Rolling bearing fault detection method and system based on mechanism and data fusion

Abstract

The invention discloses a rolling bearing fault detection method and system based on mechanism and data fusion, and relates to the technical field of equipment state monitoring and fault diagnosis, and the method comprises the steps: collecting a multi-source operation state signal and an operation condition parameter of a rolling bearing in real time; constructing and solving oscillatory differential equations under different fault types based on the signals and the parameters in combination with a rolling bearing dynamical model, and generating a mechanism feature library; denoising vibration signals in the multi-source operation state signals, and performing data reconstruction on the denoised vibration signals based on a gamma hybrid model; generating a high-dimensional data feature set according to the reconstructed vibration signal, and performing dimension reduction on the high-dimensional data feature set by adopting a matrix t-SNE algorithm to obtain low-dimensional data features; constructing a joint feature set according to the low-dimensional data features and the mechanism features, and calculating a health index; and the severity and position of the fault are judged based on the joint feature set and the health index, so that the precise detection requirement of the early-stage tiny fault of the rolling bearing under the complex working condition can be met.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of equipment condition monitoring and fault diagnosis technology, and in particular to a rolling bearing fault detection method and system based on mechanism and data fusion. Background Technology

[0002] As the core transmission component of rotating machinery, the operating status of rolling bearings directly determines the reliability of the equipment and the safety of production. According to industrial failure statistics, about 40% of rotating machinery failures are caused by rolling bearing failure. If early minor faults are not detected in time, they may develop into serious faults in a short period of time, leading to an increase of more than 30% in equipment downtime maintenance costs, or even catastrophic consequences such as production interruption and safety accidents.

[0003] Existing rolling bearing fault detection technologies are mainly divided into two categories: mechanism-driven and data-driven. However, both have significant technical bottlenecks under complex industrial conditions. Mechanism-driven methods rely on rolling bearing dynamics and Hertzian contact theory to construct fault models. However, in actual working conditions, equipment vibration signals are often superimposed with noise from base vibration, electromagnetic interference, and multi-component coupled vibration. The theoretical model is difficult to accurately match complex dynamic interference, resulting in the masking of early, minor fault characteristics and a diagnostic accuracy of less than 50%. Furthermore, the mechanism model needs to be re-derived for different bearing models and under different working conditions, resulting in poor adaptability. Data-driven methods suffer from core pain points such as insufficient noise robustness, incomplete feature extraction, poor adaptability to non-Gaussian distributions, and information loss during dimensionality reduction.

[0004] In summary, there is an urgent need for fault detection technologies that can adapt to non-Gaussian fault characteristics, have strong noise robustness, and retain multi-dimensional features in a coordinated manner. Summary of the Invention

[0005] This invention addresses the shortcomings of existing rolling bearing fault detection technologies, such as low accuracy in early fault diagnosis under complex operating conditions, poor noise robustness, insufficient adaptability to non-Gaussian features, and severe loss of dimensionality reduction information. It provides a rolling bearing fault detection method and system based on mechanism and data fusion to meet the need for accurate detection of early minor faults in rolling bearings under complex operating conditions.

[0006] In a first aspect, embodiments of the present invention provide a rolling bearing fault detection method based on mechanism and data fusion, comprising: The rolling bearing's multi-source operating status signals and operating condition parameters are collected in real time using a sensor array, wherein the multi-source operating status signals include at least vibration signals; Based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, vibration differential equations under different fault types are constructed and solved to generate a mechanism feature library; The vibration signal is subjected to wavelet threshold denoising, and the denoised vibration signal is reconstructed by a variational autoencoder based on a gamma mixture model. Multi-domain feature extraction is performed on the reconstructed vibration signal to generate a high-dimensional data feature set, and the matrix-type t-SNE algorithm is used to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features. A joint feature set is constructed based on the low-dimensional data features and the mechanism features in the mechanism feature library that correspond to the current operating condition parameters, and the health index of the joint feature set is calculated in the kernel Markov space. Based on the joint feature set and health index, the severity and specific location of the fault are determined in conjunction with the mechanism feature library.

[0007] In a preferred embodiment, the step of constructing and solving vibration differential equations under different fault types based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, to generate a mechanism feature library includes: Based on the operating condition parameters, determine the system parameters and external excitation in the rolling bearing dynamic model, and construct the vibration differential equation under normal conditions; Based on the multi-source operating status signals and preset different fault types, the vibration differential equation under the normal state is modified to obtain the modified vibration differential equations corresponding to different fault types. The modified vibration differential equation is solved using numerical simulation to obtain simulated vibration response data for each fault type under the current operating conditions. Feature extraction is performed on the simulated vibration response data of each fault type under the current operating conditions to determine the characteristic mechanism features of each type of fault and generate a mechanism feature library.

[0008] In a preferred embodiment, the step of modifying the vibration differential equation under normal conditions based on the multi-source operating state signal and preset different fault types to obtain modified vibration differential equations corresponding to different fault types includes: Based on the multi-source operating status signals, calculate the fault characteristic frequencies corresponding to different fault types; Using the fault characteristic frequency as the core parameter, the impact of different fault types on the bearing system is quantified into a time-varying stiffness perturbation term characterizing the periodic change of system stiffness and an additional external excitation term characterizing the impact of the fault. The vibration differential equation under normal conditions is modified based on the time-varying stiffness perturbation term and the additional external excitation term to obtain the modified vibration differential equation corresponding to each fault type.

[0009] In a preferred embodiment, the vibration signal is subjected to wavelet threshold denoising, and the denoised vibration signal is reconstructed using a variational autoencoder based on a gamma mixture model, including: The vibration signal was decomposed into multiple scales using wavelet transform, and the wavelet coefficients were thresholded to remove noise. The variational autoencoder based on the gamma mixture model maps the denoised vibration signal into intermediate distribution parameters; The intermediate distribution parameters are converted into the distribution parameters of the gamma mixture model through a distribution transformation mechanism; The vibration signal is reconstructed using a decoder based on the converted distributed parameters.

[0010] In a preferred embodiment, multi-domain feature extraction is performed on the reconstructed vibration signal to generate a high-dimensional data feature set, and a matrix-type t-SNE algorithm is used to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features, including: Extract time-domain features, frequency-domain features, and time-frequency-domain features from the reconstructed vibration signal; A high-dimensional data feature set is constructed based on the time-domain features, frequency-domain features, and time-frequency-domain features; Based on the high-dimensional data feature set, the similarity distribution of the sample dimension and the similarity distribution of the feature dimension are defined respectively; Based on the similarity distribution of the sample dimension and the similarity distribution of the feature dimension, the objective function of the matrix-type t-SNE algorithm is solved to obtain low-dimensional data features.

[0011] In a preferred embodiment, the step of solving the objective function of the matrix-type t-SNE algorithm based on the similarity distribution of the sample dimension and the similarity distribution of the feature dimension to obtain low-dimensional data features includes: Construct a convex combination cost function that fuses the similarity distributions based on the similarity distributions of the sample dimension and the feature dimension. By solving the problem of minimizing the convex combination cost function, low-dimensional data features that preserve the sample classification structure and feature association patterns are obtained.

[0012] In a preferred embodiment, a joint feature set is constructed based on the low-dimensional data features and the mechanistic features in the mechanistic feature library corresponding to the current operating condition parameters, and a health index of the joint feature set is calculated in the kernel Markov space, including: Based on the current operating parameters, the corresponding mechanism feature vectors are matched and extracted from the mechanism feature library; The low-dimensional data features are concatenated with the mechanism feature vectors to construct a joint feature set; The joint feature set is mapped to a high-dimensional kernel space using a kernel function; In the high-dimensional kernel space, the Mahalanobis distance between the joint feature vector of the test sample and the healthy sample set is calculated, and the health index of the joint feature set is calculated based on the Mahalanobis distance.

[0013] In a preferred embodiment, the step of determining the severity and specific location of the fault based on the joint feature set and health index, combined with the mechanistic feature library, includes: The joint feature set is input into a pre-trained improved convolutional neural network model, which outputs the probability distribution of fault categories. Based on the probability distribution of the fault categories, the fault category with the highest probability is selected as the fault type identification result; Based on the preset mapping relationship between health index threshold range and fault severity, determine the fault severity level corresponding to the current health index; Based on the fault type identification result, the corresponding fault feature frequency is queried from the mechanism feature library to locate the specific location of the fault.

[0014] In a preferred embodiment, the structure of the improved convolutional neural network model includes, in sequence: The system consists of an input layer, a first convolutional layer, a Leaky ReLU activation layer, a batch normalization layer, a second convolutional layer, a Leaky ReLU activation layer, a batch normalization layer, a max pooling layer, a dropout layer, a first fully connected layer, a Leaky ReLU activation layer, a second fully connected layer, and a softmax output layer.

[0015] Secondly, embodiments of the present invention also provide a rolling bearing fault detection system based on mechanism and data fusion, comprising: The data acquisition module is used to acquire multi-source operating status signals and operating condition parameters of the rolling bearing in real time through a sensor array, wherein the multi-source operating status signals include at least vibration signals; The mechanism feature library generation module is used to construct and solve vibration differential equations under different fault types based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, to generate the mechanism feature library. The data preprocessing module is used to perform wavelet threshold denoising on the vibration signal and to reconstruct the denoised vibration signal based on a variational autoencoder with a gamma mixture model. The data feature set dimensionality reduction module is used to extract multi-domain features from the reconstructed vibration signal to generate a high-dimensional data feature set, and then use a matrix-type t-SNE algorithm to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features. The joint feature set generation module is used to construct a joint feature set based on the low-dimensional data features and the mechanistic features in the mechanistic feature library that correspond to the current operating condition parameters, and to calculate the health index of the joint feature set in the kernel Markov space. The fault location and judgment module is used to determine the severity and specific location of the fault based on the joint feature set and health index, combined with the mechanism feature library.

[0016] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising: One or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the rolling bearing fault detection method based on mechanism and data fusion as described in any embodiment of the present invention.

[0017] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the rolling bearing fault detection method based on mechanism and data fusion as described in any embodiment of the present invention.

[0018] The present invention achieves the following beneficial effects: (1) To address the problem that measured vibration signals contain strong noise and early minor fault features are easily masked, this invention designs a variational autoencoder based on a gamma mixture model. By utilizing the accurate modeling capability of gamma distribution for asymmetric skewed data, the vibration signal is reconstructed to highlight early fault features, thereby improving the fault feature signal-to-noise ratio of the reconstructed signal and solving the problem of distortion in modeling non-Gaussian fault features by traditional autoencoders. (2) To address the technical challenge that the VAE reparameterization technique cannot be directly applied to the gamma distribution, this invention proposes a Gaussian-gamma distribution transformation to achieve stable reparameterization. A Gaussian-gamma distribution transformation mechanism is designed, which converts the Gaussian parameters output by the encoder into gamma parameters through Softmax normalization and adjustment coefficients. This ensures that the latent space features not only conform to the non-Gaussian distribution characteristics of the fault features, but also achieve stable gradient propagation through Gaussian reparameterization, thereby avoiding gradient vanishing during model training and improving the training convergence speed.

[0019] (3) In view of the problem that traditional VAE only focuses on data reconstruction error and ignores the rationality of latent space distribution, this invention proposes ELBO optimization objective to achieve collaborative optimization of reconstruction and distribution. By constructing an ELBO optimization objective that integrates data reconstruction error and KL divergence, the ELBO expression of the gamma mixture model is derived by approximation through double integral. While ensuring the accuracy of vibration signal reconstruction, the matching degree between latent space feature distribution and fault mechanism features is improved by 40%, thus solving the problem of disconnect between reconstruction features and fault mechanism.

[0020] (4) The present invention proposes a matrix-type t-SNE algorithm to achieve accurate dimensionality reduction of high-dimensional features. At the same time, it defines the similarity of sample dimension (row dimension) and feature dimension, effectively filtering out redundant information in high-dimensional features, while retaining key fault differentiation information, thereby improving the model's computational efficiency.

[0021] (5) This invention proposes a convex combination cost function to achieve collaborative preservation of dual-dimensional similarity. It designs a convex combination cost function based on weight coefficients and adaptively determines the optimal value by maximizing the sum of squares between groups / total sum of squares (SSB / SST), ensuring that after dimensionality reduction, different fault types can be clearly distinguished and the fault representation correlation between features can be preserved. Attached Figure Description

[0022] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of the rolling bearing fault detection method based on mechanism and data fusion provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the overall framework of the rolling bearing fault detection method based on mechanism and data fusion provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of matrix-type t-SNE two-dimensional dimensionality reduction provided in an embodiment of the present invention; Figure 4 This is a flowchart of health index calculation and fault diagnosis provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the rolling bearing fault detection system based on mechanism and data fusion provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0024] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0025] Example 1 like Figure 1 The diagram shows a flowchart of a rolling bearing fault detection method 100 based on mechanism and data fusion provided in Embodiment 1 of the present invention. The method 100 specifically includes the following steps: S110. Real-time acquisition of multi-source operating status signals and operating condition parameters of rolling bearings through a sensor array, wherein the multi-source operating status signals include at least vibration signals.

[0026] Specifically, in addition to vibration signals, the aforementioned multi-source operating status signals should also include speed signals, temperature signals, etc. These signals should be acquired in real time using a sensor array (such as vibration acceleration sensors, speed sensors, and temperature sensors). Vibration acceleration sensor is used to collect vibration signals of rolling bearings in the X, Y, and Z directions; A speed sensor is used to monitor the rotational speed of the bearing in real time. Temperature sensors are used to monitor changes in bearing operating temperature.

[0027] The above sensor signals are all synchronously acquired through a data acquisition card. The specific sampling frequency can be set to fs=20×fmax (fmax is the maximum fault characteristic frequency to ensure that the Nyquist sampling theorem is satisfied). The operating parameters of the equipment (such as load R and ambient temperature Tenv) are recorded synchronously. The sampling period is set to 1 second, and the continuous acquisition time covers at least one complete working cycle of the equipment to ensure that the acquired data can fully reflect the operating status of the equipment under different operating conditions.

[0028] S120. Based on the multi-source operating state signals and operating condition parameters, and combined with the rolling bearing dynamics model, construct and solve the vibration differential equations under different fault types to generate a mechanism feature library.

[0029] In some embodiments, the system parameters and external excitation in the rolling bearing dynamics model are first determined based on the above operating condition parameters, and the vibration differential equation under normal conditions is constructed. Next, based on the multi-source operating status signals and the preset different fault types, the vibration differential equation under the normal state is modified to obtain the modified vibration differential equations corresponding to different fault types. Then, the modified vibration differential equation is solved by numerical simulation to obtain the simulated vibration response data of each fault type under the current operating conditions. Finally, feature extraction is performed on the simulated vibration response data of each fault type under the current operating conditions to determine the characteristic mechanism features of each type of fault and generate a mechanism feature library.

[0030] Specifically, based on multi-source operating state signals and preset different fault types, the vibration differential equation under normal conditions is modified to obtain modified vibration differential equations corresponding to different fault types, including: Based on multi-source operating status signals, calculate the fault characteristic frequencies corresponding to different fault types; Using the fault characteristic frequency as the core parameter, the impact of different fault types on the bearing system is quantified into a time-varying stiffness perturbation term characterizing the periodic change of system stiffness and an additional external excitation term characterizing the impact of faults. The vibration differential equation under normal conditions is modified by using the time-varying stiffness perturbation term and the additional external excitation term to obtain the modified vibration differential equation corresponding to each fault type.

[0031] The above content will be explained in detail below with reference to specific vibration differential equations and related parameters. Specifically: (1) First, construct the dynamic model of the bearing under normal conditions, and establish the vibration differential equation based on Newton's second law: in, Equivalent mass of rolling bearing system (unit: kg); Rolling bearing vibration acceleration (unit: ); Damping coefficient of rolling bearing system (unit: ); First derivative of the vibration displacement of a rolling bearing; Equivalent stiffness of rolling bearing system (unit: It mainly consists of the contact stiffness between the bearing rolling elements and the inner and outer rings, and the bending stiffness of the shaft; : Vibration displacement of rolling bearing (unit: m); External excitation force of rolling bearing (unit: N).

[0032] (2) For the preset fault state (such as early spalling of the outer ring), analyze the law of stiffness change caused by the fault, and convert the fault effect into the basic excitation and local stiffness disturbance, and modify the vibration differential equation as follows: in, Fluctuation in rolling bearing stiffness caused by fault (unit: ), over time Changes, such as early spalling failure of the outer ring, can cause periodic abrupt changes in the contact stiffness between the rolling element and the outer ring. The fluctuation frequency of this change is similar to the characteristic frequency of the outer ring failure. Consistency was determined through fault mechanism analysis and fitting of experimental data; Impact excitation of rolling bearing failure (unit: N), as a function of time The change is caused by the local impact of the fault, and its pulse amplitude is positively correlated with the severity of the fault. It can be obtained by modeling the impact signal collected by the fault simulation experiment.

[0033] (3) Solve the modified differential equations through numerical simulation (such as the Runge-Kutta method) to obtain the time-domain waveforms and frequency-domain spectra of the vibration response corresponding to different fault types, and determine the mechanism characteristics (such as the outer ring fault). (Frequency components, impact pulse interval), and generate a mechanism feature library based on the mechanism features obtained from the solution.

[0034] S130. The vibration signal is denoised using wavelet thresholding, and the denoised vibration signal is reconstructed using a variational autoencoder based on a gamma mixture model.

[0035] In some embodiments, the vibration signal is first decomposed into multiple scales using wavelet transform, and the wavelet coefficients are thresholded to remove noise. A variational autoencoder based on a gamma mixture model maps the denoised vibration signal to intermediate distribution parameters. Then, the intermediate distribution parameters are converted into distribution parameters of the gamma mixture model through a distribution transformation mechanism. Finally, the vibration signal is reconstructed using a decoder based on the converted distribution parameters.

[0036] Specifically, firstly, wavelet thresholding is performed on the collected vibration signal. The db4 wavelet basis is selected, the number of decomposition layers is set to 5, and the wavelet coefficients of each layer are processed using a soft thresholding function to obtain the denoised signal.

[0037] Secondly, the variational autoencoder (GamMM-VAE) algorithm based on the gamma mixture model is used to reconstruct the data of the denoised signal. The specific steps are as follows: (a) Constructing the GamMM-VAE encoder The encoder is a 3-layer fully connected network (input dimension) Number of signal sampling points The dimensions of the hidden layers are as follows: The output is a Gaussian distribution parameter. and ), Gamma-mixed prior layers are composed of It consists of multiple gamma distributions, with the probability density function being: in, During the reconstruction of rolling bearing vibration signals, the latent spatial variable The probability density function is used to describe the distribution law of latent features and is the core prior distribution for modeling fault features in the GamMM-VAE algorithm; z: the latent space variable of GamMM-VAE, which corresponds to the low-dimensional abstract features of rolling bearing vibration signals and can capture key fault-related patterns. : The mixing coefficient vector of the gamma mixture model ,in Indicates the first The weights of each gamma component in the hybrid model; The number of components in a gamma-ray hybrid model; latent space variables Dimension index D: latent space variable The total dimension is determined by the number of sampling points of the rolling bearing vibration signal. Determined in relation to feature compression requirements; : No. The first gamma component 3D shape parameters ( This is used to control the asymmetry of the latent feature distribution in this dimension, and can adapt to the non-Gaussian distribution of rolling bearing fault characteristics; : No. The first gamma component dimensional rate parameter ( This is used to control the concentration of the latent feature distribution in this dimension; the more significant the fault features, the more concentrated they become. The larger the value, the more concentrated the distribution; Gamma function in The value at; latent space variables The Dimensional components correspond to the quantized values ​​of a certain type of abstract feature of the rolling bearing vibration signal, such as... An increase may indicate an increase in the intensity of this type of fault characteristic.

[0038] (b) Reparameterization is achieved using Gaussian-Gamma distribution transformation. Gaussian distribution parameters output by the GamMM-VAE encoder Normalized to using the Softmax function And multiply by the adjustment factor , and then combine Convert to gamma parameters: in, Shape parameters of the transformed gamma distribution This is used to model the asymmetric distribution of latent features in rolling bearings. The larger the value, the closer the distribution is to a symmetrical normal distribution, and the more suitable it is for describing significant fault characteristics; Rate parameters of the transformed gamma distribution ( This is used to control the degree of dispersion of the latent characteristic distribution of rolling bearings. The larger the size and the more concentrated the distribution, the better the ability to identify subtle fault characteristics. Gaussian mean of encoder output The result after Softmax normalization has a range of values. ; : Adjustment coefficients for Gaussian-Gamma distribution transformation, used to scale the normalized distribution. Ensure the converted and Within a reasonable range It needs to be calibrated based on the amplitude range and fault characteristic intensity of the rolling bearing vibration signal; The Gaussian distribution log-variance (dimension and variance) of the GamMM-VAE encoder output Consistent) The actual variance is used to describe the uncertainty of the vibration characteristics of rolling bearings. The larger the variance, the more severe the noise interference in this dimension.

[0039] (c) Constructing the GamMM-VAE decoder The decoder is a 3-layer fully connected network (input dimension) Latin spatial dimensions The hidden layer dimensions are 2000, 500, and 500 respectively, and the output dimension is... The network is optimized by minimizing the evidence lower bound (ELBO) composed of the reconstruction error and the KL divergence: in, Monte Carlo sampling times, i.e., for GamMM-VAE latent spatial variables The number of sampling times; Monte Carlo sampling index ; Characteristic dimensions of rolling bearing vibration signals; Vibration signal feature dimension index; The original vibration signal of the rolling bearing was at the first... The true value of the dimension; : No. Under the Monte Carlo sampling, the decoder outputs the first... Dimensional reconstruction of signal values; The rolling bearing vibration signal samples were assigned to the first... Clustering assignment probability of gamma components ,satisfy ), The larger the value, the more it indicates that the sample belongs to the first... The higher the probability of a fault type, the more important it is as a key intermediate parameter for fault classification; D: the dimension of the GamMM-VAE latent space; latent spatial dimension index Each dimension corresponds to a key fault characteristic of rolling bearings, such as Corresponding time-domain impact intensity, Corresponding frequency domain fault frequency energy; : No. The first gamma component 3D shape parameters ( ); : No. The first gamma component dimensional rate parameter ; Gamma function in The value at; : No. The latent feature is the average shape parameter across all gamma components. ; : No. The latent feature is the average rate parameter across all gamma components. ; : Double gamma function in The value at is used to calculate the expected logarithm of the latent feature distribution. It is a key mathematical component for deriving the ELBO gradient and optimizing the model. It is calculated using numerical approximation methods. In the GamMM-VAE variational posterior distribution, the th The shape parameter of the latent feature is used to approximate the true posterior distribution, balancing the exploratory nature of the model with the accuracy of fault feature capture; In the GamMM-VAE variational posterior distribution, the th Rate parameters of the latent feature; Prior mixing coefficients of gamma mixture models ( The probability of rolling bearing failure is determined by statistics and used to guide the model to prioritize learning the feature distribution of high-incidence failures.

[0040] Finally, the reconstructed signal is output through the trained GamMM-VAE decoder, which can effectively highlight fault characteristics (such as impulse pulses) and reduce noise interference.

[0041] S140. Multi-domain feature extraction is performed on the reconstructed vibration signal to generate a high-dimensional data feature set, and the matrix-type t-SNE algorithm is used to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features.

[0042] In some embodiments, time-domain features, frequency-domain features, and time-frequency-domain features are first extracted from the reconstructed vibration signal to construct a high-dimensional data feature set. Then, based on the high-dimensional data feature set, similarity distributions of the sample dimension and feature dimension are defined respectively. A convex combination cost function that fuses the similarity distributions is constructed according to the similarity distributions of the sample dimension and feature dimension. Finally, by solving the minimization problem of the convex combination cost function, low-dimensional data features that retain the sample classification structure and feature association pattern are obtained.

[0043] Specifically, firstly, multi-domain feature extraction is performed on the reconstructed signal obtained in step S130 above: peak value, root mean square, peak-to-peak value, impulse index, and margin index are extracted in the time domain; fault feature frequency amplitude and sideband energy proportion are extracted in the frequency domain; and wavelet entropy and Hilbert marginal spectrum peak value are extracted in the time-frequency domain, forming a high-dimensional data feature set. .

[0044] High-dimensional feature sets are processed using a matrix-based t-SNE algorithm. The specific steps for dimensionality reduction are as follows: (a) Define the similarity between the row dimension (sample dimension) and the column dimension (feature dimension): Row-dimensional high-dimensional similarity: Calculate the Euclidean distance between samples and obtain the conditional probability using a Gaussian kernel function. : in, Rolling bearing vibration signal sample With sample High-dimensional Euclidean distance (unit: ), For the sample High-dimensional feature vectors (including time-domain, frequency-domain, and time-frequency-domain features); : No. The high-dimensional feature vector of each rolling bearing vibration signal sample is extracted during the data preprocessing stage, reflecting the comprehensive fault characteristics of the sample. : No. The high-dimensional feature vector of a rolling bearing vibration signal sample, and Same dimensions; : In the row dimension, samples The variance of the Gaussian kernel function; Rolling bearing vibration signal sample index , The total number of samples ( (This includes samples with normal operation, outer ring faults, and inner ring faults). Rolling bearing vibration signal sample index ; Sample index traversal variables ; Sample index traversal variables ,and Combined with traversing all non- Sample, ensure the denominator is the sample The probability is normalized by summing the conditional probabilities of all other samples.

[0045] Next, based on conditional probability Symmetric generation joint probability : in, :sample For the sample The row-dimensional conditional probability, and Both can reflect the bidirectional similarity between the two samples, avoiding misjudgment of similarity caused by unidirectional probability; :sample For the sample The row-dimensional conditional probability, the calculation logic and They are identical, except for the interchangeable sample indices; Total number of samples of rolling bearing vibration signals ( This is used to normalize the joint probability, ensuring that the sum of the joint probabilities of all sample pairs conforms to the probability distribution requirements.

[0046] Column-dimensional high-dimensional similarity: calculate the cosine distance between features, and similarly obtain the joint probability. : in, Characteristics of rolling bearing vibration signals in column dimension With features The joint similarity probability, with a range of values. The higher the probability, the more similar the two features are in their ability to characterize faults; Features in column dimension Features The conditional probability of considering it as a nearest neighbor; Features in column dimension Features The conditional probability of considering it as a nearest neighbor; The total number of high-dimensional features of the rolling bearing vibration signal is used to normalize the joint probability and ensure that the joint probability of the feature pair meets the distribution requirements. Features in column dimension The variance of the Gaussian kernel function, expressed as perplexity in the column dimension. Sure( ), The larger the value, the stronger the tolerance for similar features, avoiding misjudgment of similarity due to minor differences in features; Rolling bearing vibration signal characteristic index ,and Together they form feature pairs, which are used to measure the similarity of fault representations between features.

[0047] Row-dimensional low-dimensional similarity: defining low-dimensional embeddings Calculate the low-dimensional similarity of the row dimension: in, : Rolling bearing vibration signal samples in row dimension With sample The similarity probability in the low-dimensional embedding space, with a range of values. ; Rolling bearing vibration signal sample In a low-dimensional embedding space, the first Coordinates of each feature dimension The coordinate values ​​reflect the intensity of the fault features of the sample on that low-dimensional feature, corresponding to the feature dimension. Rolling bearing vibration signal sample In a low-dimensional embedding space, the first The coordinates of each feature dimension, and Samples with the same dimension collectively constitute a low-dimensional space. and coordinate pairs; In a low-dimensional embedding space, the sample With sample In the Euclidean distance in each feature dimension (unit: low-dimensional coordinate units). : Feature dimension index of low-dimensional embedding space, the first dimension can correspond to the impact feature intensity, and the second dimension can correspond to the fault frequency energy; The total number of feature dimensions in the low-dimensional embedding space ( (i.e., 2D visualization space), used to traverse all lower dimensions to calculate the sum of squared distances.

[0048] Column-Dimensional Low-Dimensional Similarity: Defining Low-Dimensional Embeddings Calculate low-dimensional similarity of columns: in, Characteristics of rolling bearing vibration signals in column dimension With features The similarity probability in the low-dimensional embedding space, with a range of values. ; Rolling bearing vibration signal sample In a low-dimensional embedding space, the first Coordinates of each feature dimension Reflecting the sample In features Fault characteristic representation; Rolling bearing vibration signal sample In a low-dimensional embedding space, the first The coordinates of each feature dimension, and Having the same dimensions, they collectively constitute the sample. In features and Low-dimensional coordinate pairs; In a low-dimensional embedding space, the sample In features and The Euclidean distance on the two features (unit: low-dimensional coordinate units) indicates that the smaller the distance, the more similar the fault feature expressions of the samples are on the two features; The total number of samples in the rolling bearing vibration signal; traversing all samples can comprehensively reflect the characteristics. and The fault characterization is consistent across all samples.

[0049] (b) Construct a convex combinatorial cost function based on the similarity distribution of the sample dimension and the similarity distribution of the feature dimension. And minimize the solution: in, Weighting coefficients for similarity between row and column dimensions , : Optimal weighting coefficient.

[0050] (c) By solving the minimization problem of the convex combination cost function, low-dimensional data features that preserve the sample classification structure and feature association patterns are obtained. Specifically, the low-dimensional embedding is updated by gradient descent. The gradient formula is: in, Cost function For low-dimensional embedding The partial derivatives; Rolling bearing vibration signal sample In a low-dimensional embedding space, the first Dimensional coordinates; : The coefficient of the row dimension gradient term; : Row-dimensional similarity difference item; In a low-dimensional embedding space, the sample With sample In the The Euclidean distance of dimensionality; :sample and In low dimensions The difference in coordinates of the two dimensions; : Column-dimensional similarity differences; :feature and In low dimensions Coordinate differences on the sample.

[0051] Finally, the low-dimensional data features are obtained. ( (Total number of samples), removing redundant features while retaining key data patterns.

[0052] S150. Construct a joint feature set based on the low-dimensional data features and the mechanism features in the mechanism feature library that correspond to the current operating condition parameters, and calculate the health index of the joint feature set in the kernel Markov space.

[0053] In some embodiments, based on the current operating condition parameters, the corresponding mechanism feature vectors are matched and extracted from the mechanism feature library; the low-dimensional data features are concatenated with the mechanism feature vectors to construct a joint feature set; and the joint feature set is mapped to a high-dimensional kernel space through a kernel function; in the high-dimensional kernel space, the Mahalanobis distance between the joint feature vector of the test sample and the healthy sample set is calculated, and the health index of the joint feature set is calculated based on the Mahalanobis distance.

[0054] Specifically, firstly, the fault characteristic frequency amplitude and vibration mode parameters under the current operating condition are extracted from the mechanism feature library and compared with low-dimensional data features. splicing to construct a joint feature set ( (Mechanism feature vector).

[0055] Furthermore, the joint feature set is computed in the kernel Markov space. Health Index First, the joint features are processed through a kernel function. The data is mapped to the kernel space, the Mahalanobis distance between the sample and the healthy sample set is calculated, and finally the health index is obtained through normalization. (range of values) 1 represents perfect health, 1 represents a serious malfunction.

[0056] S160. Based on the joint feature set and health index, and combined with the mechanism feature library, determine the severity and specific location of the fault.

[0057] In some embodiments, the joint feature set is input into a pre-trained improved convolutional neural network (CNN) model to output the probability distribution of fault categories; based on the probability distribution of fault categories, the fault category with the highest probability is selected as the fault type identification result; according to the mapping relationship between the preset health index threshold range and the fault severity, the fault severity level corresponding to the current health index is determined; based on the fault type identification result, the corresponding fault feature frequency is queried from the mechanism feature library to locate the specific location of the fault.

[0058] Specifically, the structure of the improved convolutional neural network (CNN) model is as follows: input layer (input dimension) Joint feature set (dimensions) First convolutional layer indivual Convolution kernel, stride 1, padding 1 Leaky ReLU activation layer (slope) ) Batch Normalization Layer First convolutional layer indivual Convolution kernel, stride 1, padding 1 LeakyReLU activation layer Batch Normalization Layer Max pooling layer ( Pooling core, step size 2) Dropout layer (dropoutrate) Fully connected layer 1 (256 neurons) Leaky ReLU activation layer Fully connected layer 2 (output dimension) Number of fault categories, such as 4 categories: normal, outer ring fault, inner ring fault, rolling element fault. Softmax layer (outputs the probability of each category).

[0059] The training process for the improved convolutional neural network (CNN) model includes: An improved CNN network is trained using a labeled sample set (covering samples of different fault types and severity). The loss function is multi-class cross-entropy loss. Loss in, The total number of labeled rolling bearing samples used to train and improve CNN networks; The total number of categories for rolling bearing failure classification is determined by both the failure type and severity. The category classification must cover common rolling bearing failure modes in industrial scenarios. Rolling bearing failure category index Each index corresponds to a specific type of fault (e.g.) Corresponding to the normal state, Corresponding to early failures in the outer ring, (corresponding to severe faults in the outer ring), which is the core classification basis for labels and prediction results; Rolling bearing sample Corresponding category The true label uses one-hot encoding—if the sample The actual fault category is ,but The remaining categories corresponding ; Improve CNN networks for sample processing Category The predicted probability is calculated by the Softmax function of the network output layer, satisfying... (as in the sample) In the predicted probability This indicates that the model determines the probability that the sample represents an early outer ring fault. ); The optimizer used during training employs Adam (learning rate). ,momentum Weight decay The training rounds are set to 200, with the first 50 rounds using mini-batch gradient descent (batch size...). In the following 150 rounds, the size of the batch was increased to 64 to stabilize the training.

[0060] The joint feature set collected and processed in real time Input the trained and improved CNN network, output the probability distribution of fault categories, and select the category with the highest probability as the fault type identification result; at the same time, combine the health index. The severity of the fault is determined by comparing the degradation patterns in the mechanism feature library, such as... This is an early-stage failure. This is a mid-term failure. The fault is considered a serious fault, and the specific location of the fault, such as the outer ring or the inner ring, is located based on the transmission law of the fault characteristic frequency.

[0061] According to embodiments of the present invention, the following beneficial effects are achieved: (1) To address the problem that measured vibration signals contain strong noise and early minor fault features are easily masked, this invention designs a variational autoencoder based on a gamma mixture model. By utilizing the accurate modeling capability of gamma distribution for asymmetric skewed data, the vibration signal is reconstructed to highlight early fault features, thereby improving the fault feature signal-to-noise ratio of the reconstructed signal and solving the problem of distortion in modeling non-Gaussian fault features by traditional autoencoders. (2) To address the technical challenge that the VAE reparameterization technique cannot be directly applied to the gamma distribution, this invention proposes a Gaussian-gamma distribution transformation to achieve stable reparameterization. A Gaussian-gamma distribution transformation mechanism is designed, which converts the Gaussian parameters output by the encoder into gamma parameters through Softmax normalization and adjustment coefficients. This ensures that the latent space features not only conform to the non-Gaussian distribution characteristics of the fault features, but also achieve stable gradient propagation through Gaussian reparameterization, thereby avoiding gradient vanishing during model training and improving the training convergence speed.

[0062] (3) In view of the problem that traditional VAE only focuses on data reconstruction error and ignores the rationality of latent space distribution, this invention proposes ELBO optimization objective to achieve collaborative optimization of reconstruction and distribution. By constructing an ELBO optimization objective that integrates data reconstruction error and KL divergence, the ELBO expression of the gamma mixture model is derived by approximation through double integral. While ensuring the accuracy of vibration signal reconstruction, the matching degree between latent space feature distribution and fault mechanism features is improved by 40%, thus solving the problem of disconnect between reconstruction features and fault mechanism.

[0063] (4) The present invention proposes a matrix-type t-SNE algorithm to achieve accurate dimensionality reduction of high-dimensional features. At the same time, it defines the similarity of sample dimension (row dimension) and feature dimension, effectively filtering out redundant information in high-dimensional features, while retaining key fault differentiation information, thereby improving the model's computational efficiency.

[0064] (5) This invention proposes a convex combination cost function to achieve collaborative preservation of dual-dimensional similarity. It designs a convex combination cost function based on weight coefficients and adaptively determines the optimal value by maximizing the sum of squares between groups / total sum of squares (SSB / SST), ensuring that after dimensionality reduction, different fault types can be clearly distinguished and the fault representation correlation between features can be preserved.

[0065] Example 2 Figure 5 This is a schematic diagram of the structure of a rolling bearing fault detection system based on mechanism and data fusion provided in Embodiment 2 of the present invention, as shown below. Figure 5 As shown, the system includes: The data acquisition module 510 is used to acquire multi-source operating status signals and operating condition parameters of the rolling bearing in real time through a sensor array, wherein the multi-source operating status signals include at least vibration signals; The mechanism feature library generation module 520 is used to construct and solve vibration differential equations under different fault types based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, to generate a mechanism feature library. The data preprocessing module 530 is used to perform wavelet threshold denoising on the vibration signal and to reconstruct the data of the denoised vibration signal based on a variational autoencoder with a gamma mixture model. The data feature set dimensionality reduction module 540 is used to extract multi-domain features from the reconstructed vibration signal to generate a high-dimensional data feature set, and to use a matrix-type t-SNE algorithm to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features. The joint feature set generation module 550 is used to construct a joint feature set based on the low-dimensional data features and the mechanism features in the mechanism feature library that correspond to the current operating condition parameters, and to calculate the health index of the joint feature set in the kernel Markov space. The fault location and judgment module 560 is used to determine the severity and specific location of the fault based on the joint feature set and health index, combined with the mechanism feature library.

[0066] The rolling bearing fault detection system based on mechanism and data fusion provided in this embodiment of the invention can execute the rolling bearing fault detection method based on mechanism and data fusion provided in any of the above embodiments of the invention. It has the corresponding functions and beneficial effects of executing the rolling bearing fault detection method based on mechanism and data fusion. For detailed process, please refer to the relevant operations of the rolling bearing fault detection method based on mechanism and data fusion in the foregoing embodiments.

[0067] Example 3 Figure 6 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0068] like Figure 6As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0069] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0070] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 executes the dynamic prediction method for disaster-prone processes based on a multiphysics coupling model described above.

[0071] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0072] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A rolling bearing fault detection method based on mechanism and data fusion, characterized in that, include: The rolling bearing's multi-source operating status signals and operating condition parameters are collected in real time using a sensor array, wherein the multi-source operating status signals include at least vibration signals; Based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, vibration differential equations under different fault types are constructed and solved to generate a mechanism feature library; The vibration signal is subjected to wavelet threshold denoising, and the denoised vibration signal is reconstructed by a variational autoencoder based on a gamma mixture model. Multi-domain feature extraction is performed on the reconstructed vibration signal to generate a high-dimensional data feature set, and the matrix-type t-SNE algorithm is used to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features. A joint feature set is constructed based on the low-dimensional data features and the mechanism features in the mechanism feature library that correspond to the current operating condition parameters, and the health index of the joint feature set is calculated in the kernel Markov space. Based on the joint feature set and health index, the severity and specific location of the fault are determined in conjunction with the mechanism feature library.

2. The method according to claim 1, characterized in that, Based on the multi-source operating state signals and operating condition parameters, and combined with the rolling bearing dynamics model, vibration differential equations under different fault types are constructed and solved to generate a mechanism feature library, including: Based on the operating condition parameters, determine the system parameters and external excitation in the rolling bearing dynamic model, and construct the vibration differential equation under normal conditions; Based on the multi-source operating status signals and preset different fault types, the vibration differential equation under the normal state is modified to obtain the modified vibration differential equations corresponding to different fault types. The modified vibration differential equation is solved using numerical simulation to obtain simulated vibration response data for each fault type under the current operating conditions. Feature extraction is performed on the simulated vibration response data of each fault type under the current operating conditions to determine the characteristic mechanism features of each type of fault and generate a mechanism feature library.

3. The method according to claim 2, characterized in that, The step of modifying the vibration differential equation under normal conditions based on the multi-source operating state signals and preset different fault types to obtain modified vibration differential equations corresponding to different fault types includes: Based on the multi-source operating status signals, calculate the fault characteristic frequencies corresponding to different fault types; Using the fault characteristic frequency as the core parameter, the impact of different fault types on the bearing system is quantified into a time-varying stiffness perturbation term characterizing the periodic change of system stiffness and an additional external excitation term characterizing the impact of the fault. The vibration differential equation under normal conditions is modified based on the time-varying stiffness perturbation term and the additional external excitation term to obtain the modified vibration differential equation corresponding to each fault type.

4. The method according to claim 3, characterized in that, The vibration signal is subjected to wavelet threshold denoising, and the denoised vibration signal is reconstructed using a variational autoencoder based on a gamma mixture model, including: The vibration signal was decomposed into multiple scales using wavelet transform, and the wavelet coefficients were thresholded to remove noise. The variational autoencoder based on the gamma mixture model maps the denoised vibration signal into intermediate distribution parameters; The intermediate distribution parameters are converted into the distribution parameters of the gamma mixture model through a distribution transformation mechanism; The vibration signal is reconstructed using a decoder based on the converted distributed parameters.

5. The method according to claim 4, characterized in that, The reconstructed vibration signal undergoes multi-domain feature extraction to generate a high-dimensional data feature set. A matrix-based t-SNE algorithm is then used to reduce the dimensionality of this high-dimensional data feature set, yielding low-dimensional data features, including: Extract time-domain features, frequency-domain features, and time-frequency-domain features from the reconstructed vibration signal; A high-dimensional data feature set is constructed based on the time-domain features, frequency-domain features, and time-frequency-domain features; Based on the high-dimensional data feature set, the similarity distribution of the sample dimension and the similarity distribution of the feature dimension are defined respectively; Based on the similarity distribution of the sample dimension and the similarity distribution of the feature dimension, the objective function of the matrix-type t-SNE algorithm is solved to obtain low-dimensional data features.

6. The method according to claim 5, characterized in that, The objective function of the matrix-type t-SNE algorithm is solved based on the similarity distribution of the sample dimension and the similarity distribution of the feature dimension to obtain low-dimensional data features, including: Construct a convex combination cost function that fuses the similarity distributions based on the similarity distributions of the sample dimension and the feature dimension. By solving the problem of minimizing the convex combination cost function, low-dimensional data features that preserve the sample classification structure and feature association patterns are obtained.

7. The method according to claim 1, characterized in that, A joint feature set is constructed based on the low-dimensional data features and the mechanistic features in the mechanistic feature library corresponding to the current operating condition parameters, and a health index of the joint feature set is calculated in the kernel Markov space, including: Based on the current operating parameters, the corresponding mechanism feature vectors are matched and extracted from the mechanism feature library; The low-dimensional data features are concatenated with the mechanism feature vectors to construct a joint feature set; The joint feature set is mapped to a high-dimensional kernel space using a kernel function; In the high-dimensional kernel space, the Mahalanobis distance between the joint feature vector of the test sample and the healthy sample set is calculated, and the health index of the joint feature set is calculated based on the Mahalanobis distance.

8. The method according to claim 1, characterized in that, The method of determining the severity and specific location of a fault based on the joint feature set and health index, combined with the mechanistic feature library, includes... The joint feature set is input into a pre-trained improved convolutional neural network model, which outputs the probability distribution of fault categories. Based on the probability distribution of the fault categories, the fault category with the highest probability is selected as the fault type identification result; Based on the preset mapping relationship between health index threshold range and fault severity, determine the fault severity level corresponding to the current health index; Based on the fault type identification result, the corresponding fault feature frequency is queried from the mechanism feature library to locate the specific location of the fault.

9. The method according to claim 8, characterized in that, The structure of the improved convolutional neural network model includes, in sequence: The system consists of an input layer, a first convolutional layer, a Leaky ReLU activation layer, a batch normalization layer, a second convolutional layer, a Leaky ReLU activation layer, a batch normalization layer, a max pooling layer, a dropout layer, a first fully connected layer, a Leaky ReLU activation layer, a second fully connected layer, and a softmax output layer.

10. A rolling bearing fault detection system based on mechanism and data fusion, characterized in that, include: The data acquisition module is used to acquire multi-source operating status signals and operating condition parameters of the rolling bearing in real time through a sensor array, wherein the multi-source operating status signals include at least vibration signals; The mechanism feature library generation module is used to construct and solve vibration differential equations under different fault types based on the multi-source operating state signals and operating condition parameters, combined with the rolling bearing dynamics model, to generate the mechanism feature library. The data preprocessing module is used to perform wavelet threshold denoising on the vibration signal and to reconstruct the denoised vibration signal based on a variational autoencoder with a gamma mixture model. The data feature set dimensionality reduction module is used to extract multi-domain features from the reconstructed vibration signal to generate a high-dimensional data feature set, and then use a matrix-type t-SNE algorithm to reduce the dimensionality of the high-dimensional data feature set to obtain low-dimensional data features. The joint feature set generation module is used to construct a joint feature set based on the low-dimensional data features and the mechanistic features in the mechanistic feature library that correspond to the current operating condition parameters, and to calculate the health index of the joint feature set in the kernel Markov space. The fault location and judgment module is used to determine the severity and specific location of the fault based on the joint feature set and health index, combined with the mechanism feature library.

Images