Bearing fault diagnosis method based on multi-feature fusion
By employing a multi-feature fusion method, combining directional gradient histogram, local binary pattern, and an improved CVGG16 network, the problem of accurate bearing fault identification in complex environments was solved, achieving efficient and automated fault diagnosis and improving the accuracy and efficiency of diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENYANG INST OF AUTOMATION - CHINESE ACAD OF SCI
- Filing Date
- 2024-11-22
- Publication Date
- 2026-05-22
AI Technical Summary
Existing technologies struggle to accurately identify rolling bearing faults in complex working environments. Traditional methods rely on expert experience and are expensive, while single feature extraction techniques may miss important information, affecting the comprehensiveness and accuracy of the diagnosis.
A multi-feature fusion bearing fault diagnosis method is adopted, which combines directional gradient histogram, local binary pattern and improved CVGG16 network to extract features, and classifies them by support vector machine to establish WHLVS model for fault identification.
It improves the accuracy and automation of bearing fault diagnosis, can capture weak signal changes at different speeds, reduces computation, and enhances feature representation and classifier performance.
Smart Images
Figure CN122072702A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of industrial fault diagnosis, specifically a bearing fault diagnosis method based on multi-feature fusion. Background Technology
[0002] Rolling bearings, as key components in mechanical equipment, play a vital role in critical positions such as gearboxes, differentials, wheel axles, power take-off shafts, and auxiliary parts of engines in rotating machinery. Bearings not only perform the basic functions of supporting rotating shafts, reducing friction, and ensuring precise positioning of components, but they also withstand complex axial loads transmitted from the machine body. Their performance and stability directly affect the operating efficiency and service life of the entire mechanical equipment.
[0003] Due to their complex and variable working environment, mechanical equipment often experiences fluctuations in bearing speed, which in turn affects the accurate detection of fault characteristics, making bearing damage identification particularly difficult. Timely diagnosis of bearing faults is crucial for preventing equipment failures, ensuring operational safety, and improving production efficiency, impacting the overall performance and lifespan of the equipment. When rolling bearings fail, they not only disrupt normal equipment operation but can also trigger a series of adverse consequences, leading to significant economic burdens. Traditional bearing fault diagnosis relies on expert experience, using methods such as on-site observation and auditory evaluation. This method has significant limitations, making it difficult to comprehensively and accurately identify various fault types, and is also costly. To overcome these challenges, researchers have introduced high-sensitivity sensor technology to collect bearing operating status data in real time and accurately, and combined this with advanced data analysis algorithms to achieve early detection, accurate diagnosis, and timely warning of bearing faults. This has significant research value and practical application value for promoting information technology development.
[0004] The key to rolling bearing fault diagnosis lies in the extraction of fault features and the identification of fault categories. To extract more valuable information from the one-dimensional signals acquired by sensors, these signals often require complex preprocessing steps, including but not limited to conversion into two-dimensional image patterns, to more comprehensively capture fault features. With the development of machine learning and deep learning, these advanced algorithms provide powerful tools for automatically extracting features from raw data. After effective features are extracted, a series of mature classification model algorithms can identify the fault categories of rolling bearings. Currently, vibration signal-based detection methods include: 1) Li Xueyuan et al. used variational mode decomposition to obtain two-dimensional images of the signals and extracted image features using directional gradient histograms to achieve bearing fault diagnosis. This method suffers from problems such as pre-selected parameters and the singleness of features. 2) Zhou Zhetao et al. utilized the advantages of the Transformer encoder-decoder structure and self-attention mechanism to predict the memory degradation problem caused by excessively long sequences in bearings. This method has excessively high computational costs. Detection methods based on acoustic emission signals include: 1) Luan Xiaochi et al. proposed a bearing fault diagnosis method based on acoustic emission characteristic parameters and multi-parameter fusion to identify typical bearing fault types. This method relies on multiple reference signals, and selecting suitable reference signals is difficult. 2) Sha Yundong et al. combined wavelet packet decomposition, kurtosis index, and Hilbert envelope demodulation to extract characteristic frequencies for bearing fault diagnosis. This method has errors when processing high-frequency signals. In summary, most domestic and international research on bearing faults relies on the analysis of vibration signals, and using only a single feature extraction technique may miss some important fault information, affecting the comprehensiveness and accuracy of the diagnosis; although a small number use highly sensitive acoustic emission signals, they still rely on parameter analysis, resulting in low automation. Summary of the Invention
[0005] In view of the defects and deficiencies in the prior art, the purpose of this invention is to provide a method for detecting bearing faults caused by slight changes in bearing signals due to different working environments and rotational speeds.
[0006] The technical solution adopted by the present invention to achieve the above objectives is as follows:
[0007] A bearing fault diagnosis method based on multi-feature fusion performs the following steps to diagnose faults in unknown bearing signals and outputs a fault category:
[0008] A bearing fault diagnosis model WHLVS was established to preprocess the collected bearing acoustic emission signals. An HLV bearing fault feature extraction network was established based on the directional gradient histogram, local binary mode and improved CVGG16 network. The features extracted by the HLV model were then input into the support vector machine algorithm for fault classification.
[0009] The system collects actual acoustic emission signals from the site, preprocesses the data using the WHLVS model, and automatically identifies features to classify faults.
[0010] The bearing acoustic emission signal is captured in real time by an acoustic emission sensor installed at the bearing to capture the signal generated by the bearing during operation or an open-source dataset.
[0011] The data preprocessing includes at least one of data cleaning, overlap sampling, wavelet transform, resolution adjustment, and grayscale conversion, which converts one-dimensional signals into two-dimensional image features.
[0012] The bearing fault diagnosis model WHLVS includes:
[0013] The data preprocessing module preprocesses the one-dimensional data of the acquired acoustic emission signal before inputting it into the feature extraction module;
[0014] The feature extraction module uses at least two branch networks from the Histogram of Oriented Gradients, Local Binary Pattern, and CVGG16 network to extract features, and then inputs them into the Principal Component Analysis (PCA) network to perform feature dimensionality reduction on the extracted fused features before inputting them into the classification and recognition module.
[0015] The classification and recognition module uses the support vector machine method to classify and recognize the reduced acoustic emission features and determine whether the current signal is faulty.
[0016] The directional gradient histogram feature extraction method uses N×N gradient values to combine multiple adjacent cells into a block, and normalizes the histogram within the block using the Gamma correction method to improve the robustness of the features. Finally, the normalized histograms of all blocks are combined to form the HOG feature descriptor for bearing faults.
[0017] The local binary pattern feature extraction method refers to using an equivalent pattern extension method to reduce the dimensionality of features for feature extraction.
[0018] The improved CVGG16 network structure includes convolutional layers, pooling layers, and attention layers arranged sequentially; a dual-channel attention mechanism (CBAM) is introduced to perform in-depth analysis and feature extraction of the acoustic emission signal of the bearing.
[0019] The multi-feature fusion combines the two types of traditional features extracted with the features extracted by the deep neural network to generate a comprehensive feature vector output for each image. Each feature descriptor is used to describe different types of information, and the combination improves the performance of the classifier.
[0020] The classification and recognition module uses a support vector machine to classify data by finding an optimal hyperplane. This maximizes the distance between the support vectors of the two categories and the hyperplane, making the originally linearly inseparable data linearly separable in the new space, thereby classifying the fault types.
[0021] The bearing fault diagnosis system based on multi-feature fusion includes a signal acquisition and sensing device and a host computer. The host computer is equipped with a processor, a memory, and a database. The signal acquisition and sensing device is located at the end of the bearing to acquire the bearing's acoustic emission signal. The memory stores a data preprocessing module, a feature extraction module, and a classification and recognition module. The database stores the acquired data, fault classification labels, and ideal model parameters. The processor schedules each module in the memory to execute the steps described in claim 1, realizing the steps of pre-training to establish the ideal model and the steps of actual fault diagnosis, and finally outputs several classifications based on the bearing speed and fault location.
[0022] Further signal acquisition sensors are installed on the rolling elements, outer ring, and inner ring of the bearing; the damage width and depth are both 0.1 mm, and the damage is classified into 9 categories according to the bearing speed and fault location.
[0023] The present invention has the following beneficial effects and advantages:
[0024] 1. This invention proposes a bearing fault diagnosis method based on multi-feature fusion, which is suitable for capturing fault features in bearing acoustic emission signals. Compared with other methods, this invention has higher accuracy.
[0025] 2. Establish a lightweight CVGG16 model to extract deeper information, enhance feature representation capabilities, and improve prediction accuracy.
[0026] 3. This invention proposes an HLV multi-feature fusion model, which combines traditional feature extraction methods with deep neural networks to extract high-quality features from limited samples and perform serial fusion to obtain comprehensive and multi-level fault features.
[0027] 4. This invention uses principal component analysis to reduce dimensionality and screen out effective features, thereby reducing computational load and improving efficiency while maintaining high accuracy. Attached Figure Description
[0028] Figure 1 This is a flowchart of the bearing fault diagnosis data preprocessing of the present invention;
[0029] Figure 2 This is the bearing fault diagnosis model of the present invention;
[0030] Figure 3 This invention relates to the CVGG16 feature extraction network for bearing fault diagnosis. Detailed Implementation
[0031] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.
[0033] In a first aspect, the present invention provides a bearing fault diagnosis method based on multi-feature fusion, comprising the following steps:
[0034] 1) Collect the acoustic emission signal of the bearing and send it to the host computer for processing;
[0035] 2) The host computer performs data preprocessing on the acquired signals, such as... Figure 1 The diagram shows the data preprocessing flowchart, which includes five parts: data cleaning, overlap sampling, wavelet transform, resolution adjustment, and grayscale conversion.
[0036] 3) In the host computer, an HLV bearing fault feature extraction model is established based on the directional gradient histogram, local binary pattern, and CVGG16 network. The model is trained using preprocessed samples as input and fault classification labels as output. The model is as follows: Figure 2 As shown;
[0037] 4) Collect actual acoustic emission signals on site and automatically identify fault classification using the optimized bearing fault diagnosis model.
[0038] The bearing acoustic emission signal is captured in real time by an acoustic emission sensor installed at the bearing to capture the signal generated by the bearing during operation or an open-source dataset.
[0039] The overlapping sampling involves dividing the collected data into 2048 points with an overlap sampling rate of 30%, ensuring that the sample contains a complete cycle and preventing missing edge information.
[0040] The wavelet transform converts the samples after overlapping sampling into a two-dimensional image form, thereby obtaining richer and more comprehensive bearing fault information.
[0041] The resolution adjustment involves resizing the image to 64×64 pixels. This reduces computational burden and memory consumption while preserving key image features to some extent, enabling the model to learn and infer effectively within an acceptable computational cost.
[0042] Grayscale conversion refers to converting an RGB three-channel image into a single-channel grayscale image. This not only preserves sufficient edge and texture information for feature extraction but also reduces computational complexity and memory consumption.
[0043] Fault diagnosis models such as Figure 2 As shown, the bearing fault diagnosis model includes a data preprocessing module, a feature extraction module, and a classification and recognition module.
[0044] The data preprocessing module cleans the collected one-dimensional data, performs overlapping sampling, wavelet transform, resolution adjustment, and grayscale conversion before inputting it into the feature extraction module;
[0045] The feature extraction module uses three types of features—orientation gradient histogram, local binary pattern, and CVGG16 network—to be fused sequentially. After fusion, the extracted features are input into principal component analysis (PCA) to reduce their dimensionality and then input into the classification and recognition module.
[0046] The classification and recognition module uses a support vector machine to classify and recognize acoustic emission features to determine whether the current signal is faulty.
[0047] The directional gradient histogram feature extraction method uses 8×8 (64) gradient values to combine multiple adjacent cells into a block. The histogram within each block is then normalized using Gamma correction, which helps improve the robustness of the features. Finally, the normalized histograms of all blocks are combined to form the HOG feature descriptor for bearing faults.
[0048] The local binary pattern feature extraction method refers to using an equivalent pattern extension method to reduce the dimensionality of features while maintaining sufficient discriminative power.
[0049] The improved CVGG16 network structure includes a convolutional layer (conv), a pooling layer (fc), and an attention layer arranged sequentially. This method introduces a dual-channel attention mechanism to perform in-depth analysis and feature extraction of the acoustic emission signal of the bearing, such as... Figure 3 As shown.
[0050] By setting five convolutional layers with progressively smaller sizes at the front end, a deeper receptive field is provided, enabling the extraction of complex image features step by step. Three dual-channel attention (CBAM) modules are introduced after the convolutional layers to enhance the network's ability to perceive fault features, strengthen key features, and suppress unimportant features. The final output is obtained by connecting a pooling layer (fc) to normalize the features and concatenating them.
[0051] The multi-feature fusion refers to combining two types of features extracted by traditional feature extraction methods (histogram of oriented gradients and local binary patterns) with features extracted by a deep neural network. This HLV multi-feature fusion model forms a comprehensive feature vector for each image, which is then passed as input to the next model. Each feature descriptor provides different types of information, and their combination can improve the performance and robustness of the classifier.
[0052] Principal component analysis reduces feature dimensions while maximizing variance contribution, thereby increasing speed and efficiency.
[0053] The classification and recognition module uses a support vector machine to classify data by finding an optimal hyperplane. This maximizes the distance between the support vectors of the two categories and the hyperplane, making the originally linearly inseparable data linearly separable in the new space, thereby classifying the fault types.
[0054] Secondly, the present invention also provides a bearing fault diagnosis system based on multi-feature fusion.
[0055] The bearings diagnosed by this invention have three types of defect locations: rolling element damage, outer ring damage, and inner ring damage. The rotational speeds are set to 600 r / min, 900 r / min, and 1200 r / min, respectively, and the damage diameter is 0.1 mm. There are a total of nine damage states, which are divided into nine categories according to the bearing rotational speed and fault location.
[0056] This invention compares the proposed method with VHS, Transformer, and MACNN. The results are shown in Table 1. Table 1 shows that the HOG algorithm did not extract effective features from the image preprocessed by VMD in VHS; the Transformer model performed poorly on the dataset presented in this paper; and the mAP of the MACNN model was 94.54%, lower than the 95.57% mAP of the WHLVS model. Experimental results indicate that the proposed model has superior diagnostic performance.
[0057] Table 1 Comparison of Bearing Fault Diagnosis Models
[0058]
[0059] 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 descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A bearing fault diagnosis method based on multi-feature fusion, characterized in that, The method for diagnosing unknown bearing signals and outputting the fault category includes the following steps: A bearing fault diagnosis model WHLVS was established to preprocess the collected bearing acoustic emission signals. An HLV bearing fault feature extraction network was established based on the directional gradient histogram, local binary mode and improved CVGG16 network. The features extracted by the HLV model were then input into the support vector machine algorithm for fault classification. The system collects actual acoustic emission signals from the site, preprocesses the data using the WHLVS model, and automatically identifies features to classify faults.
2. The bearing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, The bearing acoustic emission signal is captured in real time by an acoustic emission sensor installed at the bearing to capture the signal generated by the bearing during operation or an open-source dataset.
3. The bearing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, The data preprocessing includes at least one of data cleaning, overlap sampling, wavelet transform, resolution adjustment, and grayscale conversion, which converts one-dimensional signals into two-dimensional image features.
4. The bearing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, The bearing fault diagnosis model WHLVS includes: The data preprocessing module preprocesses the one-dimensional data of the acquired acoustic emission signal before inputting it into the feature extraction module; The feature extraction module uses at least two branch networks from the Histogram of Oriented Gradients, Local Binary Pattern, and CVGG16 network to extract features, and then inputs them into the Principal Component Analysis (PCA) network to perform feature dimensionality reduction on the extracted fused features before inputting them into the classification and recognition module. The classification and recognition module uses the support vector machine method to classify and recognize the reduced acoustic emission features and determine whether the current signal is faulty.
5. The bearing fault diagnosis method based on multi-feature fusion according to claim 1 or 4, characterized in that, The directional gradient histogram feature extraction method uses N×N gradient values to combine multiple adjacent cells into a block, and normalizes the histogram within the block using the Gamma correction method to improve the robustness of the features. Finally, the normalized histograms of all blocks are combined to form the HOG feature descriptor for bearing faults.
6. The bearing fault diagnosis method based on multi-feature fusion according to claim 1 or 4, characterized in that, The local binary pattern feature extraction method refers to using an equivalent pattern extension method to reduce the dimensionality of features for feature extraction.
7. The bearing fault diagnosis method based on multi-feature fusion according to claim 1 or 4, characterized in that, The improved CVGG16 network structure includes convolutional layers, pooling layers, and attention layers arranged sequentially; a dual-channel attention mechanism (CBAM) is introduced to perform in-depth analysis and feature extraction of the acoustic emission signal of the bearing.
8. The bearing fault diagnosis method based on multi-feature fusion according to claim 4, characterized in that, The multi-feature fusion combines the two types of traditional features extracted with the features extracted by the deep neural network to generate a comprehensive feature vector output for each image. Each feature descriptor is used to describe different types of information, and the combination improves the performance of the classifier.
9. The bearing fault diagnosis method based on multi-feature fusion according to claim 7, characterized in that, The classification and recognition module uses a support vector machine to classify data by finding an optimal hyperplane. This maximizes the distance between the support vectors of the two categories and the hyperplane, making the originally linearly inseparable data linearly separable in the new space, thereby classifying the fault types.
10. A bearing fault diagnosis system based on multi-feature fusion, characterized in that, The system includes a signal acquisition and sensing device and a host computer. The host computer is equipped with a processor, a memory, and a database. The signal acquisition and sensing device is located at the end of the bearing to acquire the bearing's acoustic emission signal. The memory stores a data preprocessing module, a feature extraction module, and a classification and recognition module. The database stores the acquired data, fault classification labels, and ideal model parameters. The processor schedules each module in the memory to execute the steps described in claim 1, realizing the steps of pre-training to establish the ideal model and the steps of actual fault diagnosis, and finally outputting several classifications based on the bearing speed and fault location.