Gearbox fault diagnosis method, device and equipment based on multi-source data fusion and storage medium

Abstract

The application discloses a gearbox fault diagnosis method and device based on multi-source data fusion, equipment and a storage medium, relates to the technical field of wind power generation, and the gearbox fault diagnosis method based on multi-source data fusion at least comprises the following steps: collecting multi-source data of a gearbox of a wind power generation equipment; respectively pre-processing and / or time-frequency analyzing vibration signal data, sound signal data and temperature signal data in the multi-source data to obtain corresponding initial dynamic characteristics; respectively decomposing the initial dynamic characteristics into mode sharing characteristics and mode specific characteristics related to faults, and fusing the mode sharing characteristics to obtain a global feature vector; inputting the global feature vector and the mode specific characteristics into a preset fault diagnosis classifier after weighted fusion; and outputting a gearbox fault diagnosis result by the fault diagnosis classifier. The application improves the accuracy of the gearbox fault diagnosis of the wind power generation equipment.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to a method, apparatus, equipment and computer-readable storage medium for gearbox fault diagnosis based on multi-source data fusion. Background Technology

[0002] As a critical transmission component, the gearbox in a wind turbine generator set directly impacts the unit's operational safety and maintenance costs through fault diagnosis. However, traditional gearbox fault diagnosis methods have significant limitations: on the one hand, they rely on data from a single sensor (such as vibration signal analysis, which is sensitive to gear wear but easily affected by environmental noise, and temperature monitoring, which reflects lubrication abnormalities but has a delayed response), resulting in incomplete capture of gearbox fault characteristics; on the other hand, existing multi-source data fusion diagnostic schemes typically employ static weight fusion strategies, which cannot adapt to the dynamic changes in the importance of features under varying operating conditions, and the fault diagnosis thresholds are highly dependent on expert subjective experience, easily introducing human bias, ultimately leading to insufficient objectivity and a high false alarm rate in gearbox fault diagnosis results.

[0003] The information disclosed in this background section is only for understanding the background technology of the present application concept, and therefore may contain information that does not constitute prior art. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, and computer-readable storage medium for gearbox fault diagnosis based on multi-source data fusion, aiming to improve the accuracy of gearbox fault diagnosis for wind power generation equipment.

[0005] To achieve the above objectives, this application provides a gearbox fault diagnosis method based on multi-source data fusion, the method comprising: Collect multi-source data from the gearbox of a wind power generation device, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data; The vibration signal data, the sound signal data, and the temperature signal data are preprocessed and / or subjected to time-frequency analysis to obtain their respective initial dynamic characteristics. Each of the initial dynamic features is decomposed into fault-related modal shared features and modal specific features, and the modal shared features are fused to obtain a global feature vector; The global feature vector and each modality-specific feature are weighted and fused, and then input into a preset fault diagnosis classifier, which outputs the gearbox fault diagnosis result.

[0006] In one embodiment, the initial dynamic features include at least the vibration time-frequency features corresponding to the vibration signal data, the frequency domain feature matrix corresponding to the sound signal, and the temperature fluctuation features corresponding to the temperature signal; The step of preprocessing and / or performing time-frequency analysis on the vibration signal data, the sound signal data, and the temperature signal data respectively to obtain the corresponding initial dynamic features includes: The vibration signal data is subjected to time-frequency analysis using a preset synchronous compressed wavelet transform algorithm to obtain the corresponding vibration time-frequency characteristics; The sound signal is processed by Fast Fourier Transform and Discrete Cosine Transform to obtain the frequency domain feature matrix; The temperature signal is preprocessed to eliminate low-frequency trends and noise. The preprocessing includes detrending processing and threshold noise reduction processing. The preprocessed temperature signal is transformed using a preset empirical wavelet transform algorithm to obtain temperature fluctuation characteristics.

[0007] In one embodiment, the step of decomposing each of the initial dynamic features into fault-related mode-shared features and mode-specific features includes: The abstract features corresponding to each of the initial dynamic features are input into a preset feature decoupling network. The encoder of the feature decoupling network processes each of the initial dynamic features to obtain the corresponding feature vector. Each of the aforementioned feature vectors is input into the decoupling layer of the feature decoupling network. The decoupling layer separates the features in each feature vector into two different representation spaces, thereby obtaining the modal shared features and modal specific features corresponding to each feature vector. The modal shared features are used to characterize the fault features of the gearbox, and the modal specific features are used to characterize the inherent background information or inherent noise information of the sensor.

[0008] In one embodiment, the step of fusing the shared features of each modality to obtain a global feature vector includes: Each modality-shared feature is input into a preset neural network or fully connected layer to determine the attention score corresponding to each modality-shared feature. The attention scores corresponding to each modality-shared feature are used as the weights corresponding to each modality-shared feature. The weighted sums of the modality-shared features are then performed to obtain the global feature vector.

[0009] In one embodiment, the step of weightedly fusing the global feature vector and each modality-specific feature and then inputting the result to a preset fault diagnosis classifier, and having the fault diagnosis classifier output the gearbox fault diagnosis result, includes: The global feature vector and each modality-specific feature are normalized to obtain the normalized global feature vector and each modality-specific feature. Based on the global feature vector and the feature information entropy of each modality-specific feature, the weights corresponding to the global feature vector and each modality-specific feature are determined, wherein the smaller the feature information entropy, the larger the corresponding weight. Based on the weights of the global feature vector and each modal-specific feature, a linear weighted sum is performed on the normalized global feature vector and each modal-specific feature to obtain the fused feature vector; The fused feature vector is input into a preset fault diagnosis classifier, which then outputs the gearbox fault diagnosis result.

[0010] In one embodiment, the fault diagnosis classifier includes at least a regression network and a classification network, and the gearbox fault diagnosis result includes at least the fault severity, fault type, and fault location. The step of inputting the fused feature vector into a preset fault diagnosis classifier and having the fault diagnosis classifier output the gearbox fault diagnosis result includes: The fused feature vector is input into the regression network, which predicts the severity of the gearbox failure. The severity of the failure includes at least one of the following: number of failures, failure area, and failure offset. The fused feature vector is input into the classification network, which then predicts the fault type and location of the gearbox.

[0011] In one embodiment, before the step of inputting the weighted fusion of the global feature vector and each of the modality-specific features into a preset fault diagnosis classifier, the method further includes: Obtain the health baseline features of the data sources corresponding to each model-specific feature under the gearbox health state, wherein the health baseline features include at least the mean and standard deviation; Calculate the degree of deviation between each modality-specific feature and the health baseline features of each data source to obtain the residual sequence corresponding to each modality-specific feature; An initial feature selection threshold is obtained through initialization. Based on the changing trend of the residual sequence, the initial feature selection threshold is optimized to obtain a target feature selection threshold. Specifically, when the changing trend is increasing, the initial feature selection threshold is decreased; when the changing trend is stable or fluctuating, the initial feature selection threshold is increased. The residuals in the residual sequence that are below the target feature selection threshold are marked as the corresponding model-specific features as suppression-specific features; The suppressed specific features are deleted from each of the model-specific features to obtain updated model-specific features, wherein the updated model-specific features are used for weighted fusion with the global features.

[0012] Furthermore, this application also provides a gearbox fault diagnosis device, the gearbox fault diagnosis device comprising: The data acquisition module is used to acquire multi-source data from the gearbox of the wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data. The feature processing module is used to preprocess and / or perform time-frequency analysis on the vibration signal data, the sound signal data, and the temperature signal data respectively to obtain the corresponding initial dynamic features. The feature fusion module is used to decompose each of the initial dynamic features into fault-related modal shared features and modal specific features, and fuse the modal shared features to obtain a global feature vector; The fault diagnosis module is used to perform weighted fusion of the global feature vector and each modality-specific feature, and then input the result to a preset fault diagnosis classifier, which outputs the gearbox fault diagnosis result.

[0013] In addition, this application also provides a gearbox fault diagnosis device, which includes at least: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the gearbox fault diagnosis method of multi-source data fusion applied to the gearbox fault diagnosis device as described above.

[0014] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the gearbox fault diagnosis method based on multi-source data fusion as described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the gearbox fault diagnosis method with multi-source data fusion as described above.

[0016] This application provides a gearbox fault diagnosis method based on multi-source data fusion. The method includes: firstly, collecting multi-source data of the gearbox of a wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data; then, preprocessing and / or time-frequency analysis of the vibration signal data, sound signal data, and temperature signal data respectively to obtain corresponding initial dynamic features. This application uses multi-source data for gearbox fault diagnosis, which can comprehensively obtain relevant fault features of the gearbox and avoid affecting the accuracy of the fault diagnosis results. Furthermore, each initial dynamic feature is decomposed into fault-related modal shared features and modal specific features, and the modal shared features are fused to obtain a global feature vector. Finally, the global feature vector and each modal specific feature are weighted and fused and input into a preset fault diagnosis classifier, which outputs the gearbox fault diagnosis result. In this application's technical solution, feature extraction is divided into fault-related modal shared features and sensor-specific features. Modal shared features enable the diagnostic process to no longer rely on a single sensor, but rather through cross-validation of multimodal sensor information (vibration signals, sound signals, temperature signals), significantly improving the accuracy and robustness of gearbox fault identification. Modal specific features reflect the inherent specific information of each mode, record the contextual information during data acquisition, facilitate the evaluation of the reliability of multi-source data, and provide an adaptive data basis for outputting accurate fault diagnosis results. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a gearbox fault diagnosis method based on multi-source data fusion applied to edge devices in an embodiment of this application. Figure 2 This is a flowchart illustrating the time-frequency analysis and preprocessing of multi-source data in an embodiment of this application. Figure 3 This is a flowchart illustrating the optimization process for model-specific features in an embodiment of this application. Figure 4This is a schematic diagram of the gearbox fault diagnosis device in the embodiments of this application; Figure 5 This is a schematic diagram of the hardware operating environment of the equipment involved in the gearbox fault diagnosis method of multi-source data fusion in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] This application provides a gearbox fault diagnosis method based on multi-source data fusion, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the gearbox fault diagnosis method based on multi-source data fusion applied to edge devices, as described in this application. The gearbox fault diagnosis method based on multi-source data fusion may include: Step S10: Collect multi-source data from the gearbox of the wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data; Various sensors can be installed in the gearbox to collect multi-source data of different modes, such as vibration signal data, sound signal data, and temperature signal data. Vibration signal data reflects the periodic force or impact generated by mechanical motion, and its frequency range is usually between 0 and 20 kHz. Sound signal data reflects the sound wave signal that propagates through the air medium, including sound waves that can be heard by the human ear (20 Hz to 20 kHz) and ultrasound (20 kHz to 1 MHz). Temperature signal data reflects the temperature rise caused by heat energy generated by friction, loss, etc., and its trend is a slow-changing signal that is close to DC.

[0025] For example, in the actual diagnostic process of a gearbox, these three signals reflect the fault state from different levels and can complement each other to form a complete fault characteristic data network. For instance, combining vibration and sound signals: vibration analysis is used to diagnose mid-to-late stage faults that have already developed, while sound signals are good at capturing weak early signs of fault emergence. Combining the two can achieve full-cycle monitoring from early warning to accurate diagnosis. Combining vibration / sound signals with temperature signals: when vibration or acoustic emission signals indicate that the gearbox may have wear or lubrication problems, temperature signals can corroborate whether there is a temperature rise due to increased friction, forming cross-validation. For example, if abnormal vibration is accompanied by a temperature rise, the fault risk level will be significantly increased.

[0026] Step S20: Preprocess and / or perform time-frequency analysis on the vibration signal data, sound signal data, and temperature signal data respectively to obtain the corresponding initial dynamic characteristics. For example, during the preprocessing process, the vibration signal can be subjected to time-frequency analysis using a synchronous compressed wavelet transform algorithm to extract high-resolution time-frequency features; and the sound signal can be processed by fast Fourier transform and discrete cosine transform to obtain its feature matrix.

[0027] Furthermore, in a feasible embodiment, the initial dynamic features include at least the vibration time-frequency features corresponding to the vibration signal data, the frequency domain feature matrix corresponding to the sound signal, and the temperature fluctuation features corresponding to the temperature signal. like Figure 2 As shown, the step of preprocessing and / or performing time-frequency analysis on vibration signal data, sound signal data, and temperature signal data to obtain corresponding initial dynamic characteristics may include: Step S21: Perform time-frequency analysis on the vibration signal data using a preset synchronous compressed wavelet transform algorithm to obtain the corresponding vibration time-frequency characteristics; Step S22: Perform Fast Fourier Transform and Discrete Cosine Transform on the sound signal to obtain the frequency domain feature matrix; Step S23: Preprocess the temperature signal to eliminate low-frequency trends and noise. The preprocessing includes detrending processing and threshold noise reduction processing. Step S24: The preprocessed temperature signal is transformed using a preset empirical wavelet transform algorithm to obtain temperature fluctuation characteristics.

[0028] Before performing fault diagnosis using vibration, sound, and temperature signal data, targeted preprocessing and time-frequency analysis are essential. Temperature signals often contain slowly changing trends and noise, and directly analyzing their characteristics can mask fault features. Therefore, preprocessing processes such as trend removal and noise reduction are necessary prerequisites for extracting effective fault information. Unlike vibration and sound signals, temperature signals change slowly, and their fault characteristics often manifest as specific low-frequency fluctuation patterns. The Empirical Wavelet Transform (EWT) can adaptively construct filter banks based on the characteristics of the signal itself, making it particularly suitable for extracting weak, fault-related frequency components from such signals, outperforming traditional methods with fixed basis functions.

[0029] Specifically, the vibration signal data is processed primarily using a synchronous compressed wavelet transform algorithm for time-frequency analysis to extract high-resolution time-frequency features. For the sound signal, Fast Fourier Transform and Discrete Cosine Transform are performed to obtain its frequency domain feature matrix. Furthermore, the temperature signal requires preprocessing and time-frequency analysis. Before preprocessing, the temperature signal undergoes detrending processing to eliminate low-frequency trend terms introduced by slow changes in ambient temperature or overall equipment temperature rise, and wavelet threshold denoising is used to suppress measurement noise. Moreover, given the relatively slow temperature change and its nonlinear, non-stationary characteristics, an empirical wavelet transform algorithm is used for time-frequency analysis. Through the technical solution of this application embodiment, frequency bands can be adaptively divided, accurately separating the temperature fluctuation characteristics of specific frequency components caused by faults such as internal friction and abnormal lubrication conditions in the gearbox.

[0030] Step S30: Decompose each initial dynamic feature into fault-related modal shared features and modal specific features, and fuse the modal shared features to obtain a global feature vector; After obtaining the initial dynamic characteristics of each mode, the feature decoupling network can be used to accurately separate the mixed feature signals, and obtain the mode-shared features of the core signal reflecting the essence of the fault and the mode-specific features reflecting the characteristics of each mode sensor.

[0031] It should be noted that before decomposing the initial dynamic features, the preprocessed raw signal needs to be deeply processed by an encoder (such as a convolutional neural network or a fully connected layer). This step aims to transform the raw waveform or data sequence into higher-level, more abstract features. For example, specific frequency components in a vibration signal (such as the sidebands of gear meshing frequencies) or specific resonance peaks in a sound signal are captured by the encoder and transformed into abstract features.

[0032] Furthermore, in a feasible embodiment, the step of decomposing each initial dynamic feature into fault-related modal-shared features and modal-specific features may include: Step S31: Input the abstract features corresponding to each initial dynamic feature into the preset feature decoupling network, and process each initial dynamic feature through the encoder of the feature decoupling network to obtain the corresponding feature vector. Step S32: Input each feature vector into the decoupling layer of the feature decoupling network. The decoupling layer separates the features in each feature vector into two different representation spaces, obtaining the modal shared features and modal specific features corresponding to each feature vector. The modal shared features are used to characterize the fault features of the gearbox, and the modal specific features are used to characterize the inherent background information or inherent noise information of the sensor.

[0033] Specifically, the encoded abstract features are then fed into the decoupling layer at the core of the feature decoupling network. This decoupling layer uses pre-set mechanisms (such as attention mechanisms, gating units, or specialized decoupling loss functions) to forcibly separate the mixed features into two different representation spaces, resulting in modality-shared features and modality-specific features.

[0034] For example, the decoupling layer may include attention mechanisms and gating units, where these mechanisms can learn a weight distribution to dynamically determine which feature dimensions are strongly correlated with faults (and should be classified as modality-shared features) and which feature dimensions are more affected by sensor type or environmental interference (and should be classified as modality-specific features). Dedicated loss functions primarily ensure separation effectiveness by optimizing multiple loss objectives. For example, similarity loss is used to make modality-shared features extracted by different sensors as similar as possible; dissimilarity loss is used to make modality-shared features and modality-specific features as different as possible, avoiding information redundancy.

[0035] To facilitate understanding, modal shared features and modal specific features are further explained. Modal shared features can be understood as the "universal language" or "essential characteristics" of faults. They refer to information components extracted from different gearbox sensor data (vibration, sound, temperature, etc.) that commonly point to the same root cause of the fault. They strip away external factors such as sensor type, installation location, and measurement principle, directly reflecting the abnormal state inside the equipment. For example, whether it's a vibration signal or a sound signal, when a bearing experiences spalling, it will generate impact components related to the fault frequency. Modal shared features capture this core notification that is consistent across modes and directly related to the physical mechanism of the fault. Modal specific features contain various types of interference information, such as: the inherent noise of the sensor itself (e.g., the inherent noise of a current sensor), unique interference from the measurement environment (e.g., background noise collected by a sound sensor), and the inherent differences in the characteristics of different physical quantities (temperature signals change slowly, vibration signals change rapidly). Analyzing modal specific features makes it easier to evaluate the quality and reliability of multi-source signal data. For example, these features can be used to determine whether the current sensor signal is subject to strong interference, thereby dynamically adjusting the diagnostic strategy. For example, when the system detects significant environmental noise from the specific characteristics of sound signals, it can automatically reduce the weight of the sound signals and rely more on vibration signals that are less affected by interference for diagnosis. This enables adaptive and fault-tolerant adjustment of the weights of various modal signal data, thereby improving the accuracy of gearbox fault diagnosis.

[0036] In another feasible embodiment, the step of fusing the shared features of each modality to obtain a global feature vector includes: Step S33: Input the shared features of each modality into a preset neural network or fully connected layer to determine the attention score corresponding to each shared feature of each modality; Step S34: The attention scores corresponding to each modal shared feature are used as the weights corresponding to each modal shared feature. The weighted sum of each modal shared feature is then performed to obtain the global feature vector.

[0037] This application provides a flexible feature fusion strategy that dynamically adjusts the weights of shared features across various modalities to achieve dynamic weighted feature fusion, adapting to various gearbox fault states.

[0038] For example, shared features across modes may include vibration feature vectors. Sound feature vector and temperature eigenvectors Then the vibration feature vector Sound feature vector and temperature eigenvectors The inputs are fed into a pre-trained small neural network or fully connected layer, and attention scores are calculated for each type of feature vector. These attention scores are used to ensure the importance of each type of feature vector in the context. Then, a pre-defined Softmax (normalization exponent) function is used to convert each attention score into a corresponding weight, ensuring that the sum of all weights is 1. Finally, based on the determined weights, the shared features of each modality are weighted and summed to obtain the global feature vector.

[0039] Through the aforementioned feature fusion method, this application achieves dynamic adjustment of the emphasis based on specific contextual information such as vibration signals, sound signals, and temperature signals. For example, when high-frequency vibration characteristics are significant, the weight of the modal sharing features of the vibration signal is automatically increased; when abnormal temperature rise becomes the dominant signal, the weight of the modal sharing features corresponding to the temperature signal is increased.

[0040] Step S40: The global feature vector and modal-specific features are weighted and fused, and then input into the preset fault diagnosis classifier. The fault diagnosis classifier outputs the gearbox fault diagnosis result.

[0041] After obtaining the global feature vector reflecting the characteristics of gearbox faults and the modal-specific features of each modal sensor data, they are weighted and fused. The preset fault diagnosis classifier can be a pre-trained deep learning model that can output the corresponding gearbox fault diagnosis results based on the input feature data. The gearbox fault diagnosis results can include the number of faults, the severity of the faults, and the fault type.

[0042] The weighted fusion process can be performed using fixed weights or dynamically adjusted weights to fuse global feature vectors and modality-specific features.

[0043] Furthermore, in a feasible embodiment, the step of weightedly fusing the global feature vector and modal-specific features and then inputting them into a preset fault diagnosis classifier, and having the fault diagnosis classifier output the gearbox fault diagnosis result, may include: Step S41: Normalize the global feature vector and modal-specific features to obtain the normalized global feature vector and modal-specific features. The purpose of normalization is to eliminate the differences in the dimensions and numerical ranges of different feature vectors, and to ensure the fairness of feature fusion.

[0044] For example, the optimized specific features (e.g., high-frequency resonant components from vibration signals) and the globally consistent feature vector (e.g., multimodal common features that integrate vibration, sound, and temperature) can be processed by max-min normalization or Z-score (Z-score) standardization, respectively.

[0045] Step S42: Determine the weights corresponding to the global feature vector and each modality-specific feature based on the feature information entropy of the global feature vector and each modality-specific feature, wherein the smaller the feature information entropy, the larger the corresponding weight. Among them, feature information entropy reflects the dispersion of feature vectors. The more dispersed the feature vectors, the smaller the information entropy, the richer the discriminative information, and the higher the weight. For each dimension j of the feature vectors and modality-specific features, the feature information entropy of each feature vector or modality-specific feature can be calculated using the following formula. :

[0046] in, It refers to the proportion of the value of the i-th sample in the j-th feature to the total sum of that feature, where n is the number of samples.

[0047] The weight of feature j The calculation formula is:

[0048] Where k is an integer between 1 and m, and m is the total number of feature dimensions.

[0049] Step S43: Based on the weights of the global feature vector and each modal-specific feature, perform a linear weighted summation on the normalized global feature vector and each modal-specific feature to obtain the fused feature vector; The purpose of weighted fusion of modality-specific features and global feature vectors is to adaptively integrate feature information from different modalities. The global feature vector captures common fault features across modalities, while modality-specific features retain the subtle fault features unique to each individual modality. Through weighted fusion, the system can highlight features that contribute more to the current gearbox fault diagnosis task, ultimately forming a robust and discriminative fused feature vector.

[0050] The weights of the global feature vector and each modality-specific feature can be calculated by calculating the attention scores of the global feature vector and the modality-specific features.

[0051] For example, this can be achieved by using the global feature vector Feature vectors of modality-specific features The input is fed into a fully connected layer to obtain the normalized attention score, calculated using the following formula:

[0052] Where Score is the normalized attention score, Softmax is the normalization function, and W and b are learnable parameters.

[0053] The attention mechanism can dynamically adjust the weights based on real-time data to ensure that the weights of the features of each mode match the actual working conditions. For example, when the gearbox suffers local damage, the weights of vibration-specific features (such as impact components) are automatically increased.

[0054] The formula for linearly weighting and summing the normalized eigenvectors according to their weights can be expressed as:

[0055] in, To fuse feature vectors, and The weight coefficients for the model-specific features and the global feature vectors, respectively (which need to satisfy...) + =1).

[0056] In one feasible embodiment, a flexible weight allocation strategy can also be adopted, such as in high-confidence scenarios: if the global consistency feature is stable in historical data (e.g., abnormal meshing frequency is detected across modalities), a weight allocation strategy can be set. > Robustness is improved by relying on common information. In sudden failure scenarios: if a strong anomaly is detected by specific features (such as a momentary scream in a sound signal), robustness is temporarily enhanced. This allows for the capture of unique fault clues. Furthermore, the dimensionality of the fused feature vector remains the same as the original features, but it incorporates weighted information.

[0057] Step S44: Input the fused feature vector into the preset fault diagnosis classifier, and the fault diagnosis classifier outputs the gearbox fault diagnosis result.

[0058] Furthermore, the fault diagnosis classifier includes at least a regression network and a classification network, and the gearbox fault diagnosis result includes at least the fault severity, fault type, and fault location. The step of inputting the fused feature vector into the preset fault diagnosis classifier and having the fault diagnosis classifier output the gearbox fault diagnosis result may include: Step S441: Input the fused feature vector into the regression network, and use the regression network to predict the severity of the gearbox failure. The severity of the failure includes at least one of the following: number of failures, failure area, and failure offset. The regression network is used to map the fused feature vector of the input onto one or more continuous values, thereby quantifying the severity of gearbox failure.

[0059] Specifically, regression networks can learn a nonlinear mapping from high-dimensional features to specific numerical values ​​(such as vibration amplitude, temperature increment, and predicted wear depth) through multi-layer computation (e.g., fully connected layers). It's important to note that the output of a regression network is typically a real number. For example, for gear faults, it might output a value representing the percentage of tooth surface wear or the proportion of pitting area; for bearing faults, it might output a predicted clearance increment (i.e., fault offset) or an effective value of vibration acceleration. These values ​​themselves provide the quantitative information basis for gearbox fault diagnosis, are crucial for predictive maintenance of gearboxes, and can help engineers determine the remaining service life of the equipment and rationally plan maintenance schedules.

[0060] Step S442: Input the fused feature vector into the classification network, which then predicts the fault type and location of the gearbox.

[0061] Specifically, the core task of the classification network in this embodiment is to answer the questions of "what kind of fault" and "where," and is responsible for identifying and distinguishing gearbox fault modes. Specifically, the classification network typically takes a fused feature vector as input, and after a series of calculations, outputs a probability distribution vector through a Softmax function or a Sigmoid function (activation function). Each element of the probability distribution vector represents the probability of belonging to a certain preset fault category (such as "normal," "broken gear tooth," "bearing outer ring fault," etc.).

[0062] For example, in mutually exclusive classification scenarios (such as single fault scenarios or situations where two fault categories cannot exist simultaneously), the Softmax function is used to ensure that the sum of the probabilities of all output elements is 1. Therefore, the fault category in the final fault diagnosis result is determined to be the fault category with the highest probability.

[0063] For example, in complex fault diagnosis scenarios (i.e., multiple faults may occur simultaneously), the Sigmoid function can be used to independently determine the existence of each fault category. The network outputs a probability value between 0 and 1 for each fault category. When the probability of a certain category exceeds a set threshold (e.g., 0.5), the fault is determined to have occurred. This diagnostic method can effectively diagnose complex situations such as "gear pitting and bearing cage fracture occurring simultaneously."

[0064] In another feasible embodiment, after the regression network and the classification network output fault diagnosis results respectively, the fault diagnosis results of the two can be fused to generate a comprehensive and readable gearbox fault diagnosis report.

[0065] In the fault diagnosis results fused from the regression network and classification network, information can be complementary and output accordingly. For example, the classification network determines the nature and location of the fault (e.g., "diagnosed as gear pitting"), while the regression network quantifies the severity of the fault (e.g., "pitting area accounts for approximately 15%, belonging to moderate wear"). After information fusion, a highly readable conclusion text needs to be output, such as "Gearbox has a minor pitting fault (severity level: 2)" or "Bearing outer ring is severely worn; immediate shutdown and inspection are recommended." Simultaneously, the system will include the confidence level of the fault diagnosis result (usually derived from the highest or average probability output by the classification network) for maintenance personnel to reference and make decisions regarding whether immediate repair is necessary.

[0066] In one feasible embodiment, such as Figure 3 As shown, before the step of inputting the weighted fusion of the global feature vector and the modality-specific features into a preset fault diagnosis classifier, the method may further include: Step A10: Obtain the health baseline features of the data sources corresponding to the specific features of each model under the gearbox health state. The health baseline features include at least the mean and standard deviation. Step A20: Calculate the degree of deviation between each modality-specific feature and the health baseline features of each data source, and obtain the residual sequence corresponding to each modality-specific feature; Step A30: Initialize to obtain the initial feature selection threshold. Based on the changing trend of the residual sequence, optimize the initial feature selection threshold to obtain the target feature selection threshold. Specifically, when the changing trend is increasing, decrease the initial feature selection threshold; when the changing trend is stable or fluctuating, increase the initial feature selection threshold. Step A40: Mark the residuals in the residual sequence that are below the target feature selection threshold as the corresponding model-specific features as suppression-specific features; Step A50: Delete the suppressed specific features from each model specific feature to obtain the updated model specific features. The updated model specific features are used for weighted fusion with the global features.

[0067] The purpose of steps A10 to A50 is to optimize the specific features of each model and to dynamically optimize the feature selection threshold by analyzing the changing trend of the residual sequence. This achieves the effect of retaining effective features that are sensitive to early faults while filtering out invalid or interfering features caused by environmental noise, sensor interference, etc.

[0068] First, it is necessary to determine the health baseline features corresponding to each mode of the gearbox under healthy conditions. The health baseline features are reflected in the statistical distribution of model-specific features (such as mean and standard deviation) under normal conditions for each mode. The residual sequence reflects the deviation of each determined modality-specific feature from the monitoring baseline features. The specific calculation method can be obtained by calculating the Z-score.

[0069] Where R(t) is the residual sequence, t is time, and F(t) is the eigenvalue of the modal-specific feature at time point t. The mean value among the health baseline characteristics. This represents the standard deviation of the health baseline characteristics. The above residual sequences represent the degree of abnormality of each modality-specific characteristic compared to the healthy state.

[0070] The initial feature selection threshold can be initialized based on the aforementioned health benchmark features, for example, twice the average. The trend of the residual sequence can be determined by using a preset event sliding window (e.g., including the most recent 50 data points) and statistical residual statistics (e.g., mean, variance, or standard deviation) within that window. By analyzing the residual statistics within each window, it can be determined whether the trend of the residual sequence is stationary, fluctuating randomly, or increasing.

[0071] For example, when a significant upward trend is detected in the residual sequence, the initial feature selection threshold is lowered. This is because even anomalous features with small absolute values ​​may contain important fault information, and lowering the initial feature selection threshold can improve sensitivity and avoid missed detections. When the residuals are stable for a long period or fluctuate randomly, the initial feature selection threshold is raised. This helps suppress noise interference, reduce the false alarm rate, and improve the system's anti-interference capability. Finally, the target feature selection threshold is obtained. The target feature selection threshold is used to filter out modal specific features with relatively large changes in the residual sequence from the current modal specific features. Therefore, the residual sequences corresponding to each modal specific feature can be compared with those greater than the target feature selection threshold. For modal specific features corresponding to residuals below the target feature selection threshold, the changes are small and can be regarded as normal background noise (i.e., suppressed specific features). For modal specific features corresponding to residuals above the target feature selection threshold, the changes are large and can be understood as relatively significant anomalous features. Such features can be retained and passed to subsequent feature fusion and fault diagnosis processes. Filtering out suppressed specific features that have little impact on the fault diagnosis results can reduce the amount of data processing and improve processing efficiency.

[0072] This application also provides a gearbox fault diagnosis device, such as... Figure 4 As shown, the gearbox fault diagnosis device includes: The data acquisition module 10 is used to acquire multi-source data from the gearbox of the wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data and temperature signal data; Feature processing module 20 is used to preprocess and / or perform time-frequency analysis on the vibration signal data, the sound signal data and the temperature signal data respectively to obtain the corresponding initial dynamic features; The feature fusion module 30 is used to decompose each of the initial dynamic features into fault-related modal shared features and modal specific features, and fuse each of the modal shared features to obtain a global feature vector; The fault diagnosis module 40 is used to perform weighted fusion of the global feature vector and each modality-specific feature and input it to a preset fault diagnosis classifier, and the fault diagnosis classifier outputs the gearbox fault diagnosis result.

[0073] The gearbox fault diagnosis device provided in this application adopts the multi-source data fusion gearbox fault diagnosis method in the above embodiments, which can improve the accuracy of gearbox fault diagnosis for wind power generation equipment. Compared with the prior art, the beneficial effects of the gearbox fault diagnosis device provided in this application are the same as those of the multi-source data fusion gearbox fault diagnosis method provided in the above embodiments, and other technical features in this gearbox fault diagnosis device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0074] This application also provides a gearbox fault diagnosis device, which includes at least: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the multi-source data fusion gearbox fault diagnosis method described in the above embodiments.

[0075] The following is for reference. Figure 5 It shows a structural schematic diagram of a gearbox fault diagnosis device suitable for implementing the embodiments of this application. Figure 5 The gearbox fault diagnosis device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0076] like Figure 5As shown, the gearbox fault diagnosis device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the gearbox fault diagnosis device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the gearbox fault diagnosis device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a gearbox fault diagnosis device with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0077] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments of this application.

[0078] The gearbox fault diagnosis device provided in this application adopts the multi-source data fusion gearbox fault diagnosis method in the above embodiments, which can improve the accuracy of gearbox fault diagnosis for wind power generation equipment. Compared with the prior art, the beneficial effects of the gearbox fault diagnosis device provided in this application are the same as those of the multi-source data fusion gearbox fault diagnosis method provided in the above embodiments, and other technical features in this gearbox fault diagnosis device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0079] It should be understood that various parts of the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0080] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of the embodiments of this application. Therefore, the protection scope of the embodiments of this application should be determined by the protection scope of the above claims.

[0081] This application also provides a computer-readable storage medium storing a computer program that can run on a processor. The computer program is used to execute the gearbox fault diagnosis method with multi-source data fusion described in the above embodiments.

[0082] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0083] The aforementioned computer-readable storage medium may be included in the gearbox fault diagnosis device; or it may exist independently and not assembled into the gearbox fault diagnosis device.

[0084] The aforementioned computer-readable storage medium carries one or more programs. When the one or more programs are executed by the gearbox fault diagnosis device, the gearbox fault diagnosis device performs the following actions: collects multi-source data from the gearbox of the wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data; preprocesses and / or performs time-frequency analysis on the vibration signal data, the sound signal data, and the temperature signal data respectively to obtain corresponding initial dynamic features; decomposes each initial dynamic feature into fault-related modal shared features and modal specific features, and fuses each modal shared feature to obtain a global feature vector; weightedly fuses the global feature vector and each modal specific feature and inputs it to a preset fault diagnosis classifier, which then outputs the gearbox fault diagnosis result.

[0085] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0086] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0087] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0088] The computer-readable storage medium provided in this application embodiment stores computer-readable program instructions for executing the gearbox fault diagnosis method based on multi-source data fusion described above, which can improve the accuracy of gearbox fault diagnosis for wind power generation equipment. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application embodiment are the same as the beneficial effects of the gearbox fault diagnosis method based on multi-source data fusion provided in the above embodiments, and will not be repeated here.

[0089] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the gearbox fault diagnosis method with multi-source data fusion as described above.

[0090] The computer program product provided in this application can improve the accuracy of gearbox fault diagnosis for wind power generation equipment. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the gearbox fault diagnosis method with multi-source data fusion provided in the above embodiments, and will not be repeated here.

[0091] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A gearbox fault diagnosis method based on multi-source data fusion, characterized in that, The gearbox fault diagnosis method based on multi-source data fusion includes: Collect multi-source data from the gearbox of a wind power generation device, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data; The vibration signal data, the sound signal data, and the temperature signal data are preprocessed and / or subjected to time-frequency analysis to obtain their respective initial dynamic characteristics. Each of the initial dynamic features is decomposed into fault-related modal shared features and modal specific features, and the modal shared features are fused to obtain a global feature vector; The global feature vector and each modality-specific feature are weighted and fused, and then input into a preset fault diagnosis classifier, which outputs the gearbox fault diagnosis result. The initial dynamic features include at least the vibration time-frequency features corresponding to the vibration signal data, the frequency domain feature matrix corresponding to the sound signal, and the temperature fluctuation features corresponding to the temperature signal; The step of preprocessing and / or performing time-frequency analysis on the vibration signal data, the sound signal data, and the temperature signal data respectively to obtain the corresponding initial dynamic features includes: The vibration signal data is subjected to time-frequency analysis using a preset synchronous compressed wavelet transform algorithm to obtain the corresponding vibration time-frequency characteristics; The sound signal is processed by Fast Fourier Transform and Discrete Cosine Transform to obtain the frequency domain feature matrix; The temperature signal is preprocessed to eliminate low-frequency trends and noise. The preprocessing includes detrending processing and threshold noise reduction processing. The preprocessed temperature signal is transformed using a preset empirical wavelet transform algorithm to obtain temperature fluctuation characteristics; The step of decomposing each of the initial dynamic features into fault-related modal shared features and modal specific features includes: The abstract features corresponding to each of the initial dynamic features are input into a preset feature decoupling network. The encoder of the feature decoupling network processes each of the initial dynamic features to obtain the corresponding feature vector. Each of the aforementioned feature vectors is input to the decoupling layer of the feature decoupling network. The decoupling layer separates the features in each feature vector into two different representation spaces, thereby obtaining the modal shared features and modal specific features corresponding to each feature vector. The modal shared features are used to characterize the fault features of the gearbox, and the modal specific features are used to characterize the inherent background information or inherent noise information of the sensor.

2. The gearbox fault diagnosis method based on multi-source data fusion as described in claim 1, characterized in that, The step of fusing the shared features of each modality to obtain a global feature vector includes: Each modality-shared feature is input into a preset neural network or fully connected layer to determine the attention score corresponding to each modality-shared feature. The attention scores corresponding to each modality-shared feature are used as the weights corresponding to each modality-shared feature. The weighted sums of the modality-shared features are then performed to obtain the global feature vector.

3. The gearbox fault diagnosis method based on multi-source data fusion as described in claim 1, wherein the step of weightedly fusing the global feature vector and each modality-specific feature and then inputting the result to a preset fault diagnosis classifier, and the fault diagnosis classifier outputting the gearbox fault diagnosis result, includes: The global feature vector and each modality-specific feature are normalized to obtain the normalized global feature vector and each modality-specific feature. Based on the global feature vector and the feature information entropy of each modality-specific feature, the weights corresponding to the global feature vector and each modality-specific feature are determined, wherein the smaller the feature information entropy, the larger the corresponding weight. Based on the weights of the global feature vector and each modal-specific feature, a linear weighted sum is performed on the normalized global feature vector and each modal-specific feature to obtain the fused feature vector; The fused feature vector is input into a preset fault diagnosis classifier, which then outputs the gearbox fault diagnosis result.

4. The gearbox fault diagnosis method based on multi-source data fusion as described in claim 3, characterized in that, The fault diagnosis classifier includes at least a regression network and a classification network, and the gearbox fault diagnosis result includes at least the fault severity, fault type, and fault location. The step of inputting the fused feature vector into the preset fault diagnosis classifier and having the fault diagnosis classifier output the gearbox fault diagnosis result includes: The fused feature vector is input into the regression network, which predicts the severity of the gearbox failure. The severity of the failure includes at least one of the following: number of failures, failure area, and failure offset. The fused feature vector is input into the classification network, which then predicts the fault type and location of the gearbox.

5. The gearbox fault diagnosis method based on multi-source data fusion as described in claim 3, characterized in that, Before the step of inputting the weighted fusion of the global feature vector and each of the modality-specific features into a preset fault diagnosis classifier, the method further includes: Obtain the health baseline features of the data sources corresponding to each modality-specific feature under the gearbox health state, wherein the health baseline features include at least the mean and standard deviation; Calculate the degree of deviation between each modality-specific feature and the health baseline features of each data source to obtain the residual sequence corresponding to each modality-specific feature; An initial feature selection threshold is obtained through initialization. Based on the changing trend of the residual sequence, the initial feature selection threshold is optimized to obtain a target feature selection threshold. Specifically, when the changing trend is increasing, the initial feature selection threshold is decreased; when the changing trend is stable or fluctuating, the initial feature selection threshold is increased. Residuals in the residual sequence that are below the target feature selection threshold are labeled as corresponding modality-specific features as suppression-specific features. The suppressed specific features in each of the modality-specific features are deleted to obtain the updated modality-specific features, wherein the updated modality-specific features are used for weighted fusion with the global features.

6. A gearbox fault diagnosis device, characterized in that, The gearbox fault diagnosis device includes: The data acquisition module is used to acquire multi-source data from the gearbox of the wind power generation equipment, wherein the multi-source data includes at least vibration signal data, sound signal data, and temperature signal data. The feature processing module is used to preprocess and / or perform time-frequency analysis on the vibration signal data, the sound signal data, and the temperature signal data respectively to obtain the corresponding initial dynamic features. The feature fusion module is used to decompose each of the initial dynamic features into fault-related modal shared features and modal specific features, and fuse the modal shared features to obtain a global feature vector; The fault diagnosis module is used to perform weighted fusion of the global feature vector and each modality-specific feature and input it to a preset fault diagnosis classifier, which then outputs the gearbox fault diagnosis result. The initial dynamic features include at least the vibration time-frequency features corresponding to the vibration signal data, the frequency domain feature matrix corresponding to the sound signal, and the temperature fluctuation features corresponding to the temperature signal; the feature processing module is further configured to: perform time-frequency analysis on the vibration signal data using a preset synchronous compressed wavelet transform algorithm to obtain the corresponding vibration time-frequency features; perform fast Fourier transform and discrete cosine transform on the sound signal to obtain the frequency domain feature matrix; preprocess the temperature signal to eliminate low-frequency trends and noise, the preprocessing including detrending processing and threshold noise reduction processing; and transform the preprocessed temperature signal using a preset empirical wavelet transform algorithm to obtain the temperature fluctuation features; The feature fusion module is further configured to: input the abstract features corresponding to each of the initial dynamic features into a preset feature decoupling network, process each of the initial dynamic features through the encoder of the feature decoupling network to obtain the corresponding feature vector; input each of the feature vectors into the decoupling layer of the feature decoupling network, separate the features in each of the feature vectors into two different representation spaces through the decoupling layer to obtain the modal shared features and modal specific features corresponding to each feature vector, wherein the modal shared features are used to characterize the fault features of the gearbox, and the modal specific features are used to characterize the inherent background information or inherent noise information of the sensor.

7. A gearbox fault diagnosis device, characterized in that, The gearbox fault diagnosis device includes at least: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the gearbox fault diagnosis method of multi-source data fusion as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for implementing a gearbox fault diagnosis method based on multi-source data fusion, the program being executed by a processor to implement the steps of the gearbox fault diagnosis method based on multi-source data fusion as described in any one of claims 1 to 5.

Images