Equipment fault monitoring method and monitoring system based on bearing monitoring

Abstract

The invention provides an equipment fault monitoring method and monitoring system based on bearing monitoring, and relates to the technical field of fault diagnosis, and the method comprises the steps of data collection, signal decoupling, first model construction, fault prediction, second model construction, fault propagation, equipment fault prediction and the like. The system comprises a data acquisition module, a signal decoupling module, a first model construction module, a fault prediction module, a second model construction module, a fault propagation module and an equipment fault prediction module. According to the invention, the health states of multiple bearings can be monitored in real time, the propagation process of faults among different bearings can be predicted, potential equipment faults can be identified in advance, and accurate fault monitoring and prediction of multi-bearing equipment can be realized.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to an equipment fault monitoring method and a monitoring system based on bearing monitoring. Background Art

[0002] With the continuous improvement of industrialization and automation levels, the operating efficiency and reliability of mechanical equipment play a vital role in modern manufacturing. As one of the core components of mechanical equipment, bearings are subjected to multiple pressures such as rotation, load and friction. Their health directly affects the performance and service life of the equipment.

[0003] The Chinese invention patent with the application publication date of July 12, 2024 and publication number CN118329450A provides a bearing fault diagnosis method, system, device and storage medium. The patent obtains the original bearing vibration data to construct a bearing vibration sample set; constructs a bearing fault diagnosis model, and uses the unlabeled data and labeled data in the bearing vibration sample set to perform unsupervised pre-training and supervised fine-tuning training on the bearing fault diagnosis model; obtains the bearing vibration data to be classified, and uses the supervised fine-tuning trained bearing fault diagnosis model to perform end-to-end bearing fault diagnosis on the bearing vibration data to be classified to obtain the bearing fault classification result. The present invention performs unsupervised pre-training and supervised fine-tuning training on unlabeled data and labeled data based on time series modeling technology and masked autoencoding technology, so that the trained bearing fault diagnosis model can perform end-to-end high-precision bearing fault classification on bearing vibration data.

[0004] The above-mentioned technical solutions focus on fault diagnosis of a single bearing, ignoring the mutual influence and fault propagation between multiple bearings. When a bearing of a multi-bearing device fails, the fault signals are easily intertwined and superimposed, making it difficult to accurately identify the source of the fault from the signal collected from a single sensor. In addition, the fault diagnosis of a single bearing may ignore the evolution of the entire equipment fault and cannot accurately predict the overall health status of the equipment. Summary of the invention

[0005] In order to improve the accuracy of bearing fault monitoring in a multi-bearing system, the present application provides an equipment fault monitoring method and a monitoring system based on bearing monitoring.

[0006] In the first aspect, the present application provides a method for monitoring equipment failures based on bearing monitoring, which adopts the following technical solution: A method for monitoring equipment failure based on bearing monitoring comprises the following steps: Data collection: Collect the test data of the bearing group of the equipment; the test data includes vibration signal, temperature, lubricating oil metal powder, speed and load; Signal decoupling: Use blind source separation technology to perform signal decoupling on the collected detection data to obtain the detection data of each bearing in the bearing group; First model construction: build a fault prediction model based on deep learning algorithm; Fault prediction: predict the fault type label of each bearing based on the detection data of each bearing; Second model construction: construct the fault propagation model of the bearing group; Fault propagation: According to the fault type label of each bearing, the faulty bearings are screened out to determine whether the number of faulty bearings exceeds the preset threshold: If yes, the device is judged to be faulty and the current time is output as the device fault time; If not, the detection data of the selected bearings are input into the fault propagation model, the fault evolution results are output, and the equipment fault prediction steps are executed; Equipment failure prediction: predict the remaining life of the equipment based on the failure evolution results.

[0007] By adopting the above technical solution, the collected detection data is decoupled by blind source separation technology to ensure that the fault detection data of each bearing is independent and accurate, thereby improving the accuracy of bearing fault detection in multi-bearing systems. In addition, by establishing a fault propagation model for the bearing group, it is possible to monitor the health status of multiple bearings in real time, predict the propagation process of faults between different bearings, identify potential equipment failures in advance, and achieve accurate fault monitoring and prediction of multi-bearing equipment.

[0008] Optionally, before executing the step of data collection, the method further includes: Sensor type selection: according to the type of bearing detection data to be collected, set the sensor that collects the corresponding data type, and regard a group of sensors corresponding to the detection data type as a sensor group; Define the optimization goal: minimize the number of sensor groups and maximize the fault detection coverage, and construct the objective function; Define constraints: Take the operating characteristics, failure modes, and interference between sensors of each bearing in the equipment's bearing group as constraints of the objective function; Sensor position optimization: According to the objective function and constraints, the optimization algorithm is used to obtain the number of sensor groups and the corresponding installation positions.

[0009] By adopting the above technical solution, the number, position and arrangement of sensors are optimized to ensure that the fault detection coverage of the bearing group of the equipment is maximized. At the same time, the working characteristics and failure modes of the bearings are taken into consideration. While reducing costs, it is not only possible to effectively monitor the status of each bearing, but also to effectively reduce the signal distortion caused by sensor interference, thereby improving the quality of detection data, reducing noise and redundant information, and helping to improve the accuracy of fault diagnosis and the reliability of equipment.

[0010] Optionally, after performing the step of data acquisition and before performing the step of signal decoupling, the method further includes: Time domain analysis: The time domain analysis method is used to extract the time domain features in the detection data of the bearing group, and the extracted time domain features and the detection data of the bearing group are used as new detection data of the bearing group.

[0011] By adopting the above technical solution, the redundant information and noise in the original data are compressed into the feature data, the interference signal in the decoupling process is reduced, and the accuracy of signal decoupling is improved. In addition, the features extracted by time domain analysis improve the recognition of each bearing signal, which helps to separate the signals of each bearing from the mixed overall signal during the signal decoupling process, improve the decoupling efficiency, and enhance the accuracy of signal decoupling and data quality.

[0012] Optionally, after performing the signal decoupling step and before performing the first model building step, the method further includes: Data fusion: performing time series matching on the detection data of each bearing in the bearing group and the detection data of the same data type belonging to the same bearing, and recording the matched data as the first data; Feature extraction: The detection data belonging to the same sensor group in the detection data of each bearing is taken as a group of sensor group data, and feature extraction is performed on the sensor group data of each bearing to obtain several sensor data features of each bearing; Feature fusion: The sensor data features of each bearing are fused to obtain the fusion feature of each bearing, which is recorded as the first feature; Data splicing: Splice the first data with the first feature, and use the spliced ​​data as new detection data for each bearing.

[0013] By adopting the above technical solution, the detection data of the same data type of each bearing is aligned by using time series matching, avoiding analysis errors caused by asynchronous or incomplete data, thereby ensuring the accuracy of subsequent bearing fault analysis. In addition, ensuring that all data are synchronized on the time axis helps to further strengthen the model's capture of bearing state changes and avoid diagnostic problems caused by time dislocation or omissions. In addition, through feature extraction and feature fusion of detection data, complex raw data is converted into efficient features that are easy to analyze, which helps to extract more comprehensive and accurate fault information from multi-sensor data, enhance the multi-dimensional capability of bearing fault detection, improve the learning effect of the fault diagnosis model, and improve the accuracy of bearing fault diagnosis.

[0014] Optionally, after executing the step of building the first model and before executing the step of fault prediction, the method further includes: Constructing a first sample training set: acquiring historical detection data and corresponding fault type labels of each bearing in a bearing group of the equipment, and constructing a first sample training set based on the acquired historical detection data and corresponding fault type labels; the fault type labels include fault type and fault severity; First model training: input the first sample training set into the fault prediction model. The shared layer of the fault prediction model extracts the features of the historical detection data and the fault type labels corresponding to the historical detection data, which are recorded as shared features. Task-specific branches are created according to the fault type labels. The task-specific branches include fault type classification tasks and fault severity classification tasks. The fault prediction loss function is defined, and the shared features are input into the task-specific branches. The tasks are processed, and the fault prediction model is iteratively optimized by minimizing the fault prediction loss function. The optimized fault prediction model is used as the new fault prediction model.

[0015] By adopting the above technical solution, not only the fault type is classified, but also the severity of the fault is considered. The dual-task structure is designed so that the bearing fault prediction model can not only identify the specific type of equipment fault, but also evaluate its severity, thereby improving the accuracy of bearing fault identification.

[0016] Optionally, after executing the step of fault prediction and before executing the step of building the second model, the method further includes: Define node features: take each bearing as a node, the detection data of each bearing as the initial feature of the node, and the fault type label of each bearing as the target feature of the node; Define edge features: Each bearing is connected by an edge, and define the propagation probability and propagation strength as edge features for each edge; Constructing an adjacency matrix: According to the structure and working principle of the equipment, the connection relationship between the bearings in the bearing group is obtained, and the adjacency matrix is ​​constructed according to the connection relationship between the bearings.

[0017] By adopting the above technical solution, each bearing is regarded as a node, and its detection data is used as the initial feature of the node, and the fault type label is used as the target feature. This enables each node to carry rich state information, which can better capture the working state of the bearing and the possible fault type, thereby improving the sensitivity to the state changes of each node (bearing). In addition, by defining the propagation probability and propagation strength for the connection between each pair of bearings, the propagation process of the fault between the bearings can be simulated. By defining node features, edge features and constructing an adjacency matrix, the state of each bearing in the equipment and the fault propagation process between them can be effectively modeled, which can accurately capture the structural characteristics of the equipment and the interaction between the bearings, and improve the accuracy of fault prediction.

[0018] Optionally, after executing the step of building the second model and before executing the step of fault propagation, the method further includes: Constructing a second sample training set: obtaining historical detection data of each bearing in the bearing group of the equipment, the corresponding fault type label, and the corresponding historical fault propagation record, and constructing a second sample training set; Define loss function: Define fault propagation loss function; Constructing a second sample training set: obtaining historical detection data of each bearing in the bearing group of the equipment, the corresponding fault type label, and the corresponding historical fault propagation record, and constructing a second sample training set; Second model training: Input the second sample training set into the fault propagation model to perform model inference. Calculate the output of the fault propagation model through forward propagation, compare the difference between the predicted result and the actual result, calculate the model loss, update the fault propagation model parameters through backward calculation, iteratively optimize the fault propagation model by minimizing the fault propagation loss function, and use the optimized fault propagation model as the new fault propagation model.

[0019] By adopting the above technical solution, an accurate fault propagation model is established, which can identify and simulate the propagation process of faults from one bearing to other mutually coupled bearings. It can not only predict the failure of a single bearing, but also analyze the mutual influence and fault path between different components, helping the system design a more robust fault-tolerant strategy.

[0020] Optionally, after executing the step of fault propagation and before executing the step of device fault prediction, the method further includes: First fault identification: Based on the obtained fault evolution results, the predicted fault time of each bearing in the bearing group is obtained. From the current moment, each bearing is sorted in the order of predicted fault time from small to large, and the number of bearing faults is accumulated in order. When the accumulated number of bearing faults exceeds the preset number threshold, the predicted fault time of the last accumulated bearing is recorded as the first time, and the predicted fault time of the main bearing is obtained and recorded as the second time. It is determined whether the first time is greater than the second time: If yes, the second time is used as the equipment failure time, and the equipment failure prediction step is performed; If not, the first time is taken as the equipment failure time, and the equipment failure prediction step is performed.

[0021] By adopting the above technical solution, the predicted failure time of the main bearing is used as a reference first, which can avoid the wrong prediction of the equipment failure time due to other non-main bearing failures. In addition, by sorting the predicted failure time of all bearings and accumulating the number of failures, this method can effectively identify which bearing failure will have the most serious impact on the equipment. Especially when multiple bearings fail, it can more accurately reflect the overall failure trend of the equipment.

[0022] In a second aspect, the present invention provides an equipment fault monitoring system based on bearing monitoring, which adopts the following technical solution: An equipment fault monitoring system based on bearing monitoring, comprising: The data acquisition module is used to collect the detection data of the bearing group of the equipment; the detection data includes vibration signal, temperature, lubricating oil metal powder, speed and load; A signal decoupling module is used to perform signal decoupling on the collected detection data using a blind source separation technique to obtain detection data of each bearing in the bearing group; A first model building module, used to build a fault prediction model based on a deep learning algorithm; A fault prediction module is used to predict the fault type label of each bearing based on the detection data of each bearing; A second model building module is used to build a fault propagation model of the bearing group; The fault propagation module is used to filter out the faulty bearings according to the fault type label of each bearing and determine whether the number of faulty bearings exceeds the preset number threshold: If the number of failed bearings exceeds the preset threshold, the equipment is judged to be faulty and the current time is output as the equipment fault time; If the number of faulty bearings does not exceed the preset number threshold, the detection data of the selected bearings will be input into the fault propagation model, the fault evolution results will be output, and the equipment fault prediction module will be triggered; The equipment failure prediction module is used to predict the remaining life of the equipment based on the failure evolution results.

[0023] By adopting the above technical solution, the collected detection data is decoupled by the blind source separation technology of the signal decoupling module, ensuring that the fault detection data of each bearing is independent and accurate, thereby improving the accuracy of bearing fault detection in the multi-bearing system. In addition, by establishing a fault propagation model of the bearing group through the second model building module, it is possible to monitor the health status of multiple bearings in real time, predict the propagation process of faults between different bearings, identify potential equipment failures in advance, and realize accurate fault monitoring and prediction of multi-bearing equipment.

[0024] In summary, the present application includes at least one of the following beneficial technical effects: 1. Through the blind source separation technology, the collected detection data is signal decoupled to ensure that the fault detection data of each bearing is independent and accurate, which improves the accuracy of bearing fault detection in multi-bearing systems. In addition, by establishing a fault propagation model for the bearing group, it is possible to monitor the health status of multiple bearings in real time, predict the propagation process of faults between different bearings, identify potential equipment failures in advance, and achieve accurate fault monitoring and prediction of multi-bearing equipment.

[0025] 2. Optimize the number, location and arrangement of sensors to ensure maximum fault detection coverage of the equipment's bearing groups, while taking into account the operating characteristics and failure modes of the bearings. This not only effectively monitors the status of each bearing while reducing costs, but also effectively reduces signal distortion caused by sensor interference, thereby improving the quality of detection data, reducing noise and redundant information, and helping to improve the accuracy of fault diagnosis and the reliability of equipment.

[0026] 3. Use timing matching to align the detection data of the same data type for each bearing to avoid analysis errors caused by asynchronous or incomplete data, thereby ensuring the accuracy of subsequent bearing fault analysis. In addition, ensuring that all data is synchronized on the timeline helps to further enhance the model's capture of bearing state changes and avoid diagnostic problems caused by time dislocation or omissions. In addition, through feature extraction and feature fusion of detection data, complex raw data is converted into efficient features that are easy to analyze, which helps to extract more comprehensive and accurate fault information from multi-sensor data, enhance the multi-dimensional capabilities of bearing fault detection, improve the learning effect of the fault diagnosis model, and improve the accuracy of bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of Example 1 of the present application; Figure 2 is a flow chart of the S1 sensor layout of Example 1 of the present application; Figure 3 This is a flowchart of the first data processing in S41 of Example 1 of the present application; Figure 4 It is a flowchart of the first model construction of S5 in Example 1 of the present application; Figure 5 It is a flowchart of constructing the second model of S7 in Example 1 of the present application. DETAILED DESCRIPTION

[0028] The following combination Figures 1 to 5 This application is described in further detail.

[0029] Embodiment 1: This embodiment discloses a method for monitoring equipment failure based on bearing monitoring, such as Figure 1 As shown, the monitoring method includes: collecting detection data of the bearing group of the equipment, using blind source separation technology to perform signal decoupling on the collected detection data, obtaining detection data of each bearing in the bearing group, building a fault prediction model, predicting the fault type label of each bearing according to the detection data of each bearing, building a fault propagation model of the bearing group, screening out the faulty bearings according to the fault type label of each bearing, and judging whether the number of faulty bearings exceeds a preset number threshold: if so, judging that the equipment is faulty, and outputting the current moment as the equipment fault time; if not, inputting the detection data of the screened bearings into the fault propagation model, outputting the fault evolution result, and predicting the remaining life of the equipment according to the fault evolution result. This embodiment includes the following steps: S1 sensor layout: including S11 sensor type selection, S12 optimization goal definition, S13 constraint definition and S14 sensor location optimization, such as Figure 2 shown.

[0030] S11 sensor type selection: according to the type of bearing detection data to be collected, set the vibration sensor, temperature sensor, magnetic sensor, speed sensor and piezoelectric sensor, and regard a group of vibration sensors, temperature sensors, magnetic sensors, speed sensors and piezoelectric sensors as a sensor group.

[0031] Among them, the vibration sensor is used to collect bearing vibration signals, the temperature sensor is used to collect bearing temperature, the magnetic sensor is used to collect bearing lubricating oil metal powder data, the speed sensor is used to collect bearing speed, and the piezoelectric sensor is used to collect bearing load.

[0032] S12 defines the optimization goal: to minimize the number of sensors and maximize the fault detection coverage, and construct the objective function.

[0033] S13 defines constraints: the working characteristics of each bearing in the bearing group of the equipment, the failure mode and the interference between sensors are used as constraints of the objective function.

[0034] Working characteristic constraints: at least one sensor group is set at the main bearing of the bearing group; at least one sensor group is set at the high-load bearing; at least one sensor group is set at the bearing that is closely coupled with other bearings; at least one sensor group is set at the bearing in the vulnerable part; the installation position of the sensor is set according to the working conditions of each sensor in the sensor group. For example, vibration sensors are arranged at key positions such as the bearing contact surface, outer ring, inner ring and rolling element, and temperature sensors are set at the outer ring of the bearing or the nearby heat source.

[0035] Fault mode constraints: The sensors in the sensor group must be able to cover the data variation range of different bearing failure modes.

[0036] Interference constraint between sensors: The sum of interferences between the selected sensors is less than a preset interference threshold.

[0037] S14 Sensor location optimization: Based on the objective function and constraints, a genetic algorithm is used for optimization to obtain the number of sensor groups and the corresponding installation locations. This includes S141 initializing the population, S142 defining the fitness function, S143 selecting operations, S144 crossover operations, S145 mutation operations, S146 updating the population, and S147 outputting the optimal solution.

[0038] S141 Initialize population: Generate the initial population, each individual can be represented by a binary string or integer sequence, where each gene bit represents the installation position or selection of a sensor.

[0039] S142 defines the fitness function: The fitness function is defined as: .

[0040] in, Indicates The number of sensors for each solution, Indicates The failure mode coverage of each solution, Indicates The sensor interference of each solution is express The weight of express The weight of express The weight of .

[0041] S143 selection operation: calculate the fitness of each individual according to the fitness function, allocate the selection probability according to the proportion of the fitness of all individuals, perform roulette selection based on the selection probability, and select the parent individual.

[0042] S144 Crossover operation: Randomly select two parent individuals from the selected parent individuals, randomly select a crossover point, exchange the genes of parent 1 and parent 2 at the crossover point, generate two offspring, and place the newly generated offspring individuals into the next generation population.

[0043] S145 mutation operation: randomly select an individual, and according to the set mutation probability, randomly select a gene in the individual to mutate (for example, change the position of a sensor, or switch the type of sensor), re-evaluate the fitness of the mutated individual, and put it into the next generation population.

[0044] S146 Update population: After selection, crossover, and mutation, a new generation population is generated, the fitness value of each individual in the new generation population is calculated, the individuals of the parent and child generations are merged, the individuals with the best fitness are retained, and the individuals in the updated population are used as the new generation population.

[0045] S147 outputs the optimal solution: iterative updates are performed continuously until the maximum number of iterations is reached or the fitness function converges, the algorithm terminates, and the optimal solution is output, that is, the number of sensor groups and the corresponding installation positions are obtained.

[0046] S2 Data Collection: Collect the test data of the bearing group of the equipment; the test data includes vibration signal, temperature, lubricating oil metal powder, speed and load.

[0047] According to the S1 sensor layout, the number of sensor groups and the corresponding installation positions are obtained to install sensors. The vibration signals, temperature, lubricating oil metal powder, speed and load of the bearing group of the equipment are collected according to the installed sensors.

[0048] In this embodiment, when collecting the detection data of the bearing group of the equipment, the vibration signal, temperature, lubricating oil metal powder, rotation speed and load of the bearing group collected at the same time point are used as a group of data of the bearing group. Since multiple sensor groups are provided to collect the detection data of the bearing group, the detection data of the bearing group collected at the same time point includes data of several sensor groups.

[0049] In this embodiment, after the detection data of the bearing group is collected, the collected detection data needs to be preprocessed. The steps of data preprocessing include data cleaning, data normalization and feature extraction.

[0050] Data cleaning: De-noising, processing of missing values, and removal of outliers are performed on the collected test data.

[0051] Denoising includes using smoothing filters (such as sliding average) to eliminate small fluctuations in temperature, speed, and load data; using wavelet transform to decompose the vibration signal data, reconstructing after removing the noise component; and using smoothing filters to remove measurement fluctuations in lubricating oil metal powder data. Processing missing values ​​includes directly deleting data containing missing values. Removing outliers includes detecting outliers in the data based on the 3-sigma rule or box plot, and deleting the detected outlier data.

[0052] Data normalization: Normalize the collected test data.

[0053] S3 Time Domain Analysis: The time domain analysis method is used to extract the time domain features in the test data of the bearing group. The time domain features of each test data are extracted as follows: Time domain characteristics of vibration signals: mean, RMS value, maximum value, peak-to-peak value, variance, skewness and kurtosis.

[0054] Time domain characteristics of temperature signals: mean, maximum, standard deviation and rate of change.

[0055] Time domain characteristics of lubricating oil metal dust: mean, maximum and variance.

[0056] Time domain characteristics of the speed signal: mean, standard deviation and maximum value.

[0057] Time domain characteristics of the load signal: mean, maximum value and standard deviation.

[0058] The extracted time domain features of each detection data and the detection data of the bearing group are used as new detection data of the bearing group.

[0059] S4 signal decoupling: Use blind source separation technology to perform signal decoupling on the collected detection data to obtain the detection data of each bearing in the bearing group.

[0060] In the collected detection data, each sensor has a corresponding signal value at a certain moment. A data matrix containing M sensors and L time points is constructed, which is recorded as a mixed signal matrix. The mixed signal matrix is ​​de-meaned to remove the DC component of the signal. The signal is then whitened so that the signal has unit variance in all directions and is uncorrelated with each other. Independent component analysis is used to extract independent components to obtain the detection data of each bearing.

[0061] In this embodiment, after executing S4 signal decoupling and before executing S5 first model building, S41 first data processing is included: including S411 data fusion, S412 feature extraction, S413 feature fusion and S414 data splicing, such as Figure 3 shown.

[0062] S411 data fusion: perform time-series matching on the detection data of each bearing in the bearing group and belonging to the same data type of the same bearing, and record the matched data as the first data.

[0063] S412 Feature extraction: The detection data belonging to the same sensor group in the detection data of each bearing is taken as a group of sensor group data, and feature extraction is performed on the sensor group data of each bearing to obtain a number of sensor data features of each bearing.

[0064] S413 Feature fusion: feature fusion is performed on the sensor data features of each bearing to obtain the fusion feature of each bearing, which is recorded as the first feature. For example, the bearing detection data includes the detection data of sensor group 1 of bearing 1, the detection data of sensor group 1 of bearing 2, the detection data of sensor group 1 of bearing 3, and the detection data of sensor group 2 of bearing 1. The detection data of sensor group 1 of bearing 1 and the detection data of sensor group 2 of bearing 1 are used as the sensor group data of bearing 1. All the detection data in sensor group 1 of bearing 1 and all the detection data in sensor group 2 of bearing 1 are feature extracted respectively to obtain the data features of sensor group 1 and the data features of sensor group 2 of bearing 1. The data features of sensor group 1 and the data features of sensor group 2 are feature fused to obtain the first feature of bearing 1.

[0065] S414 data splicing: splicing the first data with the first feature, and using the spliced ​​data as new detection data for each bearing.

[0066] S5: First model construction: build a fault prediction model based on deep learning algorithm. This includes S51: model construction, S52: first sample training set construction and S53: first model training. Figure 4 shown.

[0067] S51 builds the model: The Transformer model is selected as the basic model of the fault prediction model, so a part of the Transformer encoder layers are used as shared layers of the fault prediction model, and the remaining Transformer encoder layers and the fully connected layers that perform specific tasks in the Transformer model are used as task-specific layers of the fault prediction model.

[0068] S52 constructs a first sample training set: obtains historical detection data and corresponding fault type labels of each bearing in the bearing group of the equipment, and constructs a first sample training set based on the acquired historical detection data and corresponding fault type labels. The fault type labels include fault type and fault severity.

[0069] S53 First model training: Input the first sample training set into the fault prediction model. The shared layer of the fault prediction model extracts the features of the historical detection data and the fault type labels corresponding to the historical detection data, which are recorded as shared features. Task-specific branches are created according to the fault type labels. The task-specific branches include fault type classification tasks and fault severity classification tasks. A fault prediction loss function is defined. The shared features are input into the task-specific branches. The tasks are processed. The fault prediction model is iteratively optimized by minimizing the fault prediction loss function. The optimized fault prediction model is used as the new fault prediction model.

[0070] The fault prediction loss function is: in, represents the fault prediction loss function, is the cross entropy loss for the fault type classification task, is the cross entropy loss for the fault severity classification task, and are weight coefficients, .

[0071] S6 Fault Prediction: Predict the fault type label of each bearing based on the detection data of each bearing.

[0072] S7 Second model construction: construct the fault propagation model of the bearing group. It includes S71 defining node features, S72 defining edge features, S73 constructing the adjacency matrix, S74 fault propagation model construction, S75 constructing the second sample training set, S76 defining the loss function and S77 second model training, such as Figure 5 shown.

[0073] S71 defines node features: each bearing is taken as a node, the detection data of each bearing is taken as the node initial feature, the fault type label of each bearing is taken as the node target feature, and the node feature matrix is ​​obtained according to the node initial features and node target features of all nodes.

[0074] S72 defines edge features: each bearing is connected by an edge, and a propagation probability and a propagation strength are defined for each edge as edge features.

[0075] Considering the coupling relationship between bearings, bearing workload and the influence of bearing fault types, a propagation probability formula is constructed.

[0076] The propagation probability formula is: .

[0077] in, Indicates bearing With bearings The probability of propagation between Indicates bearing With bearings The coupling weight between Indicates bearing With bearings The load weight between Indicates bearing With bearings The fault weight.

[0078] Based on the historical fault data of the equipment, the fault propagation between different couplings of different bearings is counted, and the coupling weight coefficient between bearings is set. Using the historical load data and fault data of the bearing, the impact of the bearing load on the fault propagation is determined through regression analysis, and the load weight coefficient between bearings is set. According to the use history or experimental data of the equipment passed by experts, the propagation speed and impact of different types of faults are evaluated based on experience, and the fault weight of the bearing is given.

[0079] S73 builds an adjacency matrix: According to the structure and working principle of the equipment, the connection relationship between the bearings in the bearing group is obtained, and the adjacency matrix is ​​built according to the connection relationship between the bearings and the propagation probability. With Node , then at the corresponding position of the adjacency matrix Assign a value of 1, otherwise 0. Since each bearing is used as a node, the bearing It's the node , bearings It's the node .

[0080] S74 fault propagation model construction: Based on the node feature matrix and adjacency matrix, a graph convolutional network is used to define an initial faulty node and build a fault propagation model.

[0081] In the fault propagation model, the graph convolution layer of each layer aggregates the node features with the neighbor node features through the graph convolution operation: .

[0082] in, Indicates The node feature matrix of the graph convolutional layer, represents an activation function, and the activation function may be an activation function such as a ReLU activation function and a Sigmoid activation function. In this embodiment, a ReLU activation function is used. represents the normalized adjacency matrix, Indicates The node feature matrix of the graph convolutional layer, Indicates Model weight matrix for graph convolutional layers.

[0083] S75 constructs a second sample training set: obtains historical detection data of each bearing in the bearing group of the equipment, a corresponding fault type label, and a corresponding historical fault propagation record, and constructs a second sample training set.

[0084] S76 defines a loss function: defines a fault propagation loss function as a cross entropy loss function.

[0085] The fault propagation loss function is: .

[0086] in, represents the fault propagation loss function, represents the number of samples in the second sample training set, Indicates The actual fault label of each bearing, No. The predicted failure probability of each bearing.

[0087] S77 Second model training: Input the second sample training set into the fault propagation model, perform model inference, calculate the output of the fault propagation model through forward propagation, compare the difference between the predicted result and the actual result, calculate the model loss, update the fault propagation model parameters through backward calculation, iteratively optimize the fault propagation model by minimizing the fault propagation loss function, and use the optimized fault propagation model as the new fault propagation model.

[0088] S8 fault propagation: According to the fault type label of each bearing, the faulty bearings are screened out to determine whether the number of faulty bearings exceeds the preset threshold.

[0089] If so, the device is judged to be faulty and the current time is output as the device failure time.

[0090] If not, the detected data of the selected bearings are input into the fault propagation model, the fault evolution result is output, and the first fault determination in S81 is performed.

[0091] S81 First fault judgment: Based on the obtained fault evolution results, obtain the predicted fault time of each bearing in the bearing group. From the current moment, sort the bearings in the order of predicted fault time from small to large to obtain the first sorting table. Accumulate the number of bearing faults in the order of the first sorting table. When the accumulated number of bearing faults exceeds the preset number threshold, record the predicted fault time of the last accumulated bearing as the first time, obtain the predicted fault time of the main bearing, record it as the second time, and determine whether the first time is greater than the second time.

[0092] If yes, the second time is used as the equipment failure time, and the first equipment life prediction is performed in S94.

[0093] If not, the first time is taken as the equipment failure time, and S95 second equipment life prediction is executed.

[0094] S9 Equipment Failure Prediction: Predict the remaining life of the equipment based on the failure evolution results. This includes S91 building a remaining life prediction model, S92 building a third sample training set, S93 third model training, and S94 equipment life prediction.

[0095] S91: construct a remaining life prediction model: use a recurrent neural network to construct a remaining life prediction model. In this embodiment, LSTM (Long Short-Term Memory Network) is used as a basic model to construct the remaining life prediction model.

[0096] S92 constructs a third sample training set: obtains the fault data of each bearing in the bearing group and the corresponding equipment status data, and constructs a third sample training set. The equipment status data includes the equipment operating time and the equipment physical status.

[0097] S93 Third model training: Input the third sample training set into the remaining life prediction model for model inference, define the remaining life loss function, iteratively optimize the remaining life prediction model by minimizing the remaining life loss function, and use the optimized remaining life prediction model as the new remaining life prediction model.

[0098] The remaining life loss function is: .

[0099] in, represents the remaining life loss function, represents the number of samples in the third sample training set, Indicates The actual lifespan of the samples, No. The predicted life span of samples.

[0100] S94 First equipment life prediction: According to the first ranking table, the bearing detection data before the main bearing and the main bearing detection data are input into the remaining life prediction model to obtain the remaining life of the equipment.

[0101] S95 Second Equipment Life Prediction: The bearing detection data accumulated in the first fault judgment in S81 is input into the remaining life prediction model to obtain the remaining life of the equipment.

[0102] In this embodiment, the collected detection data is signal decoupled by blind source separation technology to ensure that the fault detection data of each bearing is independent and accurate, thereby improving the accuracy of bearing fault detection in a multi-bearing system. In addition, by establishing a fault propagation model for a bearing group, it is possible to monitor the health status of multiple bearings in real time, predict the propagation process of faults between different bearings, identify potential equipment failures in advance, and achieve accurate fault monitoring and prediction of multi-bearing equipment.

[0103] Example 1: There are four interrelated bearings in an automobile transmission, including the main bearing A1, the secondary bearing A2, the auxiliary bearing A3 and the small bearing A4. Bearing A1 is a rolling bearing located at both ends of the transmission main shaft; the secondary bearing A2 is a deep groove ball bearing near the input shaft or output shaft of the transmission; the auxiliary bearing A3 is a needle bearing installed on the auxiliary drive shaft of the transmission; and the small bearing A4 is a sliding bearing installed in the gearbox in the transmission.

[0104] The fault detection of the above four bearings is as follows: 1. Sensor Layout 1. Select a vibration sensor, a temperature sensor, a magnetic sensor, a speed sensor, and a piezoelectric sensor as a sensor group.

[0105] Determine the working characteristics of each bearing: main bearing A1, heavy load, prone to wear and fatigue failure; secondary bearing A2, medium load, prone to overheating; auxiliary bearing A3, light load; small bearing A4, light load, easily affected by the environment.

[0106] Therefore, a sensor group is provided on the main bearing A1, a vibration sensor and a temperature sensor are provided on the secondary bearing A2, a vibration sensor is provided on the auxiliary bearing A3, and a vibration sensor and a temperature sensor are provided on the small bearing A4.

[0107] 2. Use genetic algorithm to optimize and obtain the sensor installation position.

[0108] In the initial stage, multiple candidate sensor placement schemes are generated, and each scheme can be represented by a binary string, for example, whether each sensor is installed at a different position or not.

[0109] Considering the minimum number of sensors, all failure modes of each bearing are adequately covered and the interference between sensors is below the threshold.

[0110] The fitness of each individual is calculated according to the fitness function formula, the selection probability is allocated according to the proportion of the fitness of all individuals, and the roulette wheel selection is performed based on the selection probability to select the parent individual.

[0111] Two parent individuals are randomly selected and crossover is performed to exchange their gene positions to generate new offspring. For example, some sensor positions of parent 1 are exchanged with other sensor positions of parent 2 to form a new sensor layout.

[0112] Randomly select individuals from the newly generated offspring and change the value of a gene (i.e., change the position or type of a sensor).

[0113] Update the population, retain the individuals with the best fitness, and generate a new generation of population.

[0114] After multiple rounds of iterations, the optimal sensor group layout solution is output, that is, the final number, location and type of sensors.

[0115] The sensor layout is as follows: The vibration sensor of bearing A1 is installed on the inner ring, the temperature sensor is installed on the outer ring, the piezoelectric sensor is installed on the outer ring, the speed sensor is installed near the transmission main shaft, and the magnetic sensor is installed on the bearing seat of bearing A1.

[0116] The vibration sensor of bearing A2 is installed on the outer ring, the temperature sensor is installed on the outer ring, and the speed sensor is installed near the transmission output shaft.

[0117] The vibration sensor of bearing A3 is installed on the outer ring.

[0118] The vibration sensor of bearing A4 is installed on the outer ring, and the temperature sensor is installed on the outer ring.

[0119] 3. Collect data according to the set sensors and perform signal decoupling on the collected data.

[0120]

[0121] Build a file containing The data matrix of sensors and L time points is recorded as a mixed signal matrix. , the number of time points 3. Build a signal matrix.

[0122] The signal matrix is ​​whitened and the whitened signal matrix is ​​used as an independent component analysis algorithm to decouple the whitened signal into independent detection data for each bearing.

[0123] The detection data of bearing A1, bearing A2, bearing A3 and bearing A4 are subjected to the steps of data fusion, feature extraction, feature fusion and data splicing to obtain the detection data of each bearing.

[0124] 4. Build a fault prediction model to predict whether the bearing is faulty based on the collected bearing inspection data.

[0125] The historical detection data of bearing A1, bearing A2, bearing A3 and bearing A4 and the corresponding fault type labels are collected as sample data for training the fault prediction model to obtain a trained fault prediction model.

[0126] The inspection data of bearings A1, A2, A3 and A4 are input into the trained fault prediction model, and the fault type labels of bearings A1, A2, A3 and A4 are output, among which bearing A1 has a slight wear fault, bearing A2 has a moderate wear fault, and bearings A3 and A4 have no fault.

[0127] 5. Screen out the faulty bearings and determine whether the car's transmission is faulty; Filter out the faulty bearings and determine whether the number of faulty bearings exceeds 3.

[0128] At this time, the faulty bearings are A1 and A2, and the number of bearing faults is less than 2. Therefore, a fault propagation model is constructed, and the collected historical detection data of bearings A1, A2, A3 and A4, the corresponding fault type labels and the corresponding historical fault propagation records are used as sample data to train the fault propagation model. The screened detection data of bearings A1 and A2 and the fault labels of the mild wear fault of bearing A1 and the moderate wear fault of bearing A2 are input into the fault propagation model, and the fault evolution results are output. The estimated time for bearing A1 to stop working is February 15, 2025, the estimated time for bearing A2 to stop working is March 1, 2025, the estimated time for bearing A3 to stop working is May 10, 2025, and the estimated time for bearing A4 to stop working is June 20, 2025.

[0129] 6. Sort the prediction results, obtain the corresponding sorting table, and determine the failure time of the automobile transmission.

[0130] The order of stopping working of bearing A1, bearing A2, bearing A3 and bearing A4 is bearing A1, bearing A2, bearing A3, bearing A4. The stopping time of main bearing A1 is shorter than that of other bearings, so the stopping time of main bearing A1 is taken as the failure time of automobile transmission. If main bearing A1 is not repaired in time, the automobile transmission will fail on February 15, 2025.

[0131] Embodiment 2: This embodiment discloses an equipment fault monitoring system based on bearing monitoring, and the monitoring system includes: The sensor layout module includes a sensor type selection unit and a sensor position determination unit.

[0132] The sensor type selection unit is used to determine the required sensor according to the input detection data type that needs to be collected by the bearing.

[0133] The sensor position determination unit is used to output the installation position of each sensor according to the working characteristics and failure modes of the bearing that need to be collected and the required sensors.

[0134] The data acquisition module is communicatively connected with the sensor layout module, and includes a vibration sensor, a temperature sensor, a magnetic sensor, a speed sensor and a piezoelectric sensor, and is used to collect the detection data of the bearing group of the equipment; the detection data includes vibration signals, temperature, lubricating oil metal powder, speed and load.

[0135] The signal decoupling module is used to perform signal decoupling on the collected detection data using blind source separation technology to obtain detection data of each bearing in the bearing group.

[0136] The first model building module is used to build a fault prediction model based on a deep learning algorithm.

[0137] The fault prediction module is used to predict the fault type label of each bearing based on the detection data of each bearing.

[0138] The second model building module is used to build a fault propagation model of the bearing group.

[0139] The fault propagation module is used to screen out the faulty bearings according to the fault type label of each bearing and determine whether the number of faulty bearings exceeds a preset number threshold.

[0140] If the number of faulty bearings exceeds the preset number threshold, the equipment is judged to be faulty and the current time is output as the equipment failure time.

[0141] If the number of faulty bearings does not exceed the preset threshold, the detection data of the screened bearings will be input into the fault propagation model, the fault evolution results will be output, and the equipment fault prediction module will be triggered.

[0142] The equipment failure prediction module is used to predict the remaining life of the equipment based on the failure evolution results.

[0143] In this embodiment, the collected detection data is signal decoupled by the blind source separation technology of the signal decoupling module to ensure that the fault detection data of each bearing is independent and accurate, thereby improving the accuracy of bearing fault detection in a multi-bearing system. In addition, by establishing a fault propagation model of the bearing group through the second model building module, the health status of multiple bearings can be monitored in real time, the propagation process of faults between different bearings can be predicted, potential equipment failures can be identified in advance, and accurate fault monitoring and prediction of multi-bearing equipment can be achieved.

[0144] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for monitoring equipment failure based on bearing monitoring, characterized in that: include: Data collection: Collect the test data of the bearing group of the equipment; the test data includes vibration signal, temperature, lubricating oil metal powder, speed and load; Signal decoupling: Collect the test data of the bearing group of the equipment to obtain the test data of each bearing in the bearing group; First model construction: build a fault prediction model based on deep learning algorithm; Fault prediction: predict the fault type label of each bearing based on the detection data of each bearing; Second model construction: construct the fault propagation model of the bearing group; Fault propagation: According to the fault type label of each bearing, the faulty bearings are screened out to determine whether the number of faulty bearings exceeds the preset threshold: If yes, the device is judged to be faulty and the current time is output as the device fault time; If not, the detection data of the selected bearings are input into the fault propagation model, the fault evolution results are output, and the equipment fault prediction steps are executed; Equipment failure prediction: predict the remaining life of the equipment based on the failure evolution results.

2. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: Before executing the data collection step, it also includes: Sensor type selection: according to the type of bearing detection data to be collected, set the sensor that collects the corresponding data type, and regard a group of sensors corresponding to the detection data type as a sensor group; Define the optimization goal: minimize the number of sensor groups and maximize the fault detection coverage, and construct the objective function; Define constraints: Take the operating characteristics, failure modes, and interference between sensors of each bearing in the equipment's bearing group as constraints of the objective function; Sensor position optimization: According to the objective function and constraints, the optimization algorithm is used to obtain the number of sensor groups and the corresponding installation positions.

3. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: After the step of performing data acquisition and before the step of performing signal decoupling, the method further includes: Time domain analysis: The time domain analysis method is used to extract the time domain features in the detection data of the bearing group, and the extracted time domain features and the detection data of the bearing group are used as new detection data of the bearing group.

4. The equipment fault monitoring method based on bearing monitoring according to claim 2 is characterized in that: After executing the step of signal decoupling and before executing the step of building the first model, the method further includes: Data fusion: performing time series matching on the detection data of each bearing in the bearing group and the detection data of the same data type belonging to the same bearing, and recording the matched data as the first data; Feature extraction: The detection data belonging to the same sensor group in the detection data of each bearing is taken as a group of sensor group data, and feature extraction is performed on the sensor group data of each bearing to obtain several sensor data features of each bearing; Feature fusion: The sensor data features of each bearing are fused to obtain the fusion feature of each bearing, which is recorded as the first feature; Data splicing: Splice the first data with the first feature, and use the spliced ​​data as new detection data for each bearing.

5. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: After executing the step of building the first model and before executing the step of fault prediction, the method further includes: Constructing a first sample training set: acquiring historical detection data and corresponding fault type labels of each bearing in a bearing group of the equipment, and constructing a first sample training set based on the acquired historical detection data and corresponding fault type labels; the fault type labels include fault type and fault severity; First model training: input the first sample training set into the fault prediction model. The shared layer of the fault prediction model extracts the features of the historical detection data and the fault type labels corresponding to the historical detection data, which are recorded as shared features. Task-specific branches are created according to the fault type labels. The task-specific branches include fault type classification tasks and fault severity classification tasks. The fault prediction loss function is defined, and the shared features are input into the task-specific branches. The tasks are processed, and the fault prediction model is iteratively optimized by minimizing the fault prediction loss function. The optimized fault prediction model is used as the new fault prediction model.

6. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: After executing the step of fault prediction and before executing the step of building the second model, the method further includes: Define node features: take each bearing as a node, the detection data of each bearing as the initial feature of the node, and the fault type label of each bearing as the target feature of the node; Define edge features: Each bearing is connected by an edge, and define the propagation probability and propagation strength as edge features for each edge; Constructing an adjacency matrix: According to the structure and working principle of the equipment, the connection relationship between the bearings in the bearing group is obtained, and the adjacency matrix is ​​constructed according to the connection relationship between the bearings.

7. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: After executing the step of building the second model and before executing the step of fault propagation, the method further includes: Constructing a second sample training set: obtaining historical detection data of each bearing in the bearing group of the equipment, the corresponding fault type label, and the corresponding historical fault propagation record, and constructing a second sample training set; Define loss function: Define fault propagation loss function; Second model training: Input the second sample training set into the fault propagation model to perform model inference. Calculate the output of the fault propagation model through forward propagation, compare the difference between the predicted result and the actual result, calculate the model loss, update the fault propagation model parameters through backward calculation, iteratively optimize the fault propagation model by minimizing the fault propagation loss function, and use the optimized fault propagation model as the new fault propagation model.

8. The equipment fault monitoring method based on bearing monitoring according to claim 1 is characterized in that: After the step of performing fault propagation and before the step of performing equipment fault prediction, the method further includes: First fault identification: Based on the obtained fault evolution results, the predicted fault time of each bearing in the bearing group is obtained. From the current moment, each bearing is sorted in the order of predicted fault time from small to large, and the number of bearing faults is accumulated in order. When the accumulated number of bearing faults exceeds the preset number threshold, the predicted fault time of the last accumulated bearing is recorded as the first time, and the predicted fault time of the main bearing is obtained and recorded as the second time. It is determined whether the first time is greater than the second time: If yes, the second time is used as the equipment failure time, and the equipment failure prediction step is performed; If not, the first time is taken as the equipment failure time, and the equipment failure prediction step is performed.

9. An equipment fault monitoring system based on bearing monitoring, the system is applicable to the method according to any one of claims 1 to 8, characterized in that: include: The data acquisition module is used to collect the detection data of the bearing group of the equipment; the detection data includes vibration signal, temperature, lubricating oil metal powder, speed and load; A signal decoupling module is used to perform signal decoupling on the collected detection data using a blind source separation technique to obtain detection data of each bearing in the bearing group; A first model building module, used to build a fault prediction model based on a deep learning algorithm; A fault prediction module is used to predict the fault type label of each bearing based on the detection data of each bearing; A second model building module is used to build a fault propagation model of the bearing group; The fault propagation module is used to filter out the faulty bearings according to the fault type label of each bearing and determine whether the number of faulty bearings exceeds the preset number threshold: If the number of failed bearings exceeds the preset threshold, the equipment is judged to be faulty and the current time is output as the equipment fault time; If the number of faulty bearings does not exceed the preset number threshold, the detection data of the selected bearings will be input into the fault propagation model, the fault evolution results will be output, and the equipment fault prediction module will be triggered; The equipment failure prediction module is used to predict the remaining life of the equipment based on the failure evolution results.

Images