Equipment fault diagnosis method and device based on multi-domain feature fusion and storage medium

Abstract

The invention discloses an equipment fault diagnosis method and device based on multi-domain feature fusion and a storage medium. Comprising the following steps: collecting multi-modal data of the rotary mechanical equipment; performing time domain feature extraction and classification on the multi-modal data to obtain a first fault classification result; performing frequency domain and amplitude domain feature extraction and classification on the multi-modal data to obtain a second fault classification result; and performing decision fusion on the first fault classification result and the second fault classification result by adopting a DS evidence theory to obtain an equipment fault diagnosis result. Through extraction and selection of the time domain, frequency domain and amplitude domain features of the multi-source signal, the multi-modal features of the signal are deeply analyzed, the equipment fault diagnosis range is expanded, and the fault diagnosis accuracy and reliability are improved. And by simulating a fault diagnosis process based on expert experience, the interpretability of the model is improved.

Description

Technical Field

[0001] The present application relates to the field of fault testing technology, and in particular to a device fault diagnosis method, apparatus, and storage medium based on multi-domain feature fusion. Background Art

[0002] In today's industrial production equipment, rotating machinery, as a key source of initial power, plays a vital role and is an indispensable core equipment in industrial production. It is widely used in key fields such as electricity, petroleum, chemical industry, aerospace, etc. The operating status of these rotating machinery equipment is directly related to production efficiency, product quality and safe production.

[0003] Due to the complexity of the operating environments, diverse loads, and volatile operating conditions of various rotating equipment, various faults are inevitable during long-term operation. If these faults are not discovered and addressed promptly, they can lead to decreased equipment performance, production interruptions, and even serious safety accidents. Therefore, condition monitoring and fault diagnosis of mechanical equipment are crucial. Implementing equipment condition monitoring and fault diagnosis helps identify potential equipment failures early, enabling appropriate countermeasures to ensure safe and stable system operation, reduce repair and maintenance costs, and improve production efficiency.

[0004] Traditionally, manufacturers in the fault diagnosis field, driven by cost considerations, have primarily relied on single sensors and signal processing technologies, such as vibration analysis, oil analysis, and acoustic emission detection. While these methods can identify equipment faults to a certain extent, a single sensor can only capture a single aspect of the equipment, failing to fully reflect its overall condition. Furthermore, a single sensor is susceptible to environmental noise and signal interference, resulting in low fault diagnosis accuracy. Therefore, the current industry practice of using a single sensor for fault diagnosis often suffers from limitations such as information scarcity and limited diagnostic accuracy, making it insufficient for multi-fault diagnosis. Summary of the Invention

[0005] The embodiments of the present application provide a device fault diagnosis method, apparatus, and storage medium based on multi-domain feature fusion, to at least solve the technical problem in related technologies that it is difficult to comprehensively and accurately diagnose device faults.

[0006] According to one aspect of an embodiment of the present application, a device fault diagnosis method based on multi-domain feature fusion is provided, comprising:

[0007] Collect multimodal data of rotating machinery;

[0008] Extracting and classifying time-domain features of the multimodal data to obtain a first fault classification result;

[0009] Extracting and classifying frequency domain and amplitude domain features of the multimodal data to obtain a second fault classification result;

[0010] The first fault classification result and the second fault classification result are subjected to decision fusion using DS evidence theory to obtain an equipment fault diagnosis result.

[0011] In an optional embodiment, collecting multimodal data of rotating machinery includes:

[0012] Collect temperature, vibration data, inlet and outlet pressure, and flow multimodal data of the rotating mechanical equipment.

[0013] In an optional embodiment, performing time domain feature extraction and classification on the multimodal data to obtain a first fault classification result includes:

[0014] The collected vibration data is input into the SDP (Symmetrized Dot Pattern) data fusion algorithm to obtain the SDP image;

[0015] The SDP image is input into a convolutional neural network for feature extraction and classification to obtain the first fault classification result.

[0016] In an optional embodiment, frequency domain and amplitude domain feature extraction and classification are performed on the multimodal data to obtain a second fault classification result, including:

[0017] Performing frequency domain and amplitude domain feature extraction on the temperature, vibration data, inlet and outlet pressure, and flow multimodal data to obtain a plurality of extracted target feature data;

[0018] The target feature data is input into a pre-trained LassoNet network model for fault classification to obtain the second fault classification result.

[0019] In an optional embodiment, the LassoNet network model performs feature selection on multiple target feature data, assigns high weights to important features, assigns low weights or zero weights to unimportant features, and makes decision classifications based on the features.

[0020] In an optional embodiment, the plurality of target feature data include at least each harmonic amplitude and phase, root mean square value, variance, root square amplitude, peak-to-peak value, kurtosis index, and threshold index.

[0021] In an optional embodiment, after collecting multimodal data of the rotating mechanical equipment, the method further includes:

[0022] The collected multimodal data are cleaned, aligned, enhanced, standardized and normalized to obtain preprocessed multimodal data.

[0023] According to another aspect of an embodiment of the present application, a device fault diagnosis apparatus based on multi-domain feature fusion is provided, comprising:

[0024] An acquisition module, used to acquire multimodal data of rotating mechanical equipment;

[0025] a first classification module, configured to extract and classify time domain features of the multimodal data to obtain a first fault classification result;

[0026] A second classification module is used to extract and classify frequency domain and amplitude domain features of the multimodal data to obtain a second fault classification result;

[0027] The fusion module is used to perform decision fusion on the first fault classification result and the second fault classification result using the DS evidence theory to obtain an equipment fault diagnosis result.

[0028] In an optional embodiment, it further includes:

[0029] The data preprocessing module is used to perform data cleaning, data alignment, data enhancement, standardization and normalization on the collected multimodal data to obtain preprocessed multimodal data.

[0030] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned device fault diagnosis method based on multi-domain feature fusion when running.

[0031] The technical solutions provided by the embodiments of the present application may have the following beneficial effects:

[0032] The equipment fault diagnosis method of the present application extracts the time domain, frequency domain and amplitude domain features of multi-source signals, deeply analyzes the multimodal characteristics of the signals, performs classification based on multimodal data, expands the scope of equipment fault diagnosis, and uses DS evidence theory to make decision fusion between the two classifications, ultimately achieving overall fault diagnosis of the equipment and improving the efficiency and reliability of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 is a flowchart of an optional device fault diagnosis method based on multi-domain feature fusion according to an embodiment of the present application;

[0035] Figure 2is a schematic diagram of a device fault diagnosis method based on multi-domain feature fusion according to an embodiment of the present application;

[0036] Figure 3 is a schematic diagram of a device fault diagnosis method according to an embodiment of the present application;

[0037] Figure 4 is a schematic diagram of a device fault diagnosis method according to an embodiment of the present application;

[0038] Figure 5 3 is a schematic diagram of an equipment fault diagnosis device based on multi-domain feature fusion according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention 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 in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0040] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] Deep learning models are currently widely used in rotating machinery fault diagnosis. Their powerful feature extraction and classification capabilities have significantly improved fault diagnosis accuracy. However, deep learning models are essentially black-box models, with complex and opaque internal structures and decision-making processes, resulting in limited interpretability. Consequently, when using deep learning models for fault diagnosis, practitioners struggle to understand how the models reach their conclusions and which factors play a key role in the diagnostic process. This lack of interpretability has, to a certain extent, limited the further application and development of deep learning models in rotating machinery fault diagnosis.

[0042] Multimodal data can be used to capture complementary features of current equipment faults. Currently, numerous methods exist for fusing multimodal data, including channel splicing for data fusion and improved 1D-CNN for feature extraction and fusion. However, these methods focus solely on the signal's time or frequency domain information and fail to fully describe the multimodal information of multi-source signals. Currently, the industry utilizes expert experience for fault diagnosis, primarily employing three analytical methods: time domain analysis, frequency domain analysis, and amplitude domain analysis. These methods extract typical features from these three domains, then leverage expert experience to arrive at fault diagnosis results.

[0043] This application proposes a device fault diagnosis method based on multi-domain feature extraction and fusion. By extracting time, frequency, and amplitude domain features from multimodal signals, the method achieves comprehensive device fault diagnosis. First, time domain feature extraction is performed on the X, Y, and Z axis vibration signals. Specifically, the SDP data fusion algorithm is used to fuse the three-axis signals, effectively preserving the time domain feature information. Feature extraction and classification are then performed using a CNN. Typical frequency and amplitude domain features are extracted from the vibration, temperature, inlet and outlet pressure, and flow rate data, including harmonic amplitude and phase, RMS value, variance, RMS amplitude, peak-to-peak value, kurtosis index, and threshold index. This method mimics the process of fault diagnosis using expert experience. LassoNet is then used for feature selection and classification. Model training assigns high weights to important features and low or zero weights to unimportant features, thus addressing the common interpretability limitations of current deep learning models. Decision fusion between the two classifications is then performed using DS evidence theory, ultimately achieving comprehensive device fault diagnosis and improving fault diagnosis efficiency and reliability.

[0044] The following is a detailed description of the device fault diagnosis method based on multi-domain feature fusion according to the embodiment of the present application with reference to the accompanying drawings. Figure 1 As shown, the method mainly includes the following steps:

[0045] S101 collects multimodal data of rotating mechanical equipment.

[0046] In an embodiment of the present application, collecting multimodal data of the rotating mechanical equipment includes collecting temperature, vibration data, inlet and outlet pressure, and flow multimodal data of the rotating mechanical equipment.

[0047] Specifically, two temperature and vibration sensors are used to obtain equipment temperature and vibration signals. These sensors are installed on the bearing housing and the pump body, respectively. Each sensor can obtain temperature data and vibration data along the X, Y, and Z axes. Sensors are installed at the pump inlet and outlet to obtain pump inlet pressure, outlet pressure, and flow rate data.

[0048] Furthermore, data preprocessing is performed on the multimodal data. In one embodiment, data cleaning, data alignment, data enhancement, standardization and normalization are performed on the collected multimodal data.

[0049] Specifically, the collected multimodal data is cleaned to process missing values, outliers and duplicate values in the data; normalized cross-correlation technology is used to align the time series signals of the multimodal data; overlapping sampling or random starting sampling is used to slice the sampled data to achieve data enhancement and obtain several signal segments with a sample length of 2048; variational mode decomposition (VMD) is used to perform modal decomposition and reconstruction on the obtained signal segments to obtain denoised signals; standardization and normalization operations are performed to eliminate the dimensional influence between different features and accelerate the convergence of the model.

[0050] S102 extracts and classifies the multimodal data in the time domain to obtain a first fault classification result.

[0051] In one embodiment, time-domain features of the data are first extracted. This includes inputting the collected vibration data into an SDP (Symmetrized Dot Pattern) data fusion algorithm to generate an SDP image. The SDP image is then input into a convolutional neural network for feature extraction and classification, resulting in a first fault classification result.

[0052] Specifically, the X, Y, and Z axis vibration data obtained by the two sensors are fused through the SDP data fusion algorithm to obtain two SDP images, which effectively retains the time domain information of the fault signal.

[0053] The SDP data fusion algorithm is a method for converting multi-channel time series data (such as vibration signals) into a two-dimensional graph. The SDP algorithm maps one-dimensional time series data into a polar coordinate space, forming a symmetrical dot pattern (a snowflake-like pattern). Through this mapping, multi-channel data can be fused into the same image. The specific steps include: normalizing the vibration data of the X, Y, and Z axes to eliminate dimension and amplitude differences. Mapping the vibration signal of each channel into a polar coordinate space to form a symmetrical dot pattern. The data of each channel occupies a different area in the image, for example, the X-axis data is mapped horizontally, the Y-axis data is mapped vertically, and the Z-axis data is mapped diagonally. The SDP images of the three channels are merged into a complete image, preserving the characteristic information of each channel.

[0054] The SDP algorithm converts time series data into a graphical representation, which can intuitively reflect the dynamic changes of the signal. Due to the special characteristics of polar coordinate images, SDP can clearly show the periodicity, mutation and trend changes of the signal, thereby effectively preserving the time domain information of the fault signal.

[0055] Furthermore, the SDP image is sent to a convolutional neural network (CNN) to extract features and perform classification to obtain a first fault classification result, which is a time domain information fault classification result.

[0056] Through the SDP data fusion algorithm, multi-channel vibration data can be effectively fused, and the time domain information of the fault signal can be retained in a graphical manner, which can significantly enhance the fault characteristics and improve the accuracy of the diagnosis model.

[0057] S103 extracts and classifies the multimodal data in frequency domain and amplitude domain to obtain a second fault classification result.

[0058] In an optional embodiment, frequency domain and amplitude domain feature extraction and classification are performed on the multimodal data to obtain a second fault classification result, including:

[0059] Frequency and amplitude domain feature extraction is performed on multimodal data including temperature, vibration, inlet and outlet pressure, and flow rate to obtain multiple target feature data. These target feature data include at least harmonic amplitude and phase, root mean square value, variance, root mean square amplitude, peak-to-peak value, kurtosis index, and threshold index. These target feature data are input into a pre-trained LassoNet network model for fault classification, yielding a second fault classification result.

[0060] Specifically, frequency domain and amplitude domain feature extraction is performed on X, Y, and Z axis vibration data, temperature data, inlet and outlet pressure data, and flow data to obtain typical features in the current industry fault diagnosis methods, including harmonic amplitude and phase, root mean square value, variance, root square amplitude, peak-to-peak value, kurtosis index, threshold index, etc.

[0061] Furthermore, these features are input into the LassoNet network to perform feature selection on multiple categories of features, assigning high weights to important features and low or zero weights to unimportant features. Decision classification is performed based on the features, and finally the frequency domain and amplitude domain information fault classification results are obtained, which is the second fault classification result.

[0062] LassoNet is a model that combines Lasso regression sparsity and neural networks, and can perform feature selection and parameter optimization simultaneously during training. The specific steps are as follows:

[0063] Input features: Input frequency domain and amplitude domain features (such as harmonic amplitude, phase, RMS value, variance, RMS amplitude, peak-to-peak value, kurtosis, etc.) into the LassoNet network.

[0064] Feature weight assignment: LassoNet allows features to have non-zero weights in the hidden layer by introducing a skip connection from input to output, but only when its skip connection is activated. This mechanism allows important features to be given high weights, while unimportant features are given low weights or zero weights.

[0065] By performing variable selection and parameter updating simultaneously, the entire solution path will be obtained during the model training process, thereby obtaining different feature combinations.

[0066] Select the optimal feature combination: Select the optimal feature combination based on the performance of different feature combinations on the solution path (such as AUC value).

[0067] Furthermore, based on the features selected by LassoNet, the process of fault classification is as follows:

[0068] Feature selection results: The feature combination selected by LassoNet is used for subsequent classification.

[0069] Classification model training: Use the selected features to train classification models (such as random forest, SVM, GBDT, etc.) to improve classification performance.

[0070] Fault classification results: The fault classification results of the frequency domain and amplitude domain information are finally obtained, that is, the second fault classification results.

[0071] Model advantages: sparsity and interpretability: LassoNet improves the interpretability of the model by selecting only important features through sparsity constraints.

[0072] Performance improvement: Experiments show that LassoNet outperforms traditional methods in feature selection and classification tasks, and can effectively improve classification accuracy.

[0073] Flexibility: LassoNet can be combined with a variety of classifiers to further improve the robustness and generalization ability of the model.

[0074] LassoNet is used for feature selection and decision classification, which can effectively utilize frequency domain and amplitude domain features to provide reliable classification results for fault diagnosis.

[0075] S104 uses DS evidence theory to perform decision fusion on the first fault classification result and the second fault classification result to obtain an equipment fault diagnosis result.

[0076] In fault diagnosis, DS evidence theory is an effective method for integrating multiple fault classification results into a decision-making fusion, thereby obtaining more reliable diagnostic conclusions. By integrating the first and second fault classification results through DS evidence theory, we can effectively integrate multi-source information, reduce uncertainty, and obtain more reliable equipment fault diagnosis results.

[0077] Optionally, when performing actual equipment fault diagnosis, model training is also included. A laboratory-based test bench is used to acquire multi-source sensor data. First, equipment vibration, temperature, inlet and outlet pressure, and flow rate data are collected. Data cleaning, alignment, and enhancement are then performed. VMD is used to denoise the signal and perform standardization and normalization. Training and test sets are then divided and labels are added. Feature extraction is performed on the experimental data in the time, frequency, and amplitude domains. A model is built based on LassoNet and the information fusion algorithm. The model is trained and performance evaluated using the dataset. The model hyperparameters are optimized to determine whether optimality has been achieved and obtain the optimal model.

[0078] In order to facilitate understanding of the device fault diagnosis method provided in the embodiment of the present application, the following Figure 2-4 Further description.

[0079] like Figure 2 As shown, the X-axis vibration signal, Y-axis vibration signal, Z-axis vibration signal, temperature signal, inlet pressure signal, outlet pressure signal and flow signal are collected by the set sensors.

[0080] Based on the X-axis vibration signal, Y-axis vibration signal, and Z-axis vibration signal, the SDP method is used to generate the SDP graph. The SDP image features are extracted and classified based on CNN. The collected temperature signal, inlet pressure signal, outlet pressure signal, and flow signal are extracted in the frequency domain and amplitude domain. LassoNet is used to select important features and classify them. The DS evidence theory is used to perform decision fusion on the two classification results, and the fault diagnosis results are output. The fault diagnosis results include: normal, bearing fault, imbalance, misalignment, mechanical looseness, and impeller fault.

[0081] like Figure 3 As shown, the overall framework of this application includes the following steps:

[0082] S301 collects multimodal data from the device;

[0083] S302 preprocesses the multimodal data;

[0084] S303 extracts the time domain, frequency domain and amplitude domain features of the multimodal data.

[0085] S304 builds and trains the model;

[0086] S305 device multi-fault diagnosis.

[0087] This application uses a centrifugal pump fault diagnosis test bench built in the laboratory to collect vibration, temperature, inlet and outlet pressure and flow signals.

[0088] The collected multimodal data was cleaned to address missing values, outliers, and duplicates. Normalized cross-correlation was used to align the multimodal data for time series signals. Overlapping or random starting sampling was used to slice the sampled data for data enhancement, resulting in several signal segments with a sample length of 2048. Variational mode decomposition (VMD) was used to decompose and reconstruct the obtained signal segments to obtain denoised signals. Standardization and normalization were performed to eliminate dimensionality effects between different features and accelerate model convergence. The resulting experimental dataset was then divided into training and test sets with a 4:1 ratio.

[0089] The time-aligned X, Y, and Z axis vibration data is fused using the SDP data fusion algorithm to generate an SDP image, effectively preserving the time domain information of the fault signal. Frequency and amplitude domain feature extraction is performed on the X, Y, and Z axis vibration data, temperature data, inlet and outlet pressure data, and flow rate data, yielding typical features used in current industry fault diagnosis methods.

[0090] A neural network structure was built based on LassoNet and the information fusion algorithm. The model was trained using the training set and its performance was evaluated using the test set. Hyperparameters were adjusted to obtain the optimal model. A CNN model was then trained to extract SDP image features and classify them.

[0091] The obtained optimal model is used to diagnose fault data obtained from real equipment to verify the accuracy and efficiency of the model.

[0092] like Figure 4 As shown in the figure, we first collect data on equipment vibration, temperature, inlet and outlet pressure, and flow rate. Next, we perform data cleaning, alignment, and augmentation. We use VMD to denoise the signals and perform standardization and normalization. Next, we divide the data into training and test sets and add labels. We then perform feature extraction on the experimental data in the time, frequency, and amplitude domains. We then build a model based on LassoNet and the information fusion algorithm. We then use the dataset to train and evaluate the model's performance, optimize model hyperparameters, determine if the model is optimal, and ultimately obtain the optimal model.

[0093] Real equipment vibration, temperature, inlet and outlet pressure, and flow rate data are collected and cleaned, aligned, and enhanced. VMD is then used to denoise the signals and perform standardization and normalization. Feature extraction is performed on the collected data in the time, frequency, and amplitude domains. This data is then input into the optimal model for multi-fault diagnosis. The resulting fault diagnosis results include: normal, bearing fault, imbalance, misalignment, mechanical looseness, and impeller fault.

[0094] This application uses the SDP data fusion algorithm to perform data fusion on multi-source signals. The SDP data fusion method can retain the time series information of the signal, and cooperate with CNN to extract features from the SDP image, which can effectively extract the time domain features of the signal. The frequency domain and amplitude domain features of the multimodal signal are extracted by traditional signal processing methods, and LassoNet is used to perform feature selection on the extracted frequency domain and amplitude domain features. LassoNet is a neural network architecture that combines the feature selection advantages of Lasso and the nonlinear modeling capabilities of deep learning. It enables the network to learn more important features by introducing jump connections with L1 penalties between network layers. It has great advantages in processing complex, high-dimensional and nonlinear data.

[0095] By extracting and selecting features from the time, frequency, and amplitude domains of multi-source signals, we deeply analyze the multimodal characteristics of the signals, expand the scope of equipment fault diagnosis, and improve the accuracy and reliability of fault diagnosis. By mimicking the fault diagnosis process based on expert experience, we increase the interpretability of the model.

[0096] According to another aspect of the embodiment of the present application, there is also provided a device fault diagnosis apparatus based on multi-domain feature fusion for implementing the above-mentioned device fault diagnosis method based on multi-domain feature fusion. Figure 5 As shown, the device includes:

[0097] The acquisition module 501 is used to acquire multimodal data of rotating mechanical equipment;

[0098] A first classification module 502 is configured to extract and classify time domain features of the multimodal data to obtain a first fault classification result;

[0099] The second classification module 503 is used to extract and classify the multimodal data in the frequency domain and amplitude domain to obtain a second fault classification result;

[0100] The fusion module 504 is configured to perform decision fusion on the first fault classification result and the second fault classification result using the DS evidence theory to obtain a device fault diagnosis result.

[0101] In an optional embodiment, it further includes:

[0102] The data preprocessing module is used to perform data cleaning, data alignment, data enhancement, standardization and normalization on the collected multimodal data to obtain preprocessed multimodal data.

[0103] It should be noted that the device fault diagnosis apparatus based on multi-domain feature fusion provided in the above embodiment only uses the division of the above-mentioned functional modules as an example when executing the device fault diagnosis method based on multi-domain feature fusion. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device fault diagnosis apparatus based on multi-domain feature fusion provided in the above embodiment and the device fault diagnosis method based on multi-domain feature fusion embodiment are based on the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0104] According to another aspect of the embodiments of the present application, a computer-readable storage medium corresponding to the device fault diagnosis method based on multi-domain feature fusion provided in the aforementioned embodiments is also provided, on which a computer program (i.e., a program product) is stored. When the computer program is run by the processor, it will execute the device fault diagnosis method based on multi-domain feature fusion provided in any of the aforementioned embodiments.

[0105] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0106] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the device fault diagnosis method based on multi-domain feature fusion provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0107] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0108] The above embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A device fault diagnosis method based on multi-domain feature fusion, characterized in that: include: Collect multimodal data of rotating machinery; Extracting and classifying time-domain features of the multimodal data to obtain a first fault classification result; Extracting and classifying frequency domain and amplitude domain features of the multimodal data to obtain a second fault classification result; The first fault classification result and the second fault classification result are subjected to decision fusion using DS evidence theory to obtain an equipment fault diagnosis result.

2. The method according to claim 1, characterized in that Collect multimodal data from rotating machinery, including: Collect temperature, vibration data, inlet and outlet pressure, and flow multimodal data of the rotating mechanical equipment.

3. The method according to claim 2, characterized in that Extracting and classifying time domain features of the multimodal data to obtain a first fault classification result includes: The collected vibration data is input into the SDP (Symmetrized Dot Pattern) data fusion algorithm to obtain the SDP image; The SDP image is input into a convolutional neural network for feature extraction and classification to obtain the first fault classification result.

4. The method according to claim 2, characterized in that Extracting and classifying frequency domain and amplitude domain features of the multimodal data to obtain a second fault classification result includes: Performing frequency domain and amplitude domain feature extraction on the temperature, vibration data, inlet and outlet pressure, and flow multimodal data to obtain a plurality of extracted target feature data; The target feature data is input into a pre-trained LassoNet network model for fault classification to obtain the second fault classification result.

5. The method according to claim 4, characterized in that The LassoNet network model performs feature selection on multiple target feature data, assigns high weights to important features, low weights or zero weights to unimportant features, and makes decision classifications based on the features.

6. The method according to claim 4, characterized in that The multiple target feature data include at least each harmonic amplitude and phase, root mean square value, variance, root square amplitude, peak-to-peak value, kurtosis index, and threshold index.

7. The method according to claim 1, characterized in that After collecting multimodal data of rotating machinery, it also includes: The collected multimodal data are cleaned, aligned, enhanced, standardized and normalized to obtain preprocessed multimodal data.

8. A device for equipment fault diagnosis based on multi-domain feature fusion, characterized in that: include: An acquisition module, used to acquire multimodal data of rotating mechanical equipment; a first classification module, configured to extract and classify time domain features of the multimodal data to obtain a first fault classification result; A second classification module is used to extract and classify frequency domain and amplitude domain features of the multimodal data to obtain a second fault classification result; The fusion module is used to perform decision fusion on the first fault classification result and the second fault classification result using the DS evidence theory to obtain an equipment fault diagnosis result.

9. The device according to claim 8, characterized in that Also includes: The data preprocessing module is used to perform data cleaning, data alignment, data enhancement, standardization and normalization on the collected multimodal data to obtain preprocessed multimodal data.

10. A computer-readable medium, characterized in that Computer-readable instructions are stored thereon, and the computer-readable instructions are executed by a processor to implement a device fault diagnosis method based on multi-domain feature fusion as described in any one of claims 1 to 7.

Images