Training method and device, detection method and device, equipment and medium

Abstract

The invention discloses a training method, a detection method, devices, equipment and a medium. The training method comprises the following steps: constructing a vibration simulation physical model, an initial fault detection model and a fault vibration source group set; the vibration simulation physical model comprises a vibration sensor array and a plurality of vibration sources distributed around the vibration sensor array, the number of vibration sensors in the vibration sensor array is less than the number of the vibration sources, and the initial fault detection model comprises at least one sub-model; each fault vibration source group in the fault vibration source group set comprises at least one vibration source; simulating fault vibration wave data in the vibration simulation physical model by using the fault vibration source group set based on preset fault vibration simulation parameters; constructing a fault detection training set according to vibration simulation sensing data corresponding to the simulation fault vibration wave data received by the vibration simulation physical model simulation vibration sensor array; and training the initial fault detection model according to the fault detection training set to obtain a fault detection model.

Description

Technical Field

[0001] This application relates to the field of industrial testing technology, and in particular to a training method, a testing method, an apparatus, equipment, and a medium. Background Technology

[0002] In modern industrial manufacturing, PVD vacuum coating equipment and CNC machining workshops are typical complex systems with multiple vibration sources. PVD equipment typically contains moving parts such as a rotating base, target rotation mechanism, and high-power vacuum pump, while CNC workshops involve various power mechanisms including spindle drive, turret conversion, and multi-axis feed systems. These vibration sources may simultaneously fail during continuous operation, and the resulting vibration signals often superimpose and couple, forming a complex composite vibration response. Related technologies can place a vibration sensor at each vibration source, collecting data from each sensor for fault diagnosis. However, this approach requires repeated sensor disassembly and reassembly, or the deployment of a large number of sensors, making the operation complex. Furthermore, it cannot handle multiple superimposed faults, and the amplitude of a single signal is easily affected by the environment, resulting in significant noise interference. It is also difficult to obtain a large amount of data on different fault scenarios in a short time, hindering the application of fault data. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application provides a training method, detection method, apparatus, electronic device, and computer-readable storage medium for a fault detection model.

[0004] This application provides a method for training a fault detection model, including: A vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups are constructed. The vibration simulation physical model includes a vibration sensor array and multiple vibration sources. The multiple vibration sources are distributed around the vibration sensor array. The number of vibration sensors in the vibration sensor array is less than the number of vibration sources. The initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model. Each fault vibration source group in the set of fault vibration source groups includes one or more of the aforementioned vibration sources. Based on preset fault vibration simulation parameters, fault vibration wave data are simulated in the vibration simulation physical model using the set of fault vibration sources. In the vibration simulation physical model, vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the vibration sensor array are simulated to construct a fault detection training set. The initial fault detection model is trained based on the fault detection training set to obtain the fault detection model.

[0005] This application also provides a fault detection model detection method, which is applied to a multi-vibration source device. The multi-vibration source device includes a machine vibration sensor array and multiple machine vibration sources. The multiple machine vibration sources are distributed around the machine vibration sensor array. The number of machine vibration sensors in the machine vibration sensor array is less than the number of machine vibration sources. Acquire the vibration detection data received by the machine tool vibration sensor array; Based on the vibration detection data, the fault detection model obtained using the training method described above is used to acquire the fault location data, fault category data, and / or fault intensity data of the machine vibration source.

[0006] Based on the same inventive concept described above, this application also provides a training device, comprising: The first construction module is used to construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups; wherein, the vibration simulation physical model includes a vibration sensor array and multiple vibration sources; the multiple vibration sources are distributed around the vibration sensor array; the number of vibration sensors in the vibration sensor array is less than the number of vibration sources; the initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model; each fault vibration source group in the set of fault vibration source groups includes one or more of the vibration sources; The simulation module is used to simulate fault vibration wave data in the vibration simulation physical model based on preset fault vibration simulation parameters and using the set of fault vibration sources. The second construction module is used to simulate the vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the vibration sensor array in the vibration simulation physical model, and construct a fault detection training set. The training module is used to train the initial fault detection model based on the fault detection training set and obtain the fault detection model.

[0007] Based on the same inventive concept described above, this application also provides an electronic device, including a processor, a memory, and a computer program, wherein the computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the training method or the detection method.

[0008] Based on the same inventive concept described above, this application also provides a non-volatile computer-readable storage medium containing a computer program that, when executed by a processor, causes the processor to perform the training method or the detection method described in any one of the above embodiments.

[0009] In the training method of this application, a vibration simulation physical model, an initial fault detection model, and a set of fault vibration sources are constructed. Simulation data is generated based on the set of fault vibration sources and the vibration simulation physical model. The initial fault detection model is then trained based on the simulation data to obtain a fault detection model capable of identifying multiple faults. This effectively overcomes the limitations of related technologies that rely on single-point sensor deployment and are difficult to handle concurrent multi-source faults. By using fewer sensors than the number of vibration sources, the complexity of on-site deployment is significantly simplified, reducing reliance on repeated disassembly and assembly or large-scale hardware installation. It can also effectively decouple the vibration characteristics of each fault source from mixed vibration signals. In addition, the simulation generates rich and diverse fault scenario data, which makes up for the scarcity of fault samples and the inability to cover multiple fault combinations in actual working conditions. This enhances the accuracy of the fault detection model in the face of noise and environmental interference, thereby achieving more accurate positioning, identification, and intensity assessment in the diagnosis of concurrent multi-vibration source faults, providing reliable technical support for fault detection in complex industrial scenarios.

[0010] Additional aspects and advantages of embodiments of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of embodiments of this application. Attached Figure Description

[0011] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the training method of some embodiments of this application; Figure 2 This is a schematic diagram of a training device according to certain embodiments of this application; Figure 3 This is a scene diagram of the vibration simulation physical model of some embodiments of this application; Figure 4-8 This is a flowchart illustrating the training method of some embodiments of this application; Figure 9 This is a flowchart illustrating the detection method of some embodiments of this application; Figure 10 This is a schematic diagram of a training device according to certain embodiments of this application.

[0012] Main component reference numerals: Training device 10, first construction module 110, simulation module 120, second construction module 130, training module 140, generation module 150, training module 160, detection device 20, acquisition module 210, and detection module 220. Detailed Implementation

[0013] The embodiments of the present application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present application, and should not be construed as limiting the implementation of the present application.

[0014] Please see Figure 1 This application provides a training method for a fault detection model, the training method including the following steps: 01. Construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups; wherein, the vibration simulation physical model includes a vibration sensor array and multiple vibration sources; multiple vibration sources are distributed around the vibration sensor array; the number of vibration sensors in the vibration sensor array is less than the number of vibration sources; the initial fault detection model includes one or more of the following: a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model; each fault vibration source group in the set of fault vibration source groups includes one or more vibration sources; 02. Based on the preset fault vibration simulation parameters, the fault vibration wave data is simulated in the vibration simulation physical model using the set of fault vibration sources. 03. In the vibration simulation physical model, the vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the simulated vibration sensor array is used to construct a fault detection training set. 04. Based on the fault detection training set, train the initial fault detection model and obtain the fault detection model.

[0015] Please see Figure 2 This application provides a training device 10 for a fault detection model. The training device 10 includes a first construction module 110, a simulation module 120, a second construction module 130, and a training module 140. Step 01 can be implemented by the first construction module 110, step 02 by the simulation module 120, step 03 by the second construction module 130, and step 04 by the training module 140.

[0016] Alternatively, the first construction module 110 can be used to construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups; wherein, the vibration simulation physical model includes a vibration sensor array and multiple vibration sources; the multiple vibration sources are distributed around the vibration sensor array; the number of vibration sensors in the vibration sensor array is less than the number of vibration sources; the initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model; each fault vibration source group in the set of fault vibration source groups includes one or more vibration sources; the simulation module 120 can be used to simulate fault vibration wave data in the vibration simulation physical model based on preset fault vibration simulation parameters and using the set of fault vibration source groups; the second construction module 130 can be used to simulate the vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the vibration sensor array in the vibration simulation physical model, and construct a fault detection training set; the training module 140 can be used to train the initial fault detection model according to the fault detection training set and obtain the fault detection model.

[0017] This application also provides an electronic device, including a processor, a memory, and a computer program, wherein the computer program is stored in the memory, and when the computer program is executed by the processor, the processor performs the above-described training method. Specifically, the processor is used to construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups; wherein the vibration simulation physical model includes a vibration sensor array and multiple vibration sources; the multiple vibration sources are distributed around the vibration sensor array; the number of vibration sensors in the vibration sensor array is less than the number of vibration sources; the initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model; each fault vibration source group in the fault vibration source set includes one or more vibration sources, and based on preset fault vibration simulation parameters, fault vibration wave data is simulated in the vibration simulation physical model using the fault vibration source set; in the vibration simulation physical model, vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the simulated vibration sensor array is simulated to construct a fault detection training set; based on the fault detection training set, the initial fault detection model is trained to obtain the fault detection model.

[0018] The training method, training device, and electronic equipment for the fault detection model in this application construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration sources. Vibration simulation sensing data is obtained based on the set of fault vibration sources and the vibration simulation physical model. The initial fault detection model is then trained based on the vibration simulation sensing data, resulting in a fault detection model capable of recognizing multiple faults. This effectively overcomes the limitations of related technologies that rely on single-point sensor deployment and struggle to handle concurrent multi-source faults. By utilizing fewer sensors than the number of vibration sources, the complexity of on-site deployment is significantly simplified, reducing reliance on repeated disassembly and reassembly or extensive hardware installation. Furthermore, the simulation generates rich and diverse fault scenario data. On one hand, this compensates for the scarcity of fault samples and the difficulty in covering multiple fault combinations in actual working conditions, enhancing the accuracy of the fault detection model. This allows the trained fault detection model to achieve more precise location, identification, and intensity assessment in the diagnosis of concurrent multi-vibration source faults. On the other hand, it avoids actual on-site data collection, significantly reducing data acquisition costs.

[0019] In some implementations, the training device 10 may be part of an electronic device. Alternatively, the electronic device may include the training device 10. As hardware, the training device 10 may be a standalone component or added to the electronic device as an additional peripheral element. The training device 10 may also be integrated into the electronic device; for example, when the training device 10 is part of the electronic device, it may be integrated into the processor.

[0020] It is worth noting that the fault detection model is used to detect vibration anomalies in multi-vibration-source equipment. The content of vibration anomaly detection can include the location, type, and intensity of the faulty vibration source in the multi-vibration-source equipment. Multi-vibration-source equipment refers to a device that integrates two or more independent vibration generating units (i.e., vibration sources), and multiple vibration sources may simultaneously fail during operation, generating superimposed vibration signals. Multi-source vibration equipment can be, but is not limited to, PVD equipment, CNC machine tools, etc.

[0021] The fault detection model can be trained from the initial fault detection model (a blank model that has not been trained with data and only has the basic algorithm framework). The initial fault detection model can contain one or more sub-models. The sub-models belong to the machine learning models in artificial intelligence models. The sub-models can include, but are not limited to, one or more of the following: fault vibration source localization sub-model, fault type identification sub-model, and fault intensity detection sub-model. That is to say, the fault detection model can be trained from one or more of the following sub-models: fault vibration source localization sub-model, fault type identification sub-model, and fault intensity detection sub-model.

[0022] For example, the initial fault detection model may only include the fault vibration source location sub-model. Alternatively, the initial fault detection model may only include the fault vibration source location sub-model and the fault type identification sub-model. Yet another example is that the initial fault detection model may include the fault vibration source location sub-model, the fault type identification sub-model, and the fault intensity detection sub-model.

[0023] The fault vibration source localization sub-model is used to locate the position of the fault vibration source (e.g., its spatial coordinates). This sub-model can use either a Convolutional Neural Network (CNN) or a Graph Neural Network (GNN) depending on the arrangement of the vibration sensor array. A CNN is used when the vibration sensor array is regular, while a GNN is used when it is irregular. The fault type identification sub-model is used to identify the fault type of the vibration source. This sub-model can employ classification models such as Gradient Boosting Tree (GBDT) or Random Forest. The fault intensity detection sub-model is used to identify the fault intensity of the vibration source.

[0024] A vibration simulation physical model is a digital model built upon physical laws, material properties, and system operating mechanisms to simulate the behavior and characteristics of a real physical system with multiple vibration sources. The vibration simulation physical model may include multiple vibration sources and a vibration sensor array. The multiple vibration sources are distributed around the vibration sensor array, which may include multiple vibration sensors, arranged either in a regular or irregular array; and the number of vibration sensors is less than the number of vibration sources. For example, such as... Figure 3 As shown, the number of vibration sources can be 8, and the number of vibration sensors in the vibration sensor array is 3.

[0025] Vibration simulation physical models can be constructed in a virtual environment. For example, based on digital twin technology, a geometric model can be constructed using CAD / 3D modeling software, and material parameters, boundary conditions, and vibration sensor characteristics can be input into a simulation platform (such as ANSYS, COMSOL, or MATLAB / Simulink) to form a computable vibration simulation physical model.

[0026] A fault vibration source set refers to a collective classification carrier used to coordinate all fault vibration sources in a vibration simulation physical model. A fault vibration source set can include multiple fault vibration source groups, each representing a fault scenario. The composition of each fault vibration source group satisfies the basic rule of "single vibration source forming an independent group or multiple vibration sources forming a coupled group." For independent faults in the vibration simulation physical model, where the fault is caused by an abnormal operating condition of a single vibration source, this vibration source with independent fault-causing attributes can be separately classified as a fault vibration source group. For coupled faults in the vibration simulation physical model, where the fault is jointly induced by the coordinated abnormal operating conditions of multiple vibration sources, all vibration sources with fault correlation need to be integrated into the same fault vibration source group. It can be understood that by dividing fault vibration source groups, both accurate localization of independent faults of a single vibration source and complete coverage of complex fault scenarios under the coupled action of multiple vibration sources can be achieved, ensuring the comprehensiveness and accuracy of fault diagnosis.

[0027] For example, the vibration simulation physical model includes three vibration sources A, B, and C. The set of fault vibration source combinations includes a total of 7 fault vibration source groups (representing 7 fault scenarios). Among them, a fault vibration source group is formed when only vibration source A is faulty, a fault vibration source group is formed when only vibration source B is faulty, a fault vibration source group is formed when only vibration source C is faulty, a fault vibration source group is formed when only vibration source A and vibration source B are faulty, a fault vibration source group is formed when only vibration source A and vibration source C are faulty, a fault vibration source group is formed when only vibration source B and vibration source C are faulty, and a fault vibration source group is formed when vibration source A, vibration source B, and vibration source C are all faulty.

[0028] Preset fault vibration simulation parameters refer to a set of quantitative simulation indicators defined in advance based on the actual fault type, characteristics, and operating conditions of the vibration source. Preset fault vibration simulation parameters include, but are not limited to, fault characteristic parameters of the vibration source, basic physical parameters of the vibration wave (such as vibration frequency range, amplitude threshold, and phase difference range), and simulation environment parameters (such as load intensity and temperature field distribution).

[0029] For each group of fault vibration sources, multiple sets of fault vibration simulation parameters can be set (each set of fault vibration simulation parameters corresponds to a certain operating condition), so that simulation can be carried out in the vibration simulation physical model to obtain multiple fault vibration wave data of each group of fault vibration sources under different operating conditions.

[0030] Vibration simulation sensing data refers to data generated by a vibration sensor array that monitors simulated fault vibration waves excited by a group of faulty vibration sources in real time. Vibration sensors can acquire data synchronously in time, ensuring that monitoring data from vibration sensors at different locations can be correlated and analyzed based on the time dimension.

[0031] After obtaining the vibration simulation sensing data, it can be cleaned, labeled, feature extracted, and formatted to form a fault detection training set that meets the requirements for training machine learning models. For example, labels such as fault type, fault severity (mild / moderate / severe), and operating conditions (no load / full load) can be added to the vibration simulation sensing data. At the same time, operations such as invalid data removal, feature dimension filtering, and dataset partitioning are performed to finally produce a standardized dataset that can be directly used for model training.

[0032] In step 04, the vibration simulation sensing data in the fault detection training set can be divided into an input training set and a validation set. The input training set is fed into the initial fault vibration detection model for training. The initial fault vibration detection model will repeatedly learn the mapping relationship between feature data and fault labels in the input training set through a preset learning algorithm. For example, the weight parameters of the neural network are adjusted through backpropagation, and the weak learner combination method of the ensemble model is optimized through gradient boosting to gradually reduce the error between the prediction result and the true label. When the prediction error of the initial fault vibration detection model on the input training set drops to a preset threshold and the generalization performance on the validation set meets the standard, the training process terminates, thereby obtaining the fault detection model.

[0033] In addition, during model training, data augmentation strategies such as noise perturbation and amplitude scaling can be introduced to improve robustness to real working conditions. Furthermore, cross-validation can be used to evaluate positioning error and classification accuracy, ensuring controllable accuracy under multi-source vibration conditions.

[0034] Please see Figure 4 In some implementations, step 02 includes the following sub-steps: 021. Based on the preset fault vibration simulation parameters, generate fault vibration wave data using the vibration simulation physical model; wherein, the preset fault vibration simulation parameters include one or more of the following: fault location, vibration amplitude, vibration frequency, vibration phase, and vibration propagation attenuation coefficient of the vibration source in the fault vibration source group.

[0035] In some implementations, sub-step 021 can be implemented by simulation module 120, or simulation module 120 can be used to generate fault vibration wave data using a vibration simulation physical model based on preset fault vibration simulation parameters; wherein, the preset fault vibration simulation parameters include one or more of the following: fault location of vibration source in fault vibration source group, vibration amplitude, vibration frequency, vibration phase and vibration propagation attenuation coefficient.

[0036] In some implementations, the processor can be used to generate fault vibration wave data using a vibration simulation physical model based on preset fault vibration simulation parameters; wherein the preset fault vibration simulation parameters include one or more of the following: fault location of the vibration source in the fault vibration source group, vibration amplitude, vibration frequency, vibration phase, and vibration propagation attenuation coefficient.

[0037] Specifically, based on preset fault vibration simulation parameters, the corresponding fault vibration source group can be matched from the set of fault vibration source groups. If it is an independent fault, a fault vibration source group composed of a single vibration source is called; if it is a coupled fault, a fault vibration source group composed of multiple vibration sources coupled together is called. This clarifies the vibration source objects and combination methods that need to be subjected to fault excitation in the simulation, avoiding omissions or mismatches of simulation objects. Then, the correlation between the preset fault vibration simulation parameters and the fault vibration source group is substituted into the vibration simulation physical model. Through the physical law calculation of the preset fault vibration simulation parameters, the vibration transmission process of the vibration source under the fault state is simulated, and finally the fault vibration wave data corresponding to each fault vibration source group is obtained.

[0038] Thus, by introducing preset fault vibration simulation parameters that include multiple key physical propagation parameters such as fault location, vibration amplitude, frequency, phase, and propagation attenuation coefficient, the vibration simulation physical model is driven to generate fault vibration wave data. On the one hand, at the simulation level, fault vibration simulation in real industrial scenarios is realized, thereby generating simulation data with broad coverage and clear physical meaning, effectively overcoming the problem of scarce real fault samples and difficulty in covering all fault scenarios.

[0039] Please see Figure 5 In some implementations, step 03 includes the following sub-steps: 031. Based on the vibration wave propagation path, vibration time delay data and / or vibration energy attenuation data from the vibration source to the target vibration sensor in the vibration sensor array, the target vibration simulation sensing data of the target vibration sensor is obtained by simulating the fault vibration wave data. 032, Perform data processing on the target vibration simulation sensing data to obtain target fault characteristic data related to the faulty components of the faulty vibration source group in the target vibration simulation sensing data; 033. Based on the target fault characteristic data, obtain the fault characteristic ratio data corresponding to the vibration sensor array; 034. Construct a fault detection training set based on the proportion of fault characteristics.

[0040] In some implementations, sub-steps 031-034 can be implemented by the second building module 130. In other words, the second building module 130 can be used to simulate and obtain target vibration simulation sensing data of the target vibration sensor based on the vibration wave propagation path, vibration time delay data, and / or vibration energy attenuation data from the vibration source to the target vibration sensor in the vibration sensor array, using fault vibration wave data; perform data processing on the target vibration simulation sensing data to obtain target fault feature data related to the faulty components of the faulty vibration source group in the target vibration simulation sensing data; obtain fault feature ratio data corresponding to the vibration sensor array based on the target fault feature data; and construct a fault detection training set based on the fault feature ratio data.

[0041] In some implementations, the processor can be used to simulate and obtain target vibration simulation sensing data of the target vibration sensor based on the vibration wave propagation path from the vibration source to the target vibration sensor in the vibration sensor array, vibration time delay data, and / or vibration energy attenuation data, using fault vibration wave data; perform data processing on the target vibration simulation sensing data to obtain target fault feature data related to the faulty components of the faulty vibration source group in the target vibration simulation sensing data; obtain fault feature ratio data corresponding to the vibration sensor array based on the target fault feature data; and construct a fault detection training set based on the fault feature ratio data.

[0042] Specifically, finite element analysis or multibody dynamics can be used to simulate the propagation process of fault vibration wave data from the fault vibration source to the target vibration sensor in the vibration sensor array, based on the vibration wave propagation path, vibration time delay data, and / or vibration energy attenuation data. This results in simulated target vibration sensing data that closely matches the real-world scenario. Understandably, the propagation path determines the transmission medium of the vibration wave and the signal superposition logic, the time delay reflects the timing difference of the vibration from the fault source to the sensor, and the energy attenuation data corrects for the amplitude loss during propagation. Therefore, this avoids the data obtained by the sensor from being out of sync with the actual operating conditions due to neglecting vibration propagation characteristics, laying a data foundation for subsequent feature extraction.

[0043] Data processing may include, but is not limited to, denoising, time-domain kurtosis calculation, frequency-domain feature extraction, and time-frequency domain analysis, thereby decomposing the target vibration simulation sensing data into signals to extract target fault feature data related to the faulty components of the faulty vibration source group from the target vibration simulation sensing data.

[0044] After obtaining the target fault feature data of each vibration sensor in the vibration sensor array, feature vectors can be constructed from all the target fault feature data in the vibration sensor array to obtain the fault feature ratio data of the fault vibration source group. The fault feature ratio data eliminates the influence of overall amplitude changes and environmental load fluctuations, while retaining the spatial distribution information of the array, ensuring that the model input features are stable and consistent with the actual collected features.

[0045] For example, a vibration sensor array includes vibration sensor 1, vibration sensor 2, and vibration sensor 3, where the target fault characteristic data of vibration sensor 1 is... The target fault characteristic data of vibration sensor 2 is The target fault characteristic data of vibration sensor 3 are Then the fault characteristic ratio data is .

[0046] Finally, the fault proportion features are combined with at least one of the following labels: fault location (e.g., spatial coordinates), fault type, and fault intensity, to form a training sample, ultimately creating a high-precision fault detection training set.

[0047] In this way, by simulating the physical propagation process of vibration waves from the fault source to the sensor, the original fault vibration wave data is transformed into simulated sensing data that closely resembles the real measurement scenario. Then, features related to the faulty component are extracted and their distribution ratio in the sensor array is calculated. Training samples that can characterize the spatial distribution and coupling relationship of the fault are constructed, which enhances the robustness of the subsequent fault detection model to interference factors such as background noise, signal attenuation and fluctuations in the environment.

[0048] Please see Figure 6 In some implementations, after step 031, the training method further includes: 05. Obtain Gaussian noise data and / or background vibration data; 06. Update the target vibration simulation sensing data based on Gaussian noise data and / or background vibration data to make the target vibration simulation sensing data closer to the actual measurement value.

[0049] In some implementations, steps 05 and 06 can be implemented by the second building module 130. In other words, the second building module 130 can be used to acquire Gaussian noise data and / or background vibration data, and update the target vibration simulation sensing data based on the Gaussian noise data and / or background vibration data, so that the target vibration simulation sensing data is close to the actual measurement value.

[0050] In some implementations, the processor can be used to acquire Gaussian noise data and / or background vibration data, and update the target vibration simulation sensing data based on the Gaussian noise data and / or background vibration data, so that the target vibration simulation sensing data closely approximates the actual measured value.

[0051] It is worth noting that Gaussian noise data can be generated based on the inherent noise characteristics of sensors in multi-source vibration equipment. The amplitude distribution of Gaussian noise data follows a normal distribution law, which can accurately simulate the random errors caused by circuit thermal noise and electromagnetic interference during the signal acquisition process of sensors. Background vibration data can be derived from the baseline vibration signal under normal operating conditions of multi-vibration source equipment. It can include fault-free interference components such as mechanical resonance, motor power frequency vibration, and environmental vibration during no-load operation of multi-vibration source equipment. It can be obtained by fault-free operating condition acquisition of multi-vibration source equipment or by fitting baseline data of similar equipment.

[0052] During the process of updating the target vibration simulation sensing data, Gaussian noise data and / or background vibration data can be superimposed on the target vibration simulation sensing data according to the interference intensity of the actual monitoring scenario and a preset weight. For example, for the simulation data corresponding to high-precision sensors, only low-amplitude Gaussian noise is superimposed; while for the simulation data of industrial strong interference scenarios, medium-amplitude Gaussian noise and equipment baseline background vibration are superimposed at the same time, so that the updated sensing data retains the core characteristics of the fault and has the noise basis of the actual monitoring data.

[0053] In this way, by injecting Gaussian noise and / or background vibration data into the target vibration sensing data, the data distribution difference between the ideal simulation environment and the complex real industrial scenario is effectively bridged, thereby enhancing the robustness of the subsequent model to the complex acoustic and vibration background on site, and ensuring that it can still maintain high accuracy and high reliability in fault diagnosis when facing noisy measurement signals during real deployment.

[0054] Please see Figure 7 In some implementations, step 032 includes: 0321, Preprocess the target vibration simulation sensing data to obtain target preprocessed data; 0322, Perform signal decomposition on the target preprocessed data to obtain the set of intrinsic mode functions corresponding to the target preprocessed data; 0323, perform Hilbert transform on each intrinsic mode function in the intrinsic mode function set to obtain the target fault frequency domain data related to the faulty component of the fault vibration source group; 0324. Based on the target fault frequency domain data and the resonant frequency data of the faulty components of the fault vibration source group, obtain the target fault characteristic data.

[0055] In some implementations, sub-steps 0321-0324 can be implemented by the second building module 130. In other words, the second building module 130 can be used to preprocess the target vibration simulation sensing data to obtain target preprocessed data; perform signal decomposition on the target preprocessed data to obtain the intrinsic mode function set corresponding to the target preprocessed data; perform Hilbert transform on each intrinsic mode function in the intrinsic mode function set to obtain target fault frequency domain data related to the faulty components of the faulty vibration source group; and obtain target fault characteristic data based on the target fault frequency domain data and the resonant frequency data of the faulty components of the faulty vibration source group.

[0056] In some implementations, the processor can be used to preprocess the target vibration simulation sensing data to obtain target preprocessed data; perform signal decomposition on the target preprocessed data to obtain the intrinsic mode function set corresponding to the target preprocessed data; perform Hilbert transform on each intrinsic mode function in the intrinsic mode function set to obtain target fault frequency domain data related to the faulty component of the faulty vibration source group; and obtain target fault characteristic data based on the target fault frequency domain data and the resonant frequency data of the faulty component of the faulty vibration source group.

[0057] Preprocessing can include DC removal and normalization. This involves performing DC removal and normalization on the target vibration simulation sensing data to obtain preprocessed target data. DC removal refers to identifying and eliminating the DC component using a preset algorithm. This process ensures the target vibration simulation sensing data fluctuates around zero mean, eliminating baseline offset interference for subsequent signal analysis (such as frequency domain filtering and feature extraction). Normalization maps the original data (such as vibration signal amplitude, resonant frequency intensity characteristics, etc.) to a fixed numerical range (e.g., [0,1] or [-1,1]) according to preset rules. This eliminates the magnitude differences between different dimensions of data, giving each data dimension equal weight, thus completing the standardized preprocessing and ultimately obtaining time-continuous, interference-free target preprocessed data.

[0058] After obtaining the target preprocessed data, the Hilbert-Huang Transform (HHT) can be used to process the target preprocessed data to obtain the time spectrum. Then, the target fault frequency domain data related to the faulty components of the faulty vibration source group can be extracted from the time spectrum. The Hilbert-Huang Transform method first decomposes the original data into intrinsic mode functions (IMFs) through Empirical Mode Decomposition (EMD), and then performs Hilbert Transform (HT) on each IMF component to obtain the time spectrum.

[0059] Specifically, the target preprocessed data is first decomposed into several Intrinsic Mode Functions (IMFs) through Empirical Mode Decomposition (EMD), thus obtaining the set of Intrinsic Mode Functions (IMFs); the decomposition process is as follows: (1) Traverse the target preprocessed data s （t） Mark all local maxima and minima, and fit the upper envelope using cubic spline interpolation. and lower envelope ; (2) Calculate the mean of the envelope. and preprocess the target data s(t) Subtracting the mean from the mean yields the preliminary components. ; (3) Repeat steps (1)-(2) for... The process is iterated until the iteratively processed component satisfies the two criteria of the IMF (the number of extreme points equals or the difference between the number of zero-crossing points is ≤1; the mean of the upper and lower envelopes is 0 at any time). At this point, the first IMF component is obtained. That is, the intrinsic mode functions of this application.

[0060] (4) Preprocessing data from the target s(t) The first IMF component was separated from it. The residual signal was obtained. Then The above process is repeated as new input to obtain the 2nd to nth IMF components in sequence. , , ... ; (5). When the residual signal When the function is monotonic or constant, the decomposition stops, and the final target preprocessed data satisfies the following decomposition relationship:

[0061] in, Let i be the i-th IMF component (intrinsic mode function). This is the final residual component.

[0062] (6) Extract all intrinsic mode functions and construct the intrinsic mode function set IMFs, i.e., IMFs = .

[0063] A Hilbert transform is performed on each intrinsic mode function (IMF) in the intrinsic mode function set (IMFs) to achieve the feature conversion from the time domain to the instantaneous frequency domain. This allows the acquisition of target fault frequency domain data related to the faulty components of the faulty vibration source group. The processing procedure is as follows: (1) For the i-th eigenmode function Perform Hilbert transform Defined as:

[0064] (2) Combining the original eigenmode function and the transformation result, construct the analytic signal:

[0065] Among them, amplitude The amplitude of the response signal at time t; phase The instantaneous frequency can be obtained by differentiating the phase.

[0066] By integrating the instantaneous frequencies and amplitudes of all IMF components, the Hilbert spectrum can be obtained. H(w,t), Further integration over time yields the marginal spectrum. This enables a complete characterization of the time-frequency distribution features.

[0067] (3) Target fault frequency domain data related to the faulty components of the faulty vibration source group are selected from the time-frequency distribution characteristics to obtain the target fault frequency domain data. .

[0068] Thus, by preprocessing, performing intrinsic mode decomposition, and Hilbert transform on the target vibration simulation sensing data, the complex coupled vibration signal is converted into a set of time-frequency components with clear physical meaning. This allows for the accurate capture of the target fault frequency domain data caused by the faulty component. Finally, by matching and filtering the extracted fault frequency domain data with the known component resonant frequencies, background vibration and propagation path interference can be effectively filtered out, and core feature vectors that are highly sensitive to fault type and location and have low dependence on absolute signal strength can be extracted.

[0069] Please see Figure 8 In some implementations, sub-step 0324 includes: 03241, Based on the target fault frequency domain data, obtain the fault vibration intensity data of the faulty component; 03242, Based on the resonant frequency data of the faulty component, obtain the fundamental frequency data of the faulty component's vibration; 03243. Based on the fault vibration intensity data and vibration fundamental frequency data, obtain the target fault characteristic data.

[0070] In some implementations, sub-steps 03241-03243 can be implemented by the second building module 130. In other words, the second building module 130 can be used to obtain the fault vibration intensity data of the faulty component based on the target fault frequency domain data; obtain the vibration fundamental frequency data of the faulty component based on the resonant frequency data of the faulty component; and obtain the target fault characteristic data based on the fault vibration intensity data and the vibration fundamental frequency data.

[0071] In some implementations, the processor can be used to obtain fault vibration intensity data of the faulty component based on the target fault frequency domain data; obtain vibration fundamental frequency data of the faulty component based on the resonant frequency data of the faulty component; and obtain target fault characteristic data based on the fault vibration intensity data and the vibration fundamental frequency data.

[0072] For example, the target fault frequency domain data can be the resonant frequency signal of the faulty component (such as a bearing), and the fault vibration intensity data of the faulty component can be the root mean square amplitude (RMS amplitude). Then, the expression for calculating the fault vibration intensity number can be:

[0073] in, The final fault vibration intensity data is obtained, where T is the time length (the duration of the target fault frequency domain data). For the target fault frequency domain data, The energy integral of the target fault frequency domain data over the time region (reflecting the total energy of the target fault frequency domain data).

[0074] The calculation expression for the target fault characteristic data can be:

[0075] in, For target fault characteristic data, For fault vibration intensity data, This refers to the fundamental frequency data of the faulty component.

[0076] In this way, by combining the fault vibration intensity data with the vibration fundamental frequency data derived from the resonant frequency to construct the target fault feature data, not only is intensity information that can characterize the severity of the fault extracted, but also fundamental frequency information that reflects the inherent physical characteristics of the faulty component is deeply integrated, eliminating the influence of different calibration values ​​between different sensors, and further improving the robustness and stability of the extracted features.

[0077] Please see Figure 9This application also provides a detection method implemented by the above-described fault detection model. The detection method is applied to a multi-vibration-source device, which includes a machine vibration sensor array and multiple machine vibration sources. The multiple machine vibration sources are distributed around the machine vibration sensor array. The number of machine vibration sensors in the machine vibration sensor array is less than the number of machine vibration sources. The detection method includes: Step 201: Obtain vibration detection data received by the machine tool vibration sensor array; Step 202: Based on the vibration detection data, use the fault detection model to obtain the fault location data, fault category data, and / or fault intensity data of the machine vibration source.

[0078] Please see Figure 10 This application also provides a fault detection model detection device 20, which includes an acquisition module 210 and a detection module 220. Step 201 can be implemented by the acquisition module 210, and step 202 can be implemented by the detection module 220. Alternatively, the acquisition module 210 is used to acquire vibration detection data received by the machine tool vibration sensor array; the detection module 220 is used to acquire, based on the vibration detection data and using the fault detection model, fault location data, fault category data, and / or fault intensity data of the machine tool vibration source.

[0079] This application also provides an electronic device, including a processor, a memory, and a computer program. The computer program is stored in the memory, and when executed by the processor, it causes the processor to perform the aforementioned detection method. Specifically, the processor acquires vibration detection data received by the machine vibration sensor array, and based on the vibration detection data, uses a fault detection model to acquire fault location data, fault category data, and / or fault intensity data of the machine vibration source.

[0080] In the detection method, detection device, and electronic equipment of this application, by deploying a machine vibration sensor array with fewer sensors than the vibration source, and using a fault detection model obtained through simulation training to process the vibration data collected by the array, it is possible to monitor and diagnose equipment with multiple vibration sources, effectively overcoming the problems of dense sensor deployment and high cost in related technologies; at the same time, the fault detection model can simultaneously analyze the precise location, specific type, and severity of the fault source from the mixed vibration signals received by the array, significantly improving the diagnostic capability and efficiency for multiple fault scenarios.

[0081] It is worth noting that the arrangement of the machine tool vibration sensor array and multiple machine tool vibration sources in the multi-vibration source equipment is the same as the arrangement of the vibration sensor array and multiple vibration sources in the vibration simulation physical model mentioned above. The machine tool vibration sensor array can be a triaxial MEMS vibration sensor array (such as ADXL355, LIS3DH).

[0082] The machine tool vibration sensor array can collect raw vibration detection data during the operation of equipment with multiple vibration sources. The raw vibration detection data is then processed through rectification, amplification, analog filtering, etc., to obtain digital vibration detection data.

[0083] After obtaining the vibration detection data, a real-time fault feature ratio vector corresponding to the machine vibration sensor array can be constructed based on the vibration detection data. Then, the real-time fault feature ratio vector is input into the fault detection model, and the fault detection model outputs the fault location data, fault category data, and / or fault intensity data of the machine vibration source.

[0084] The construction method of the fault feature proportion vector is consistent with the construction method of the fault feature proportion data in the training phase (i.e., preprocessing, decomposition, Hilbert transform, and feature extraction are performed sequentially to obtain the target fault frequency domain data, and then the target fault frequency domain data is used to construct the fault feature proportion data), which will not be elaborated here. Understandably, because the construction method of the fault feature proportion vector is consistent with that in the training phase, the fault detection model can achieve seamless transfer between virtual and real data.

[0085] This application also provides a non-volatile computer-readable storage medium containing a computer program, which, when executed by a processor, causes the processor to perform any of the training methods or detection methods described in the above embodiments.

[0086] In the readable storage medium of this application's embodiments, a vibration simulation physical model, an initial fault detection model, and a set of fault vibration sources are constructed. Vibration simulation sensing data is obtained based on the set of fault vibration sources and the vibration simulation physical model. The initial fault detection model is then trained based on the vibration simulation sensing data, resulting in a fault detection model capable of recognizing multiple faults. This effectively overcomes the limitations of related technologies that rely on single-point sensor deployment and struggle to handle concurrent multi-source faults. By utilizing fewer sensors than the number of vibration sources, the complexity of on-site deployment is significantly simplified, reducing reliance on repeated disassembly and assembly or extensive hardware installation. Furthermore, the vibration characteristics of each fault source can be effectively decoupled from mixed vibration signals. In addition, the simulation generates rich and diverse fault scenario data, compensating for the scarcity of fault samples and the inability to cover multiple fault combinations in actual working conditions. This enhances the accuracy of the fault detection model. Thus, the trained fault detection model can achieve more accurate location, identification, and intensity assessment in the diagnosis of concurrent multi-vibration source faults, providing reliable technical support for complex industrial scenarios.

[0087] In some embodiments, the non-volatile computer-readable storage medium may be a storage medium built into an electronic device, such as a memory, or a storage medium that can be plugged into an electronic device, such as an SD card.

[0088] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any other combination. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0090] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A training method for a fault detection model, characterized in that, The training method includes: A vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups are constructed. The vibration simulation physical model includes a vibration sensor array and multiple vibration sources. The multiple vibration sources are distributed around the vibration sensor array. The number of vibration sensors in the vibration sensor array is less than the number of vibration sources. The initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model. Each fault vibration source group in the set of fault vibration source groups includes one or more of the aforementioned vibration sources. Based on preset fault vibration simulation parameters, fault vibration wave data are simulated in the vibration simulation physical model using the set of fault vibration sources. In the vibration simulation physical model, the vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the vibration sensor array is simulated to construct a fault detection training set. The initial fault detection model is trained based on the fault detection training set to obtain the fault detection model.

2. The training method according to claim 1, characterized in that, The process of simulating fault vibration wave data in the vibration simulation physical model based on preset fault vibration simulation parameters and utilizing the set of fault vibration sources includes: Based on preset fault vibration simulation parameters, the fault vibration wave data is generated using the vibration simulation physical model; wherein, the preset fault vibration simulation parameters include one or more of the following: fault location, vibration amplitude, vibration frequency, vibration phase, and vibration propagation attenuation coefficient of the vibration source in the fault vibration source group.

3. The training method according to claim 1, characterized in that, In the vibration simulation physical model, the vibration simulation sensor data corresponding to the simulated fault vibration wave data received by the vibration sensor array is used to construct a fault detection training set, including: Based on the vibration wave propagation path from the vibration source to the target vibration sensor in the vibration sensor array, vibration time delay data, and / or vibration energy attenuation data, the target vibration simulation sensing data of the target vibration sensor is obtained by simulating the fault vibration wave data. Data processing is performed on the target vibration simulation sensing data to obtain target fault feature data related to the faulty components of the faulty vibration source group in the target vibration simulation sensing data; Based on the target fault characteristic data, obtain the fault characteristic ratio data corresponding to the vibration sensor array; The fault detection training set is constructed based on the fault feature ratio data.

4. The training method according to claim 3, characterized in that, After using the fault vibration wave data to simulate and obtain the target vibration simulation sensing data of the target vibration sensor, the training method further includes: Acquire Gaussian noise data and / or background vibration data; Update the target vibration simulation sensing data based on the Gaussian noise data and / or the background vibration data to make the target vibration simulation sensing data closer to the actual measurement value.

5. The training method according to claim 3, characterized in that, The step of processing the target vibration simulation sensing data to obtain target fault feature data related to the faulty components of the faulty vibration source group includes: The target vibration simulation sensing data is preprocessed to obtain target preprocessed data; The target preprocessed data is decomposed to obtain the set of intrinsic mode functions corresponding to the target preprocessed data; Perform a Hilbert transform on each intrinsic mode function in the intrinsic mode function set to obtain the target fault frequency domain data related to the faulty component of the fault vibration source group; Based on the target fault frequency domain data and the resonant frequency data of the faulty components of the fault vibration source group, the target fault characteristic data are obtained.

6. The training method according to claim 5, characterized in that, The step of obtaining the target fault characteristic data based on the target fault frequency domain data and the resonant frequency data of the faulty components of the fault vibration source group includes: Based on the target fault frequency domain data, obtain the fault vibration intensity data of the faulty component; Based on the resonant frequency data of the faulty component, obtain the fundamental frequency data of the faulty component. Based on the fault vibration intensity data and the vibration fundamental frequency data, the target fault characteristic data are obtained.

7. A detection method for a fault detection model, characterized in that, The detection method is applied to a multi-vibration-source device; the multi-vibration-source device includes a machine vibration sensor array and multiple machine vibration sources; the multiple machine vibration sources are distributed around the machine vibration sensor array; the number of machine vibration sensors in the machine vibration sensor array is less than the number of machine vibration sources; the detection method includes: Acquire the vibration detection data received by the machine tool vibration sensor array; Based on the vibration detection data, the fault detection model obtained by the training method as described in any one of claims 1 to 6 is used to obtain the fault location data, fault category data, and / or fault intensity data of the machine vibration source.

8. A training device for a fault detection model, characterized in that, include: The first construction module is used to construct a vibration simulation physical model, an initial fault detection model, and a set of fault vibration source groups; wherein, the vibration simulation physical model includes a vibration sensor array and multiple vibration sources; the multiple vibration sources are distributed around the vibration sensor array; the number of vibration sensors in the vibration sensor array is less than the number of vibration sources; the initial fault detection model includes one or more of a fault vibration source location sub-model, a fault type identification sub-model, and a fault intensity detection sub-model; each fault vibration source group in the set of fault vibration source groups includes one or more of the vibration sources; The simulation module is used to simulate fault vibration wave data in the vibration simulation physical model based on preset fault vibration simulation parameters and using the set of fault vibration sources. The second construction module is used to simulate the vibration simulation sensing data corresponding to the simulated fault vibration wave data received by the vibration sensor array in the vibration simulation physical model, and construct a fault detection training set. The training module is used to train the initial fault detection model based on the fault detection training set and obtain the fault detection model.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program, wherein the computer program is stored in the memory, and when the computer program is executed by the processor, the processor performs the training method of any one of claims 1-6 or the detection method of claim 7.

10. A non-volatile computer-readable storage medium, characterized in that, The system includes a computer program that, when executed by a processor, causes the processor to perform the training method of any one of claims 1-6 or the detection method of claim 7.

Images