Self-adaptive motor fault diagnosis system and fault early warning method

Abstract

The invention discloses a self-adaptive motor fault diagnosis system and a fault early warning method, and relates to the technical field of motor fault diagnosis and self-adaption. Firstly, various sensors are installed at key parts of a motor, multi-source data such as vibration, temperature, current and voltage are synchronously collected at the sampling frequency of 10 kHz, the data are cleaned and subjected to normalization preprocessing, and the data are stored in a database; time domain, frequency domain and time-frequency domain features are extracted and screened, a CNN and RNN mixed model is used for training, the learning rate is adaptively adjusted in the training process, data processed in real time are input into the model, the fault type and probability are output, and if an early warning threshold value exceeds 80%, a signal is sent, and information is transmitted to operation and maintenance personnel. According to the invention, multi-source data acquisition, accurate feature extraction, multi-model training diagnosis, self-adaption to working conditions and new faults, accurate diagnosis of motor faults, timely early warning and severity evaluation can be realized; and the cloud platform is convenient to manage, guarantees data safety, improves performance after optimization, reduces cost, and guarantees stable operation of the motor.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis and adaptive technology, and in particular to an adaptive motor fault diagnosis system and fault early warning method. Background Technology

[0002] In modern industrial systems, electric motors are crucial power equipment, and their stable operation is of great significance to numerous fields. In manufacturing, motors drive various production equipment, from precision electronics manufacturing to heavy machining; motor performance directly affects product quality and production efficiency. In transportation, motors are used in electric vehicles, rail transit, and other applications, impacting traffic safety and travel convenience. In the energy sector, motors are used in power generation and transmission, ensuring a stable energy supply.

[0003] However, motors face numerous factors that can lead to malfunctions during long-term operation. Mechanically, the bearings, gears, and other components of the motor experience wear and fatigue over time, resulting in abnormal vibration and increased noise. Electrically, faults such as short circuits, open circuits, and insulation aging in the windings occur frequently, affecting the motor's normal performance. Environmental factors cannot be ignored either; harsh environments such as high temperatures, humidity, and dust can accelerate the damage to motor components.

[0004] Traditional motor fault diagnosis methods mainly rely on human experience and simple instrument testing. Manual inspection is inefficient, makes it difficult to monitor motor status in real time, and the results are greatly influenced by human experience and subjective judgment. For example, in large factories with numerous motors, manual inspection is not only time-consuming and labor-intensive but may also miss potential faults. While simple instrument testing can obtain some basic parameters, its diagnostic capabilities for complex faults are limited, and it cannot accurately determine the type and severity of the fault.

[0005] With the development of technology, fault diagnosis methods based on signal processing and machine learning have gradually emerged. These methods diagnose faults by collecting signals such as motor vibration, current, and temperature, and then performing feature extraction and pattern recognition. However, they have certain limitations. Signal processing often requires manually designed features, which are difficult to comprehensively and accurately reflect fault information for complex and variable motor faults. Machine learning models have high requirements for data quality; the actual collected data may contain noise, missing values, and other issues, affecting the model's training effect and diagnostic accuracy. Moreover, these models lack adaptability; when the motor's operating conditions change or new fault modes appear, the model's diagnostic performance will significantly decrease.

[0006] Existing fault warning methods are also inadequate. Most rely on fixed thresholds for warnings, failing to dynamically adjust these thresholds based on the motor's real-time operating status and historical data, leading to frequent false alarms or missed alarms. False alarms increase the workload and costs for maintenance personnel, while missed alarms can result in serious equipment damage and production accidents.

[0007] Therefore, developing an adaptive motor fault diagnosis system and fault early warning method is of significant practical importance. This system and method can diagnose motor faults accurately and in real time, issue timely early warning signals, and adaptively adjust according to changes in motor operating conditions and fault modes, thereby improving the reliability and safety of motor operation, reducing maintenance costs, and meeting the needs of modern industry for efficient motor management. Summary of the Invention

[0008] The present invention proposes an adaptive motor fault diagnosis system and fault early warning method to solve the problems mentioned in the prior art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive motor fault diagnosis system and fault early warning method, comprising: Multi-source data acquisition: Vibration, temperature, current, and voltage sensors are installed at key parts of the motor. The vibration sensor uses a piezoelectric accelerometer to collect motor vibration signals; the temperature sensor uses a thermocouple to collect motor temperature changes; and the current and voltage sensors use Hall effect sensors to collect data, thus constructing a multi-source data sample set. Data preprocessing: Cleaning the collected data to remove outliers. A median filtering algorithm is used, taking the median of the data surrounding a given data point as the processed value. The window size is adjusted according to the data's fluctuation characteristics, using the formula... Normalization, where X is the original data, X min and X max These are the minimum and maximum values ​​of the data, which are then mapped to the [0,1] interval to eliminate dimensional differences. Feature extraction and selection: Time-domain, frequency-domain, and time-frequency-domain features are extracted from the preprocessed data. A feature selection method based on correlation analysis and information gain is used to select feature subsets according to a set threshold. The correlation analysis uses the Pearson correlation coefficient formula. , and For the values ​​of two features, and The characteristic mean; Adaptive fault diagnosis model training: A hybrid model of deep learning, namely Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), is used, with CNN and RNN handling spatial and temporal features of the data, respectively; the cross-entropy loss function is used during training. The evaluation is conducted where N is the sample size and C is the number of fault categories. For the true label of the sample, To predict probabilities, weights and biases are adjusted using the backpropagation algorithm, while an adaptive learning rate adjustment strategy is employed. , Let be the learning rate for the t-th iteration. The initial learning rate, For coefficients; Fault diagnosis and early warning: Real-time collected and processed data is input into the adaptive fault diagnosis model. The model outputs the fault type and probability. When the fault probability exceeds the early warning threshold, the system issues an early warning signal and sends the fault information to the maintenance personnel's terminal to remind them to handle the problem.

[0010] Furthermore, it includes a model adaptive update step, which involves collecting new motor operation data and fault cases to form a new dataset. An incremental learning method is then used to combine the new dataset with the original training set and further divide the dataset into new training, validation, and test sets. When updating the model, some convolutional layers of the CNN are frozen, and the parameters of the RNN layers and some fully connected layers are fine-tuned. A regularization term is also introduced, as shown in the formula. L is the original loss function. The regularization coefficient is . M represents the model parameters, and M is the number of parameters, used to prevent overfitting.

[0011] Furthermore, the process includes a multi-model fusion diagnostic step, constructing a fault diagnosis model based on Support Vector Machine (SVM) and Random Forest (RF). SVM classifies motor faults; RF determines the fault category; and a weight-based multi-model fusion method is used, assigning corresponding weights to each model based on its performance metrics on the test set. The fusion formula is as follows: ,in The result is the fused fault category prediction. This represents the prediction result of the i-th model, where i=1 indicates a CNN-RNN hybrid model. Let be the weights of the i-th model, and .

[0012] Furthermore, the multi-source data acquisition step also includes environmental parameter acquisition. Temperature and humidity sensors and air pressure sensors are installed in the motor operating environment to collect environmental parameters of the motor operation. These parameters are then collected and stored synchronously with the motor's vibration, current, and voltage data. In subsequent data processing, the environmental parameters are input as additional features into the fault diagnosis model.

[0013] Furthermore, the feature extraction and selection steps also include feature dimensionality reduction based on principal component analysis (PCA). PCA analysis is performed on the extracted feature subset to solve for the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvectors corresponding to the k largest eigenvalues ​​are selected based on the magnitude of the eigenvalues. The original feature subset is projected into the low-dimensional space formed by these eigenvectors to obtain the dimensionality-reduced feature subset. The feature subset is then screened using a feature selection method that combines correlation analysis and information gain.

[0014] Furthermore, the adaptive fault diagnosis model training step also includes model fusion and ensemble learning, training multiple sub-models with different structures and parameters, and using a stacked ensemble learning method, using the outputs of these sub-models as inputs to construct a new meta-model. The meta-model training employs the mean squared error (MSE) loss function. ,in For actual fault category labels, The result is the prediction of the meta-model, where n is the number of samples. The parameters of the meta-model are adjusted using the backpropagation algorithm.

[0015] Furthermore, the fault diagnosis and early warning steps also include fault severity assessment. Based on the fault type and probability output by the fault diagnosis model, and combined with the historical fault parameters of the motor, a fault severity assessment model is established. Using the fuzzy comprehensive evaluation method, each indicator is assigned a corresponding weight. The actual values ​​of each indicator are mapped to a fuzzy set through the fuzzy membership function. Fuzzy synthesis and defuzzification are then performed to obtain the fault severity level. Early warning strategies and plans are formulated according to different levels to provide decision support for operation and maintenance personnel.

[0016] Furthermore, it also includes the steps of building a remote monitoring and fault diagnosis cloud platform. This involves constructing a cloud platform based on cloud computing technology, uploading data collected from the motor site to the cloud platform via a wireless network, and processing and analyzing the motor operation data. Fault diagnosis models and related algorithms are deployed on the cloud platform to enable remote diagnosis of multiple motors. At the same time, the cloud platform provides a visual interface, allowing maintenance personnel to view motor operation information through a mobile application for convenient maintenance.

[0017] Furthermore, it includes data security and privacy protection steps. During the data acquisition phase, the data collected by the sensors is encrypted using a symmetric encryption algorithm to generate encryption keys. During the data storage phase, the encrypted data is stored in a database with access permissions set for different users. In the model training and fault diagnosis process, differential privacy technology is used to add noise to the data.

[0018] Furthermore, it also includes system performance evaluation and optimization steps, regularly evaluating the performance of the adaptive motor fault diagnosis system, analyzing system problems based on the evaluation results, and taking corresponding measures.

[0019] Compared with existing technologies, the beneficial effects of this invention are: In terms of fault diagnosis accuracy, the system employs multi-source data acquisition, integrating various signals such as vibration, temperature, current, and voltage, and also incorporating environmental parameters to comprehensively reflect the motor's operating status. Through advanced feature extraction and selection methods, the most representative features are selected, and combined with CNN and RNN hybrid models and multi-model fusion technology, it can accurately identify various motor faults, significantly improving diagnostic accuracy.

[0020] It possesses strong adaptive capabilities. New data is collected regularly, and the model is updated using incremental learning methods. At the same time, some convolutional layers are frozen for fine-tuning. Combined with regularization to prevent overfitting, the model can quickly adapt to changes in motor operating conditions and new fault modes, and always maintain good diagnostic performance.

[0021] The fault warnings are timely and accurate. Based on the fault diagnosis results, the severity of the fault is assessed using fuzzy comprehensive evaluation, and different warning strategies and maintenance plans are formulated. This not only issues timely warning signals but also provides maintenance personnel with detailed fault information and targeted decision support, preventing production interruptions and equipment damage caused by faults.

[0022] System performance is effectively guaranteed. A remote monitoring and fault diagnosis cloud platform has been built to achieve centralized management and remote real-time monitoring of multiple motors, allowing maintenance personnel to monitor motor status anytime, anywhere. At the same time, data security and privacy protection are emphasized, employing encryption algorithms and differential privacy technology to ensure the security and privacy of data during transmission and storage.

[0023] Through regular system performance evaluation and optimization, continuous adjustment of model structure and parameters, reselection of features, and updating of training data, the fault diagnosis performance and reliability of the system are continuously improved, the false alarm rate and false negative rate are reduced, the operation and maintenance costs are reduced, and the stability and safety of motor operation are improved, providing a strong guarantee for the efficient and stable operation of industrial production. Attached Figure Description

[0024] Figure 1 This is a schematic block diagram of an adaptive motor fault diagnosis system and fault early warning method proposed in this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0027] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0028] Reference Figure 1 An adaptive motor fault diagnosis system and fault early warning method, comprising the following steps: Multi-source data acquisition: Various types of sensors, including vibration sensors, temperature sensors, current sensors, and voltage sensors, are installed at different key locations on the motor, such as bearings and stator windings. The vibration sensors are piezoelectric accelerometers with an excellent frequency response range of 0.5Hz-20kHz. Whether it's low-frequency vibration caused by rotor imbalance or high-frequency vibration generated by high-speed rotating components, both can be accurately captured, providing crucial clues for analyzing mechanical faults in the motor.

[0029] The temperature sensor employs a high-precision thermocouple sensor, with a measurement range of -50℃ to 200℃ and an accuracy of ±0.5℃. During motor operation, temperature changes in critical components are crucial indicators for assessing its operational status. This temperature sensor can accurately monitor the temperature of various critical components of the motor at all times, promptly detecting abnormal temperature increases caused by problems such as insulation aging or winding short circuits.

[0030] Both the current and voltage sensors utilize Hall effect sensors. In the electrical system of a motor, current and voltage parameters are core data reflecting the motor's load and electrical performance. Hall effect sensors, with their high precision and fast response, can accurately acquire current and voltage data from the motor, providing reliable data support for analyzing motor energy consumption, operating efficiency, and determining the presence of electrical faults. To ensure the accuracy and consistency of the acquired data, data from each sensor is collected synchronously at a high sampling frequency of 10kHz. This multi-dimensional, high-frequency synchronous acquisition method constructs a large and rich multi-source data sample set. This data sample set covers various aspects of the motor's operation, including mechanical vibration characteristics, temperature change trends, and electrical parameter fluctuations, providing a solid data foundation for subsequent in-depth fault diagnosis, condition assessment, and the development of intelligent operation and maintenance strategies.

[0031] Data preprocessing: Raw data often contains numerous outliers due to sensor malfunctions or external interference. If these outliers are not processed, they will severely interfere with the accuracy of data analysis. Therefore, the primary task is to meticulously clean the raw data and accurately remove these outliers, laying the foundation for subsequent processing. Median filtering plays a crucial role in data smoothing. This algorithm analyzes the data within a defined window surrounding each data point and takes the median as the processed value for that point. The window size is not fixed but dynamically adjusted based on the data's fluctuation characteristics. When data fluctuations are significant, the window is appropriately enlarged to better filter noise; when data fluctuations are small, the window is reduced to preserve the data's detailed features.

[0032] Then, normalization is performed using the formula. Where X is the original data, X min and X max The minimum and maximum values ​​of the data are represented by these values, and the data is mapped to the [0,1] interval. Data collected by different sensors often have different dimensions, and normalization can effectively eliminate this difference, making all types of data on the same scale, which facilitates subsequent unified analysis and processing, thereby improving the accuracy and reliability of the motor fault diagnosis model and providing strong support for accurately judging the motor's operating status.

[0033] Feature extraction and selection: Time-domain, frequency-domain, and time-frequency-domain features are extracted from the preprocessed data. Time-domain features include mean, variance, and peak value; frequency-domain features are obtained through Fast Fourier Transform (FFT), such as spectral amplitude and dominant frequency; time-frequency-domain features are extracted using wavelet transform, such as wavelet coefficient energy. A feature selection method based on correlation analysis and information gain is adopted to calculate the correlation and information gain value between each feature and the motor fault type. The most representative and discriminative feature subset is selected based on a set threshold. The correlation analysis uses the Pearson correlation coefficient formula. ,in and These are the values ​​of the two features, and These are the means of the two features, respectively.

[0034] Adaptive fault diagnosis model training: A hybrid model combining deep learning-based convolutional neural networks (CNNs) and recurrent neural networks (RNNs) is employed. CNNs are used to extract spatial features of the data, while RNNs are used to process the time-series features. The feature subset is divided into training, validation, and test sets, with proportions of 70%, 15%, and 15%, respectively. During training, the cross-entropy loss function is used. Where N is the number of samples and C is the number of fault categories. The true label (0 or 1) for sample i belonging to category j. To predict the probability that sample i belongs to class j, the model's weights and biases are continuously adjusted using the backpropagation algorithm. Simultaneously, an adaptive learning rate adjustment strategy is employed, dynamically adjusting the learning rate based on the model's performance on the validation set. The formula is as follows: ,in Let be the learning rate for the t-th iteration. The initial learning rate, To adjust the coefficients and improve the model's convergence speed and generalization ability.

[0035] Fault Diagnosis and Early Warning: Multi-source data from the motor is collected in real time via sensors, encompassing vibration, temperature, current, and voltage information. The collected data undergoes preprocessing, including outlier removal, unit standardization, and feature extraction to select the most representative information. The processed data is then input into a pre-trained adaptive fault diagnosis model. This model, based on deep learning technology, intelligently analyzes the data and outputs the possible fault types and corresponding probabilities of the motor. When the model's output fault probability exceeds a set warning threshold, such as 80%, it indicates a high risk of motor failure. The system then reacts quickly, immediately issuing a fault warning signal. Simultaneously, detailed fault information, including fault type, probability, and related fault characteristics, is sent to maintenance personnel's commonly used terminal devices, such as mobile phones and computers. Upon receiving this information, maintenance personnel can quickly take appropriate maintenance measures based on this accurate and timely information to prevent the fault from escalating and ensure the normal operation of the motor.

[0036] This invention also includes a model adaptive update step. New motor operation data and fault cases are collected periodically to form a new dataset. When the number of samples in the new dataset reaches a certain threshold (e.g., 500), the model update process is initiated. An incremental learning method is used to merge the new dataset with the original training set, and the mixed dataset is then re-partitioned to obtain new training, validation, and test sets. When updating the model, some convolutional layers in the CNN are frozen, and only the parameters of the RNN layers and some fully connected layers are fine-tuned to reduce computation and training time. Simultaneously, a regularization term, such as L2 regularization, is introduced, with the formula: Where L is the original loss function, The regularization coefficient is . M represents the parameters of the model, and M is the number of parameters. This prevents the model from overfitting and improves the model's adaptability to new failure modes and its diagnostic accuracy.

[0037] This invention also includes a multi-model fusion diagnostic step. In addition to a hybrid model based on CNN and RNN, a fault diagnosis model based on Support Vector Machine (SVM) and Random Forest (RF) is constructed. The SVM model classifies motor faults by finding the optimal classification hyperplane; the RF model determines the final fault category by constructing multiple decision trees and voting on the outputs of these decision trees. A weight-based multi-model fusion method is employed, assigning corresponding weights to each model based on its performance metrics such as accuracy, recall, and F1 score on the test set. The fusion formula is as follows: ,in The result is the fused fault category prediction. Let i be the prediction result of the i-th model (i=1 represents the CNN-RNN hybrid model, i=2 represents the SVM model, and i=3 represents the RF model). Let be the weights of the i-th model, and By fusing multiple models, the accuracy and reliability of fault diagnosis can be improved.

[0038] In this invention, the multi-source data acquisition step also includes environmental parameter acquisition. The motor's operating performance is closely linked to its surrounding environment; subtle changes in environmental factors can be reflected in the motor's operating state. Therefore, to achieve precise control over the motor's operating status, it is crucial to rationally place temperature and humidity sensors and air pressure sensors in the motor's operating environment. Temperature and humidity sensors exhibit extremely high measurement accuracy, with a temperature measurement accuracy of ±0.3℃ and a humidity measurement accuracy of ±1%RH, enabling them to sensitively and accurately detect subtle changes in temperature and humidity in the environment. Air pressure sensors also possess excellent performance, with a measurement accuracy of ±0.05kPa, accurately acquiring environmental air pressure parameters. During actual acquisition, a strict synchronous acquisition and storage mechanism is implemented for environmental parameters and the motor's own vibration, current, and voltage data. This synchronous operation ensures that the collected data forms an organic whole, comprehensively reflecting the overall operating status of the motor. In subsequent data processing and feature extraction stages, the potential impact of environmental parameters on the motor's operating state is fully considered. Environmental parameters such as temperature, humidity, and air pressure are input as additional feature dimensions into the fault diagnosis model. In this way, the fault diagnosis model can be analyzed and calculated based on a broader information foundation, thereby significantly improving the ability to diagnose motor faults under different environmental conditions, providing solid and powerful support for the reliable operation and efficient maintenance of motors, and enabling motors to achieve accurate condition monitoring and fault prediction even in complex and ever-changing environments.

[0039] In this invention, the feature extraction and selection steps further include feature dimensionality reduction based on Principal Component Analysis (PCA). PCA analysis is performed on the extracted feature subset to calculate the covariance matrix of the features, and then the eigenvalues ​​and eigenvectors of the covariance matrix are solved. Based on the magnitude of the eigenvalues, the eigenvectors corresponding to the k largest eigenvalues ​​are selected, and the original feature subset is projected into a low-dimensional space formed by these eigenvectors to obtain the dimensionality-reduced feature subset. The dimensionality-reduced feature subset retains most of the information of the original features while reducing the feature dimensionality, thus reducing the computational complexity of the model and improving the training and diagnostic efficiency. Simultaneously, feature selection methods combining correlation analysis and information gain are used to further screen the dimensionality-reduced feature subset, ensuring that the features finally input into the fault diagnosis model have high relevance and discriminative power.

[0040] In this invention, the adaptive fault diagnosis model training step also includes model fusion and ensemble learning. In addition to the CNN-RNN hybrid model, multiple sub-models with different structures and parameters are trained, such as CNN models with different numbers of layers and neurons, and RNN models with different numbers of hidden layer units. A stacked ensemble learning method is used, taking the outputs of these sub-models as input to construct a new meta-model. The meta-model can be a simple fully connected neural network. By training the meta-model, the complex relationships between the outputs of the sub-models are learned, further improving the accuracy and stability of fault diagnosis. The mean squared error (MSE) loss function is used when training the meta-model. ,in For actual fault category labels, The result is the prediction of the meta-model, where n is the number of samples. The parameters of the meta-model are adjusted using the backpropagation algorithm.

[0041] In this invention, the fault diagnosis and early warning steps also include fault severity assessment. After the fault diagnosis model outputs the fault type and fault probability, to more comprehensively and accurately grasp the motor fault status, a fault severity assessment model needs to be built by combining historical fault data accumulated by the motor and real-time operating parameters. This model uses fuzzy comprehensive evaluation method to comprehensively consider multiple factors. Among them, the fault type reflects the nature of the fault, and different types of faults have different impacts on the motor; the fault probability reflects the likelihood of the fault occurring; the length of motor operation time affects the potential degree of fault development; and the load condition relates to the driving effect of the motor's current working intensity on the fault. These factors are set as evaluation indicators, and corresponding weights are assigned according to their influence on the fault severity. Then, using fuzzy membership functions, the actual values ​​of each indicator are cleverly mapped to fuzzy sets to complete the fuzzification process. Subsequently, fuzzy synthesis is performed through a specific algorithm to integrate the fuzzy information of each indicator. Finally, after defuzzification, a clear fault severity level is obtained, which is divided into mild, moderate, and severe faults. Based on different levels, targeted early warning strategies and maintenance plans are developed to provide maintenance personnel with accurate and practical decision-making basis and help to efficiently solve motor failure problems.

[0042] This invention also includes a step of building a remote monitoring and fault diagnosis cloud platform. A cloud platform based on cloud computing technology is constructed, uploading data collected from the motor site in real time via wireless networks (such as 4G, 5G, or Wi-Fi) to the cloud platform. The cloud platform possesses powerful computing and storage capabilities, enabling centralized processing and analysis of large amounts of motor operating data. Fault diagnosis models and related algorithms are deployed on the cloud platform to achieve remote real-time monitoring and fault diagnosis of multiple motors. Simultaneously, the cloud platform provides a visual interface, allowing maintenance personnel to view the motor's operating status, fault diagnosis results, and early warning information anytime, anywhere via a web browser or mobile application, facilitating unified management and maintenance of the motors.

[0043] This invention also includes data security and privacy protection steps. In the initial data acquisition stage, the various motor operation data collected by sensors contain critical information, and leakage could lead to serious consequences. Therefore, symmetric encryption algorithms, such as the widely used and highly secure AES algorithm, are used to encrypt this data. Specifically, a unique encryption key is generated, which is used to encrypt the original data, ensuring that the data flows in ciphertext form during transmission, effectively resisting the risk of theft during transmission and comprehensively ensuring data transmission security. In the data storage stage, the encrypted data is securely stored in a high-security database. This database is equipped with an advanced access control mechanism, meticulously classifying and setting different levels of access permissions for different types of users, such as data administrators, maintenance engineers, and researchers, based on their actual work needs and responsibilities. This prevents unauthorized users from illegally accessing data from the source, greatly reducing the risk of data leakage. In the core processes of model training and fault diagnosis, differential privacy technology is innovatively adopted to further protect the privacy of motor operation data while ensuring that the model's diagnostic performance is not affected. This technology cleverly obscures individual data details by adding carefully calculated specific noise to the data without changing the key features and distribution of the data. Even if the data is accidentally leaked, it is difficult to accurately restore the original sensitive information, effectively protecting data privacy and helping the system achieve a balance between security and efficiency.

[0044] This invention also includes system performance evaluation and optimization steps. Motors often face complex and changing operating conditions, with numerous uncertainties in their operating environment. Therefore, it is essential to conduct comprehensive and in-depth evaluations of system performance regularly. The evaluation utilizes a series of scientific and comprehensive indicators. Fault diagnosis accuracy accurately measures the system's correct identification of real faults; the higher the accuracy, the stronger the reliability of the diagnosis. Recall rate focuses on the system's overall ability to capture actual faults; a higher recall rate avoids missing faults. The F1 score combines accuracy and recall, presenting a more objective picture of the overall system performance. False alarm rate directly reflects the frequency of erroneous fault alarms issued by the system; a lower false alarm rate reduces interference with operation and maintenance. False negative rate measures the likelihood that the system fails to detect actual faults; a lower value is more beneficial for stable motor operation. Through in-depth analysis of these evaluation indicators, the problems and deficiencies of the system can be accurately located. For example, when faced with new and complex operating conditions, the model may lack generalization ability, leading to deviations in fault diagnosis; the feature selection process may also be unreasonable, failing to select the most critical and representative information from a large amount of raw data, thus affecting the diagnostic effect.

[0045] To address these issues, a series of targeted optimization strategies will be formulated and implemented. The model structure and parameters will be finely adjusted based on system performance feedback data to enhance the model's adaptability to complex data. The feature selection process will be streamlined and optimized, using advanced algorithms to select features more valuable for fault diagnosis and improve data quality. Simultaneously, training data will be actively updated to promptly incorporate newly emerging fault types and motor operating data under different conditions, allowing the model to continuously learn new knowledge and improve its adaptability. Through continuous performance evaluation and optimization, the system's fault diagnosis performance and operational reliability will be continuously improved, ensuring long-term stable operation under various complex environments and providing strong support for the safe and efficient operation of the motor.

[0046] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An adaptive motor fault diagnosis system and fault early warning method, characterized in that, include: Multi-source data acquisition: Vibration, temperature, current, and voltage sensors are installed at key parts of the motor. The vibration sensor uses a piezoelectric accelerometer to collect motor vibration signals; the temperature sensor uses a thermocouple to collect motor temperature changes; and the current and voltage sensors use Hall effect sensors to collect data, thus constructing a multi-source data sample set. Data preprocessing: Cleaning the collected data to remove outliers. A median filtering algorithm is used, taking the median of the data surrounding a given data point as the processed value. The window size is adjusted according to the data's fluctuation characteristics, using the formula... Normalization, where X is the original data, X min and X max These are the minimum and maximum values ​​of the data, respectively. The data is mapped to the [0,1] interval to eliminate the difference in units. Feature extraction and selection: Time-domain, frequency-domain, and time-frequency-domain features are extracted from the preprocessed data. A feature selection method based on correlation analysis and information gain is used to select feature subsets according to a set threshold. The correlation analysis uses the Pearson correlation coefficient formula. , and For the values ​​of two features, and The characteristic mean; Adaptive fault diagnosis model training: A hybrid model of deep learning convolutional neural network (CNN) and recurrent neural network (RNN) is adopted, with CNN and RNN respectively handling spatial and temporal features of the data; Cross-entropy loss function is used during training. The evaluation is conducted where N is the sample size and C is the number of fault categories. For the true label of the sample, To predict probabilities, weights and biases are adjusted using the backpropagation algorithm, while an adaptive learning rate adjustment strategy is employed. , Let be the learning rate for the t-th iteration. The initial learning rate, For coefficients; Fault diagnosis and early warning: Real-time collected and processed data is input into the adaptive fault diagnosis model. The model outputs the fault type and probability. When the fault probability exceeds the early warning threshold, the system issues an early warning signal and sends the fault information to the maintenance personnel's terminal to remind them to handle the problem.

2. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, It also includes a model adaptive update step, which involves collecting new motor operation data and fault cases to form a new dataset. An incremental learning method is then used to combine the new dataset with the original training set and further divide the dataset into new training, validation, and test sets. When updating the model, some convolutional layers of the CNN are frozen, and the parameters of the RNN layers and some fully connected layers are fine-tuned. A regularization term is also introduced, as shown in the formula: L is the original loss function. The regularization coefficient is . M represents the model parameters, and M is the number of parameters, used to prevent overfitting.

3. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, It also includes a multi-model fusion diagnostic step, constructing a fault diagnosis model based on Support Vector Machine (SVM) and Random Forest (RF). SVM classifies motor faults; RF determines the fault category; and a weight-based multi-model fusion method is used, assigning corresponding weights to each model based on its performance metrics on the test set. The fusion formula is as follows: ,in The result is the fused fault category prediction. This represents the prediction result of the i-th model, where i=1 indicates a CNN-RNN hybrid model. Let be the weights of the i-th model, and .

4. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, The multi-source data acquisition step also includes environmental parameter acquisition. Temperature and humidity sensors and air pressure sensors are installed in the motor operating environment to collect environmental parameters of the motor operation. These parameters are collected and stored synchronously with the motor's vibration, current, and voltage data. In subsequent data processing, the environmental parameters are input as additional features into the fault diagnosis model.

5. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, The feature extraction and selection steps also include feature dimensionality reduction based on principal component analysis (PCA). PCA analysis is performed on the extracted feature subset to solve for the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvectors corresponding to the k largest eigenvalues ​​are selected according to the magnitude of the eigenvalues. The original feature subset is projected into the low-dimensional space formed by these eigenvectors to obtain the dimensionality-reduced feature subset. The feature subset is then screened by combining correlation analysis and information gain feature selection methods.

6. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, The adaptive fault diagnosis model training step also includes model fusion and ensemble learning, training multiple sub-models with different structures and parameters, and using a stacked ensemble learning method to construct a new meta-model by using the outputs of these sub-models as inputs. The meta-model is trained using the mean squared error (MSE) loss function. ,in For actual fault category labels, The result is the prediction of the meta-model, where n is the number of samples. The parameters of the meta-model are adjusted using the backpropagation algorithm.

7. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, The fault diagnosis and early warning steps also include fault severity assessment, which establishes a fault severity assessment model based on the fault type and fault probability output by the fault diagnosis model and combined with the historical fault parameters of the motor. The fuzzy comprehensive evaluation method is adopted, and each indicator is assigned a corresponding weight. The actual value of each indicator is mapped to a fuzzy set through the fuzzy membership function. Fuzzy synthesis and defuzzification are performed to obtain the severity level of the fault. Early warning strategies and plans are formulated according to different levels to provide decision support for operation and maintenance personnel.

8. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, It also includes the steps of building a remote monitoring and fault diagnosis cloud platform. The platform is built based on cloud computing technology. Data collected on-site from the motor is uploaded to the cloud platform via wireless network. The cloud platform processes and analyzes the motor operation data. Fault diagnosis models and related algorithms are deployed on the cloud platform to enable remote diagnosis of multiple motors. At the same time, the cloud platform provides a visual interface, allowing maintenance personnel to view the motor operation information through a mobile application for convenient maintenance.

9. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, It also includes data security and privacy protection steps. During the data acquisition phase, the data collected by the sensors is encrypted using a symmetric encryption algorithm to generate encryption keys. During the data storage phase, the encrypted data is stored in a database with access permissions set for different users. During model training and fault diagnosis, differential privacy technology is used to add noise to the data.

10. The adaptive motor fault diagnosis system and fault early warning method according to claim 1, characterized in that, It also includes system performance evaluation and optimization steps, regularly evaluating the performance of the adaptive motor fault diagnosis system, analyzing system problems based on the evaluation results, and taking corresponding measures.

Images