Fault diagnosis method, device and equipment of micro-impact test bench and storage medium

Abstract

The application discloses a kind of micro-impact test bench fault diagnosis method, device, equipment and storage medium, belong to equipment state monitoring technical field;Method includes obtaining the multi-source monitoring data of micro-impact test bench;Extract the time-domain statistical characteristics and frequency-domain statistical characteristics of each kind of monitoring data, and the fusion of each statistical characteristics extracted, form the feature vector representing the health state of micro-impact test bench;The feature vector is input into the diagnosis model trained in advance, to obtain the fault information of micro-impact test bench output by the diagnosis model. Through the multi-source monitoring data such as electrical data, mechanical component mechanical response data and motion data when impact occurs of micro-impact test bench, based on the information complementarity between the multi-source monitoring data, more comprehensive master equipment operating state, solve the problem that existing micro-impact test bench fault diagnosis accuracy is low.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, specifically to a fault diagnosis method, apparatus, equipment, and storage medium for a miniature impact test bench. Background Technology

[0002] A miniature impact test bench is a dynamic mechanical testing device used to simulate the impact environment that a product suffers during transportation and use, in order to test the product's impact resistance and dynamic mechanical properties. It is widely used in aerospace, precision electronics, automotive parts and other fields.

[0003] The stability of the working state of the miniature impact test bench directly determines the accuracy of the test data. In order to improve the test accuracy of the miniature impact test bench, it is necessary to diagnose its health status, that is, whether there is a fault.

[0004] Currently, the fault diagnosis methods for micro impact test benches typically include: acquiring the operating data of the micro impact test bench, comparing the operating data with a preset threshold, and identifying a fault when the operating data exceeds the threshold (or is less than the threshold, which may be an upper or lower limit depending on the operating data).

[0005] Because the micro impact test bench can experience significant fluctuations in its operating data under different working conditions (such as different impact forces and frequencies), even if the micro impact test bench is not in a faulty state, the above-mentioned method of using thresholds for judgment has the problem of misjudgment. Summary of the Invention

[0006] The purpose of this invention is to provide a fault diagnosis method, device, equipment, and storage medium for a miniature impact test bench. By using multi-source monitoring data such as electrical data, mechanical response data of mechanical components, and motion data during impact, and based on the information complementarity between these multi-source monitoring data, the operating status of the equipment can be more comprehensively understood, thus solving the problem of low fault diagnosis accuracy of existing miniature impact test benches.

[0007] This invention is achieved through the following technical solution:

[0008] The first aspect of this application provides a fault diagnosis method for a miniature impact testing bench, including:

[0009] Acquire multi-source monitoring data from a micro impact test bench; the multi-source monitoring data includes electrical data of the micro impact test bench, mechanical response data of mechanical components, and motion data during impact.

[0010] The time-domain and frequency-domain statistical features of various monitoring data are extracted and then fused to form a feature vector characterizing the health status of the micro impact test bench.

[0011] The feature vector is input into a pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model.

[0012] The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

[0013] In one feasible implementation, the method further includes: training the diagnostic model using incremental data to update the diagnostic model; specifically including:

[0014] The multi-source monitoring data acquired at the current sampling time is taken as the first sample, and the multi-source monitoring data acquired at the previous sampling time is taken as the second sample. The first similarity and the second similarity between the first sample and the second sample are calculated respectively.

[0015] The sample bias is determined based on the values ​​of the first similarity and the second similarity.

[0016] Based on the sample bias and the preset bias, the confidence level of the multi-source monitoring data acquired at the current sampling time is evaluated, and when the confidence level exceeds the preset confidence threshold, the first sample is stored in the safe sample library; the second sample is stored in the safe sample library.

[0017] When the diagnostic model needs to be updated, samples are selected from the security sample library to perform incremental training on the diagnostic model.

[0018] In one feasible implementation, the source domain model is a fault diagnosis model based on depthwise separable convolution: the construction of the depthwise separable convolution includes:

[0019] Based on the input signal channel, calculate the product of the input signal at the corresponding position in the convolution kernel and the weight at that position, and sum the products corresponding to each position in the convolution kernel as the intermediate feature of the channel.

[0020] By performing pointwise convolution, the weighted product of the intermediate feature and the corresponding channel's pointwise convolution kernel is calculated. This product is then summed over all channels of the input signal to obtain the output feature.

[0021] In one feasible implementation, the extraction of time-domain and frequency-domain statistical features of various monitoring data includes:

[0022] The time-domain statistical characteristics of the monitoring data include the peak value representing the impact intensity, the root mean square value representing the impact energy, and the kurtosis representing the steepness of the signal distribution shape.

[0023] The frequency domain statistical characteristics of the monitoring data include the centroid frequency of the location where the spectral energy is concentrated, and the mean square frequency that characterizes the distribution range of frequency components.

[0024] In one feasible implementation, the step of fusing the extracted statistical features to form a feature vector characterizing the health status of the micro-impact testing bench includes:

[0025] Based on the extracted statistical features, an original feature matrix is ​​formed;

[0026] Principal component analysis is performed on the original feature matrix to determine the eigenvectors characterizing the micro-impact test bench.

[0027] In one feasible implementation, the diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench, and fine-tuning the parameters of some convolutional layers of the source domain model, specifically including:

[0028] The source domain model is trained using sample data from a micro-impact test bench, and the parameters of the basic feature extraction layer in the pre-trained source domain model are retained during the training process, while the parameters of the classification layer are fine-tuned.

[0029] In one feasible implementation, the diagnostic model is deployed on the edge computing device of the micro-impact test bench.

[0030] A second aspect of this application provides a fault diagnosis device for a miniature impact testing bench, comprising:

[0031] The monitoring data acquisition unit is used to acquire multi-source monitoring data of the micro impact test bench; the multi-source monitoring data includes electrical data of the micro impact test bench, mechanical response data of mechanical components, and motion data when the impact occurs;

[0032] The feature vector forming unit is used to extract time-domain and frequency-domain statistical features of various monitoring data, and to fuse the extracted statistical features to form a feature vector characterizing the health status of the micro impact test bench.

[0033] The fault diagnosis unit is used to input the feature vector into a pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model.

[0034] The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

[0035] A third aspect of this application provides an electronic device, including: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the above-described method.

[0036] A fourth aspect of this application provides a storage medium, comprising: storing a program or instructions on the storage medium, wherein the program or instructions, when executed by a processor, implement the steps of the above-described method.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] This application embodiment acquires multi-source monitoring data, such as electrical data, mechanical response data of mechanical components, and motion data during impact, from a micro impact test bench. This multi-source monitoring data is used for feature extraction to form a feature vector for input diagnostic model. Due to the complementary nature of the information among the multi-source monitoring data, the feature vector contains comprehensive information on the operating status of the micro impact test bench, thereby improving the fault identification accuracy of the diagnostic model. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0040] Figure 1 A flowchart illustrating a fault diagnosis method for a miniature impact testing bench provided in an embodiment of this application;

[0041] Figure 2 A flowchart illustrating a specific implementation of a fault diagnosis method for a miniature impact testing bench provided in this application embodiment;

[0042] Figure 3 A schematic diagram of the structure of a fault diagnosis device for a miniature impact testing bench provided in an embodiment of this application;

[0043] Figure 4This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for explanation only and are not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0045] As will be known to those skilled in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0046] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, product, or apparatus.

[0047] Example 1:

[0048] Embodiment 1 of this application provides a fault diagnosis method for a miniature impact test bench, which solves the problem of low accuracy in fault diagnosis of existing miniature impact test benches.

[0049] The subject executing this method can be any computing device capable of implementing the method, such as a server, mobile phone, personal computer, smart wearable device, smart robot, etc.

[0050] Furthermore, the embodiments of this application do not limit the execution order of different steps. When using the method provided in the embodiments of this application, the execution order of different steps can be adjusted according to actual needs.

[0051] For ease of description, the following uses a fault diagnosis device for a miniature impact test bench as the subject of this method to provide a detailed description of the method provided in this application embodiment.

[0052] like Figure 1 The diagram shown is a detailed implementation flowchart of a fault diagnosis method for a miniature impact testing platform provided in this application embodiment, including the following steps 11-13:

[0053] Step 11: Acquire multi-source monitoring data from the micro impact test bench.

[0054] The multi-source monitoring data includes electrical data of the micro-impact test bench, mechanical response data of mechanical components, and motion data at the time of impact.

[0055] A sensor network is deployed on a miniature impact test bench to acquire multi-source monitoring data.

[0056] The sensor network includes current sensors, high-frequency accelerometers, and high dynamic range accelerometers.

[0057] The current sensor is used to collect the three-phase current signal of the drive motor, and the acquired current signal can be used to analyze faults such as motor torque fluctuation and electrical imbalance.

[0058] High-frequency accelerometers are installed on key mechanical components such as bearing housings, guide rails, and impact heads to collect broadband vibration acceleration signals.

[0059] A high dynamic range accelerometer is mounted on the impact head to collect transient waveforms for each impact event.

[0060] In one feasible implementation, this embodiment includes preprocessing the acquired raw monitoring data signals, specifically including denoising the one-dimensional sequence signal by wavelet thresholding, filtering the vibration signal by downsampling, and reconstructing the signal by wavelet transform, thereby removing interference signals with a sampling confidence of 0.

[0061] Denoising a one-dimensional sequence signal using wavelet thresholding can be achieved by: using wavelet transform to decompose the signal into low-frequency approximate components (effective signal) and high-frequency detail components (noise), and then using thresholding to suppress high-frequency noise while retaining the effective signal.

[0062] Downsampling can be used to filter vibration signals. For vibration signals from high-frequency accelerometers and high dynamic range accelerometers, the number of sampling points is reduced to decrease the amount of signal data, while suppressing high-frequency aliasing, thus achieving the purpose of dimensionality reduction and low-pass filtering.

[0063] Reconstructing a signal using wavelet transform can be achieved by: determining the confidence level based on the deviation between the signal amplitude and the historical normal signal amplitude; filtering valid signal components based on the confidence level and eliminating completely invalid interference signals (confidence level = 0).

[0064] Step 12: Extract the time-domain and frequency-domain statistical features of various monitoring data, and fuse the extracted statistical features to form a feature vector characterizing the health status of the micro impact test bench.

[0065] Time-domain statistical characteristics reflect the statistical characteristics of the signal amplitude of each monitoring data as it changes over time.

[0066] The time-domain statistical characteristics of the monitoring data include the peak value representing the impact intensity, the root mean square value representing the impact energy, and the kurtosis representing the steepness of the signal distribution shape.

[0067] peak Specifically, the maximum amplitude of the signal representing each monitoring data point of the micro-impact test bench, reflecting the impact strength, can be expressed as:

[0068] ;

[0069] In the formula, Indicates the first The signal value of each sampling point, also known as the monitoring data value.

[0070] Root mean square value Specifically, it characterizes the time-domain signals of various monitoring data from the micro-impact test bench. The average energy is sensitive to wear-related failures; it can be expressed as:

[0071] ;

[0072] Where N is the number of signal sampling points.

[0073] cliff Specifically, it characterizes the steepness of the sampling signal distribution shape of a miniature impact test bench and is extremely sensitive to impact failures; it can be expressed as:

[0074] ;

[0075] in, This represents the average value of the time-domain signal X.

[0076] Frequency domain statistical characteristics of the frequency components and energy distribution of the signal are obtained by performing a Fast Fourier Transform on the time domain signal. .

[0077] The frequency domain statistical characteristics of the monitoring data include the centroid frequency of the location where the spectral energy is concentrated, and the mean square frequency that characterizes the distribution range of frequency components.

[0078] The centroid frequency FC can be expressed as:

[0079] ;

[0080] In the formula, Indicates the first The frequencies corresponding to the spectral lines; Represents frequency domain signal The amplitude, K represents the total number of spectral lines.

[0081] Mean square frequency , can be represented as:

[0082] .

[0083] The extracted statistical features are fused to form a feature vector characterizing the health status of the micro-impact testing platform, including steps 1201-1202:

[0084] Step 1201: Based on the extracted statistical features, form the original feature matrix.

[0085] The rows in the original feature matrix can be the number of samplings (or the number of tests), that is, each row in the original feature matrix represents the feature corresponding to one sample of data; the columns of the original feature matrix can be the statistical features extracted based on one sample.

[0086] Step 1202: Perform principal component analysis on the original feature matrix to determine the eigenvectors characterizing the micro-impact test bench.

[0087] The features in the original feature matrix are standardized, specifically by unifying features with different units (dimensions) and numerical ranges to a zero mean and unit variance scale, so as to avoid the principal component analysis results being biased towards features with large numerical values ​​due to differences in dimensionality.

[0088] Based on the standardized original feature matrix, the covariance matrix is ​​calculated; each element in the covariance matrix represents the correlation between two features in the original feature matrix.

[0089] Based on the covariance matrix, a set of direction vectors orthogonal to the covariance matrix are found, and the variance characterization value is determined based on the variance of the element values ​​in the covariance matrix in the corresponding direction. Each vector in this set of direction vectors represents a principal component direction. The larger the variance characterization value, the more original information the corresponding principal component contains. The variance characterization values ​​are sorted in descending order.

[0090] Calculate the proportion of the variance representation value of a single principal component to the sum of all variance representation values ​​to determine the contribution of a single principal component; if the contribution of the top k principal components in the ranking exceeds a preset contribution threshold, retain the top k principal components.

[0091] The standardized original feature matrix is ​​mapped onto the retained first k principal component directions to obtain the fused feature vector representing the health status of the micro-impact test bench.

[0092] The normalized original feature matrix is ​​mapped to the retained first k principal component directions, which can be achieved by performing projection through matrix multiplication. That is, the normalized original feature matrix is ​​multiplied by the matrix formed by the first k principal component directions.

[0093] Step 13: Input the feature vector into the pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model.

[0094] The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

[0095] Training the source domain model specifically includes:

[0096] Acquire source domain data (training samples for the source domain model); source domain data consists of fault-related data from the selected impact testing machine or its simulation model. The source domain data must contain comprehensive labeled data, including normal data and fault data. The fault data must include at least the sensor monitoring data corresponding to faults such as bearing wear, misalignment, and rubbing.

[0097] Furthermore, the similarity between the physical mechanisms of the impact testing machine and the micro impact testing platform can be quantified by the structural and functional homology (core components and energy paths), dynamic characteristic matching degree, and failure mode correlation of the impact testing machine and the micro impact testing platform. Specifically, the above indicators can be weighted and summed to determine the value of similarity. When the physical mechanism similarity exceeds a preset threshold, the two are considered to have similar physical mechanisms, and the monitoring data of the impact testing machine can be used as training samples for the source domain model.

[0098] The source domain model is trained based on source domain data. The feature extractor in the source domain model is a fault diagnosis model based on depthwise separable convolution. By selecting depthwise convolution to build a lightweight backbone network, the computational cost is greatly reduced compared to traditional convolutional networks.

[0099] The feature extraction of traditional convolutional networks is as follows:

[0100] ;

[0101] In the formula:

[0102] Indicates the output feature map at position ,aisle The value;

[0103] This indicates that the input feature map is at the location ,aisle The value;

[0104] Indicates the position of the convolution kernel Input channel Output channel The weights;

[0105] This indicates the kernel size.

[0106] As can be seen from the above formula, the feature extraction process of traditional convolutional networks requires summation based on position and channel respectively.

[0107] The separable convolution in this embodiment decomposes the traditional convolution described above into two independent operations.

[0108] The first step is to establish a depthwise convolution: Based on the input signal channels, calculate the product of the input signal at the corresponding position within the convolution kernel and the weight at that position, and sum the products at each position within the convolution kernel to obtain the intermediate features for that channel. Specifically, this can be represented as:

[0109]

[0110] In the formula:

[0111] Indicates the intermediate feature map at position ,aisle The value;

[0112] Indicates the position of the depthwise convolution kernel ,aisle The weights;

[0113] Indicates the position of the convolution kernel Input channel Output channel The weights;

[0114] As can be seen from the formula, this embodiment only sums the positions.

[0115] Depthwise separable convolution performs a pointwise convolution step on the second step of standard convolution: Through pointwise convolution, the weights of the intermediate features and the corresponding channel's pointwise convolution kernel are calculated. This product is then summed over all channels of the input signal to obtain the output features. Specifically, this can be represented as:

[0116] ;

[0117] In the formula:

[0118] Identify the pointwise convolution kernel in the input channel Output channel The weight.

[0119] In this embodiment, the input signal length is determined by the actual sampling, and the input channel... =1, the network structure is a 5-layer depthwise separable convolution.

[0120] Using the aforementioned depthwise separable convolution as a feature extractor, feature vectors are extracted from the source domain data, ultimately yielding the output of the source domain model. Based on the value of the loss function determined by the deviation between the output and the sample labels, the model parameters are adjusted. When the parameters converge or the set number of training iterations is reached, the trained source domain model is determined.

[0121] In order to obtain a diagnostic model capable of diagnosing faults in a micro-impact test bench, the pre-trained source domain model (except for the final classification layer) was transferred to the diagnostic task of the micro-impact test bench.

[0122] The transfer process includes: training the source domain model using sample data from a micro-impact test bench, retaining the parameters of the basic feature extraction layer in the pre-trained source domain model during training, and fine-tuning the parameters of the classification layer.

[0123] During transfer learning, the first three layers of the pre-trained source domain model are frozen, and only the parameters of the last two layers are fine-tuned. The three retained layers are general feature extraction layers that can extract basic features such as edges and gradients. The two fine-tuned classification layers can adapt to the number of fault categories on the micro-impact test bench. By using a small learning rate, only the parameters of the higher-level network and the new classification layer are updated to avoid overfitting.

[0124] When using a trained diagnostic model for fault diagnosis of a micro-impact test bench, mechanical fatigue and wear of the equipment inevitably lead to significant differences between the latest sampling and the samples used in transfer training, resulting in model diagnostic drift. To address this issue, this embodiment combines incremental learning and meta-learning concepts, enabling the model to have online learning capabilities.

[0125] The diagnostic model is trained using incremental data to update the diagnostic model; specifically, steps 1401-1404 are included:

[0126] Step 1401: Take the multi-source monitoring data obtained at the current sampling time as the first sample and the multi-source monitoring data obtained at the previous sampling time as the second sample, and calculate the first similarity and the second similarity between the first sample and the second sample respectively.

[0127] The first similarity can be represented by the Euclidean distance between the first and second samples, and the second similarity can be represented by the Manhattan distance between the first and second samples.

[0128] Step 1402: Determine the sample bias based on the value of the first similarity and the value of the second similarity.

[0129] Sample bias It can be expressed as a weighted sum of the first similarity and the second similarity, specifically:

[0130] ;in, , Indicates the weight.

[0131] Step 1403: Based on the sample bias and the preset bias, assess the confidence level of the multi-source monitoring data acquired at the current sampling time, and when the confidence level exceeds the preset confidence level threshold, store the first sample in the safe sample library; the second sample is stored in the safe sample library.

[0132] Specifically, if the sample deviation is not greater than the preset deviation (i.e., the deviation between the current sample and the historical normal sample is within the allowable range, and the data is reliable), then the confidence level is higher than the set confidence level threshold (e.g., confidence level ≥ preset confidence level threshold 0.8), and the current sample is stored in the safe sample library; if the sample deviation is greater than the preset deviation (the sample deviation is too large, which may be a fault signal or abnormal interference), then the confidence level is lower than the threshold, and the sample is removed.

[0133] In one feasible implementation, to improve the accuracy of the calculated sample bias, multiple historical samples stored in the safe sample library prior to the current sampling time can be selected, and the mean of the sample biases determined based on each historical sample can be used as the sample bias to be compared with the preset bias.

[0134] Step 1404: When the diagnostic model needs to be updated, select samples from the security sample library and perform incremental training on the diagnostic model.

[0135] The update requirement for the diagnostic model can be a preset interval, such as during equipment operation, where 20% of new samples are extracted from the safety sample library every hour and mixed with historical samples to perform incremental learning training on the diagnostic model, thereby constraining the update of important model parameters and avoiding model drift and inaccuracy.

[0136] In addition, the diagnostic model can be updated based on the current diagnostic results. If the diagnostic model frequently diagnoses faults during normal equipment operation, incremental training of the diagnostic model is required.

[0137] The fault information output by the diagnostic model characterizes the health status of the miniature impact test bench. When the model diagnoses a specific fault mode in the miniature impact bench, it locates the most likely faulty component by analyzing the characteristic signals collected by different sensors and the contribution of these characteristic signals to the model's decision-making. When an abnormality is diagnosed or the health score falls below a threshold, an alarm is automatically triggered, and maintenance suggestions are generated to guide maintenance personnel in making precise interventions.

[0138] Fault modes include torque fluctuation, electrical imbalance, wear, looseness, misalignment, and rubbing. Different fault modes often correspond to different sensitive sensors. For example, current sensors are often sensitive to torque fluctuation faults and electrical imbalance faults; high-frequency acceleration sensors are often sensitive to misalignment and rubbing faults.

[0139] Based on the fault modes output by the diagnostic model, the characteristic signals of the sensitive sensors under the fault modes are analyzed, as well as the contribution of the characteristic signals to the model decision. When the contribution exceeds the threshold, the component corresponding to the signal collected by the sensitive sensor is located as the most likely faulty component.

[0140] In addition, the diagnostic model in this embodiment is deployed on the edge computing device of the micro impact test bench, which meets the millisecond-level response requirements of online real-time monitoring and provides technical support for realizing true predictive maintenance.

[0141] like Figure 2 The diagram illustrates a specific implementation flow of this embodiment, including an offline preparation phase, an online operation phase, and a maintenance phase. The offline preparation phase includes the collection of multi-source monitoring data, feature extraction and fusion based on the collected data, and initialization of the diagnostic model through transfer learning. The online operation phase includes: based on the trained diagnostic model, inputting real-time data streams, using a lightweight model to diagnose the health status of the micro-impact test bench in real time, and evaluating the confidence level of the output health status; screening samples from the real-time collected monitoring data, storing the selected samples in a safety sample set for incremental learning by the meta-learning optimizer, where hyperparameters are optimized to improve the accuracy of the diagnostic model. The maintenance phase includes diagnosing, locating, and issuing early warnings for the fault models output by the diagnostic model, used for implementing maintenance of the micro-impact test bench.

[0142] This application embodiment acquires multi-source monitoring data, such as electrical data, mechanical response data of mechanical components, and motion data during impact, from a micro impact test bench. This multi-source monitoring data is used for feature extraction to form a feature vector for input diagnostic model. Due to the complementary nature of the information among the multi-source monitoring data, the feature vector contains comprehensive information on the operating status of the micro impact test bench, thereby improving the fault identification accuracy of the diagnostic model.

[0143] Example 2:

[0144] To address the problem of low fault diagnosis accuracy in existing micro-impact testing benches, and based on the same inventive concept as Embodiment 1, this application also provides a fault diagnosis device for a micro-impact testing bench.

[0145] A schematic diagram of the specific structure of the device is shown below. Figure 3 As shown, it includes the following functional units 31-33:

[0146] The monitoring data acquisition unit 31 is used to acquire multi-source monitoring data of the micro impact test bench; the multi-source monitoring data includes electrical data of the micro impact test bench, mechanical response data of mechanical components and motion data when the impact occurs.

[0147] The monitoring data acquisition unit is specifically used for:

[0148] The feature vector forming unit 32 is used to extract time-domain and frequency-domain statistical features of various monitoring data, and to fuse the extracted statistical features to form a feature vector characterizing the health status of the micro impact test bench.

[0149] The feature vector forming unit is also used to extract time-domain and frequency-domain statistical features of various monitoring data, specifically for:

[0150] The time-domain statistical characteristics of the monitoring data include the peak value representing the impact intensity, the root mean square value representing the impact energy, and the kurtosis representing the steepness of the signal distribution shape; the frequency-domain statistical characteristics of the monitoring data include the centroid frequency of the concentrated location of the spectral energy, and the mean square frequency representing the distribution range of the frequency components.

[0151] The feature vector forming unit is also used to form an original feature matrix based on the extracted statistical features; and to perform principal component analysis on the original feature matrix to determine the feature vectors characterizing the micro impact test bench.

[0152] The fault diagnosis unit 33 is used to input the feature vector into a pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model.

[0153] The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

[0154] The source domain model is a fault diagnosis model based on depthwise separable convolution: the fault diagnosis unit is also used to construct the depthwise separable convolution, specifically including:

[0155] Based on the input signal channels, the product of the input signal at the corresponding position in the convolution kernel and the weight at that position is calculated, and the products corresponding to each position in the convolution kernel are summed to obtain the intermediate feature of that channel; through pointwise convolution, the product of the intermediate feature and the weight of the pointwise convolution kernel of the corresponding channel is calculated, and the product of the intermediate feature and the weight of the pointwise convolution kernel of the corresponding channel is summed over all channels of the input signal to obtain the output feature.

[0156] The fault diagnosis model is also used to train the source domain model using sample data from a micro-impact test bench, and during the training process, the parameters of the basic feature extraction layer in the pre-trained source domain model are retained, and the parameters of the classification layer are fine-tuned.

[0157] The diagnostic device (including a diagnostic model) in this embodiment is deployed on the edge computing device of the micro impact test bench.

[0158] The diagnostic device in this embodiment further includes a model update unit, used to train the diagnostic model using incremental data to update the diagnostic model; specifically used for:

[0159] The multi-source monitoring data acquired at the current sampling time is used as the first sample, and the multi-source monitoring data acquired at the previous sampling time is used as the second sample. The first similarity and the second similarity between the first sample and the second sample are calculated respectively. The sample bias is determined based on the value of the first similarity and the value of the second similarity. Based on the sample bias and the preset bias, the confidence level of the multi-source monitoring data acquired at the current sampling time is evaluated. When the confidence level exceeds the preset confidence level threshold, the first sample is stored in the safe sample library. The second sample is stored in the safe sample library. When the diagnostic model has an update requirement, samples are selected from the safe sample library to perform incremental training on the diagnostic model.

[0160] This application embodiment acquires multi-source monitoring data, such as electrical data, mechanical response data of mechanical components, and motion data during impact, from a micro impact test bench. This multi-source monitoring data is used for feature extraction to form a feature vector for input diagnostic model. Due to the complementary nature of the information among the multi-source monitoring data, the feature vector contains comprehensive information on the operating status of the micro impact test bench, thereby improving the fault identification accuracy of the diagnostic model.

[0161] Based on the same inventive concept as the foregoing embodiments of this application, this application also provides a computing device.

[0162] like Figure 4As shown, the computing device includes a memory 41 and a processor 42. The memory 41 can be configured to store various other data to support operation on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device. The memory 41 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0163] The processor 42, coupled to the memory 41, is used to execute the computer program stored in the memory 41 to perform a fault diagnosis method for a miniature impact test bench as described in the foregoing embodiments.

[0164] When the processor 42 executes the computer program to perform a fault diagnosis method for a miniature impact test bench, it acquires multi-source monitoring data such as electrical data, mechanical response data of mechanical components, and motion data during impact of the miniature impact test bench. This multi-source monitoring data is used for feature extraction to form a feature vector for input diagnostic model. Due to the complementary nature of the information among the multi-source monitoring data, the feature vector contains more comprehensive information on the operating status of the miniature impact test bench, thereby improving the fault identification accuracy of the diagnostic model.

[0165] When the processor 42 executes the computer program in the memory 41, in addition to the functions described above, it can also perform other functions, as detailed in the descriptions of the preceding embodiments.

[0166] Furthermore, such as Figure 4 As shown, the computing device also includes other components such as a display 44, a communication component 43, a power supply component 45, and an audio component 46. Figure 4 The diagram only shows some components and does not mean that the computing device includes only these components. Figure 4 The components shown.

[0167] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the methods provided in the above embodiments.

[0168] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0170] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for a miniature impact testing bench, characterized in that, include: Acquire multi-source monitoring data from a micro impact test bench; the multi-source monitoring data includes electrical data of the micro impact test bench, mechanical response data of mechanical components, and motion data during impact. The time-domain and frequency-domain statistical features of various monitoring data are extracted and then fused to form a feature vector characterizing the health status of the micro impact test bench. The feature vector is input into a pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model. The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

2. The method according to claim 1, characterized in that, The method further includes: training the diagnostic model using incremental data to update the diagnostic model; specifically including: The multi-source monitoring data acquired at the current sampling time is taken as the first sample, and the multi-source monitoring data acquired at the previous sampling time is taken as the second sample. The first similarity and the second similarity between the first sample and the second sample are calculated respectively. The sample bias is determined based on the values ​​of the first similarity and the second similarity. Based on the sample bias and the preset bias, the confidence level of the multi-source monitoring data acquired at the current sampling time is evaluated, and when the confidence level exceeds the preset confidence threshold, the first sample is stored in the safe sample library; the second sample is stored in the safe sample library. When the diagnostic model needs to be updated, samples are selected from the security sample library to perform incremental training on the diagnostic model.

3. The method according to claim 1, characterized in that, The source domain model is a fault diagnosis model based on depthwise separable convolution: the construction of the depthwise separable convolution includes: Based on the input signal channel, calculate the product of the input signal at the corresponding position in the convolution kernel and the weight at that position, and sum the products corresponding to each position in the convolution kernel as the intermediate feature of the channel. By performing pointwise convolution, the weighted product of the intermediate feature and the corresponding channel's pointwise convolution kernel is calculated. This product is then summed over all channels of the input signal to obtain the output feature.

4. The method according to claim 1, characterized in that, The extraction of time-domain and frequency-domain statistical features of various monitoring data includes: The time-domain statistical characteristics of the monitoring data include the peak value representing the impact intensity, the root mean square value representing the impact energy, and the kurtosis representing the steepness of the signal distribution shape. The frequency domain statistical characteristics of the monitoring data include the centroid frequency of the location where the spectral energy is concentrated, and the mean square frequency that characterizes the distribution range of frequency components.

5. The method according to claim 4, characterized in that, The process of fusing the extracted statistical features to form a feature vector characterizing the health status of the micro-impact testing bench includes: Based on the extracted statistical features, an original feature matrix is ​​formed; Principal component analysis is performed on the original feature matrix to determine the eigenvectors characterizing the micro-impact test bench.

6. The method according to claim 1, characterized in that, The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact testing platform and fine-tuning the parameters of some convolutional layers of the source domain model. Specifically, it includes: The source domain model is trained using sample data from a micro-impact test bench, and the parameters of the basic feature extraction layer in the pre-trained source domain model are retained during the training process, while the parameters of the classification layer are fine-tuned.

7. The method according to claim 1, characterized in that, The diagnostic model is deployed on the edge computing device of the micro-impact test bench.

8. A fault diagnosis device for a miniature impact testing bench, characterized in that, include: The monitoring data acquisition unit is used to acquire multi-source monitoring data of the micro impact test bench; the multi-source monitoring data includes electrical data of the micro impact test bench, mechanical response data of mechanical components, and motion data when the impact occurs; The feature vector forming unit is used to extract time-domain and frequency-domain statistical features of various monitoring data, and to fuse the extracted statistical features to form a feature vector characterizing the health status of the micro impact test bench. The fault diagnosis unit is used to input the feature vector into a pre-trained diagnostic model to obtain the fault information of the micro impact test bench output by the diagnostic model. The diagnostic model is determined by transferring a pre-trained source domain model to the diagnostic task of a micro-impact test bench and fine-tuning the parameters of some convolutional layers of the source domain model. The source domain model is a fault diagnosis model trained using fault-related data from an impact testing machine or its simulation model as training samples, and the similarity between the physical mechanisms of the impact testing machine and the micro-impact test bench exceeds a preset threshold.

9. An electronic device, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as claimed in any one of claims 1-7.

10. A storage medium, characterized in that, include: The storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1-7.

Images