parts in the data, and extract features useful for fault prediction from time series data, such as statistical features, trend features, seasonal features, etc.

[0091] According to the requirements of the SVM sub-model and the LSTM network sub-model, the feature set is converted into the appropriate format and dimension, and the dataset is divided into training set, validation set, and test set, usually with a ratio of 70%, 15%, 15% or similar, ensuring that each dataset contains normal and faulty samples to maintain data balance.

[0092] Step S5: training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model.

[0093] It is understandable that the fault prediction model is trained based on the training set, wherein the fault prediction model includes a support vector machine SVM sub-model and a long short-term memory network LSTM sub-model. Then, the support vector machine SVM sub-model and the long short-term memory network LSTM sub-model are trained based on the training set respectively.

[0094] Exemplarily, the support vector machine SVM sub-model is trained based on the training set, and the first loss of the support vector machine SVM sub-model is calculated according to the fault prediction result and the actual fault label of each training sample after passing through the support vector machine SVM sub-model;

[0095] Based on the first loss, the model parameters of the support vector machine SVM sub-model are optimized until the first loss is less than a first preset threshold, thereby obtaining the trained support vector machine SVM sub-model.

[0096] Specifically, the support vector machine SVM sub-model is trained based on the original training set. During the training process, the support vector machine SVM sub-model parameters are adjusted by calculating the loss until the loss reaches a certain accuracy, that is, the loss is less than the set threshold, and the trained support vector machine SVM sub-model is obtained.

[0097] By selecting appropriate SVM kernel functions such as RBF, linear, etc., decisions are made according to the characteristics of the data and the complexity of the classification task. Strategies such as grid search and random search are used, combined with cross-validation such as K-fold cross-validation to optimize key parameters such as SVM C (regularization parameter) and γ (coefficient of the RBF kernel). Through multiple iterations and evaluations, the optimal parameter combination is found. The training set data is used to train the SVM model, and the validation set data is used to evaluate the classification performance of the model, including accuracy, recall rate, F1 score and other indicators. The model parameters or structure are adjusted according to the verification results until a satisfactory classification effect is achieved.

[0098] Obtaining samples in the training set that are identified as having a failure risk by the trained support vector machine SVM sub-model;

[0099] Training the LSTM sub-model based on samples with fault risks in the training set, and calculating a second loss of the LSTM sub-model based on a fault prediction result and an actual fault label of each sample with fault risk after passing through the LSTM sub-model;

[0100] Based on the second loss, the model parameters of the long short-term memory network LSTM sub-model are optimized until the second loss is less than a second preset threshold, and the trained long short-term memory network LSTM sub-model is obtained.

[0101] Among them, the LSTM sub-model of the long short-term memory network includes an input layer, an LSTM layer, a Dropout layer, a fully connected layer, and an output layer, and key parameters such as the number of LSTM layers, the number of units, and the activation function are determined. The preprocessed time series data is divided into time windows of fixed length to generate the input sequence of the LSTM network, and corresponding labels such as fault type and fault time are generated for each input sequence for supervised learning. The learning rate, optimizer such as Adam, loss function such as MSE, cross entropy and other hyperparameters are set, and appropriate adjustments are made according to task requirements and data characteristics. The LSTM sub-model is trained using the training set data, and the performance indicators on the loss function and the validation set, such as accuracy and mean square error, are monitored during the training process. The early stopping method or learning rate adjustment strategy is used to prevent overfitting and improve training efficiency. The model parameters and structure are further optimized through methods such as cross-validation.

[0102] Exemplarily, a preliminary prediction is made on the original training set based on the trained support vector machine SVM sub-model, and when the motor operating state output by the support vector machine SVM sub-model is at risk of failure, these sample sets at risk of failure are extracted. The long short-term memory network LSTM sub-model is trained based on these sample sets at risk of failure, and during the training process, the parameters of the long short-term memory network LSTM sub-model are adjusted by the calculated loss until the loss reaches a certain accuracy, that is, the loss is less than the set threshold, and then the trained long short-term memory network LSTM sub-model is obtained.

[0103] Specifically, the integration strategy of the support vector machine (SVM) sub-model and the long short-term memory (LSTM) network sub-model is designed. For example, the support vector machine (SVM) sub-model can be used to preliminarily classify the feature set data and screen out samples that may have fault risks. Then, the long short-term memory (LSTM) network sub-model is used to perform a more in-depth time series analysis on these high-risk samples, and the data transmission method and interface between the support vector machine (SVM) sub-model and the long short-term memory (LSTM) network sub-model are determined. The integrated model is trained as a whole to ensure the collaborative work between the support vector machine (SVM) sub-model and the long short-term memory (LSTM) network sub-model. The performance of the integrated model is evaluated using the validation set data, including indicators such as fault prediction accuracy, recall rate, and F1 score. The integration strategy is adjusted or the parameters of a single model are optimized based on the evaluation results. The final model is comprehensively evaluated using the test set data to ensure the generalization ability and prediction accuracy of the model. The trained fault prediction model is deployed to the actual system to realize real-time monitoring and fault warning functions. In the data preparation and preprocessing stage, the generalization ability of the model is greatly enhanced by widely collecting historical fault data and normal operation data and ensuring that the data covers a variety of operating conditions and fault types. At the same time, the integrity and accuracy of the data are carefully processed, including the identification and processing of missing values ​​and outliers, as well as the smoothing or interpolation of time series data. These measures effectively improve the data quality and lay a solid foundation for subsequent analysis.

[0104] In addition, rich features are extracted from time series data, such as statistical features, trend features, seasonal features, etc., and the data format and dimension are adjusted according to model requirements, which further enriches the input information of the model and helps improve prediction accuracy.

[0105] Secondly, in the SVM preliminary classification model construction stage, by selecting appropriate SVM kernel functions and optimization strategies and combining cross-validation technology, fine adjustment of key parameters of the support vector machine SVM sub-model is achieved, thereby ensuring the excellent performance of the model in classification tasks. This process not only improves the classification accuracy of the model, but also continuously adjusts the model parameters through evaluation and feedback of the validation set until a satisfactory classification effect is achieved, providing a reliable classification basis for the subsequent integrated model construction.

[0106] Next, in the LSTM time series analysis model construction stage, by carefully designing the LSTM network structure and adjusting the parameters in combination with the characteristics of time series data, in-depth analysis and feature extraction of time series data were achieved. The LSTM model can capture long-term dependencies in the data and is particularly suitable for the prediction task of time series data. At the same time, by monitoring the loss function and validation set performance indicators during the training process, early stopping or learning rate adjustment strategies are used to prevent overfitting, further improving the training efficiency and prediction accuracy of the model.

[0107] Finally, in the integrated model construction and evaluation stage, the integration strategy of the support vector machine (SVM) sub-model and the long short-term memory (LSTM) network sub-model was designed to achieve complementary advantages of the two models. As a preliminary classifier, the support vector machine (SVM) sub-model can quickly screen out samples that may have failure risks, while the long short-term memory (LSTM) network sub-model conducts a more in-depth time series analysis of these high-risk samples, thereby improving the accuracy and recall rate of fault prediction.

[0108] In addition, through overall training and validation set evaluation, the integration strategy and individual model parameters are continuously optimized to ensure the collaborative working ability and final performance of the integrated model. Finally, the trained integrated model is deployed to the actual system to realize real-time monitoring and fault warning functions, providing strong guarantee for the stable operation of the equipment or system.

[0109] Step S6, predicting motor operation faults based on the trained fault prediction model.

[0110] Exemplarily, the step S6, predicting the motor operation fault based on the trained fault prediction model, includes:

[0111] Based on high-sensitivity vibration sensors and temperature sensors, the vibration signals and temperature data of the motor during operation in the current set time period are collected;

[0112] Preprocessing the vibration signal and the temperature data, and extracting frequency domain features in the vibration signal and statistical features and time series features in the temperature data to form a current motor feature set;

[0113] Inputting the current motor feature set into the trained support vector machine SVM sub-model to obtain a first prediction result output by the support vector machine SVM sub-model;

[0114] If the first prediction result is no fault, then the current running state of the motor is no fault;

[0115] If the first prediction result is a fault, the current motor feature set is input into the trained long short-term memory network LSTM sub-model to obtain a second prediction result output by the long short-term memory network LSTM sub-model, and the second prediction result is the current operating state of the motor.

[0116] It is understandable that after training the support vector machine SVM sub-model and the long short-term memory network LSTM sub-model through step S5, the trained support vector machine SVM sub-model and the long short-term memory network LSTM sub-model are integrated into a complete fault prediction model to ensure that the data interface and interaction logic between the two sub-models are correct. The integrated fault prediction model is deployed to the prediction and alarm module of the industrial control system, and the performance of the model is tuned in the industrial control system to ensure the efficiency and accuracy of the model in real-time data processing and prediction. According to system resources and data processing requirements, adjust the parallelism of the model operation, memory usage and other parameters, collect equipment operation data in real time through sensors, logs and other data sources of the industrial control system, these data include key parameters such as temperature, vibration, current, pressure, etc., and pre-process the collected real-time data, including data cleaning, feature extraction and normalization.

[0117] The preprocessed data stream is input into the fault prediction model. First, the support vector machine SVM sub-model is used to preliminarily classify the data to determine whether there are signs of faults. When there are signs of faults, the long short-term memory network LSTM sub-model is used to conduct in-depth analysis of the characteristic data with fault signs to capture long-term change trends and short-term fluctuations. By integrating the trained support vector machine SVM sub-model and the long short-term memory network LSTM sub-model into a complete fault prediction model, the support vector machine SVM sub-model is good at classification tasks and can quickly and accurately identify fault signs in the data, while the long short-term memory network LSTM sub-model is good at processing time series data and can capture long-term change trends and short-term fluctuations in the data. The combination of the two enables the fault prediction system to analyze the equipment operation status more comprehensively and deeply, and improve the accuracy and reliability of the prediction. Secondly, in the process of fault prediction model deployment, it is emphasized to ensure that the data interface and interaction logic between the two sub-models are correct, which is the key to achieving efficient collaboration between models. By carefully designing the data interface and interaction logic, the smooth transmission and accurate processing of data between models are guaranteed, avoiding performance degradation or prediction errors caused by data format mismatch or transmission errors.

[0118] Furthermore, the integrated fault prediction model is deployed to the prediction and alarm module of the industrial control system, and the performance of the model is tuned in the industrial control system. This measure ensures the efficiency and accuracy of the model in real-time data processing and prediction. Industrial control systems usually have high-performance hardware resources and optimized software architecture, which can provide strong support for model operation. By performing performance tuning in the industrial control system, the potential of the model can be further explored and its performance in real-time data processing and prediction tasks can be improved.

[0119] In addition, the model's running parallelism, memory usage and other parameters are adjusted according to system resources and data processing requirements. This measure enables the fault prediction model to adapt more flexibly to different operating environments and data scales, thereby improving processing efficiency while ensuring prediction accuracy.

[0120] Finally, in terms of real-time data processing, the equipment operation data is collected in real time through the sensors, logs and other data sources of the industrial control system, and the collected data is fully preprocessed, including data cleaning, feature extraction and normalization. These steps ensure the quality of the data input into the fault prediction model and improve the prediction accuracy and stability of the fault prediction model. At the same time, the preprocessed data stream is sequentially input into the support vector machine SVM sub-model and the long short-term memory network LSTM sub-model for analysis, which not only utilizes the fast classification capability of the support vector machine SVM sub-model, but also gives play to the advantages of the long short-term memory network LSTM sub-model in time series analysis, realizing comprehensive monitoring and accurate prediction of the equipment operation status.

[0121] When conducting risk assessment on motor operation failure, a fault risk assessment algorithm is designed based on the output results of the support vector machine (SVM) sub-model and the long short-term memory (LSTM) sub-model. The algorithm comprehensively considers factors such as classification results, time series trends, and abnormal fluctuations to calculate the risk probability of equipment failure. According to historical data and business needs, a reasonable risk threshold is set. When the assessed risk probability exceeds the threshold, the equipment is considered to have a high risk of failure. When the fault risk assessment result exceeds the preset threshold, the system automatically generates an alarm message, which should include key information such as the fault type, risk level, and occurrence time. The alarm message is sent to relevant personnel through preset notification channels, which may include email, SMS, instant messaging software, or the alarm interface of the industrial control system. According to the fault type and risk level, the system automatically provides targeted maintenance suggestions, which can be formulated based on historical fault handling experience, equipment maintenance manuals, and other sources. The operating status of the fault prediction and alarm module is monitored in real time to ensure the stability of the model and the accuracy of the prediction, and a feedback mechanism is established. The feedback from relevant personnel on the alarm information and maintenance suggestions is collected, and the model algorithm and notification process are continuously optimized based on the feedback to improve the overall efficiency of the system.

[0122] See also Figure 2 , provides a motor operation fault prediction system, the system comprising:

[0123] The acquisition module 201 is used to collect vibration signals and temperature data of the motor in real time based on a high-sensitivity vibration sensor and a temperature sensor;

[0124] A preprocessing module 202, used for preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing;

[0125] An extraction module 203 is used to extract frequency domain features closely related to the fault from the vibration signal, and to extract statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set;

[0126] An acquisition module 204 is used to acquire a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a motor without faults, each of the negative samples includes a feature set of a motor with faults, and the fault type is marked;

[0127] A training module 205 is used to train a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model;

[0128] The prediction module 206 is used to predict the motor operation fault based on the trained fault prediction model.

[0129] It can be understood that the motor operation fault prediction system provided by the present invention corresponds to the motor operation fault prediction method provided by the aforementioned embodiments. The relevant technical features of the motor operation fault prediction system can refer to the relevant technical features of the motor operation fault prediction method, which will not be repeated here.

[0130] The embodiment of the present invention provides a motor operation fault prediction method and system, which collects vibration signals and temperature data in real time during operation; extracts frequency domain features, statistical features and time series features from the vibration signals and temperature data respectively to form a feature set; obtains a training set, the training set includes multiple positive samples and multiple negative samples; trains a fault prediction model based on the training set, the fault prediction model includes a support vector machine SVM sub-model and a long short-term memory network LSTM sub-model; predicts motor operation faults based on the trained fault prediction model. The present invention solves the problems of hysteresis and inaccuracy in the predictive maintenance of motor bearing faults in the prior art, and the failure of the prior method to fully combine the advantages of frequency domain features and time series analysis.

[0131] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0132] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0134] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0136] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0137] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting motor operation faults, characterized in that: include: Step S1, based on a high-sensitivity vibration sensor and a temperature sensor, real-time collection of vibration signals and temperature data during motor operation; Step S2, preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing; Step S3, extracting frequency domain features closely related to the fault from the vibration signal, and extracting statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set; Step S4, obtaining a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a motor without faults, each of the negative samples includes a feature set of a motor with faults, and the fault type is marked; Step S5, training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model; Step S6, predicting motor operation faults based on the trained fault prediction model.

2. The motor operation fault prediction method according to claim 1, characterized in that: The vibration sensor is a piezoelectric acceleration sensor, which is installed on the outside of the bearing shell of the motor to collect vibration signals during the operation of the motor; the temperature sensor is a thermocouple temperature sensor with a high response speed, which is close to the motor housing or the internal hot spot area to monitor the motor temperature in real time; The vibration signal sensed by the vibration sensor and the motor temperature data sensed by the temperature sensor are collected based on the data acquisition card, and the collected vibration signal and temperature data are transmitted to a data processing center.

3. The motor operation fault prediction method according to claim 1, characterized in that: The step S2, preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing, includes: A digital filter is used to low-pass filter the vibration signal to remove high-frequency noise in the vibration signal and retain low-frequency vibration components related to motor failure; a moving average or median filter is used to smooth temperature fluctuations; Eliminate the abnormal part of the vibration signal that exceeds the peak value of the normal vibration range, and standardize the vibration signal using a Z-score standardization method; Abnormal temperature data that exceeds the temperature range during normal operation of the motor is eliminated.

4. The motor operation fault prediction method according to claim 1, characterized in that: Step S3, extracting frequency domain features closely related to the fault from the vibration signal, including: Performing FFT fast Fourier transform on the preprocessed vibration signal to convert the time domain signal into a frequency domain signal; identifying a specific fault frequency and its harmonic components closely related to the bearing fault from the frequency domain signal; The amplitude and phase of the specific fault frequency are calculated, and the ratio of each harmonic to the fundamental frequency and the relative amplitude between each harmonic are calculated to form a frequency domain feature.

5. The motor operation fault prediction method according to claim 1, characterized in that: In step S3, extracting statistical features and time series features from the temperature data includes: Calculate the mean and variance of the temperature data within a set time period, wherein the mean and variance of the temperature data constitute statistical features; Based on the time series analysis method, the trend line of the temperature data changing over time and the rate of change of the temperature are extracted, and the trend line of the temperature data changing over time and the rate of change of the temperature constitute the time series characteristics.

6. The motor operation fault prediction method according to claim 1 or 5, characterized in that: The step S5, training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model, including: Training the support vector machine SVM sub-model based on the training set, and calculating the first loss of the support vector machine SVM sub-model according to the fault prediction result and the actual fault label of each training sample after passing through the support vector machine SVM sub-model; Optimizing the model parameters of the support vector machine SVM sub-model based on the first loss until the first loss is less than a first preset threshold, and obtaining the trained support vector machine SVM sub-model; Obtaining samples in the training set that are identified as having a failure risk by the trained support vector machine SVM sub-model; Training the LSTM sub-model based on samples with fault risks in the training set, and calculating a second loss of the LSTM sub-model based on a fault prediction result and an actual fault label of each sample with fault risk after passing through the LSTM sub-model; Based on the second loss, the model parameters of the long short-term memory network LSTM sub-model are optimized until the second loss is less than a second preset threshold, and the trained long short-term memory network LSTM sub-model is obtained.

7. The motor operation fault prediction method according to claim 1, characterized in that: The step S6, predicting the motor operation fault based on the trained fault prediction model, includes: Based on high-sensitivity vibration sensors and temperature sensors, the vibration signals and temperature data of the motor during operation in the current set time period are collected; Preprocessing the vibration signal and the temperature data, and extracting frequency domain features in the vibration signal and statistical features and time series features in the temperature data to form a current motor feature set; Inputting the current motor feature set into the trained support vector machine SVM sub-model to obtain a first prediction result output by the support vector machine SVM sub-model; If the first prediction result is no fault, then the current running state of the motor is no fault; If the first prediction result is a fault, the current motor feature set is input into the trained long short-term memory network LSTM sub-model to obtain a second prediction result output by the long short-term memory network LSTM sub-model, and the second prediction result is the current operating state of the motor.

8. A motor operation fault prediction system, characterized in that: include: The acquisition module is used to collect vibration signals and temperature data of the motor in real time based on high-sensitivity vibration sensors and temperature sensors; A preprocessing module, used for preprocessing the vibration signal and the temperature data, wherein the preprocessing includes denoising, removing outliers and standardizing; An extraction module, used to extract frequency domain features closely related to the fault from the vibration signal, and to extract statistical features and time series features from the temperature data, wherein the frequency domain features, statistical features and time series features constitute a motor feature set; An acquisition module is used to acquire a training set, wherein the training set includes a plurality of positive samples and a plurality of negative samples, each of the positive samples includes a feature set of a motor without faults, each of the negative samples includes a feature set of a motor with faults, and the fault type is marked; A training module, used for training a fault prediction model based on the training set, wherein the fault prediction model includes a support vector machine (SVM) sub-model and a long short-term memory (LSTM) network sub-model; The prediction module is used to predict the motor operation fault based on the trained fault prediction model.

Images