Device predictive maintenance method based on deep learning

Abstract

The invention relates to the field of equipment diagnosis, and particularly discloses an equipment predictive maintenance method based on deep learning, and the method comprises the steps: carrying out the time domain amplitude normalization and frequency domain weighted normalization of training data, and carrying out the dual-channel feature fusion, so as to obtain a fusion feature vector; performing feature extraction on the fused feature vector by using multiple groups of self-adaptive wavelet kernels to obtain a self-adaptive time-frequency feature vector; constructing a weight population through statistical characteristics, and screening the optimal initial weight of the initial diagnosis model; carrying out multi-scale depth feature calculation, time-frequency domain attention feature fusion, fault prototype comparative learning and a pre-constructed total loss function on the adaptive time-frequency feature vector, and carrying out iterative updating on the initial diagnosis model to obtain a diagnosis model; and the equipment is diagnosed through the diagnosis model. Multi-scale feature fusion and a double-path attention mechanism can cooperatively capture short-time impact and a long-period mode, and the limitation of a traditional method in diversified fault scenes is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of equipment diagnosis, and in particular to a device predictive maintenance method based on deep learning. BACKGROUND

[0002] In recent years, with the large-scale, complex and continuous operation of modern industrial equipment, higher requirements are put forward for the reliability and safety of equipment operation. The traditional equipment maintenance mode depends on periodic maintenance or after-service maintenance, which not only increases the maintenance cost, but also may cause shutdown due to sudden failure, resulting in serious economic loss. In order to realize early fault warning and health state evaluation of equipment, predictive maintenance has gradually become a research hotspot in the industry.

[0003] The prior art has the following disadvantages: 1. Global Min-Max or single-domain normalization is commonly used, which can easily weaken the fault-sensitive frequency band features, so that the features are overwhelmed by noise, and the diagnosis accuracy is low.

[0004] 2. The traditional wavelet transform relies on manual preset parameters, and cannot flexibly adapt to diversified fault modes, resulting in insufficient extraction of key time-frequency features.

[0005] 3. Randomly initialized weights can easily cause unstable convergence, fixed convolution kernel scales cannot take into account multi-time scale features, and traditional LSTM treats irrelevant noise equally, which reduces the diagnosis effect.

[0006] 4. The conventional classifier has low discrimination for similar fault categories and poor adaptability to new categories; at the same time, it is difficult to take into account the needs of the early and late stages of training with fixed regularization strength, which can easily lead to overfitting or underfitting. SUMMARY

[0007] In order to solve the above problems, the present application provides a device predictive maintenance method based on deep learning, wherein the method comprises: After the training data is subjected to time-domain amplitude normalization and frequency-domain weighted normalization, dual-channel feature fusion is performed to obtain a fusion feature vector; the Morlet wavelet basis function parameters are optimized, and a plurality of adaptive wavelet kernels are used to extract features from the fusion feature vector to obtain an adaptive time-frequency feature vector; a weight population is constructed through the statistical characteristics of the adaptive time-frequency feature vector, and the optimal initial weight of the initial diagnosis model is screened in combination with a differential evolution algorithm; the initial diagnosis model is iteratively updated through multi-scale deep feature calculation, time-frequency domain attention feature fusion, fault prototype comparison learning and a pre-constructed total loss function on the adaptive time-frequency feature vector to obtain a diagnosis model; and the diagnosis model is used to diagnose the fault of the equipment to be tested.

[0008] In one example, before the time-domain amplitude normalization and the frequency-domain weighted normalization are performed on the training data, the method further comprises: collecting original vibration signals of the device in a running state at a preset sampling frequency within a preset time length through a preset vibration sensor preset on the device, and synchronously recording device working condition parameters as initial sample data; the device working condition parameters include at least one of a rotating speed, a load, and a temperature; and labeling the initial sample data in combination with a device running log, a disassembly and repair report, and an oil analysis result, wherein the labeling categories include a fault mode and a health mode, and the fault mode includes at least one of a rotor imbalance, an axis misalignment, a bearing inner ring damage, a bearing outer ring damage, a gear tooth breakage, and a gear wear.

[0009] In one example, after the time-domain amplitude normalization and the frequency-domain weighted normalization are performed on the training data, two-channel feature fusion is performed, specifically including: performing amplitude normalization processing on the original vibration signals to obtain a time-domain normalized feature vector; performing fast Fourier transform on the original vibration signals to obtain an amplitude spectrum; performing weighting on the amplitude spectrum through a frequency weighting matrix to suppress high-frequency noise and enhance a fault-sensitive frequency band; performing logarithmic operation on the weighted result to compress the amplitude range and improve the feature separability to obtain a frequency-domain normalized feature; and performing a splicing operation on the time-domain normalized feature vector and the frequency-domain normalized feature vector to generate a fusion feature vector.

[0010] In one example, the Morlet wavelet basis function parameters are optimized, specifically including: based on a standard Morlet wavelet basis function, combining a scale parameter to control wavelet stretching, a translation parameter to control wavelet time shift, and a window function weight parameter to adjust the steepness of a Sigmoid window to construct a parameterized adaptive wavelet basis function.

[0011] In one example, the fusion feature vector is subjected to feature extraction using multiple sets of adaptive wavelet kernels to obtain an adaptive time-frequency feature vector, specifically including: calculating the energy of each adaptive wavelet basis function, and converting the wavelet kernel energy into a weighted coefficient through normalization processing; performing convolution operation on the fusion feature vector using multiple sets of adaptive wavelet kernels, weighting and summing the convolution results according to the energy weight coefficients after taking the modulus, and outputting an initial adaptive time-frequency feature vector; taking the variance of improving the feature distribution of different fault categories as an optimization target, and dynamically optimizing the adaptive wavelet kernel parameter set by using an evolutionary algorithm to output the adaptive time-frequency feature vector; the adaptive wavelet kernel parameter set includes at least one of a wavelet kernel scale parameter, a translation parameter, and a window weight.

[0012] In one example, the constructing a weight population through statistical characteristics of the adaptive time-frequency feature vector and screening an optimal initial weight of the initial diagnosis model in combination with a differential evolution algorithm specifically includes: performing mean vector calculation on the adaptive time-frequency feature vector to obtain a feature mean vector; performing singular value decomposition on a covariance matrix of the adaptive time-frequency feature vector to obtain a left singular matrix and a singular value diagonal matrix; generating a diversity initial weight population in combination with a random orthogonal matrix based on the feature mean vector, the left singular matrix and the singular value diagonal matrix; performing differential mutation operation on a target individual whose individual loss value is greater than a benchmark individual loss value through a mutation factor and a plurality of individual weights selected randomly to obtain an evolved weight population; evaluating loss values of all individuals in the evolved weight population and selecting a weight corresponding to an individual with the minimum loss value as the optimal initial weight of the initial diagnosis model.

[0013] In one example, the performing multi-scale deep feature calculation on the adaptive time-frequency feature vector specifically includes: performing one-dimensional convolution operation and multi-dimensional convolution operation on the adaptive time-frequency feature vector using convolution kernels of multiple sizes to extract a short-scale feature vector and a long-scale feature vector; splicing the short-scale feature vector and the long-scale feature vector and sequentially performing batch normalization processing and Gaussian error linear unit activation function operation to output a multi-scale fusion feature.

[0014] In one example, the performing multi-scale deep feature calculation on the adaptive time-frequency feature vector and time-frequency domain attention feature fusion specifically includes: converting the multi-scale fusion feature into a query matrix through a learnable query transformation matrix, processing the query matrix through a normalized exponential function and weighting the time domain feature to output a time domain attention feature vector; performing fast Fourier transform on the multi-scale fusion feature to obtain a frequency domain representation, calculating a frequency domain attention weight through amplitude weighting and a normalized exponential function, and outputting a frequency domain attention feature vector through inverse fast Fourier transform; adding the time domain attention feature vector and the frequency domain attention feature vector and outputting a fusion time-frequency domain attention feature vector through layer normalization operation.

[0015] In one example, the performing multi-scale deep feature calculation on the adaptive time-frequency feature vector, time-frequency domain attention feature fusion and fault prototype contrast learning specifically includes: inputting the fusion time-frequency domain attention feature vector into a multi-layer perceptron projection network to convert it into a low-dimensional projection feature vector; weighting historical fault prototype vectors based on a momentum coefficient, combining the mean value of the low-dimensional projection feature vectors of the same type of samples in the current batch, and updating the prototype vectors of each type of fault; calculating the cosine distance between the sample projection feature and the real class prototype vector, combining the distance between the sample projection feature and the prototype vectors of other classes, and obtaining a prototype contrast loss through exponential function and natural logarithm operation.

[0016] In one example, the initial diagnosis model is iteratively updated by multi-scale deep feature calculation, time-frequency domain attention feature fusion, fault prototype contrast learning and pre-constructed total loss function on the adaptive time-frequency feature vector to obtain a diagnosis model, specifically including: on the basis of the basic loss and the prototype contrast loss, a total loss function is constructed by combining an L2 regularization term decaying over time, wherein the regularization strength exponentially decays with the increase of the training round; new parameter values are generated using a multivariate normal distribution, and the decay rate parameter of the dynamic regularization coefficient is optimized by a covariance matrix adaptive evolution strategy; parent individuals are selected according to the fitness ranking, and the mean vector of the evolution distribution is updated by weighted average; based on the total loss function, an adaptive matrix estimator optimizer is used to perform iterative update of the model parameters.

[0017] The method proposed in the application can bring the following beneficial effects: 1. The time domain amplitude feature and the frequency domain weighted feature are spliced and fused, which not only retains the original vibration amplitude feature, but also enhances the fault sensitive frequency band feature, significantly improving the feature discrimination.

[0018] 2. The Morlet wavelet base function parameters are optimized based on the evolution algorithm, a plurality of wavelet kernels are dynamically constructed and weighted fused, and adaptive time-frequency feature capture for different fault modes is realized.

[0019] 3. The weight initialization strategy optimized by differential evolution, the multi-scale convolution fusion structure and the double-path attention mechanism are proposed, so that the model can stably converge and take into account short-time impact features and long-period patterns.

[0020] 4. The prototype representation and contrast loss function are updated by momentum to improve small sample diagnosis capability, and a regularization strategy exponentially decaying with the training process is introduced, taking into account early learning and late overfitting prevention. BRIEF DESCRIPTION OF DRAWINGS

[0021] The drawings described herein are used to provide further understanding of the application, and form a part of the application. The illustrative embodiments of the application and their descriptions serve to explain the application, and do not constitute an improper limitation on the application. In the drawings: Figure 1 a flowchart of a predictive maintenance method for equipment based on deep learning in an embodiment of the application; Figure 2 a schematic diagram of several original vibration signals in an embodiment of the application; Figure 3 a schematic diagram of several conventional frequency spectrum analyses in an embodiment of the application; Figure 4 a schematic diagram of the influence of different normalization methods on fault diagnosis accuracy in an embodiment of the application; Figure 5The diagnostic performance of different methods in various embodiments of the present application on various faults is shown in the diagram. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be described below in detail with reference to specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0023] The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the drawings.

[0024] Figure 1 A flowchart of a deep learning-based equipment predictive maintenance method provided by one or more embodiments of the present application is shown. The method can be applied to different types of industrial equipment, such as bearings, gearboxes, or motors, etc. The flowchart can be executed by a computing device (such as a control module or a cloud server, etc.) connected to the industrial equipment. Some input parameters or intermediate results in the flowchart allow manual intervention to adjust to help improve accuracy.

[0025] The analysis method related to the embodiments of the present application can be implemented by a terminal device or a server, and the present application does not make special limitations thereon. For the convenience of understanding and description, the following embodiments are described in detail with the server as an example. It should be noted that the server can be a separate device or a system composed of multiple devices, i.e., a distributed server, and the present application does not make specific limitations thereon.

[0026] As shown in Figure 1 The embodiments of the present application provide a deep learning-based equipment predictive maintenance method, which includes: S101: After time-domain amplitude normalization and frequency-domain weighted normalization of the training data, double-channel feature fusion is performed to obtain a fusion feature vector.

[0027] First, the diagnostic model needs to be trained according to different types of equipment to be measured. When training the diagnostic model, first, the training data is obtained, and after time-domain amplitude normalization and frequency-domain weighted normalization of the training data, double-channel feature fusion is performed to obtain a fusion feature vector.

[0028] In one embodiment, when acquiring training data, a three-axis vibration sensor needs to be deployed at a critical part of the device (such as a bearing, a gear box or a motor shaft end) to continuously collect original vibration signals under the running state of the device at a sampling frequency of not less than 20 kHz; the collection time length of each initial sample data is 0.2 seconds, and an original vibration signal vector with a dimension of 4096 is generated, and the device working condition parameters (such as rotating speed, load, temperature) are recorded synchronously.

[0029] When data labeling is performed on the initial sample data, it is completed by a domain expert in combination with a device running log, a disassembly and repair report and an oil analysis result, the labeling categories cover six typical fault modes (including rotor imbalance, shaft misalignment, bearing inner ring damage, bearing outer ring damage, gear tooth breakage and gear wear) and a health state, and a sample library with labels is formed.

[0030] All labeling processes need to ensure that the fault modes strictly correspond to the physical mechanism of the vibration characteristics, for example, the bearing inner ring damage labeling needs to correspond to the feature frequency sideband modulation phenomenon, and finally a standardized training data set containing time domain waveforms, working condition parameters and multi-category fault labels is constructed.

[0031] In one embodiment, the original signals and time-frequency spectrums of the health state, bearing inner ring damage and gear tooth breakage are as shown in Figure 2 The conventional frequency spectrum analysis is as shown in Figure 3 It is difficult to distinguish the characteristic frequencies of similar faults.

[0032] In one embodiment, there is a problem of mixing high-frequency noise and low-frequency fault characteristics in the vibration sensor monitoring data, and the conventional processing method such as Min-Max normalization adopts a global normalization method to eliminate the dimension influence, but it will weaken the feature performance of the fault related frequency band, resulting in that the fault characteristics are submerged by the noise and affecting the subsequent diagnosis accuracy.

[0033] The application realizes the calculation of the time domain normalized feature vector and the calculation of the frequency domain normalized feature vector, and then fuses them into a high discriminative feature vector through a splicing operation, and the specific steps are as follows: First, the original vibration signal is subjected to amplitude normalization processing to obtain a time domain normalized feature vector. Specifically, the time domain normalized feature vector with a value range of 0 to 1 is obtained by subtracting the minimum value of the vector and dividing by the difference between the maximum value and the minimum value, eliminating the dimension influence while retaining the original waveform amplitude feature, and is expressed as:

[0034] Among them, represents the original vibration signal vector, with a dimension of 4096; represents the minimum value of the vector represents the maximum value of the vector represents the maximum value of the vector the maximum value of the time domain normalized feature vector; is a time domain normalized feature vector, and the value range of .

[0035] Then the original vibration signal is subjected to frequency domain weighted normalization to output a frequency domain normalized feature vector. Specifically, a fast Fourier transform is performed on the original vibration signal vector to obtain an amplitude spectrum, the amplitude spectrum is weighted by a frequency weighting matrix to suppress high-frequency noise and enhance fault-sensitive frequency bands, and a logarithmic operation is performed on the weighted result to compress the amplitude range and improve feature separability, and finally a frequency domain normalized feature vector is output, denoted as:

[0036] In the formula, denotes a fast Fourier transform operation; denotes the amplitude spectrum of the fast Fourier transform result; is a frequency weighting matrix, defined as a diagonal matrix ; denotes an operator for constructing a diagonal matrix; denotes a-0.5 power operation on the first frequency component after Fourier transform; denotes a-0.5 power operation on the second frequency component after Fourier transform; denotes a-0.5 power operation on the 2048th frequency component after Fourier transform; is a logarithmic function with base 10, used to compress the amplitude range and improve feature separability; is a frequency domain normalized feature vector, with a dimension of 2048.

[0037] Then the time domain normalized feature vector and the frequency domain normalized feature vector need to be spliced to generate a fusion feature vector, denoted as:

[0038] In the formula, denotes a vector splicing operation; is a fusion feature vector.

[0039] Different normalization methods are analyzed and compared to evaluate the performance difference between the frequency-time domain hybrid normalization method proposed in the present application and traditional normalization techniques. For example, Figure 4The comparison results of four normalization methods on fault diagnosis accuracy are shown: the global normalization as the conventional technology performs the weakest; the time domain normalization (only retains the original waveform amplitude feature) and the frequency domain normalization (only suppresses high frequency noise) each has an improvement but the effect is limited; and the mixed normalization method of the application significantly improves the diagnosis accuracy by fusing the time domain amplitude feature and the frequency domain weighted feature. It can be seen in the figure that the accuracy column chart of the mixed normalization method is significantly higher than that of other methods, and the error bar length is the shortest, indicating that its diagnosis result stability is the best. The unit percentage of the vertical coordinate "diagnosis accuracy (%)" directly reflects the technical advantages of the method in eliminating the influence of dimension and enhancing the fault feature performance, solving the core problem that the fault feature is overwhelmed by noise in the conventional method.

[0040] S102: Optimize the Morlet wavelet base function parameters, use multiple adaptive wavelet kernels to extract features from the fusion feature vector to obtain an adaptive time-frequency feature vector.

[0041] The conventional wavelet transform needs to preset the base function parameters such as scale and translation, which is difficult to adapt to the time-frequency feature expression requirement under the variable fault mode, resulting in insufficient fault feature extraction and unable to effectively capture the fault sensitive information.

[0042] The application dynamically optimizes the wavelet base function parameters through the evolutionary algorithm to construct an adaptive wavelet kernel group, and the specific steps are as follows: first, the Morlet wavelet base function parameters need to be optimized, specifically, based on the standard Morlet wavelet base function, the scale parameter is used to control the wavelet stretching, the translation parameter is used to control the wavelet time shift, and the window function weight parameter is used to adjust the steepness of the Sigmoid window, to construct a parameterized adaptive wavelet base function, which is represented as:

[0043] In the formula, Adaptive wavelet base function is represented as Standard Morlet wavelet base function is represented as Scale parameter is represented as , control wavelet stretching Translation parameter is represented as Window function weight parameter is represented as Sigmoid activation function is represented as Time variable is represented as

[0044] After the wavelet basis function parameters are dynamically optimized by the evolutionary algorithm, a plurality of adaptive wavelet kernels are used to extract features from the fusion feature vector to obtain an adaptive time-frequency feature vector. At this time, the wavelet kernel energy weight needs to be calculated first, that is, the energy of each adaptive wavelet basis function is calculated, and the wavelet kernel energy is converted into a weighting coefficient through normalization processing, which is used for subsequent feature fusion and is expressed as:

[0045] In the formula, denotes the normalized energy weight of the th wavelet kernel; denotes the energy of the th wavelet kernel, and the calculation method is expressed as . denotes the energy of the th wavelet kernel; is the total number of wavelet kernels, such as . denotes the th adaptive wavelet basis function; is the th adaptive wavelet basis function.

[0046] Then the time-frequency feature needs to be extracted, that is, a plurality of adaptive wavelet kernels are used to perform convolution operation on the fusion feature vector, and the adaptive time-frequency feature vector is output after the convolution result is weighted and summed according to the energy weight coefficient, which is expressed as:

[0047] In the formula, is the adaptive time-frequency feature vector; denotes the convolution operation.

[0048] Finally, the wavelet basis parameters can be optimized. At this time, the evolutionary algorithm can be used to dynamically optimize the parameter set , so that the variance between fault classes is maximized. The optimization objective is to maximize the variance of the feature distribution of different fault classes, and the optimization parameters include the scale of the th wavelet kernel, the translation of the th wavelet kernel, and the window weight of the th wavelet kernel. Specifically, the standard evolutionary algorithm is used to update the parameter population through mutation, crossover and selection operations. The termination condition is to reach the preset evolution number or the fitness converges, and the adaptive time-frequency feature vector corresponding to the optimal parameter set is output .

[0049] S103: Construct a weight population by statistical characteristics of the adaptive time-frequency feature vector, and screen the optimal initial weight of the initial diagnosis model in combination with the differential evolution algorithm.

[0050] If the deep model adopts completely random weight initialization in the initial training stage, the training process will often be in shock or even unstable convergence due to the mismatch between the weight distribution and the input features, especially in the case of high-dimensional vibration features, random initialization is difficult to reflect the statistical law of the features, thereby prolonging the model convergence time and reducing the final diagnosis performance.

[0051] The application constructs a weight population by utilizing the statistical characteristics of the adaptive time-frequency feature, and screens the optimal initial weight in combination with the differential evolution algorithm, and the specific steps are as follows: First, the mean vector of the adaptive time-frequency feature vector is calculated, and the left singular matrix and the singular value diagonal matrix are obtained by performing singular value decomposition on the covariance matrix, which is represented as:

[0052]

[0053] In the formula, The mean value of the feature vector is calculated; The feature mean vector is calculated; The feature covariance matrix is calculated; The singular value decomposition operation is represented as: The left singular matrix is represented as: The singular value diagonal matrix is represented as.

[0054] Then, based on the feature mean vector, the left singular matrix and the singular value diagonal matrix, a diversity initial weight population is generated in combination with a random orthogonal matrix, which is represented as:

[0055] In the formula, The initial weight matrix of the first individual is represented as: The first random orthogonal matrix is represented as: All individual indexes are traversed, ; The population size is represented as: .

[0056] Then, for the individual whose current individual loss value is greater than the reference individual loss value, the evolutionary weight is generated by differential mutation operation of the mutation factor and three individual weights selected at random, otherwise the original weight is retained, which is represented as: ​​​​

[0057] wherein, represents the evolutionary weight of the i-th individual; represents the evolutionary weight of the i-th individual; is a randomly selected first individual index; is a randomly selected second individual index; is a randomly selected third individual index; is a mutation factor, such as, ; is a loss function, which is calculated as a cross-entropy loss based on a 5% data subset.

[0058] Finally, the loss values of all individuals in the evolutionary weight population can be evaluated, and the weight corresponding to the individual with the smallest loss value is selected as the optimal initial weight, denoted as:

[0059] wherein, is the optimal initial weight; represents finding the parameters that minimize the objective function; is the i-th evolutionary individual; is the loss value of the i-th evolutionary individual.

[0060] S104: The initial diagnosis model is iteratively updated by performing multi-scale deep feature calculation, time-frequency domain attention feature fusion, fault prototype comparison learning, and a pre-constructed total loss function on the adaptive time-frequency feature vector, to obtain a diagnosis model.

[0061] Traditional convolutional neural networks use fixed-size convolution kernels, which can only extract features at a single scale. For vibration signals, both short-time impact features and long-period patterns are included. Single-scale convolution kernels often cannot capture features at different time scales at the same time, resulting in insufficient model capabilities in multi-scale feature expression.

[0062] The present application provides a double-branch convolution structure that extracts short-scale local features and long-scale global features, respectively, and further generates multi-scale deep features that combine local and global information. The specific steps are as follows: When performing short-scale feature extraction, a small-size convolution kernel can be used for one-dimensional convolution operation on the adaptive time-frequency feature vector to extract a short-scale feature vector, denoted as:

[0063] wherein, represents the short-scale feature vector; represents one-dimensional convolution operation; is the convolution kernel size, such as, ; is the convolution step size, such as, .

[0064] In the long-scale feature extraction, a large-size convolution kernel is used for the adaptive time-frequency feature vector to perform a dilated convolution operation, and the receptive field expansion degree is controlled by a dilated rate, and a long-scale feature vector is output, denoted as:

[0065] In the formula, denotes the output long-scale feature vector; denotes the dilated convolution operation; is the dilated rate, such as , and the receptive field expansion degree is controlled.

[0066] Then, the short-scale feature vector and the long-scale feature vector are spliced, and batch normalization processing and Gaussian error linear unit activation function operation are sequentially performed, and a multi-scale fusion feature is output, denoted as:

[0067] In the formula, is the output multi-scale fusion feature; denotes the batch normalization processing; is the Gaussian error linear unit activation function.

[0068] The traditional long short-term memory network treats all time or frequency components equally when processing feature sequences. However, in the fault diagnosis scenario, key features are often concentrated in certain time segments or frequency components. Equal treatment will cause irrelevant noise to occupy a large proportion, thereby reducing the diagnosis accuracy.

[0069] The application introduces a double-channel attention mechanism based on the fusion feature, and models the time domain and frequency domain respectively, and the specific steps are as follows: First, the time domain attention is calculated, specifically: the multi-scale fusion feature is converted into a query matrix through a learnable query transformation matrix, the self-attention weight is calculated combined with the key matrix and the value matrix, and the time domain attention feature is output after being weighted with the time domain feature after normalization exponential function processing, denoted as:

[0070] In the formula, denotes the time domain attention output feature; is the learnable query transformation matrix; is the query matrix, which is obtained by linear transformation of the input feature, and the calculation method is represented as ; is the key matrix; is the value matrix; is the scaling factor, such as , to prevent the dot product from being too large; represents the normalized exponential function; express The transpose of .

[0071] Then the frequency domain attention calculation needs to be performed. Specifically, the multi-scale fusion features are fast Fourier transformed to obtain the frequency domain representation, the frequency domain attention weight is calculated by amplitude weighting and normalized exponential function, and the frequency domain attention feature is output by inverse fast Fourier transform, which is expressed as:

[0072] Where, Represents the frequency domain attention output feature; Represents element-wise multiplication; represents the inverse fast Fourier transform.

[0073] Then the attention feature fusion is performed, specifically: the time domain attention feature and the frequency domain attention feature are added, and the fused time-frequency domain attention feature vector is output through the layer normalization operation, which is expressed as:

[0074] Where, To fuse the time-frequency domain attention feature vector; Representation layer normalization operation.

[0075] Because in the small-sample fault diagnosis scenario, traditional classifiers have difficulty in effectively distinguishing similar fault modes and have poor adaptability to new fault categories, the recognition accuracy under small-sample conditions will be significantly reduced.

[0076] To this end, this application adopts the momentum update mechanism to construct the fault prototype representation, and uses the contrast loss function to bring the features of similar samples closer to their corresponding prototypes, while pushing the distance between prototypes of different categories farther. First, feature projection transformation is required. The fused time-frequency domain attention feature vector is input into the multi-layer perceptron projection network and converted into a low-dimensional projection feature vector, which is expressed as:

[0077] Where, Indicates the The fused time-frequency domain attention feature vector corresponding to the samples; Represents a multilayer perceptron projection network; For the The low-dimensional projection features corresponding to the samples.

[0078] Then, the historical fault prototype vectors are weighted based on the momentum coefficient, and the prototype vectors of each type of fault are updated by combining the mean of the projection features of the same type of samples in the current batch, which is expressed as:

[0079] Where, Indicates the Class Fault The prototype vector after the iteration update; Indicates the Class Fault The prototype vector after the iteration update; is the momentum coefficient, such as, , control the degree of historical information retention; Indicates that the current batch belongs to The sample index collection of the class; Indicates that the current batch belongs to The number of samples in the sample index set of the class; is the fault category index, ; is the total number of fault categories.

[0080] Finally, the contrast loss is calculated, that is, the cosine distance between the sample projection feature and the true category prototype vector is calculated, combined with the distance to the prototype vector of other categories, and the prototype contrast loss is obtained through the exponential function and natural logarithm operation, which is expressed as:

[0081] Where, represents the prototype contrast loss; Indicates the The samples correspond to the true Prototype vectors of categories; Indicates the The samples correspond to Prototype vectors of categories; represents the cosine distance function; is the temperature coefficient, such as, , controls the sharpness of the distribution; represents the exponential function; Represents the natural logarithm.

[0082] During model training, using a fixed regularization strength often makes it difficult to balance the needs of different training stages. In the early stages, overly strong constraints may inhibit the model's learning ability, while in the later stages, too weak constraints may lead to overfitting. Fixed strategies are insufficient in the dynamic training process, especially in vibration fault diagnosis. Based on this, this application constructs a total loss function based on the basic loss and prototype contrast loss, combined with an L2 regularization term that decays exponentially over time. First, a dynamic regularization loss function is defined. Specifically, a total loss function is constructed based on the basic loss and prototype contrast loss, combined with an L2 regularization term that decays over time. The regularization strength decays exponentially with the increase in training rounds, expressed as:

[0083] Where, represents the total loss function; represents the L2 norm; is the model weight parameter; is the time-varying regularization coefficient, which decays with time and is calculated as ; Indicates the current training round; Indicates the preset total number of training rounds; is the initial regularization strength, such as, ; is the minimum regularization strength, such as, ; Indicates the The new parameter value of the generation.

[0084] Then, to optimize the decay rate parameter, it is necessary to use the multivariate normal distribution to generate new parameter values, and optimize the decay rate parameter of the dynamic regularization coefficient through the covariance matrix adaptive evolution strategy, which is expressed as:

[0085] Where, Indicates the New parameter values ​​for the generation; represents a multivariate normal distribution; For the The mean vector of the distribution of generations; For the The step size parameter of the generation; For the The covariance matrix of the generation; Represents the evolutionary generation index.

[0086] Finally, the evolutionary distribution parameters can be updated, that is, the parent individuals are selected according to the fitness ranking, and the mean vector of the evolutionary distribution is updated by weighted averaging, which is expressed as:

[0087] in, For the The distribution mean vector of the generation is obtained by weighted average. Generation parameter distribution center; For the the first individuals in the population sorted by fitness, representing an index of the individuals in the population sorted by fitness; representing a number of parents selected; representing the first weight coefficient.

[0088] In the model iterative updating and parameter training, the adaptive moment estimation optimizer can be used to perform the iterative updating of the model parameters based on the total loss function.

[0089] In each training round, first, the preprocessed vibration feature samples are loaded in small batches, and the multi-scale fusion features, attention enhanced features and low-dimensional projection features are calculated by forward propagation; then the sum of the basic classification loss, the prototype contrast loss and the dynamic regularization loss is calculated as the total loss; the gradient of the total loss with respect to all weight parameters of the model is calculated by the back propagation algorithm, and the gradient clipping technique is applied to limit the gradient norm to stabilize the training process. The optimizer dynamically adjusts the learning rate of each parameter by combining the first moment estimation and the second moment estimation of the gradient, and simultaneously incorporates a dynamic weight decay mechanism to adapt to the time-varying regularization constraint.

[0090] During the training process, the fault diagnosis accuracy and loss value on the validation set are continuously monitored, and if the validation loss does not improve for consecutive multiple rounds, the early stopping mechanism is triggered. Finally, the model parameter snapshot with the optimal performance on the validation set is saved, and the end-to-end training of the deep learning model is completed.

[0091] In one embodiment, the multi-fault type diagnosis performance analysis is performed, and the grouping column chart as shown in Figure 5 is used to verify the generalization ability of the present application on seven typical fault types (including the healthy state). The horizontal coordinate lists all fault types, and the vertical coordinate "diagnosis accuracy (%) " quantifies the diagnosis performance. The traditional support vector machine method (left column) performs significantly weakly on complex faults such as gear tooth breakage; the conventional deep learning method (middle column) improves overall, but still has shortcomings in categories such as bearing outer ring damage; the present application method (right column) maintains the highest column height on all fault types, especially in the "bearing inner ring damage" (which requires identification of feature frequency sideband modulation) and "gear wear" (long period pattern) two types of faults, the experimental results show that the multi-scale feature fusion and dual-path attention mechanism can cooperatively capture short-term impact and long-period patterns, overcoming the limitations of traditional methods in diversified fault scenarios.

[0092] S105: performing fault diagnosis on the to-be-tested equipment through the diagnostic model.

[0093] Deploy the fully trained deep learning model to an edge computing terminal or cloud-based diagnostic platform. During the online diagnostic phase, the equipment's vibration signals are collected in real time and sequentially executed through steps S1: data acquisition, S2: frequency-time hybrid normalization, S3: adaptive time-frequency feature extraction, and S4: deep learning model forward inference. The model then outputs a probability distribution of fault categories and a confidence score. The diagnostic results are mapped to specific fault modes, such as "bearing outer ring damage, probability 92%," and a health status index is generated simultaneously.

[0094] The maintenance decision system generates the optimal maintenance strategy based on the fault type, such as "replace the bearing", pushes it to the equipment management system and links it to the spare parts inventory, achieving closed-loop management from condition monitoring, fault diagnosis to maintenance scheduling.

[0095] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0096] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0101] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0102] The memory can include non-persistent memory and / or volatile memory, such as a random access memory (RAM) including a cache area for the temporary storage of data. The memory can also include non-volatile memory, such as a read only memory (ROM), EPROM, EEPROM, or flash memory. The memory can be another form of computer-readable media.

[0103] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for the storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0104] It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0105] The above description is merely illustrative of the application, and not restrictive. Various modifications and changes can become apparent to those skilled in the art. Incorporating any modification, equivalent substitution, improvement, etc. within the spirit and principle of the application, shall be included in the scope of the claims of the application.

Claims

1. A deep learning-based equipment predictive maintenance method, characterized in that: include: After performing time domain amplitude normalization and frequency domain weighted normalization on the training data, dual-channel feature fusion is performed to obtain a fused feature vector; Optimizing Morlet wavelet basis function parameters, and using multiple sets of adaptive wavelet kernels to perform feature extraction on the fused feature vector to obtain an adaptive time-frequency feature vector; The weight population is constructed through the statistical characteristics of the adaptive time-frequency feature vector, and the optimal initial weight of the initial diagnostic model is selected by combining the differential evolution algorithm; The initial diagnosis model is iteratively updated by performing multi-scale deep feature calculation, time-frequency domain attention feature fusion, fault prototype contrast learning, and a pre-constructed total loss function on the adaptive time-frequency feature vector to obtain a diagnosis model; Fault diagnosis is performed on the device to be tested using the diagnostic model.

2. The method according to claim 1, characterized in that Before performing time domain amplitude normalization and frequency domain weighted normalization on the training data, the method further includes: By using a vibration sensor pre-installed on the equipment, the original vibration signal of the equipment in the operating state is collected at a preset sampling frequency within a preset time period, and the equipment operating condition parameters are simultaneously recorded as initial sample data; the equipment operating condition parameters include at least one of the rotational speed, load and temperature; The initial sample data is labeled in combination with the equipment operation log, disassembly and maintenance report, and oil analysis results. The labeling categories include failure modes and health modes. The failure modes include at least one of rotor imbalance, shaft misalignment, bearing inner ring damage, bearing outer ring damage, gear tooth breakage, and gear wear.

3. The method according to claim 1, characterized in that After the training data is normalized in the time domain amplitude and weighted normalized in the frequency domain, dual-channel feature fusion is performed, specifically including: Perform amplitude normalization on the original vibration signal to obtain a time-domain normalized feature vector; Performing a fast Fourier transform on the original vibration signal to obtain an amplitude spectrum; The amplitude spectrum is weighted by a frequency weighting matrix to suppress high-frequency noise and enhance the fault-sensitive frequency band; Perform logarithmic operation on the weighted results to compress the amplitude range and improve feature separability to obtain frequency domain normalized features; The time domain normalized feature vector and the frequency domain normalized feature vector are concatenated to generate a fused feature vector.

4. The method according to claim 1, wherein The optimization of the Morlet wavelet basis function parameters specifically includes: Based on the standard Morlet wavelet basis function, a parameterized adaptive wavelet basis function is constructed by combining the scale parameter to control the wavelet scaling, the translation parameter to control the wavelet time shift, and the window function weight parameter to adjust the steepness of the Sigmoid window.

5. The method according to claim 1, wherein The step of extracting features from the fused feature vector using multiple sets of adaptive wavelet kernels to obtain an adaptive time-frequency feature vector specifically includes: Calculate the energy of each adaptive wavelet basis function and convert the energy of each wavelet kernel into a weighted coefficient through normalization processing; Using multiple groups of adaptive wavelet kernels to perform convolution operation on the fused feature vector, taking the modulus of the convolution result and weighting and summing it according to the energy weight coefficient, and outputting the initial adaptive time-frequency feature vector; Taking the improvement of the variance of the characteristic distribution of different fault categories as the optimization goal, an evolutionary algorithm is used to dynamically optimize the adaptive wavelet kernel parameter set to output an adaptive time-frequency feature vector. The adaptive wavelet kernel parameter set includes at least one of a wavelet kernel scale parameter, a translation parameter, and a window weight.

6. The method according to claim 1, characterized in that The method of constructing a weight population by using the statistical characteristics of the adaptive time-frequency feature vector and selecting the optimal initial weight of the initial diagnostic model by combining the differential evolution algorithm specifically includes: Performing mean vector calculation on the adaptive time-frequency feature vector to obtain a feature mean vector; Performing singular value decomposition on the covariance matrix of the adaptive time-frequency eigenvector to obtain a left singular matrix and a singular value diagonal matrix; Based on the characteristic mean vector, the left singular matrix and the singular value diagonal matrix, a random orthogonal matrix is ​​combined to generate a diversity initial weight population; For target individuals whose individual loss value is greater than the benchmark individual loss value, differential mutation operation is performed through the mutation factor and multiple randomly selected individual weights to obtain the evolutionary weight population; The loss values ​​of all individuals in the evolutionary weight population are evaluated, and the weight corresponding to the individual with the smallest loss value is selected as the optimal initial weight of the initial diagnostic model.

7. The method according to claim 1, characterized in that The multi-scale depth feature calculation of the adaptive time-frequency feature vector specifically includes: Using convolution kernels of various sizes to perform one-dimensional convolution operations and multi-dimensional convolution operations on the adaptive time-frequency feature vectors to extract short-scale feature vectors and long-scale feature vectors; The short-scale feature vector is concatenated with the long-scale feature vector, and batch normalization and Gaussian error linear unit activation function operations are performed in sequence to output multi-scale fusion features.

8. The method according to claim 7, characterized in that The multi-scale depth feature calculation and time-frequency domain attention feature fusion of the adaptive time-frequency feature vector specifically include: The multi-scale fusion feature is converted into a query matrix through a learnable query transformation matrix, the query matrix is ​​processed by a normalized exponential function, and then weighted with the time domain feature to output a time domain attention feature vector; Performing a fast Fourier transform on the multi-scale fusion features to obtain a frequency domain representation, calculating the frequency domain attention weight by amplitude weighting and a normalized exponential function, and outputting a frequency domain attention feature vector by inverse fast Fourier transform; The time domain attention feature vector is added to the frequency domain attention feature vector, and the fused time-frequency domain attention feature vector is output through layer normalization operation.

9. The method according to claim 8, characterized in that The method of performing multi-scale deep feature calculation, time-frequency domain attention feature fusion, and fault prototype comparative learning on the adaptive time-frequency feature vector specifically includes: The fused time-frequency domain attention feature vector is input into the multi-layer perceptron projection network and converted into a low-dimensional projection feature vector; The historical fault prototype vectors are weighted based on the momentum coefficient, and the prototype vector of each type of fault is updated by combining the mean of the low-dimensional projection feature vectors of the same type of samples in the current batch; The cosine distance between the sample projection feature and the true category prototype vector is calculated, and the distance between the sample projection feature and the prototype vector of other categories is combined to obtain the prototype contrast loss through exponential function and natural logarithm operation.

10. The method according to claim 8, characterized in that The initial diagnosis model is iteratively updated by performing multi-scale deep feature calculation, time-frequency domain attention feature fusion, fault prototype contrast learning, and a pre-built total loss function on the adaptive time-frequency feature vector to obtain a diagnosis model, specifically including: Based on the base loss and prototype contrast loss, the total loss function is constructed by combining the L2 regularization term that decays over time, where the regularization strength decays exponentially with the increase of training rounds; Multivariate normal distribution is used to generate new parameter values, and the decay rate parameter of the dynamic regularization coefficient is optimized through the covariance matrix adaptive evolution strategy; Select parent individuals according to fitness ranking and update the mean vector of evolutionary distribution by weighted averaging; Based on the total loss function, an adaptive moment estimation optimizer is employed to perform iterative updates of the model parameters.

Images