Equipment fault diagnosis method and device based on incremental learning

Abstract

The invention discloses an incremental learning-based equipment fault diagnosis method and device. Obtaining related data of the sample equipment, and expanding the related data through a generator and a discriminator to obtain expanded data; inputting the extended data into a to-be-trained feature extraction model for feature extraction model training, and extracting extended feature data of the extended data through the feature extraction model; inputting the expanded feature data into a to-be-trained feature dimension reduction model to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model; inputting the expanded feature data after data dimension reduction into a to-be-trained classifier model for classifier model training; and performing equipment fault diagnosis on to-be-diagnosed equipment through the feature extraction model, the data dimension reduction model and the classifier model. According to the scheme, more accurate equipment fault diagnosis can be realized.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an equipment fault diagnosis method and device based on incremental learning. Background Art

[0002] With the improvement of industrial automation and intelligence, the operation data of equipment has become increasingly complex, including real-time monitoring data of various sensors and operation logs, etc. These data contain rich key information on equipment operation status and potential failures, but also bring challenges to data processing and analysis, especially when the data volume is large, the feature dimension is high, and the failure mode is changeable.

[0003] Existing technologies usually use traditional machine learning methods and some basic neural network technologies to deal with these problems, but they are often limited by the generalization ability of the model and the quality and quantity of training data. In practical applications, models often face problems such as insufficient samples, diverse fault types, and difficulty in quickly adapting to new fault modes. In addition, the lack of feature extraction and dimensionality reduction often leads to the inability of the model to effectively learn useful information from complex data, affecting the final diagnostic accuracy.

[0004] In summary, based on the problem of low accuracy of equipment fault diagnosis in the prior art, a more accurate equipment fault diagnosis method is needed. Summary of the invention

[0005] The present invention provides an equipment fault diagnosis method and device based on incremental learning to improve the accuracy of equipment fault diagnosis.

[0006] According to one aspect of the present invention, there is provided a device fault diagnosis method based on incremental learning, comprising:

[0007] Acquire relevant data of the sample device, and expand the relevant data through the generator and the discriminator to obtain expanded data;

[0008] Inputting the expanded data into a feature extraction model to be trained to perform feature extraction model training, and extracting expanded feature data of the expanded data through the feature extraction model;

[0009] Inputting the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model;

[0010] Inputting the expanded feature data after data dimension reduction into the classifier model to be trained to perform classifier model training;

[0011] Equipment fault diagnosis is performed on the equipment to be diagnosed through the feature extraction model, the data dimension reduction model and the classifier model.

[0012] Optionally, the relevant data includes one or more of sensor data, operation logs, maintenance records and detailed fault data.

[0013] Optionally, the generator and the discriminator are trained in the following manner:

[0014] Initialize the parameters of the generator and discriminator;

[0015] Generate initial pseudo fault data through the generator, input the real fault data and the initial pseudo fault data into the discriminator to calculate and output a discrimination result, and update the discriminator parameters according to the improved loss function;

[0016] According to the discriminator outputting the discriminant result, the generator updates the generator parameters according to the loss calculation;

[0017] According to the incremental learning strategy, new real fault data is added and the discriminator parameters and the generator parameters are updated again until the preset stop iteration condition is met.

[0018] Optionally, inputting the expanded data into a feature extraction model to be trained to perform feature extraction model training includes:

[0019] Initialize the weights, biases, and basic topology of the neural network;

[0020] Before each iteration, the topology sensor is used to evaluate the requirements of the current task on the network structure. The topology sensor is defined as a function that outputs the topology adjustment index based on the current loss and network structure.

[0021] Calculate the loss function under the current network configuration and decide whether to perform parameter optimization or topology adjustment based on the calculation results;

[0022] Execute parameter or topology adjustment according to parameter adjustment rules;

[0023] Optimize the effectiveness of topology and parameter tuning through feedback loops;

[0024] Adopting an adaptive topology reconstruction feedback mechanism, monitoring network performance feedback through performance monitoring functions, and dynamically adjusting the reconstruction strategy of the topology structure;

[0025] Repeat the iteration until the preset stop condition is met.

[0026] Optionally, the step of inputting the expanded feature data into a feature dimensionality reduction model to be trained to perform feature dimensionality reduction model training includes:

[0027] Set up the micro-adjustable strategy autoencoder neural network and initialize weights and biases;

[0028] Reconstructing the expanded feature data into high-dimensional feature data and inputting it into an encoder, propagating it through an activation function and a multi-layer network structure to generate a low-dimensional feature representation;

[0029] Calculating a total loss function, where the total loss includes a reconstruction loss and a dual loss;

[0030] According to the total loss function, the network parameters are updated using the gradient descent method;

[0031] Repeat the iteration until the preset stop condition is met.

[0032] Optionally, the step of inputting the expanded feature data after data dimension reduction into a classifier model to be trained to perform classifier model training includes:

[0033] Set up a quantum-coded spiking neural network and initialize weights and biases;

[0034] Decompose the feature vector of the expanded feature data after the input data dimension reduction into multiple quantum bits, and convert them into quantum states through quantum coding;

[0035] During the forward propagation of the spiking neural network, the input quantum state is passed through the input layer to the hidden layer;

[0036] Calculating a dual loss function, wherein the dual loss function includes a cross entropy loss and a quantum state loss;

[0037] Updating the weights and biases of the spiking neural network based on the gradient back-propagation error of the dual loss function;

[0038] Adaptively adjust the learning rate and the firing threshold of the spiking neuron according to the change of the dual loss function during the training process;

[0039] Repeat the iteration until the preset stop condition is met.

[0040] Optionally, the performing equipment fault diagnosis on the equipment to be diagnosed by using the feature extraction model, the data dimension reduction model and the classifier model includes:

[0041] The trained feature extraction model, data dimension reduction model and classifier model are used to process the relevant data of the device to be diagnosed and output the diagnosis results.

[0042] According to another aspect of the present invention, there is provided an equipment fault diagnosis device based on incremental learning, comprising:

[0043] A data expansion unit, used to obtain relevant data of the sample device, and expand the relevant data through the generator and the discriminator to obtain expanded data;

[0044] A feature extraction model training unit, used for inputting the expanded data into a feature extraction model to be trained to perform feature extraction model training, and extracting expanded feature data of the expanded data through the feature extraction model;

[0045] A feature dimension reduction model training unit, used for inputting the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model;

[0046] A classifier training unit, used for inputting the expanded feature data after data dimension reduction into a classifier model to be trained to perform classifier model training;

[0047] The model application unit is used to perform equipment fault diagnosis on the equipment to be diagnosed through the feature extraction model, the data dimension reduction model and the classifier model.

[0048] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0049] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the device fault diagnosis method based on incremental learning described in any embodiment of the present invention.

[0050] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the device fault diagnosis method based on incremental learning described in any embodiment of the present invention when executed.

[0051] According to another aspect of the present invention, a computer program product is provided, characterized in that the computer program product includes a computer program, and when the computer program is executed by a processor, the device fault diagnosis method based on incremental learning described in any embodiment of the present invention is implemented.

[0052] The technical solution of the embodiment of the present invention provides a method for training a device fault diagnosis model based on incremental learning, which mainly includes the steps of data collection and annotation, data expansion, feature extraction model training, feature dimension reduction model training, classifier model training and device fault diagnosis. Through incremental learning, topology adaptation learning algorithm, micro-adjustment strategy and quantum coding and other technologies, the model's adaptability to new fault modes, training stability and classification accuracy are improved to achieve more accurate equipment fault diagnosis.

[0053] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0055] Figure 1 is a flow chart of a device fault diagnosis method based on incremental learning provided in Embodiment 1 of the present invention;

[0056] Figure 2 It is a flowchart of a feature extraction model training provided by the first embodiment of the present invention;

[0057] Figure 3 It is a schematic diagram of a process of training a feature dimensionality reduction model provided in the first embodiment of the present invention;

[0058] Figure 4 It is a structural schematic diagram of an equipment fault diagnosis device based on incremental learning provided in Embodiment 2 of the present invention;

[0059] Figure 5 It is a structural schematic diagram of an electronic device for implementing the device fault diagnosis method based on incremental learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0062] Embodiment 1

[0063] Figure 1 This is a flow chart of a device fault diagnosis method based on incremental learning provided by the first embodiment of the present invention. This embodiment is applicable to the case of diagnosing device faults. The method can be executed by a device fault diagnosis device based on incremental learning. The device fault diagnosis device based on incremental learning can be implemented in the form of hardware and / or software. The device fault diagnosis device based on incremental learning can be configured in an electronic device. Figure 1 As shown, the method includes:

[0064] S110, obtaining relevant data of the sample device, and expanding the relevant data through the generator and the discriminator to obtain expanded data.

[0065] In the embodiment of the present invention, the relevant data includes one or more of sensor data, operation logs, maintenance records and detailed fault data.

[0066] Relevant data is collected from equipment sensor data, log records generated during equipment operation, various data recorded during equipment maintenance, and detailed data recorded when equipment fails. For example, relevant data may include:

[0067] ax1: temperature sensor reading;

[0068] ax2: pressure sensor reading;

[0069] ax3: vibration sensor reading;

[0070] ax4: running time;

[0071] ax5: operator ID;

[0072] ax6: device ID;

[0073] ax7: maintenance times;

[0074] ax8: number of historical failures;

[0075] ax9: current load;

[0076] ax10: Device status.

[0077] The attributes of the relevant data can be increased or decreased according to the actual application.

[0078] In the embodiment of the present invention, the generator and the discriminator are trained in the following manner:

[0079] Initialize the parameters of the generator and discriminator;

[0080] Generate initial pseudo-fault data through the generator, input the real fault data and the initial pseudo-fault data into the discriminator to calculate and output the discrimination result, and update the discriminator parameters according to the improved loss function;

[0081] According to the discriminator outputs the discriminant result, the generator updates the generator parameters according to the loss calculation;

[0082] According to the incremental learning strategy, new real fault data is added and the discriminator parameters and generator parameters are updated again until the preset stop iteration condition is met.

[0083] The collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The present invention adopts a generative adversarial network algorithm based on an incremental strategy. The generative adversarial network algorithm based on an incremental strategy consists of a generator and a discriminator. The generator is used to generate realistic equipment failure data, and the discriminator is used to determine whether the input data is real failure data or generated data. The present invention adopts an incremental learning strategy. During the training process of the generator and the discriminator, new data samples are gradually added to avoid overfitting of the model and enhance the adaptability of the model.

[0084] Initialize the parameters of the generator and discriminator. Specifically, the initial parameters of the generator are set to θ g0 , the initial parameters of the discriminator are set to θ d0 , the initial learning rates are α g0 and α d0 In one embodiment, θ g0 and θ d0 They are set by random initialization, α g0 and α d0 Set to 0.01 and 0.005 respectively.

[0085] The generator is used to generate initial pseudo fault data. The output of the generator can be expressed as:

[0086] bx =G(b z θ g0 );

[0087] Where b x represents the generated pseudo fault data, G() represents the generator, b z represents the input noise vector of the generator, θ g0 is the initial parameter of the generator.

[0088] During the training process of the discriminator, the real fault data and the generated pseudo fault data are input, the output is calculated by the discriminator, and the discriminator parameters are updated according to the improved loss function. The output of the discriminator can be expressed as:

[0089]

[0090] In the formula, D represents the discriminator, Sig() is the activation function, and w d and b d are the weight and bias parameters of the discriminator, θ d are the parameters of the discriminator.

[0091] The loss function of the discriminator is defined as:

[0092]

[0093] In the formula, m represents the batch size, represents the i-th real fault data sample, represents the i-th noise vector sample, θ g The parameters for the generator.

[0094] According to the loss function, the parameter update method of the discriminator can be expressed as:

[0095]

[0096] Among them, α d represents the learning rate of the discriminator, Denotes the loss function L d About the discriminator parameters θ d gradient.

[0097] For the weights and biases of the discriminator, the loss function L d The gradient calculation method can be expressed as:

[0098]

[0099] In the formula, w d and b d Represent the weight and bias parameters of the discriminator respectively, represents the i-th real fault data sample, represents the i-th noise vector sample, Represents the generated pseudo fault data.

[0100] During the training process of the generator, according to the discriminator outputs the discriminant result, the generator optimizes its own parameters according to the loss calculation. The loss function of the generator is defined as:

[0101]

[0102] According to the loss function, the parameter update method of the generator can be expressed as:

[0103]

[0104] In the formula, α g represents the learning rate of the generator, Denotes the loss function L g About the generator parameter θ g gradient.

[0105] For the weights and biases of the generator, the loss function L g The gradient calculation of is the same as that of the discriminator.

[0106] According to the incremental learning strategy, new real fault data is added and the above training process is repeated. In this process, adding new real fault data can be expressed as:

[0107]

[0108] In the formula, Indicates the newly introduced fault data, and na indicates the number of newly added data.

[0109] According to the change of loss function during training, the learning rate of generator and discriminator is adaptively adjusted, which can be expressed as:

[0110] α d ←α d ×exp(-β vh ΔL d );

[0111] α g ←α g ×exp(-γ vh ΔL g );

[0112] In the formula, β vh and γ vh is the adaptive adjustment parameter, ΔL d and ΔL g are the changes in the discriminator and generator loss functions, respectively. Preferably, βvh and γ vh Set to 0.3 and 0.7 respectively.

[0113] ΔL d The calculation method can be expressed as:

[0114] ΔL d =L d (t)-L d (t-1);

[0115] ΔL g The calculation method can be expressed as:

[0116] ΔL g =L g (t)-L g (t-1);

[0117] In the formula, β and γ are adaptive adjustment parameters, L d (t) and L g (t) are the loss function values ​​in the current iteration cycle, L d (t-1) and L g (t-1) is the loss function value in the previous iteration cycle.

[0118] Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is preferably to reach a preset maximum number of iterations, for example, the preset maximum number of iterations is set to 1000 times.

[0119] S120, inputting the expanded data into the feature extraction model to be trained to train the feature extraction model, and extracting expanded feature data of the expanded data through the feature extraction model.

[0120] The expanded data is input into the feature extraction model to train the feature extraction model. The present invention uses a 3-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, gradient disappearance, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a feature extraction model of a neural network optimized based on a topological adaptive learning algorithm, and adopts a topological dynamic adjustment mechanism, allowing the network to adjust not only weights and biases during the learning process, but also the topological structure of its internal connections.

[0121] In an embodiment of the present invention, the augmented data is input into the feature extraction model to be trained to perform feature extraction model training, including:

[0122] Initialize the weights, biases, and basic topology of the neural network;

[0123] Before each iteration, the topology sensor is used to evaluate the requirements of the current task on the network structure. The topology sensor is defined as a function that outputs the topology adjustment index based on the current loss and network structure.

[0124] Calculate the loss function under the current network configuration and decide whether to perform parameter optimization or topology adjustment based on the calculation results;

[0125] Execute parameter or topology adjustment according to parameter adjustment rules;

[0126] Optimize the effectiveness of topology and parameter tuning through feedback loops;

[0127] Adopting an adaptive topology reconstruction feedback mechanism, monitoring network performance feedback through performance monitoring functions, and dynamically adjusting the reconstruction strategy of the topology structure;

[0128] Repeat the iteration until the preset stop condition is met.

[0129] Figure 2 It is a flowchart of a feature extraction model training provided in Example 1 of the present invention.

[0130] Initialize the parameters of the neural network (weight w and bias b) and the basic topology, and define the initial weight matrix W (0) and the bias vector B (0) for:

[0131] W (0) =randn(d in ,d out );

[0132] B (0) =randn(0,d out );

[0133] Where, d in and d out They represent the input and output dimensions of the network layer respectively, and randn(,) is a randomly selected value in a specific interval.

[0134] Before each iteration, the network uses the built-in topology sensor to evaluate the requirements of the current task on the network structure. The topology sensor is defined as a function T, which outputs the topology adjustment index τ based on the current loss L and network structure S, which can be expressed as:

[0135] τ=T(L,S);

[0136]

[0137] Where Sig() is the Sigmoid activation function.

[0138] gradient The calculation method can be expressed as:

[0139]

[0140] In the formula, E is the prediction error, which is calculated by the preset Softmax function.

[0141] Calculate the loss function number L under the current network configuration. The loss function L consists of two parts: prediction error E and topology adaptability penalty P. The calculation method can be expressed as:

[0142] L=E+λ sa P;

[0143]

[0144] In the formula, λ sa is the topology weight hyperparameter, Δs j represents the change in the jth structural element (such as connection weight). Preferably, λ sa is set to 0.1.

[0145] Δs j The calculation method can be expressed as:

[0146]

[0147] In the formula, and are the structural states in consecutive iterations, respectively.

[0148] According to the values ​​of τ and L, it is decided whether to perform parameter optimization or topology adjustment. The adjustment method is implemented through the decision function D, which can be expressed as:

[0149]

[0150] In the formula, θ cp is the threshold value. Preferably, θ cp Set to 0.5.

[0151] To adjust the execution parameters or topology, the parameter adjustment rule can be expressed as:

[0152]

[0153] In the formula, w (t+1) and b (t+1) are the adjusted neural network weights and biases, w (t) and b (t+1) is the neural network weight and bias before adjustment, η vh is the learning rate. Preferably, η vh Set to 0.01.

[0154] gradient Calculated by the chain rule, it can be expressed as:

[0155]

[0156] Calculated by the chain rule, it can be expressed as:

[0157]

[0158] In the formula, y i is the actual label of the i-th sample, is the predicted label of the i-th sample output by the preset Softmax function.

[0159] The effectiveness of topology and parameter adjustment is optimized through feedback loops. Specifically, η or λ is adjusted to optimize the learning process, which can be expressed as:

[0160] η vh (t+1) =η vh (t) ×β bu ;

[0161] λ sa (t+1) =λ sa (t) +α bu ;

[0162] Where η vh (t+1) and λ sa (t+1) are the adjusted learning rate and topology weight hyperparameters, η vh (t) and λ sa (t) are the learning rate and topology weight hyperparameters before adjustment, β bu and α bu is the adjustment factor. Preferably, β bu and α bu Set to 0.9 and 0.01 respectively.

[0163] Adopting the adaptive topology reconstruction feedback mechanism, by monitoring the network performance feedback, dynamically adjusting the reconstruction strategy of the topology structure, and realizing more refined network optimization, specifically, defining the performance monitoring function R as a measure of the stability and accuracy of the network output, which can be expressed as:

[0164] R=ω 1 R s +ω 2 R p ;

[0165] In the formula, R s Represents output stability, obtained by calculating the output variance, R p Represents accuracy, measured by the inverse loss function value. Weight ω 1 and ω 2 Adjust the influence weight of stability and accuracy. Preferably, ω 1 and ω 2 Set to 0.6 and 0.4 respectively.

[0166] R s and R p The calculation method can be expressed as:

[0167]

[0168] In the formula, yes The average value, E max is the maximum possible error. Preferably, E max Set to 0.5.

[0169] Adopting adaptive adjustment factor δ bu , dynamically adjust the topology according to the value of R, which can be expressed as:

[0170]

[0171] In the formula, κ up is the upper limit of the adjustment rate, λ vu is the sensitivity of the adjustment. Preferably, κ up and λ vu Set to 0.1 and 5 respectively.

[0172] According to δ, the topology can be expressed as:

[0173] S (t+1) =S (t) +δ bu ΔS;

[0174] Where S is the topology before adjustment, S (t+1) is the adjusted topological structure, and ΔS is the topological adjustment direction derived based on the gradient or other optimization algorithms.

[0175] Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is preferably to reach a preset maximum number of iterations, for example, the preset maximum number of iterations is set to 1000 times.

[0176] S130, inputting the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model.

[0177] The data after feature extraction is input into the feature dimensionality reduction model to train the feature dimensionality reduction model. The present invention adopts a micro-adjustable strategy autoencoder neural network as the feature dimensionality reduction model. The micro-adjustable strategy autoencoder neural network includes an encoder and a decoder. The encoder is responsible for compressing the input high-dimensional features into a low-dimensional feature space, and the decoder attempts to reconstruct the original input data from this low-dimensional feature space. In order to improve the training efficiency and reconstruction quality of the model, the present invention adopts a dual loss calculation method, which not only considers the reconstruction error, but also evaluates the fitness of the encoded low-dimensional features and the target dimensionality reduction space, thereby optimizing the quality of feature representation.

[0178] In the embodiment of the present invention, the expanded feature data is input into the feature dimensionality reduction model to be trained to perform feature dimensionality reduction model training, including:

[0179] Set up the micro-adjustable strategy autoencoder neural network and initialize weights and biases;

[0180] Reconstruct the expanded feature data into high-dimensional feature data and input it into the encoder, propagate it through the activation function and multi-layer network structure to generate low-dimensional feature representation;

[0181] Calculate the total loss function, which includes reconstruction loss and dual loss;

[0182] According to the total loss function, the gradient descent method is used to update the network parameters;

[0183] Repeat the iteration until the preset stop condition is met.

[0184] Figure 3 It is a flowchart of a feature dimensionality reduction model training provided in Example 1 of the present invention.

[0185] In the initialization stage, all weights w and biases b in the network are initialized with random values. In one embodiment, Gaussian distribution is used for initialization, which can be expressed as:

[0186]

[0187] In the formula, w pij represents the connection weight from the jth neuron to the ith neuron in the autoencoder; b pi represents the bias of the i-th neuron; σ cb is the standard deviation, used to control the dispersion of the initialization. Preferably, σ cb Set to 0.01.

[0188] The high-dimensional feature data x after input feature extraction p To the encoder, it is propagated through the activation function and multi-layer network structure to generate a low-dimensional feature representation. Each layer of the encoder outputs z pl Defined as:

[0189] z pl =Sig(Σ j w plj ·z p(l-1) +b pl );

[0190] The low-dimensional features enter the decoder, which attempts to reconstruct the original high-dimensional data. The decoder output reconstructs for:

[0191]

[0192] In the formula, z pl is the output of the encoder layer l; y pk is the k-th layer input of the decoder; Sig() is the Sigmoid activation function.

[0193] Let the last layer output of the encoder be z pm , a differentiable adjustment layer is used to adjust the encoder output z pm for It can be expressed as:

[0194]

[0195] Where ⊙ represents element-wise multiplication; α p is a learnable adjustment factor; δ p is a parameter adjusted according to the gradient information. Preferably, δ p Set to 0.3.

[0196] Adjustment factor α p The calculation method can be expressed as:

[0197]

[0198] In the formula, Represents the feature after encoding from the kth to the weight of the adjustment factor; is the bias of the adjustment factor; Re() is the ReLU activation function.

[0199] Calculate the total loss L p, including reconstruction loss and dual loss. For reconstruction loss, in one embodiment, the mean square error is used to measure the difference between the reconstructed data and the original data. For dual loss, it includes evaluating the matching degree between the encoded features and the predetermined low-dimensional target features to optimize the representation ability of the features. Specifically, the total loss L p The calculation method can be expressed as:

[0200]

[0201] Where N is the number of samples; x pi is the i-th original data point; is the i-th reconstructed data point; D() is the dual loss function; z target is the target low-dimensional space; p is the weight factor of the dual loss. Preferably, λ p Set to 0.4.

[0202] The dual loss function D() is used to measure the encoded feature z pm and the target low-dimensional space z target The degree of matching between them can be calculated as:

[0203]

[0204] In the formula, z pmi is the i-th feature dimension after encoding; z targeti is the i-th feature dimension in the target low-dimensional space.

[0205] The network parameters are updated using the gradient descent method according to the total loss function. In particular, the parameters are adjusted according to the gradient information in the micro-adjustable layer to achieve better information compression and feature expression. Specifically, let the gradient and Used to update parameters, the update method of the autoencoder parameters can be expressed as:

[0206]

[0207] In the formula, and are the updated autoencoder weights and bias parameters, and are the autoencoder weights and bias parameters before updating; γ p is the learning rate. Preferably, γ p Set to 0.01.

[0208] Gradient of weights The calculation method can be expressed as:

[0209]

[0210] In the formula, is the partial derivative of the loss function with respect to the model output, It is the partial derivative of the model output with respect to the weight.

[0211] Bias gradient The calculation method can be expressed as:

[0212]

[0213] In the formula, is a constant term, preferably set to 1.

[0214] Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is preferably to reach a preset maximum number of iterations, for example, the preset maximum number of iterations is set to 1000 times.

[0215] S140, inputting the expanded feature data after data dimension reduction into the classifier model to be trained to perform classifier model training.

[0216] The reduced-dimensional data is input into the classifier model to train the classifier. The present invention adopts a pulse neural network classification algorithm based on quantum coding. The pulse neural network processes the reduced-dimensional data by simulating the pulse emission and propagation of neurons. In addition, quantum coding converts the input reduced-dimensional data into a quantum state, realizes the nonlinear transformation of the data through quantum gate operations, and improves the expression ability and classification accuracy of the classifier.

[0217] In an embodiment of the present invention, the expanded feature data after data dimension reduction is input into a classifier model to be trained to perform classifier model training, including:

[0218] Set up a quantum-coded spiking neural network and initialize weights and biases;

[0219] Decompose the feature vector of the expanded feature data after the input data dimension reduction into multiple quantum bits, and convert them into quantum states through quantum coding;

[0220] During the forward propagation of the spiking neural network, the input quantum state is passed through the input layer to the hidden layer;

[0221] Calculate the dual loss function, which includes cross entropy loss and quantum state loss;

[0222] Update the weights and biases of the spiking neural network based on the gradient of the dual loss function.

[0223] Adaptively adjust the learning rate and the firing threshold of the spiking neuron according to the changes in the dual loss function during training;

[0224] Repeat the iteration until the preset stop condition is met.

[0225] Initialize the parameters of the spiking neural network, including the weights and biases of the spiking neural network. In one embodiment, the initialization method is random initialization.

[0226] The feature vector p after the input feature is reduced in dimension x Decomposed into multiple quantum bits and converted into quantum state q through quantum coding x , to realize the conversion of the quantum state of data, which can be expressed as:

[0227] q x =U(p x θ q );

[0228] In the formula, q x is the quantum state after conversion, p x is the input feature vector, U() is the quantum coding operation, θ q is the parameter of quantum coding.

[0229] The implementation of quantum coding operation can be expressed as:

[0230] U(p x θ q )=U 2 U 1 p x ;

[0231] Where U 1 and U 2 Represent two quantum gate operations respectively, and the calculation method can be expressed as:

[0232] U 1 =exp(-iH 1 t 1 );

[0233] U 2 =exp(-iH 2 t 2 );

[0234] In the formula, H 1 and H 2 are two different Hamiltonians, t 1 and t 2 is the evolution time, and i is an imaginary unit.

[0235] In the forward propagation process of the pulse neural network, the input quantum state q x Passing from the input layer to the hidden layer, the propagation calculation can be expressed as:

[0236] c h (t) = Sig(W h q x +b h -v h (t));

[0237] In the formula, c h (t) is the pulse potential of the hidden layer at time t, W h is the weight matrix from the input layer to the hidden layer, b h is the bias vector of the hidden layer, v h (t) is the emission threshold of the hidden layer at time t.

[0238] In each spiking neuron, the firing status of the neuron is calculated based on the input potential and the firing threshold, and the pulse signal is passed to the next layer. In the output layer, the pulse signals are summarized to obtain the classification results. The calculation method of propagation from the hidden layer to the output layer can be expressed as:

[0239] c o (t) = Sig(W o c h (t)+b o -v o (t));

[0240] In the formula, c o (t) is the pulse potential of the output layer at time t, W o is the weight matrix from the hidden layer to the output layer, b o is the bias vector of the output layer, v o (t) is the emission threshold of the output layer at time t.

[0241] Calculate the dual loss function, including cross entropy loss and quantum state loss. Among them, cross entropy loss is used to measure the gap between the predicted result and the true label, and the calculation method can be expressed as:

[0242]

[0243] Where, L ce is the cross entropy loss, m is the number of samples, y i is the true label of the i-th sample, is the predicted label of the i-th sample.

[0244] Quantum state loss is used to measure the difference between the input quantum state and the output quantum state, enhancing the effectiveness of quantum coding. The calculation method can be expressed as:

[0245]

[0246] Where, L qtis the quantum state loss, is the input quantum state of the i-th sample, is the output quantum state of the i-th sample, Represents the difference in quantum states.

[0247] Quantum state differences The calculation method can be expressed as:

[0248]

[0249] In the formula, is the element of the i-th input sample quantum state, is the element of the i-th output sample quantum state, and d is the dimension of the quantum state.

[0250] Based on this, the calculation method of the dual loss function can be expressed as:

[0251] L=L ce +λ q L qt ;

[0252] In the formula, λ q is the weight coefficient of quantum state loss. Preferably, λ q Set to 0.4.

[0253] According to the gradient of the loss function, back-propagation error, update the weights and bias of the spiking neural network, the weight update method from the input layer to the hidden layer can be expressed as:

[0254]

[0255] The bias update method of the hidden layer can be expressed as:

[0256]

[0257] The weight update method from the hidden layer to the output layer can be expressed as:

[0258]

[0259] The bias update method of the output layer can be expressed as:

[0260]

[0261] In the formula, α h and α o are the learning rates for the hidden layer and the output layer, respectively.

[0262] The weight update method from the input layer to the hidden layer can be expressed as:

[0263]

[0264] Among them, the gradient The calculation method can be expressed as:

[0265]

[0266] Similarly, the weight update method from the hidden layer to the output layer can be expressed as:

[0267]

[0268] Among them, the gradient The calculation method can be expressed as:

[0269]

[0270] Preferably, α h and α o It is set in an adaptive adjustment mode, and its initial values ​​are set to 0.03 and 0.07 respectively.

[0271] According to the changes in the loss function during training, the learning rate and the firing threshold of the pulse neuron are adaptively adjusted. The learning rate adjustment method can be expressed as:

[0272] α h ←α h ×exp(-γ h ΔL);

[0273] α o ←α o ×exp(-γ o ΔL);

[0274] The emission threshold adjustment method can be expressed as:

[0275] v h (t)←v h (t)+η h ΔL;

[0276] v o (t)←v o (t)+η o ΔL;

[0277] In the formula, γ h and γ o is the learning rate adjustment parameter, ΔL is the change in the loss function, η h and η o is the emission threshold adjustment parameter. Preferably, γ h and γ o are set to 0.1 and 0.2 respectively, η h and η o Set them to 0.1 and 0.3 respectively.

[0278] Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. The preset stop iteration condition is preferably to reach a preset maximum number of iterations, for example, the preset maximum number of iterations is set to 1000 times.

[0279] S150, performing equipment fault diagnosis on the equipment to be diagnosed through a feature extraction model, a data dimension reduction model and a classifier model.

[0280] In an embodiment of the present invention, the performing of equipment fault diagnosis on the equipment to be diagnosed by using the feature extraction model, the data dimension reduction model and the classifier model includes:

[0281] The trained feature extraction model, data dimension reduction model and classifier model are used to process the relevant data of the device to be diagnosed and output the diagnosis result.

[0282] The newly collected device data is processed using the trained feature extraction model, data dimension reduction model and classifier model, and a diagnosis result is output. In one embodiment, the diagnosis result includes: initial failure, mid-term failure and late failure.

[0283] Through incremental learning and topology adaptive learning algorithms, the model of the present invention can effectively cope with new fault types and avoid the overfitting problem caused by limited data samples.

[0284] The topology adaptation mechanism and the autoencoding optimization strategy reduce instabilities during training, such as gradient disappearance and explosion, and improve the stability of network training.

[0285] Through the self-encoded differentiable adjustment layer, the information compression and feature representation of high-dimensional data are improved, which helps to identify and classify faults more effectively.

[0286] Quantum coding provides pulse neural networks with a stronger ability to process complex data structures, allowing classifiers to more accurately process and identify various types of faults.

[0287] Embodiment 2

[0288] Figure 4 Schematic diagram of the structure of an incremental learning-based device fault diagnosis apparatus provided by the second embodiment of the present invention. Figure 4 As shown, the device comprises:

[0289] A data expansion unit 410 is used to obtain relevant data of the sample device, and expand the relevant data through a generator and a discriminator to obtain expanded data;

[0290] A feature extraction model training unit 420 is used to input the expanded data into a feature extraction model to be trained to perform feature extraction model training, and extract expanded feature data of the expanded data through the feature extraction model;

[0291] A feature dimension reduction model training unit 430 is used to input the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and perform data dimension reduction on the feature dimension reduction model through the feature dimension reduction model;

[0292] A classifier training unit 440 is used to input the expanded feature data after data dimension reduction into a classifier model to be trained to perform classifier model training;

[0293] The model application unit 450 is used to perform equipment fault diagnosis on the equipment to be diagnosed by using the feature extraction model, the data dimension reduction model and the classifier model.

[0294] The device for diagnosing equipment faults based on incremental learning provided in an embodiment of the present invention can execute the device diagnosing equipment faults based on incremental learning provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0295] Embodiment 3

[0296] Figure 5 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing 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 present invention described and / or required herein.

[0297] like Figure 5As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0298] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0299] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as an equipment fault diagnosis method based on incremental learning.

[0300] In some embodiments, the device fault diagnosis method based on incremental learning can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the device fault diagnosis method based on incremental learning described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the device fault diagnosis method based on incremental learning in any other appropriate manner (for example, by means of firmware).

[0301] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0302] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0303] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0304] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0305] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0306] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0307] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0308] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A device fault diagnosis method based on incremental learning, characterized in that: include: Acquire relevant data of the sample device, and expand the relevant data through the generator and the discriminator to obtain expanded data; Inputting the expanded data into a feature extraction model to be trained to perform feature extraction model training, and extracting expanded feature data of the expanded data through the feature extraction model; Inputting the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model; Inputting the expanded feature data after data dimension reduction into the classifier model to be trained to perform classifier model training; Equipment fault diagnosis is performed on the equipment to be diagnosed through the feature extraction model, the data dimension reduction model and the classifier model.

2. The method according to claim 1, characterized in that The relevant data includes one or more of sensor data, operation logs, maintenance records and detailed fault data.

3. The method according to claim 1, characterized in that The generator and the discriminator are trained in the following way: Initialize the parameters of the generator and discriminator; Generate initial pseudo fault data through the generator, input the real fault data and the initial pseudo fault data into the discriminator to calculate and output a discrimination result, and update the discriminator parameters according to the improved loss function; According to the discriminator outputting the discriminant result, the generator updates the generator parameters according to the loss calculation; According to the incremental learning strategy, new real fault data is added and the discriminator parameters and the generator parameters are updated again until the preset stop iteration condition is met.

4. The method according to claim 1, characterized in that The step of inputting the expanded data into the feature extraction model to be trained to perform feature extraction model training includes: Initialize the weights, biases, and basic topology of the neural network; Before each iteration, the topology sensor is used to evaluate the requirements of the current task on the network structure. The topology sensor is defined as a function that outputs the topology adjustment index based on the current loss and network structure. Calculate the loss function under the current network configuration and decide whether to perform parameter optimization or topology adjustment based on the calculation results; Execute parameter or topology adjustment according to parameter adjustment rules; Optimize the effectiveness of topology and parameter tuning through feedback loops; Adopting an adaptive topology reconstruction feedback mechanism, monitoring network performance feedback through performance monitoring functions, and dynamically adjusting the reconstruction strategy of the topology structure; Repeat the iteration until the preset stop condition is met.

5. The method according to claim 1, characterized in that The step of inputting the expanded feature data into the feature dimensionality reduction model to be trained to perform feature dimensionality reduction model training includes: Set up the micro-adjustable strategy autoencoder neural network and initialize weights and biases; Reconstructing the expanded feature data into high-dimensional feature data and inputting it into an encoder, propagating it through an activation function and a multi-layer network structure to generate a low-dimensional feature representation; Calculating a total loss function, where the total loss includes a reconstruction loss and a dual loss; According to the total loss function, the network parameters are updated using the gradient descent method; Repeat the iteration until the preset stop condition is met.

6. The method according to claim 1, characterized in that The step of inputting the expanded feature data after data dimension reduction into a classifier model to be trained to perform classifier model training includes: Set up a quantum-coded spiking neural network and initialize weights and biases; Decompose the feature vector of the expanded feature data after the input data dimension reduction into multiple quantum bits, and convert them into quantum states through quantum coding; During the forward propagation of the spiking neural network, the input quantum state is passed through the input layer to the hidden layer; Calculating a dual loss function, wherein the dual loss function includes a cross entropy loss and a quantum state loss; Updating the weights and biases of the spiking neural network based on the gradient back-propagation error of the dual loss function; Adaptively adjust the learning rate and the firing threshold of the spiking neuron according to the change of the dual loss function during the training process; Repeat the iteration until the preset stop condition is met.

7. The method according to any one of claims 1 to 6, characterized in that: The performing equipment fault diagnosis on the equipment to be diagnosed by using the feature extraction model, the data dimension reduction model and the classifier model includes: The trained feature extraction model, data dimension reduction model and classifier model are used to process the relevant data of the device to be diagnosed and output the diagnosis results.

8. An equipment fault diagnosis device based on incremental learning, characterized in that: include: A data expansion unit, used to obtain relevant data of the sample device, and expand the relevant data through the generator and the discriminator to obtain expanded data; A feature extraction model training unit, used for inputting the expanded data into a feature extraction model to be trained to perform feature extraction model training, and extracting expanded feature data of the expanded data through the feature extraction model; A feature dimension reduction model training unit, used for inputting the expanded feature data into the feature dimension reduction model to be trained to perform feature dimension reduction model training, and performing data dimension reduction on the feature dimension reduction model through the feature dimension reduction model; A classifier training unit, used for inputting the expanded feature data after data dimension reduction into a classifier model to be trained to perform classifier model training; The model application unit is used to perform equipment fault diagnosis on the equipment to be diagnosed through the feature extraction model, the data dimension reduction model and the classifier model.

9. An electronic device, characterized in that: The electronic device comprises: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the incremental learning-based equipment fault diagnosis method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the device fault diagnosis method based on incremental learning according to any one of claims 1 to 7 when executed.

11. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the device fault diagnosis method based on incremental learning according to any one of claims 1 to 7.

Images