Bolt fault diagnosis model construction method, bolt fault diagnosis method, bolt fault diagnosis system and bolt fault diagnosis equipment

Abstract

The invention belongs to the technical field of mechanical fault diagnosis, and relates to a bolt fault diagnosis model construction method, a bolt fault diagnosis method, a bolt fault diagnosis system and a bolt fault diagnosis device. According to the method, a vibration signal of a bolt operation state is obtained, and a data set is constructed according to the vibration signal of the operation state; preprocessing the vibration signal in the operation state to obtain a preprocessed vibration signal; extracting fault feature data of the preprocessed vibration signal; performing hierarchical processing on each data segment of the fault feature data to obtain a plurality of low-frequency and high-frequency components, constructing a graph model for each low-frequency and high-frequency component, and calculating a graph similarity score of the graph model and the data set, combining the graph similarity scores of all the low-frequency and high-frequency components into an integral similarity index through an adaptive input weighted fusion technology; a convolutional neural network algorithm is combined with a similarity index to carry out model training on an existing data set, a bolt fault diagnosis model with relatively high precision is obtained, and accurate detection and diagnosis of a bolt fault are realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical fault diagnosis, and relates to a method for constructing a bolt fault diagnosis model and a diagnosis method, system and equipment. Background Art

[0002] In modern industrial processes, fault detection at an early stage is very important to avoid failures and safety-threatening consequences, and to reduce maintenance costs and machine downtime. Regarding bolt fault detection, it is currently possible to use the time of the detected change as the start time of the degradation process to predict the remaining service life of the bolt, and once the change is detected, the potential fault type can be identified at the initial stage based on pattern classification technology. Vibration measurement has advantages in providing a quick interpretation and indication of the health status of bearings. As a non-contact detection method, vibration detection technology can detect longitudinal waves, transverse waves, surface waves, etc. generated by bolt fault losses, and can achieve non-destructive measurement, with a wider range of applications.

[0003] However, the vibration signals currently collected from actual engineering scenarios are always complex. In addition to background noise and interference, this complexity also includes complex changes in the spectrum, nonlinear behavior, etc. In other words, the vibration signal is non-stationary. This fact leads to uncertainty and inaccuracy in statistical indicators such as mean and root mean square that can be used to describe the dynamic health state. The second reason is that in the initial stage, the appearance / existence of the change information is very weak. This weak information in the initial stage is usually drowned out by strong environmental noise, making it very difficult to identify the conditions. Therefore, various improved research works have been carried out in the prior art to overcome these challenges.

[0004] Graphical model (GM) is an effective method to describe the correlation and dependence between non-stationary signals and is used for machine monitoring. It has been proven to be useful in statistics to absorb high noise in the original vibration signal. However, the graphical model (GM) does not consider the effect of the difference between frequency components, which means that weak changes at a specific frequency may be submerged or even lost in the constructed graph, thus affecting the accuracy of bolt fault detection. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method for constructing a bolt fault diagnosis model and a diagnosis method, system and equipment. The present invention can generate a bolt fault diagnosis model with high accuracy, and the bolt fault diagnosis model can realize accurate detection and diagnosis of bolt faults.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for constructing a bolt fault diagnosis model, comprising the following steps:

[0008] Obtain the vibration signal of the bolt running state, and construct a data set according to the vibration signal of the running state;

[0009] Preprocessing the vibration signal in the running state to obtain a preprocessed vibration signal;

[0010] Extracting fault characteristic data of preprocessed vibration signals;

[0011] Each data segment of the fault feature data is processed in layers to obtain multiple low-frequency and high-frequency components. A graph model is constructed for each low-frequency and high-frequency component, and the graph similarity score between the graph model and the data set is calculated. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology.

[0012] The convolutional neural network algorithm combined with the similarity index was used to train the model on the existing data set to obtain the bolt fault diagnosis model.

[0013] In a second aspect, the present invention provides a bolt fault diagnosis method, comprising the following steps:

[0014] Obtain vibration signals of bolt operation status;

[0015] The vibration signal of the operating state is input into the bolt fault diagnosis model described in any one of claims 1 to 7 to perform bolt fault diagnosis.

[0016] In a third aspect, the present invention provides a system for constructing a bolt fault diagnosis model, comprising:

[0017] Data acquisition module: used to obtain the vibration signal of the bolt running state and build a data set according to the vibration signal of the running state;

[0018] Preprocessing module: used to preprocess the vibration signal in the running state to obtain a preprocessed vibration signal;

[0019] Feature extraction module: used to extract fault feature data of preprocessed vibration signals;

[0020] Feature fusion module: used to perform hierarchical processing on each data segment of fault feature data to obtain multiple low-frequency and high-frequency components, build a graph model for each low-frequency and high-frequency component, and calculate the graph similarity score between the graph model and the data set. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology;

[0021] Model generation module: used to use convolutional neural network algorithm combined with similarity index to train the model of the existing data set to obtain the bolt fault diagnosis model.

[0022] In a fourth aspect, the present invention provides an electronic device, comprising: a processor; a memory for storing computer program instructions; and steps for implementing a method for constructing a bolt fault diagnosis model when executing the computer program.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The method of the present invention obtains the vibration signal of the bolt running state, constructs a data set according to the vibration signal of the running state, so as to facilitate the establishment of the model; pre-processes the vibration signal of the running state to obtain a pre-processed vibration signal to eliminate the influence of the data dimension and scale, removes irrelevant signal components such as environmental noise and mechanical interference, and retains the vibration signal generated by the bolt fault, which is conducive to improving the accuracy of the generated model; extracts the fault feature data of the pre-processed vibration signal to generate a bolt fault diagnosis model; hierarchically processes each data segment of the fault feature data to obtain multiple low-frequency and high-frequency components, and constructs a graph model for each low-frequency and high-frequency component, which can better capture the complex change pattern in the signal, and calculates the graph similarity score between the graph model and the data set, and merges the graph similarity scores of all low-frequency and high-frequency components into an overall similarity index through adaptive input weighted fusion technology to obtain a more accurate health status assessment and realize feature visualization; adopts a convolutional neural network algorithm combined with a similarity index to train the model of the existing data set to obtain a bolt fault diagnosis model, which further ensures the accuracy of the model.

[0025] 2. The system of the present invention includes: a data acquisition module, a preprocessing module, a feature extraction module, a feature fusion module and a model generation module; the data acquisition module is used to obtain the vibration signal of the bolt running state, and construct a data set according to the vibration signal of the running state; the preprocessing module is used to preprocess the vibration signal of the running state to obtain the preprocessed vibration signal; the feature extraction module is used to extract the fault feature data of the preprocessed vibration signal; the feature fusion module is used to perform hierarchical processing on each data segment of the fault feature data to obtain multiple low-frequency and high-frequency components, construct a graph model for each low-frequency and high-frequency component, and calculate the graph similarity score between the graph model and the data set, and combine the graph similarity scores of all low-frequency and high-frequency components into an overall similarity index through adaptive input weighted fusion technology; the model generation module is used to use a convolutional neural network algorithm combined with a similarity index to perform model training on the existing data set to obtain a bolt fault diagnosis model. Each module cooperates with each other to generate a more accurate bolt fault diagnosis model, which can realize accurate diagnosis of bolt faults.

[0026] 3. The device of the present invention can also generate a relatively accurate bolt fault diagnosis model, which can realize accurate diagnosis of bolt faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of prediction results according to an embodiment of the present invention;

[0029] Figure 3 This is a flow chart of fault diagnosis model training and testing according to an embodiment of the present invention;

[0030] Figure 4 is a system module diagram of an embodiment of the present invention;

[0031] Figure 5 is a flow chart of a method for constructing a bolt fault diagnosis model of the present invention;

[0032] Figure 6 It is a flow chart of the bolt fault diagnosis method of the present invention;

[0033] Figure 7 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] Before introducing the implementation of the present invention, the following definitions are given for the relevant terms involved in the implementation of the present invention:

[0037] According to the principle of mass inertia, when the monitored object vibrates, the mass block and spring inside the vibration sensor will also vibrate accordingly due to inertia, generating mechanical stress, which will cause tiny deformation inside the sensor element. These deformations will be converted into electrical signals, which will be amplified and processed and then output to the data acquisition device for analysis and processing.

[0038] The convolutional neural network model is a deep learning model or a multi-layer perceptron similar to an artificial neural network, which is often used to analyze visual images. It includes a data input layer, a convolution calculation layer, a ReLU excitation layer, a pooling layer, and a fully connected layer.

[0039] Edge machine learning refers to the ability to run machine learning models locally on edge devices that are far away from the cloud or big data centers, such as real-time low-latency predictions, weak connections to the cloud, and sending big data.

[0040] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0041] See also Figure 5 The present invention discloses a method for constructing a bolt fault diagnosis model, comprising the following steps:

[0042] S1. Obtain the vibration signal of the bolt running state, and construct a data set according to the vibration signal of the running state;

[0043] Preferably, the steps of obtaining the vibration signal of the bolt running state are as follows:

[0044] According to the mass inertia and sampling principle, the cyclic stationary signal acquisition method is adopted to continuously record the vibration signal of the bolt and store it in the secure digital card, with each record lasting 100 seconds.

[0045] Preferably, the vibration signals of the bolt operation status are divided into the following five categories: vibration signals of healthy bolt, bolt wear, bolt crack, bolt deformation and corroded bolt operation status.

[0046] S2. Preprocess the vibration signal in the running state to obtain a preprocessed vibration signal, as follows:

[0047] De-noising and filtering the vibration signal in the running state;

[0048] Selecting a number of sampling data from the vibration signal after denoising and filtering for time domain analysis;

[0049] The vibration signal after time domain analysis is subjected to fast Fourier transform to obtain a preprocessed vibration signal.

[0050] S3. Extracting fault feature data of the pre-processed vibration signal, as follows:

[0051] A three-layer neural network autoencoder consisting of an input layer, a hidden layer, and an output layer is used to enhance and reconstruct the preprocessed vibration signal to obtain the fault feature data of the faulty bolt.

[0052] S4. Perform layered processing on each data segment of the fault feature data to obtain multiple low-frequency and high-frequency components, build a graph model for each low-frequency and high-frequency component, and calculate the graph similarity score between the graph model and the data set. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology:

[0053] Preferably, the fault feature data includes ten time domain features and six frequency domain features, as follows:

[0054] Six frequency domain features are extracted from the frequency domain of the preprocessed vibration signal, including: root mean square, mean, shape factor, kurtosis which measures the degree of outlier tendency of the distribution, skewness which measures the asymmetry of the data around the sample mean, impulse factor, crest factor, margin factor, variance, and median measurement;

[0055] Ten time domain features are extracted from the time domain signal of the preprocessed vibration signal, including: the maximum amplitude of the frequency extracted in the signal, the second largest amplitude of the frequency extracted in the signal, the third largest amplitude of the frequency extracted in the signal, the frequency value of the maximum amplitude, the frequency value of the second largest amplitude, and the frequency value of the third largest amplitude.

[0056] Preferably, the step of performing layered processing on each data segment of the fault characteristic data to obtain a plurality of low-frequency and high-frequency components is specifically as follows:

[0057] The vibration signal analysis method based on high-dimensional hash decomposition and graph similarity measurement analyzes the characteristics of the faulty bolts and divides the vibration signal into time series data segments;

[0058] Each data segment is processed hierarchically using high-dimensional hash decomposition to obtain multiple low-frequency and high-frequency components.

[0059] S4. The convolutional neural network algorithm is combined with the similarity index to train the model on the existing data set to obtain the bolt fault diagnosis model.

[0060] See also Figure 5 In another feasible embodiment of the present invention, the following is adaptively modified according to the circumstances.

[0061] The following steps are involved:

[0062] Obtain the vibration signal of the bolt running state, and build a data set based on the vibration signal of the running state to facilitate the establishment of the model;

[0063] The vibration signal of the running state is preprocessed to obtain the preprocessed vibration signal, so as to eliminate the influence of data dimension and scale, remove irrelevant signal components such as environmental noise and mechanical interference, and retain the vibration signal generated by the bolt failure, which is conducive to improving the accuracy of the generated model;

[0064] Extracting fault feature data of preprocessed vibration signals to generate a bolt fault diagnosis model;

[0065] Each data segment of the fault feature data is processed in layers to obtain multiple low-frequency and high-frequency components. A graph model is constructed for each low-frequency and high-frequency component to better capture the complex change pattern in the signal and calculate the graph similarity score between the graph model and the data set. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology to obtain a more accurate health status assessment and achieve feature visualization.

[0066] The convolutional neural network algorithm combined with the similarity index was used to train the model on the existing data set to obtain the bolt fault diagnosis model, which further ensured the accuracy of the model.

[0067] The present invention can generate a bolt fault diagnosis model with high precision, and the accuracy of bolt fault detection using the model of the present invention is high.

[0068] Based on the above method, the present invention also discloses a bolt fault diagnosis method, see Figure 6 , including the following steps:

[0069] S1, obtaining the vibration signal of the bolt running state;

[0070] S2. Inputting the vibration signal of the operating state into the bolt fault diagnosis model described in any one of claims 1 to 7 to perform bolt fault diagnosis.

[0071] The bolt fault diagnosis method of the present invention can accurately diagnose bolt faults, and compared with the prior art, the accuracy of bolt fault detection is improved.

[0072] Embodiment 1:

[0073] This embodiment provides a method for constructing a bolt fault diagnosis model and a diagnosis method, see Figure 1 , install vibration sensors on the bolts to capture the vibration signals of the workpiece's operating status, extract fault features through the HGM model, and then use AIW to fuse the features. Train the convolutional neural network model with the vibration signal dataset and deploy it on the edge device to perform bolt fault detection. The following is an explanation with the attached figures.

[0074] like Figure 1As shown, the method comprises the following steps:

[0075] Step 1: Install the vibration sensor. The selected vibration sensor model is WV-100, with a working efficiency of 100-1300kHz. It is arranged at the axial position of the bolt end cover by magnetic adsorption. The PCI8811 vibration data acquisition card collects the vibration sensor signal. The vibration sensor and the data acquisition card are connected by a coaxial cable.

[0076] Step 2: Vibration signal acquisition. According to the mass inertia and sampling principle, the data acquisition card adopts a periodic stationary signal acquisition method to continuously record the vibration signal of the bolt with a wide range of sampling frequencies and store it in a secure digital (SD) card, with each record lasting 100 seconds. In order to diagnose different types of faulty bolts, the workpiece vibration signal is divided into the following five categories: healthy, bolt wear, bolt crack, bolt deformation, and bolt corrosion. All known faulty bolts and normal bolts are tested, and the sampling frequency is set to 40kHz to cover all audible ranges to obtain the signals of known faulty bolts and normal bolts.

[0077] Step 3: Data preprocessing, denoising and filtering the collected signal to improve the signal-to-noise ratio of the signal; the digital bandpass filter is used as an anti-aliasing filter to remove DC offset from the signal at a cutoff frequency of 20 Hz and 20 kHz to remove irrelevant signal components such as environmental interference and mechanical interference to retain the vibration signal generated by the bolt failure; 4000 sampling data are selected from the vibration signal for time domain analysis; considering frequency resolution and spectrum leakage, the Hamming window is used to perform fast Fourier transform on the digital signal to obtain the analysis spectrum of the digital signal, that is, the preprocessed vibration signal.

[0078] Step 4: Feature extraction and fusion. For the characteristic frequencies generated by bolt failures, the HGM model is established. The characteristic frequencies are analyzed based on the vibration signal analysis method of high-dimensional hash decomposition (HD) and graph similarity measurement. The vibration signal is divided into time series data segments, and each data segment is hierarchically processed using HD to obtain multiple low-frequency and high-frequency components. Then, a corresponding graph model is constructed for each component, and its graph similarity score with the entire data set is calculated. Finally, the scores of all components are combined into an overall similarity index through AIW fusion technology, and common hypothesis testing methods are used to detect whether there is a state change.

[0079] MATLAB Coder is used to generate C functions for frequency domain feature extraction given FFT spectrum, as well as predefined significance values ​​for peak extraction and C functions for selected machine learning algorithms. To gain a deeper understanding of system faults, for each data window, the extracted bolt vibration signal feature set is shown in Table 1. Six of the features are extracted from the frequency domain, including: root mean square, mean, shape factor, kurtosis to measure the degree of outlier tendency of the distribution, skewness to measure the asymmetry of the data around the sample mean, impulse factor, crest factor, margin factor, variance, and median measurement. Ten features are extracted from time domain signal analysis, including P1, P2, P3, P4, P5, and P6. The fault domain includes time domain features and frequency domain features. After extracting all features, feature selection based on variance analysis is used to sort the features by importance. P1 is the most important feature, followed by variance σ2 and the square root of the arithmetic mean of the squares of continuous signal values ​​RMS. In order to capture transient changes in signals and enhance main features, a three-layer neural network autoencoder consisting of an input layer, a hidden layer and an output layer is introduced to enhance and reconstruct data features.

[0080] Table 1 Bolt vibration signal feature set

[0081]

[0082]

[0083] Step 5: Training the model. Based on the importance of the extracted features, the cross-validation weight value k = 6 is selected to verify the accuracy of each classification algorithm. The comparison of the recognition accuracy of each machine learning technique is shown in Table 2. Since most of the misclassifications occurred during bolt failures, when the mean, skewness, and median features were removed, the convolutional neural network algorithm was used to train the model on the existing data set, and the confusion matrix was used to evaluate the performance of the model.

[0084] Table 2 Comparison of recognition accuracy of various machine learning technologies

[0085] Serial number Classification Algorithms Accuracy 1 Convolutional Neural Networks 97.2％ 2 Bag-Tree Ensemble Classifier 96.6％ 3 Quadratic SVM 96.0％ 4 Naive Bayes 93.5％ 5 LR 92.6％

[0086] Step 6: Deploy and integrate the model, and run the fault diagnosis model directly on the edge IoT device based on the STM32 MCU. The edge device performs data preprocessing and feature extraction of selected features for each data buffer with a real-time fault diagnosis node, and performs real-time AI calculations at the edge and instant response to complete the fault diagnosis of the bolts; the final fault diagnosis results and prediction results will be sent to the display end for intuitive display.

[0087] After obtaining the trained convolutional neural network fault diagnosis model, all the collected eigenvalues ​​are used as the input of the fault diagnosis model to test the performance of the fault diagnosis model. Figure 2 As shown, the known label is the known fault type, the predicted label is the predicted fault type, and the test samples include fault 1 for health, fault 2 for bolt wear, fault 3 for bolt crack, fault 4 for bolt deformation, and fault 5 for bolt corrosion. The model accuracy remains basically consistent from the development stage to the implementation stage. The reduction of feature dimension improves the classification speed without affecting the accuracy of the model, and the application of the empirical cumulative distribution function improves the stability and accuracy of the model. The prediction accuracy of fault 1 is 99.6%, the prediction accuracy of fault 2 is 98.9%, the prediction accuracy of fault 3 is 97.2%, the prediction accuracy of fault 4 is 93.9%, and the prediction accuracy of fault 5 is 96.4%. It can be seen that the method of the present invention uses the least amount of data and extracts relevant features from the row data, and achieves the stability, accuracy and real-time performance of fault detection through a simple edge machine learning model. This shows that under low-speed and heavy-load conditions, the method of the present invention can quickly and intelligently identify bolt faults and provide efficient monitoring and diagnosis for the health status of related equipment.

[0088] Embodiment 2:

[0089] See also Figure 3 This embodiment discloses a method for constructing a bolt fault diagnosis model and a diagnosis method, comprising the following steps:

[0090] The vibration signal of the bolt is obtained by the vibration sensor and transmitted to the data acquisition card. 2048 samples are collected per second. Each sample uses 16-bit pulse code modulation. After preprocessing, the data is extracted to obtain the characteristics of the faulty bolt. Select the recording mode to collect features and record the data to the SD card. The length of each record is one minute, and each record represents a fault state.

[0091] Therefore, more than three faults can be added, such as the combined failures that may occur in the normal life cycle of the bolt, and the HGM model can be established to perform hierarchical analysis on the data features, and its weights can be automatically optimized with AIW, and the features can be fused.

[0092] The collected features are used to train the fault diagnosis model using machine learning technology. The selected model is a convolutional neural network model, which has the advantages of faster operation and higher accuracy. The trained fault diagnosis model is deployed on the edge device and restarted. In the running mode, the same preprocessing and feature extraction algorithms as in the recording mode are used to enable the parallel computing framework to process and acquire data in parallel on the edge device to achieve real-time prediction of the fault type of the bolt.

[0093] The above description mainly introduces the solution provided by the embodiment of the present invention from the perspective of the fault diagnosis method flow and step functions. According to the above description, the implementation of the embodiment of the present invention involves hardware structure and / or software modules, and through the combination of the functions of each part, a bolt fault diagnosis method based on a hierarchical graph model and adaptive input weighted fusion is finally realized. For each specific application, technicians can flexibly choose different methods to complete the functions, but these implementation methods should all fall within the scope of the present invention.

[0094] Embodiment three:

[0095] See also Figure 1 and Figure 3 This embodiment discloses a method for constructing a bolt fault diagnosis model and a diagnosis method, comprising the following steps:

[0096] The vibration signal of the bolt is captured using a vibration sensor. The raw vibration signal is then analyzed using a hierarchical decomposition (HD) method to check the bearing health status at different scales. A set of HGMs are constructed for each decomposed component produced by HD. In order to consider and utilize the differences between the constructed HGMs, an adaptive input weighted fusion is used to test the null hypothesis to make a decision. The proposed method runs periodically and continuously, so that the current health status of the bolt can be checked online. The model is trained by simulating a benchmark database. Finally, the machine learning model is deployed on the edge device for fault detection of the bolt.

[0097] The vibration sensor is arranged at one end of the bolt, one side of the sensor is close to the surface of the workpiece, and the other side is connected to the vibration signal processing and analysis unit, and then connected to the data acquisition card. The fixing method between the vibration sensor and the workpiece includes but is not limited to mechanical fixing, adhesive fixing and magnetic adsorption fixing. In addition, the vibration sensor can also be other intelligent devices that can record and save vibration signals to receive vibration signals at different positions of the vibration source. According to the mass inertia principle and the sampling theorem, the vibration data acquisition card records the vibration signal of the bolt at a wide range of periodic sampling frequency and stores it in a secure digital (SD) card.

[0098] Based on the vibration signal collected by the vibration sensor, the vibration signal is preprocessed to generate a training data set; normalization is achieved by subtracting the minimum value from each signal value and then dividing it by the difference between the maximum value and the minimum value to eliminate the impact of data dimension and scale. Irrelevant signal components such as environmental noise and mechanical interference are removed to retain the vibration signal generated by the bolt failure. The vibration signal is then converted from the time domain to the frequency domain through fast Fourier transform (FFT) to better perform spectrum analysis and feature extraction.

[0099] The characteristic frequencies generated by bolt failures are analyzed using the HGM model to better capture the complex change patterns in the signal. After extracting all the features, the AIW algorithm is used to aggregate all the graphs to obtain a more accurate health status assessment and achieve feature visualization.

[0100] The training data set and the selected features are input into the machine learning model. The machine learning model adopts a convolutional neural network model to perform model training on the existing data set and classify the fault information using a classification model. The confusion matrix is ​​used to evaluate the performance of the classification model, and cross-validation is used to evaluate the performance of the model.

[0101] Furthermore, the neural network model is run on an edge device with an IoT connection to complete the fault diagnosis of the bolts; in order to compensate for the reduced accuracy in the neural network model, a simple empirical cumulative distribution function is used for the prediction result block, and the prediction with the highest repeatability is used as the final prediction result.

[0102] The present invention uses a vibration sensor to capture the vibration signal of the bolt operation, extracts the bolt fault characteristics based on time domain and frequency domain analysis, processes the characteristic signal with HGM, fuses the characteristics with AIW technology, and uses neural network machine learning technology to perform bolt fault detection by training the machine learning model, thereby improving the efficiency and accuracy of bolt fault diagnosis. On the one hand, compared with the traditional contact monitoring method, the present invention does not need to contact the bolt, thereby avoiding damage to the bolt and reducing maintenance costs and risks. On the other hand, compared with the traditional statistical method or neural network data preprocessing fault diagnosis method, the hierarchical graph model (HGM) can better handle complex data features and improve the accuracy of fault diagnosis. In the model training process, based on the visualization image formed after the adaptive input weighting (AIW) feature fusion that has been used, the neural network model is used for training to improve the stability and accuracy of the model. In addition, the present invention deploys the machine learning model on the edge device to realize real-time fault detection of the bolt. Compared with the traditional cloud server model, the deployment of edge devices realizes effective data management, data persistence, effective delivery and content caching, can respond to faults more quickly, and reduces communication delay and dependence. The trained machine learning model can be used as a fault diagnosis model to monitor and predict the failure of bolts under operating conditions in real time based on the vibration signals of the bolts to be predicted.

[0103] Based on the above bolt fault diagnosis model construction method, the present invention also discloses a bolt fault diagnosis model construction system, see Figure 7 ,include:

[0104] Data acquisition module: used to obtain the vibration signal of the bolt running state and build a data set according to the vibration signal of the running state;

[0105] Preprocessing module: used to preprocess the vibration signal in the running state to obtain a preprocessed vibration signal;

[0106] Feature extraction module: used to extract fault feature data of preprocessed vibration signals;

[0107] Feature fusion module: used to perform hierarchical processing on each data segment of fault feature data to obtain multiple low-frequency and high-frequency components, build a graph model for each low-frequency and high-frequency component, and calculate the graph similarity score between the graph model and the data set. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology;

[0108] Model generation module: used to use convolutional neural network algorithm combined with similarity index to train the model of the existing data set to obtain the bolt fault diagnosis model.

[0109] The various modules cooperate with each other to generate a relatively accurate bolt fault diagnosis model, which can realize accurate diagnosis of bolt faults.

[0110] Embodiment 4:

[0111] See also Figure 4 This embodiment discloses a bolt fault diagnosis model construction system and a diagnosis system, including: a data acquisition module, a data processing module, a fault detection module and a test and display module.

[0112] The sensor node is used to monitor the vibration signal generated by the bolt in real time. The sensor node needs to have high sensitivity and wide-band response characteristics, and can accurately capture the weak vibration signal generated by the bolt in operation. Optionally, the sensor node can be a vibration sensor attached to one side of the bolt, or it can be an intelligent vibration signal acquisition system based on the vibration sensor.

[0113] The data acquisition module is used to receive the collected vibration signals from the sensor nodes and convert them into digital signals for processing. The data acquisition module needs to have high-speed and high-precision analog-to-digital conversion capabilities to ensure accurate collection and conversion of vibration signals. Optionally, the data acquisition module can be a PCI8811 vibration signal data acquisition card or other equipment that can collect and record bolt vibration signals.

[0114] The data processing module is used to receive the digital signal from the data acquisition module, filter, reduce noise, analyze and extract features to extract the characteristic parameters related to the bolt fault, select the features through the HGM model, and then fuse them with AIW technology to obtain the feature fusion graph. The data processing module needs to include a signal processing algorithm specifically for bolt fault diagnosis, which can effectively extract fault features and reduce other signal interference.

[0115] The fault detection module is used to determine whether the bolt has an abnormality or potential fault. The model is trained and optimized based on the characteristic parameters of the vibration signal and the corresponding fault labels in the training set. The trained model can perform fault diagnosis based on the characteristic parameters of the vibration signal and determine whether the bolt has an abnormality or potential fault. The fault detection model needs to be fully trained and verified to ensure the ability to accurately identify and judge different types of bolt faults.

[0116] The test and display module is used to receive the diagnostic results output by the fault detection model and display them to the user or other devices. The trained fault detection model is deployed on the edge device for real-time monitoring and diagnosis of the working status of the bolt. At the same time, the module can also provide data logging, alarm and remote communication functions to help users monitor the status of the bolts in real time and perform corresponding maintenance and processing. Optionally, the fault detection model can also achieve the same function on the cloud server, but deploying the model on the edge device has the advantage of being able to respond to faults more quickly and reducing communication delays and dependencies.

[0117] An electronic device comprises: a processor; a memory for storing computer program instructions; and steps for implementing a method for constructing a bolt fault diagnosis model when executing the computer program.

[0118] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0122] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for constructing a bolt fault diagnosis model, characterized in that: The following steps are involved: Obtain the vibration signal of the bolt running state, and construct a data set according to the vibration signal of the running state; Preprocessing the vibration signal in the running state to obtain a preprocessed vibration signal; Extracting fault characteristic data of preprocessed vibration signals; Each data segment of the fault feature data is processed in layers to obtain multiple low-frequency and high-frequency components. A graph model is constructed for each low-frequency and high-frequency component, and the graph similarity score between the graph model and the data set is calculated. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology. The convolutional neural network algorithm combined with the similarity index was used to train the model on the existing data set to obtain the bolt fault diagnosis model.

2. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The steps of obtaining the vibration signal of the bolt running state are specifically as follows: According to the mass inertia and sampling principle, the cyclic stationary signal acquisition method is adopted to continuously record the vibration signal of the bolt and store it in the secure digital card, with each record lasting 100 seconds.

3. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The vibration signals of the bolt operation status are divided into the following five categories: vibration signals of healthy bolt, bolt wear, bolt crack, bolt deformation and corroded bolt operation status.

4. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The steps of preprocessing the vibration signal of the running state to obtain the preprocessed vibration signal are specifically as follows: De-noising and filtering the vibration signal in the running state; Selecting a number of sampling data from the vibration signal after denoising and filtering for time domain analysis; The vibration signal after time domain analysis is subjected to fast Fourier transform to obtain a preprocessed vibration signal.

5. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The steps of extracting fault characteristic data from the pre-processed vibration signal are specifically as follows: A three-layer neural network autoencoder consisting of an input layer, a hidden layer, and an output layer is used to enhance and reconstruct the preprocessed vibration signal to obtain the fault feature data of the faulty bolt.

6. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The fault feature data includes ten time domain features and six frequency domain features, as follows: Six frequency domain features are extracted from the frequency domain of the preprocessed vibration signal, including: root mean square, mean, shape factor, kurtosis which measures the degree of outlier tendency of the distribution, skewness which measures the asymmetry of the data around the sample mean, impulse factor, crest factor, margin factor, variance, and median measurement; Ten time domain features are extracted from the time domain signal of the preprocessed vibration signal, including: the maximum amplitude of the frequency extracted in the signal, the second largest amplitude of the frequency extracted in the signal, the third largest amplitude of the frequency extracted in the signal, the frequency value of the maximum amplitude, the frequency value of the second largest amplitude, and the frequency value of the third largest amplitude.

7. The method for constructing a bolt fault diagnosis model according to claim 1, characterized in that: The steps of performing layered processing on each data segment of the fault characteristic data to obtain multiple low-frequency and high-frequency components are specifically as follows: The vibration signal analysis method based on high-dimensional hash decomposition and graph similarity measurement analyzes the characteristics of the faulty bolts and divides the vibration signal into time series data segments; Each data segment is processed hierarchically using high-dimensional hash decomposition to obtain multiple low-frequency and high-frequency components.

8. A bolt fault diagnosis method, characterized in that: The following steps are involved: Obtain vibration signals of bolt operation status; The vibration signal of the operating state is input into the bolt fault diagnosis model described in any one of claims 1 to 7 to perform bolt fault diagnosis.

9. A system for constructing a bolt fault diagnosis model for implementing the method described in any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to obtain the vibration signal of the bolt running state and build a data set according to the vibration signal of the running state; Preprocessing module: used to preprocess the vibration signal in the running state to obtain a preprocessed vibration signal; Feature extraction module: used to extract fault feature data of preprocessed vibration signals; Feature fusion module: used to perform hierarchical processing on each data segment of fault feature data to obtain multiple low-frequency and high-frequency components, build a graph model for each low-frequency and high-frequency component, and calculate the graph similarity score between the graph model and the data set. The graph similarity scores of all low-frequency and high-frequency components are combined into an overall similarity index through adaptive input weighted fusion technology; Model generation module: used to use convolutional neural network algorithm combined with similarity index to train the model of the existing data set to obtain the bolt fault diagnosis model.

10. An electronic device comprising: Processor; memory, electronic device used to store computer program instructions; characterized in that it is used to implement the steps of the method for constructing a bolt fault diagnosis model as described in any one of claims 1-7 when executing the computer program.

Images