Bearing fault diagnosis method and device

Abstract

The invention discloses a bearing fault diagnosis method and device, and relates to the technical field of bearing fault diagnosis, and the method comprises a model training step, a model integration step and a fault diagnosis step, and the model training step provides a trained diagnosis model for the model integration step. The anomaly detection model output in the model integration step is used as the model algorithm input in the fault diagnosis step, and the model integration step is used for integrating a plurality of models to improve the diagnosis accuracy and the system robustness; the model integration step comprises the steps of multi-model integration, dynamic model selection and anomaly detection. According to the bearing fault diagnosis method, the model integration step is designed, the advantage integration function of multiple models is achieved, the limitation problem when a single model processes complex and changeable fault modes is solved, fault diagnosis is difficult to comprehensively and accurately conduct, the accuracy and robustness of a bearing fault diagnosis system are improved, and the fault diagnosis efficiency is improved. Fault detection under complex working conditions can be better handled, and the probability of false alarm and missing alarm is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a bearing fault diagnosis method and device. Background Art

[0002] Bearings are key components in mechanical equipment, and their operating status directly affects the performance and life of the equipment. With the development of industrial automation and intelligent manufacturing, the demand for equipment status monitoring and fault diagnosis is increasing. However, traditional bearing fault diagnosis methods still have many problems and defects.

[0003] Most existing technologies use a single diagnostic model, such as a support vector machine or a decision tree. Single models have poor robustness and are difficult to adapt to fault diagnosis needs under different working conditions and complex environments. The model is unstable when facing noise and abnormal data, and is prone to false positives and false negatives. Traditional bearing fault diagnosis methods usually rely on experience and simple signal processing techniques, such as time domain analysis and frequency domain analysis. These methods often find it difficult to provide highly accurate diagnostic results when faced with complex working conditions and diverse failure modes. Traditional methods make insufficient use of sensor data, usually focusing only on certain single features and ignoring the comprehensive analysis of multi-dimensional features. A large amount of potential information has not been fully mined, resulting in poor diagnostic results.

[0004] Therefore, it is very necessary to propose a bearing fault diagnosis method and device in the present application to solve the limitation problem of single model diagnosis in the existing bearing diagnosis technology.

[0005] Patent document CN110689130B discloses a bearing fault diagnosis method. The above patent realizes bearing fault diagnosis through the learned target domain BN model and the existing reasoning algorithm, and makes full use of the migration mechanism to obtain data and information from similar fields. It can solve the problem of insufficient sample data for bearing diagnosis modeling, and reduce the impact of insufficient sample quantity on parameter learning accuracy and diagnosis results. It can be widely used in noisy, uncertain environments where it is difficult to obtain a large amount of target data. However, the above patent cannot realize the function of integrating the advantages of multiple models.

[0006] Patent document CN101354312B discloses an intelligent bearing fault diagnosis system. The above patent realizes the ability to monitor the status of the bearing online based on the three types of signals of the intelligent bearing on the basis of the existing intelligent bearing with a composite sensor, and to determine which component of the bearing the fault occurs in. However, the above patent cannot realize the training function of the fault diagnosis model.

[0007] Patent document CN117251680B discloses a bearing fault diagnosis network. The above patent realizes high-precision fault diagnosis and has strong anti-noise performance, but the above patent cannot realize the automatic diagnosis function of bearing faults.

[0008] Patent document CN113702041B discloses a bearing fault diagnosis method. The above patent achieves the ability to ensure the accuracy of the bearing frequency model produced, thereby increasing the accuracy of bearing detection, but the above patent cannot achieve the function of efficient data processing.

[0009] In summary, the above patents cannot realize the functions of integrating the advantages of multiple models, training the fault diagnosis model, automating the diagnosis of bearing faults, and efficiently processing data, resulting in low accuracy, poor system robustness, slow response speed of manual intervention monitoring, and poor adaptability to different working conditions in bearing fault diagnosis; To this end, the present application proposes a bearing fault diagnosis method and device that can realize the advantages of integrating multiple models, training fault diagnosis models, automated diagnosis of bearing faults and efficient data processing. Summary of the invention

[0010] The purpose of the present invention is to provide a bearing fault diagnosis method and device to solve the technical problems proposed in the above-mentioned background technology, that is, the inability to realize the advantages of integrating multiple models, the training function of fault diagnosis models, the automatic diagnosis function of bearing faults and the efficient processing function of data, resulting in low accuracy, poor system robustness, slow response speed of manual intervention monitoring and poor adaptability to different working conditions in bearing fault diagnosis.

[0011] To achieve the above-mentioned object, the present invention provides the following technical solutions: a bearing fault diagnosis method, comprising a model training step, a model integration step and a fault diagnosis step, wherein the model training step provides a trained diagnosis model for the model integration step, the abnormality detection model output by the model integration step is used as a model algorithm input of the fault diagnosis step, and the model integration step is used to integrate multiple models to improve the diagnosis accuracy and system robustness; The model integration step includes: multi-model integration, dynamic model selection and anomaly detection steps; The multi-model integration is to design an integrated learning framework, combine different classification models to fuse the prediction results of the models by voting / weighted average, output the model integration algorithm code, and integrate the model effect evaluation report; The dynamic model selection is to design a dynamic model selection algorithm based on real-time data features, select the model that performs best under the current bearing state for fault diagnosis, and output the dynamic model selection algorithm code and real-time model selection demonstration; The anomaly detection step uses an anomaly detection algorithm to establish an anomaly detection model, monitor and identify data that exceeds a normal range, and output an anomaly detection step code and an anomaly detection result analysis report.

[0012] Preferably, the model training step is used to train and evaluate different classification models; The model training steps include: training data preparation, model selection, and model training evaluation; Training data preparation is to extract sample data of normal and faulty bearing operation from actual operation data, mark them, and output sample data sets and data marking description documents; Model selection is to compare different machine learning models, select the most suitable model for bearing fault diagnosis, and output the model comparison report and the selected model and its parameters; Model training evaluation is to use the training data set to train the selected model, and use the test data set to evaluate the model's accuracy, precision and recall performance indicators, and output the trained model file and model evaluation report.

[0013] Preferably, the fault diagnosis step is used to classify bearing faults and display fault diagnosis results; The fault diagnosis steps include: fault classification, visualization interface and alarm system; Fault classification is to use the trained model to classify the real-time sensor data, identify the bearing fault type, and output the fault classification algorithm code and real-time fault diagnosis results; Visual interface means designing interactive interface, including real-time data display, fault diagnosis result display and historical data query function; The alarm system sets the threshold for fault diagnosis. When the diagnosis result exceeds the threshold, an alarm is triggered and an email, text message and sound prompt are sent.

[0014] Preferably, the sensor data is collected in real time by a sensor, and the sensor real-time collection step is used to collect and process the bearing working data in real time; The sensor real-time acquisition steps include: sensor selection and layout, data acquisition and transmission, and data preprocessing; Sensor selection and layout means selecting sensors with a wide frequency range and high sensitivity, including MEMS accelerometers, to monitor the vibration data of the bearings. Temperature sensors should be of high temperature resistance type to ensure real-time and reliable collection of bearing temperature data. Sound sensors should be of high sensitivity and excellent signal-to-noise ratio to detect abnormal sound data generated during bearing operation. Data acquisition and transmission is to use a high-speed sampling rate analog-to-digital converter to convert the sensor data into analog-to-digital data, and transmit the data to the data processing unit through the serial port and Ethernet; The data processing unit is equipped with a data preprocessing program, which reduces high-frequency noise through a low-pass filter, uses wavelet transform to remove background noise and clutter, designs a data alignment algorithm to ensure the timing consistency of data from different sensors, and outputs preprocessed data samples.

[0015] Preferably, the data processing unit is also designed with a feature extraction step, which is used to effectively extract data features from the sensor data; The feature extraction steps include: time domain feature extraction, frequency domain feature extraction and statistical feature extraction; Time domain feature extraction is to calculate the mean, standard deviation, and peak-to-peak value time domain statistical features of the bearing vibration signal, and output an example of the time domain feature extraction result; Frequency domain feature extraction is to use Fourier transform to convert time domain signals to frequency domain, extract spectrum characteristics, and output frequency domain feature extraction result examples; Statistical feature extraction is to calculate the statistical features of the autocorrelation function and cross-correlation function of the bearing vibration signal, and output an example of the statistical feature extraction result.

[0016] Preferably, the data processing unit is also designed with a feature selection step, which is used to analyze, evaluate and select different features; The feature selection steps include: feature correlation analysis, feature importance assessment, and feature selection algorithm; Feature correlation analysis uses the canonical correlation analysis method to evaluate the correlation between features, exclude highly correlated features, and output feature correlation matrix and correlation analysis report; Feature importance assessment is to use information gain and Gini index methods to assess the importance of features, select the features that have the greatest impact on classification, and output feature importance rankings and feature importance assessment reports; The feature selection algorithm uses the Wrapper method, Filter method, and Embedded method to select features and output the final selected feature list.

[0017] Preferably, the dynamic model selection algorithm performs dynamic selection based on real-time data characteristics to ensure that the most suitable model for the current bearing state is selected for fault diagnosis. The specific algorithm steps are as follows: S1. Calculate the scores of each model under the current bearing state according to the real-time data characteristics; S2, select the model with the highest score as the optimal model; S3. Return the optimal model.

[0018] Preferably, the anomaly detection model adopts an isolation forest anomaly detection model, and the specific algorithm steps are as follows: Random sampling: randomly extract samples from the data set to construct random numbers; Tree construction: randomly select features and thresholds, recursively divide the data into subspaces, and construct random trees; Isolation calculation: Calculate the depth of each data point in the tree as a measure of its isolation; Abnormality judgment: Determine whether a data point is abnormal based on isolation and threshold.

[0019] Preferably, the diagnostic device is used to implement the model training step, the model integration step, the fault diagnosis step and the sensor real-time acquisition step in the diagnostic method, and the diagnostic device includes: a sensor module and a data processing unit, and the sensor module is connected to the data processing unit through an Ethernet interface; The sensor module is used to implement the sensor real-time acquisition steps; The sensor module includes: an acceleration sensor, a temperature sensor and a sound sensor, which are used to detect the vibration, temperature and sound parameters of the bearing respectively. The sensor module is arranged around the bearing, and the sensor data is converted into a digital signal through an analog-to-digital converter and transmitted to the data processing unit through a serial port and an Ethernet interface; The data processing unit is used to implement the model training step, the model integration step, the feature extraction step and the feature selection step, and performs data processing according to the bearing operating condition information data collected by the sensor in real time, and outputs the corresponding fault diagnosis model.

[0020] Preferably, the data processing unit is connected to an actual application feedback module via a data signal line, and the actual application feedback module is used to implement the fault diagnosis steps and display the parameter information data of the diagnosis process in real time; The actual application feedback module includes: a visual interface display screen and a fault diagnosis software, wherein the fault diagnosis software is installed in the visual interface display screen; The fault diagnosis software receives the input fault diagnosis module code data and the real-time working data of the bearing, performs optimal diagnosis processing on the bearing, and transmits the diagnosis result data and real-time process parameter information to the visual interface display screen for display.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention realizes the advantage integration function of multiple models by designing a model integration step, solves the limitation problem of a single model in processing complex and changeable fault modes, and is difficult to perform fault diagnosis comprehensively and accurately, thereby improving the accuracy and robustness of the bearing fault diagnosis system, being able to better cope with fault detection under complex working conditions, and reducing the probability of false alarms and missed alarms; 2. The present invention realizes the training function of the fault diagnosis model by designing a model training step, solves the problem that the traditional method does not make sufficient use of data, is difficult to accurately establish a fault diagnosis model with a small amount of data, and has a weak generalization ability of the model. The present invention makes full use of historical data for training, constructs a model with strong generalization ability and good diagnostic effect, and improves the reliability of fault diagnosis; 3. The present invention realizes the automatic diagnosis function of bearing faults by designing a fault diagnosis step, solves the problem that the traditional fault diagnosis method relies on manual judgment, is inefficient, has a high risk of misjudgment and missed judgment, and is difficult to find faults in time, realizes automatic and real-time fault diagnosis and alarm, improves the fault response speed and accuracy, reduces equipment downtime, and reduces maintenance costs; 4. The present invention realizes efficient data processing function by designing a data processing unit, solves the problem that the original sensor data usually contains a lot of noise and interference, and directly using it for fault diagnosis will affect the accuracy of the diagnosis result. Through the preprocessing steps, the quality and reliability of the data are improved, laying a solid foundation for subsequent feature extraction and model diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the model integration steps of the present invention; Figure 2 It is a schematic diagram of the model training steps of the present invention; Figure 3 It is a schematic diagram of the fault diagnosis steps of the present invention; Figure 4 is a schematic diagram of a data processing unit of the present invention; Figure 5 It is a schematic diagram of the feature extraction step of the present invention; Figure 6 It is a schematic diagram of the feature selection steps of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of 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.

[0024] Example 1 See also Figure 1 and Figure 6, an embodiment provided by the present invention: a bearing fault diagnosis method, comprising a model training step, a model integration step and a fault diagnosis step, wherein the model training step provides a trained diagnosis model for the model integration step, the abnormality detection model output by the model integration step is used as a model algorithm input of the fault diagnosis step, and the model integration step is used to integrate multiple models to improve the diagnosis accuracy and system robustness; The model integration step includes: multi-model integration, dynamic model selection and anomaly detection steps; The multi-model integration is to design an integrated learning framework, combine different classification models to fuse the prediction results of the models by voting / weighted average, output the model integration algorithm code, and integrate the model effect evaluation report; The dynamic model selection is to design a dynamic model selection algorithm based on real-time data features, select the model that performs best under the current bearing state for fault diagnosis, and output the dynamic model selection algorithm code and real-time model selection demonstration; The anomaly detection step uses an anomaly detection algorithm to establish an anomaly detection model, monitor and identify data that exceeds the normal range, and output an anomaly detection step code and an anomaly detection result analysis report; Furthermore, we first conduct multi-model integration and build multiple fault diagnosis models of different types, such as support vector machines, random forests, and convolutional neural networks. We select appropriate features and divide the data set into a training set and a test set. We use the training set to train each model separately and optimize the model parameters to ensure that each model has high accuracy. We use the test set to evaluate the performance of each model and record the accuracy, recall, F1 value and other indicators of each model. We then perform model fusion and use the voting method: each model classifies the input data and selects the category with the most occurrences in the classification results as the final diagnosis result. We use the weighted average method: according to the performance indicators of each model, such as accuracy, we assign weights and calculate the weighted average as the final diagnosis result. By fusing the outputs of different models, we can improve the overall diagnostic accuracy and robustness. Dynamic model selection: Extract time domain features such as mean, standard deviation, frequency domain features such as main frequency, spectrum width and statistical features such as autocorrelation function from real-time sensor data, standardize the extracted features to ensure the consistency of scales of different features; pre-evaluate the performance of each model under different operating conditions, establish a feature-model mapping relationship, use the K nearest neighbor KNN algorithm or other classification algorithms, select the optimal model based on real-time features, input real-time data into the selected optimal model, and perform fault diagnosis; Anomaly detection: Build an anomaly detection model based on Isolation Forest. Use normal operation data to train the Isolation Forest model, establish a baseline for normal operation, determine model parameters such as the number of trees and subsample size, and ensure that the model has good detection performance. Input real-time sensor data into the Isolation Forest model, calculate the isolation degree of each data point, and compare the isolation degree with the set threshold. If the isolation degree is less than the threshold, the data point is judged to be abnormal. Record and alarm the detected anomaly, and notify maintenance personnel to conduct further inspection. The accuracy and robustness of fault diagnosis are improved through multi-model integration, dynamic model selection ensures the selection of the optimal model under different conditions, and anomaly detection further improves the system's ability to identify unknown faults.

[0025] Example 2 See also Figure 2 and Figure 4 , an embodiment provided by the present invention: a bearing fault diagnosis method, wherein the model training step is used to train and evaluate different classification models; The model training steps include: training data preparation, model selection, and model training evaluation; Training data preparation is to extract sample data of normal and faulty bearing operation from actual operation data, mark them, and output sample data sets and data marking description documents; Model selection is to compare different machine learning models, select the most suitable model for bearing fault diagnosis, and output the model comparison report and the selected model and its parameters; Model training evaluation is to use the training data set to train the selected model, and use the test data set to evaluate the model's accuracy, precision and recall performance indicators, and output the trained model file and model evaluation report; Furthermore, sensor data under various operating conditions are collected from the bearing equipment, including normal operating conditions and different types of fault conditions. The collected data are preprocessed, including operations such as removing outliers, data alignment and denoising, to ensure data quality. Features are extracted from the preprocessed data, including time domain features, frequency domain features and statistical features, etc., to facilitate model training. The extracted features are marked with the corresponding normal or faulty bearing conditions to construct a training data set. According to the task requirements of bearing fault diagnosis, a group of models are selected as candidates, such as support vector machine (SVM), decision tree (Decision Tree) and convolutional neural network (CNN). The training data set is divided into training set and validation set. Usually 70% of the data is used as the training set and 30% of the data is used as the validation set. The divided training set is input into each model. The training algorithm is used to learn the pattern and regularity of the data to obtain a trained model. According to the performance indicators of the validation set, such as accuracy, recall rate, F1 value, etc., the hyperparameters of the model are adjusted to optimize the model performance. Cross-validation techniques such as k-fold cross-validation are used to evaluate each model, and the model with better performance is selected as the final training model. Input the reserved test data set into the trained model for performance evaluation. Calculate the performance indicators of the model on the test data set, including accuracy, recall rate, F1 value, etc. Evaluate the diagnostic accuracy and robustness of the model, analyze the performance of the model, identify models with poor performance, and think about improvement strategies. Select the model with the best performance as the final bearing fault diagnosis model based on the evaluation results. Save the trained model as a model file for subsequent use in the bearing fault diagnosis device. Record the model's training parameters, performance indicators and other relevant information for future reference and updating.

[0026] Example 3 See also Figure 1 , Figure 2 and Figure 3 , an embodiment provided by the present invention: a bearing fault diagnosis method, wherein the fault diagnosis step is used to classify the bearing fault and display the fault diagnosis result; The fault diagnosis steps include: fault classification, visualization interface and alarm system; Fault classification is to use the trained model to classify the real-time sensor data, identify the bearing fault type, and output the fault classification algorithm code and real-time fault diagnosis results; Visual interface means designing interactive interface, including real-time data display, fault diagnosis result display and historical data query function; The alarm system sets the threshold for fault diagnosis. When the diagnosis result exceeds the threshold, the alarm is triggered and an email, text message and sound prompt are sent; Furthermore, the extracted features are input into a pre-trained fault classification model, which analyzes and classifies the data and outputs the fault type and health status of the bearing. The classification results are stored in a database for subsequent query and analysis, and the classification results are passed to the visualization interface and alarm system through an interface. Design a user-friendly interface to display the operating status of the bearing, fault diagnosis results, and real-time data information of the sensor. The interface uses graphical display methods, such as charts, dashboards, and heat maps, to make the information intuitive and easy to understand; real-time data display: display the real-time data change trend of the sensor through line charts, bar charts, etc. Fault diagnosis results: display the current diagnosed fault type and occurrence time in the form of a list or icon. Historical data query: provide query function, users can view fault records and data changes in the past period of time, refresh interface data in real time to ensure that users see the latest monitoring and diagnosis results, provide interactive operations, users can click to view detailed information, zoom in or out data charts; According to the fault classification results and sensor data, alarm conditions are set, such as temperature exceeding a certain threshold, abnormal vibration signal, etc. Users can adjust the alarm threshold in the visual interface, set the alarm level for different fault types, monitor the fault classification results and sensor data in real time, and trigger the alarm when the alarm conditions are met. The alarm information includes the fault type, occurrence time, sensor data and fault description, etc. The device sounds an alarm to remind the on-site personnel to pay attention, and the fault information is highlighted on the visual interface, with color changes or flashing reminders. Maintenance personnel are notified through SMS, email or mobile applications to ensure timely response, and the alarm information is recorded and stored in the database for subsequent analysis and processing by maintenance personnel. User confirmation processing: In the visual interface, maintenance personnel can confirm the alarm information and record the processing process and results; Through the above process, the fault diagnosis steps can efficiently classify faults, display diagnostic results and issue alarms in a timely manner, helping maintenance personnel to respond quickly and reduce equipment downtime and maintenance costs.

[0027] Example 4 See also Figure 2 and Figure 4 , an embodiment provided by the present invention: a bearing fault diagnosis method, the sensor data is collected by a sensor real-time collection step, and the sensor real-time collection step is used to collect and process the bearing working data in real time; The sensor real-time acquisition steps include: sensor selection and layout, data acquisition and transmission, and data preprocessing; Sensor selection and layout means selecting sensors with a wide frequency range and high sensitivity, including MEMS accelerometers, to monitor the vibration data of the bearings. Temperature sensors should be of high temperature resistance type to ensure real-time and reliable collection of bearing temperature data. Sound sensors should be of high sensitivity and excellent signal-to-noise ratio to detect abnormal sound data generated during bearing operation. Data acquisition and transmission is to use a high-speed sampling rate analog-to-digital converter to convert the sensor data into analog-to-digital data, and transmit the data to the data processing unit through the serial port and Ethernet; The data processing unit is equipped with a data preprocessing program, which reduces high-frequency noise through a low-pass filter, removes background noise and clutter using wavelet transform, designs a data alignment algorithm to ensure the time sequence consistency of data from different sensors, and outputs preprocessed data samples; Furthermore, the acceleration sensor is installed on the bearing seat or bearing housing to ensure that the characteristic signal of the bearing vibration can be captured. Choose a suitable installation position and direction to obtain the best vibration signal. The temperature sensor is installed near the bearing housing or bearing seat to ensure that the temperature change of the bearing can be accurately reflected. Use materials with good thermal conductivity to fix it to ensure temperature conductivity. The sound sensor is installed near the bearing to avoid interference from external environmental noise as much as possible. Choose a microphone with high sensitivity to ensure that abnormal sounds during the operation of the bearing can be captured. Calibrate the sensor before installing it; Use an analog-to-digital converter to convert the analog signals of the acceleration sensor and temperature sensor into digital signals, configure the data acquisition software, set the sampling frequency to 1KHz, ensure the synchronous acquisition of vibration and temperature data, and transmit the collected data to the data processing unit through the Ethernet interface. The data is stored in the local silver disk and the cloud server at the same time to ensure the security and accessibility of the data; Perform low-pass filtering on vibration data to remove high-frequency noise, smooth temperature data to eliminate measurement errors of temperature sensors, align vibration and temperature data, synchronize the AEB data on the time axis, extract time domain features, frequency domain features, and statistical features, standardize the extracted features, and convert them into standard normal distribution; Through the above process, sensor data can be collected and processed in real time, providing high-quality input data for subsequent fault diagnosis, ensuring the accuracy and reliability of fault diagnosis.

[0028] Example 5 See also Figure 3 , Figure 4 and Figure 5 , an embodiment provided by the present invention: a bearing fault diagnosis method, wherein the data processing unit is further designed with a feature extraction step, the feature extraction step is used to effectively extract data features from sensor data; The feature extraction steps include: time domain feature extraction, frequency domain feature extraction and statistical feature extraction; Time domain feature extraction is to calculate the mean, standard deviation, and peak-to-peak value time domain statistical features of the bearing vibration signal, and output an example of the time domain feature extraction result; Frequency domain feature extraction is to use Fourier transform to convert time domain signals to frequency domain, extract spectrum characteristics, and output frequency domain feature extraction result examples; Statistical feature extraction is to calculate the statistical features of the autocorrelation function and cross-correlation function of the bearing vibration signal, and output the statistical feature extraction result example; Further, the mean, standard deviation, peak, root mean square, skewness and kurtosis are calculated from the vibration signal to obtain a time domain feature vector. For example, the mean of the vibration signal is calculated to be 0.5, the standard deviation is 0.2, the peak is 1.5, the root mean square value is 0.7, the skewness is 0.1, and the kurtosis is 3.2. The vibration signal is subjected to Fourier transform to obtain a spectrum. The main frequency extracted from the spectrum is 50 Hz, the spectrum energy is 200, the bandwidth is 20 Hz, and the harmonic components include 100 Hz, 150 Hz, etc. The autocorrelation function of the vibration signal is calculated, the periodicity of the signal is analyzed, the power spectrum of the signal is calculated, the frequency distribution of the signal is described, the time-frequency characteristics of the signal are analyzed using short-time Fourier transform, and a time-frequency diagram is obtained. The multi-scale characteristics of the signal are analyzed using wavelet transform to obtain wavelet coefficients. Through the above process, the bearing monitoring system can extract rich time domain, frequency domain and statistical features to provide high-quality input data for subsequent fault diagnosis, ensuring the accuracy and reliability of fault diagnosis.

[0029] Example 6 See also Figure 4 , Figure 5 and Figure 6 , an embodiment provided by the present invention: a bearing fault diagnosis method, wherein the data processing unit is further designed with a feature selection step, the feature selection step is used to analyze, evaluate and select different features; The feature selection steps include: feature correlation analysis, feature importance assessment, and feature selection algorithm; Feature correlation analysis uses the canonical correlation analysis method to evaluate the correlation between features, exclude highly correlated features, and output feature correlation matrix and correlation analysis report; Feature importance assessment is to use information gain and Gini index methods to assess the importance of features, select the features that have the greatest impact on classification, and output feature importance rankings and feature importance assessment reports; The feature selection algorithm uses the Wrapper method, Filter method, and Embedded method to select features and output the final selected feature list; Further, the Pearson correlation coefficient between each pair of features was calculated, and it was found that the correlation coefficient between feature A and feature B was 0.8, and the correlation coefficient between feature A and feature C was 0.2. The mutual information between the feature and the label was calculated, and it was found that the mutual information between feature A and the label was 0.5, and the mutual information between feature B and the label was 0.3. The feature importance was calculated using the random forest model, and it was found that the importance of feature A was 0.4, and the importance of feature B was 0.2. Lasso regression was used for feature selection, and it was found that the coefficient of feature A was 0.6, and the coefficient of feature B was 0.1. The filtering method was used to select features with mutual information higher than 0.3 based on the results of feature-label correlation analysis. The recursive feature elimination method RFE was used, and the support vector machine SVM model was used for feature selection. One feature was removed each time, and finally the feature subset with the best performance was selected. Lasso regression was used for embedding feature selection, and features with non-zero coefficients were selected.

[0030] Working principle: First, the vibration, temperature and noise data of the bearing during operation are collected in real time through a variety of sensors. The collected data are converted into digital signals through an analog-to-digital converter and transmitted to the data processing unit. The data processing unit filters, denoises and standardizes the original signal, extracts time domain, frequency domain and statistical features, and provides high-quality input data for subsequent fault diagnosis; Next, among the various features extracted in the preprocessing stage, the most useful features for fault diagnosis are screened out through the feature selection step. Feature selection includes feature correlation analysis, feature importance evaluation and application of selection algorithms. Common methods include Pearson correlation coefficient, mutual information, random forest feature importance and Lasso regression. The selected features are used to train the fault diagnosis model. Machine learning algorithms such as support vector machine, random forest and gradient boosting decision tree are used for model training and evaluation to optimize the parameters and performance of the model to ensure that it has good fault identification capabilities. Finally, the trained model is integrated into the diagnostic device to analyze sensor data in real time. When the model detects abnormal patterns or fault characteristics, the system automatically classifies the fault type and displays the diagnostic results on a visual interface. The system also includes an alarm function, which notifies maintenance personnel in a timely manner to take corresponding measures through sound and light alarms, text messages or emails. The entire process achieves early detection and early warning of bearing faults, avoids the occurrence of serious faults, and improves the safety and reliability of equipment operation.

[0031] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A bearing fault diagnosis method, characterized in that: The method comprises a model training step, a model integration step and a fault diagnosis step, wherein the model training step provides a trained diagnosis model for the model integration step, the abnormality detection model output by the model integration step is used as a model algorithm input of the fault diagnosis step, and the model integration step is used to integrate multiple models to improve the diagnosis accuracy and system robustness; The model integration step includes: multi-model integration, dynamic model selection and anomaly detection steps; The multi-model integration is to design an integrated learning framework, combine different classification models to fuse the prediction results of the models by voting / weighted average, output the model integration algorithm code, and integrate the model effect evaluation report; The dynamic model selection is to design a dynamic model selection algorithm based on real-time data features, select the model that performs best under the current bearing state for fault diagnosis, and output the dynamic model selection algorithm code and real-time model selection demonstration; The anomaly detection step uses an anomaly detection algorithm to establish an anomaly detection model, monitor and identify data that exceeds a normal range, and output an anomaly detection step code and an anomaly detection result analysis report.

2. A bearing fault diagnosis method according to claim 1, characterized in that: The model training step is used to train and evaluate different classification models; The model training steps include: training data preparation, model selection, and model training evaluation; Training data preparation is to extract sample data of normal and faulty bearing operation from actual operation data, mark them, and output sample data sets and data marking description documents; Model selection is to compare different machine learning models, select the most suitable model for bearing fault diagnosis, and output the model comparison report and the selected model and its parameters; Model training evaluation is to use the training data set to train the selected model, and use the test data set to evaluate the model's accuracy, precision and recall performance indicators, and output the trained model file and model evaluation report.

3. A bearing fault diagnosis method according to claim 1, characterized in that: The fault diagnosis step is used to classify bearing faults and display fault diagnosis results; The fault diagnosis steps include: fault classification, visualization interface and alarm system; Fault classification is to use the trained model to classify the real-time sensor data, identify the bearing fault type, and output the fault classification algorithm code and real-time fault diagnosis results; Visual interface means designing interactive interface, including real-time data display, fault diagnosis result display and historical data query function; The alarm system sets the threshold for fault diagnosis. When the diagnosis result exceeds the threshold, an alarm is triggered and an email, text message and sound prompt are sent.

4. A bearing fault diagnosis method according to claim 3, characterized in that: The sensor data is collected in real time by a sensor, and the sensor real time collection step is used to collect and process the bearing working data in real time; The sensor real-time acquisition steps include: sensor selection and layout, data acquisition and transmission, and data preprocessing; Sensor selection and layout means selecting sensors with a wide frequency range and high sensitivity, including MEMS accelerometers, to monitor the vibration data of the bearings. Temperature sensors should be of high temperature resistance type to ensure real-time and reliable collection of bearing temperature data. Sound sensors should be of high sensitivity and excellent signal-to-noise ratio to detect abnormal sound data generated during bearing operation. Data acquisition and transmission is to use a high-speed sampling rate analog-to-digital converter to convert the sensor data into analog-to-digital data, and transmit the data to the data processing unit through the serial port and Ethernet; The data processing unit is equipped with a data preprocessing program, which reduces high-frequency noise through a low-pass filter, uses wavelet transform to remove background noise and clutter, designs a data alignment algorithm to ensure the timing consistency of data from different sensors, and outputs preprocessed data samples.

5. A bearing fault diagnosis method according to claim 4, characterized in that: The data processing unit is also designed with a feature extraction step, which is used to effectively extract data features from the sensor data; The feature extraction steps include: time domain feature extraction, frequency domain feature extraction and statistical feature extraction; Time domain feature extraction is to calculate the mean, standard deviation, and peak-to-peak value time domain statistical features of the bearing vibration signal, and output an example of the time domain feature extraction result; Frequency domain feature extraction is to use Fourier transform to convert time domain signals to frequency domain, extract spectrum characteristics, and output frequency domain feature extraction result examples; Statistical feature extraction is to calculate the statistical features of the autocorrelation function and cross-correlation function of the bearing vibration signal, and output an example of the statistical feature extraction result.

6. A bearing fault diagnosis method according to claim 4, characterized in that: The data processing unit is also designed with a feature selection step, which is used to analyze, evaluate and select different features; Feature selection steps Includes: feature correlation analysis, feature importance assessment and feature selection algorithm; Feature correlation analysis uses the canonical correlation analysis method to evaluate the correlation between features, exclude highly correlated features, and output feature correlation matrix and correlation analysis report; Feature importance assessment is to use information gain and Gini index methods to assess the importance of features, select the features that have the greatest impact on classification, and output feature importance rankings and feature importance assessment reports; The feature selection algorithm uses the Wrapper method, Filter method, and Embedded method to select features and output the final selected feature list.

7. A bearing fault diagnosis method according to claim 1, characterized in that: The dynamic model selection algorithm performs dynamic selection based on real-time data features to ensure that the most suitable model for the current bearing state is selected for fault diagnosis. The specific algorithm steps are as follows: S1. Calculate the scores of each model under the current bearing state according to the real-time data characteristics; S2, select the model with the highest score as the optimal model; S3. Return the optimal model.

8. A bearing fault diagnosis method according to claim 1, characterized in that: The anomaly detection model adopts the isolation forest anomaly detection model, and the specific algorithm steps are as follows: Random sampling: randomly extract samples from the data set to construct random numbers; Tree construction: randomly select features and thresholds, recursively divide the data into subspaces, and construct random trees; Isolation calculation: Calculate the depth of each data point in the tree as a measure of its isolation; Abnormality judgment: Determine whether a data point is abnormal based on isolation and threshold.

9. A bearing fault diagnosis device, applicable to a bearing fault diagnosis method according to any one of claims 1 to 8, characterized in that: The diagnostic device is used to implement the model training step, the model integration step, the fault diagnosis step and the sensor real-time acquisition step in the diagnostic method, and the diagnostic device includes: a sensor module and a data processing unit, and the sensor module is connected to the data processing unit through an Ethernet interface; The sensor module is used to implement the sensor real-time acquisition steps; The sensor module includes: an acceleration sensor, a temperature sensor and a sound sensor, which are used to detect the vibration, temperature and sound parameters of the bearing respectively. The sensor module is arranged around the bearing, and the sensor data is converted into a digital signal through an analog-to-digital converter and transmitted to the data processing unit through a serial port and an Ethernet interface; The data processing unit is used to implement the model training step, the model integration step, the feature extraction step and the feature selection step, and performs data processing according to the bearing operating condition information data collected by the sensor in real time, and outputs the corresponding fault diagnosis model.

10. A bearing fault diagnosis device according to claim 9, characterized in that: The data processing unit is connected to the actual application feedback module through a data signal line, and the actual application feedback module is used to implement the fault diagnosis steps and display the parameter information data of the diagnosis process in real time; The actual application feedback module includes: a visual interface display screen and a fault diagnosis software, wherein the fault diagnosis software is installed in the visual interface display screen; The fault diagnosis software receives the input fault diagnosis module code data and the real-time working data of the bearing, performs optimal diagnosis processing on the bearing, and transmits the diagnosis result data and real-time process parameter information to the visual interface display screen for display.

Images