Small sample process monitoring method suitable for multi-working-condition environment

Abstract

The invention discloses a small sample process monitoring method suitable for a multi-working-condition environment, which is used for solving the problem of small sample fault detection under different working conditions. According to the method, the prior knowledge of the working condition is embedded into the model-independent meta-learning framework, so that the learning ability is improved. A meta-learning framework is used to construct a plurality of small sample tasks, a convolutional neural network is used to carry out feature extraction on samples, an embedded vector representing working condition modes is constructed, a sample attention mechanism is introduced, and it is ensured that the samples containing more working condition mode information have higher weights in the process of generating the embedded vector. A plurality of condition layers are introduced into a basic classifier of meta learning, a characteristic linear modulation method is used, modulation parameters are generated by calculating embedded vectors, then the parameters of a basic classifier network are adjusted point by point, and working condition prior information is quoted.

Description

Technical Field

[0001] The present invention relates to the field of industrial monitoring technology, and in particular to a small sample process monitoring method suitable for multiple working conditions. Background Art

[0002] With the constant changes in load, temperature, humidity and other working conditions in industrial environments, traditional monitoring models often find it difficult to maintain stable performance under various working conditions, resulting in a significant decrease in the accuracy and efficiency of fault monitoring. Traditional methods are usually unable to fully adapt and generalize in complex and changing environments, which in turn affects the reliability and timeliness of fault monitoring. At the same time, data collection faces the difficulty of sample scarcity, especially in extreme or rare working conditions, where high-quality data is even more difficult to obtain. Therefore, it is particularly urgent to develop a high-precision and adaptable monitoring method for fault monitoring under small samples and multiple working conditions.

[0003] In order to solve the problems of model performance degradation, insufficient adaptability and lack of generalization ability caused by changes in working conditions, there are currently two main types of solutions. The first type is based on vibration signal processing, mainly using modal decomposition, empirical mode decomposition (EMD) and singular value decomposition (SVD) techniques to extract vibration features. In practical applications, in the process from time domain decomposition to frequency domain feature extraction, this method may have eigenmode overlap, resulting in cross-interference between signal components, thereby reducing the accuracy of feature extraction and the reliability of diagnostic results. Another method regards the change of working conditions as a migration between different domains, and proposes a fault diagnosis method based on transfer learning (TL). Transfer learning uses the knowledge in one task (source domain) to enhance learning and performance in another similar task (target domain), while reducing the demand for a large amount of labeled data and effectively improving the generalization ability of the model. However, in the actual operation of equipment, the probability of failure is low, especially the difficulty in obtaining samples of rare fault types, presenting a typical small sample learning problem.

[0004] Meta-learning (ML) provides a new training paradigm for small-sample learning and has gradually shown its important value in the field of fault detection and diagnosis. This paradigm extracts common learning patterns by simulating multiple small-sample learning tasks, so that the model can quickly adapt to new tasks and maintain high diagnostic performance even when data is scarce. Although the above-mentioned meta-learning-based framework has alleviated the small-sample learning problem to a certain extent, it still faces the challenge of insufficient generalization ability when dealing with cross-domain tasks or tasks with significant distribution differences. This limitation is mainly due to the shortcomings of existing methods in feature extraction and distribution alignment, which make it difficult to effectively capture the complex relationships between different tasks. Summary of the invention

[0005] In order to solve the limitations and defects of the prior art, the present invention provides a small sample process monitoring method applicable to multiple working conditions, comprising: Step S1, using sensors to collect industrial multi-operating condition data; Step S2, standardizing the industrial multi-operating condition data; Step S3, using the standardized industrial multi-condition data, constructing a small sample task in the meta-learning framework according to the N-way and K-shot settings, wherein the N-way means that in each meta-learning task, the model classifies from N different categories, and the K-shot means that for each category, the model randomly selects K samples of each category during training; Step S4: A sample attention mechanism is introduced into the meta-learning method based on prior knowledge of working conditions, and the sample attention mechanism is used to dynamically assign a weight to each sample and highlight the information contribution of key samples; Step S5, using a convolutional neural network to extract features from the sample set after weighted processing by the sample attention mechanism, so as to provide preset features to support the generation of a working condition embedding vector; Step S6: Based on the sample set after weighted processing by the sample attention mechanism, a mapping function is used to generate a working condition embedding vector, where the working condition embedding vector includes feature information of the sample and is used to reflect the commonalities and differences between working conditions; Step S7, a preset number of conditional layers are introduced into the meta-learning-based basic classifier, wherein the conditional layer uses a feature linear modulation method, generates modulation parameters according to the working condition embedding vector, adjusts the parameters of the basic classifier network point by point, and introduces working condition prior information.

[0006] Optionally, step S2 further includes: The expression of the standardization process is as follows: , in, is the original value, is the value after normalization. is the mean of the bearing data, is the standard deviation of the bearing data.

[0007] Optionally, step S4 further includes: The format expression of the support set samples in the meta-learning method is as follows: , in, Indicates input, Indicates the corresponding label, k For each sample in a batch; Calculate the attention score and perform softmax normalization on the samples of the support set batch size. The expression is as follows: , Among them, S i is the attention score of the i-th sample, is the corresponding attention weight, exp represents the exponential function, batch_size Indicates the total number of samples in the batch; The attention weight is assigned to the sample as follows: , in, is the input tensor, is the attention-weighted sample.

[0008] Optionally, step S5 further includes: Taking the support set samples as input, the vector representation of the working condition information is generated by embedding the network model. The expression is as follows: , in, are the embedded network parameters, To use the convolutional neural network for feature extraction, is the attention-weighted sample, y is the corresponding label, v is a vector representation of the working condition information.

[0009] Optionally, step S6 further includes: Get the working condition embedding vector The expression is as follows: , in, is the number of dimensions, It is input vector, is the load case embedding vector, is the weight matrix, is the bias parameter, is a linear transformation function or a nonlinear transformation function.

[0010] Optionally, the step S7 further includes: The conditional layer modifies the parameters of the base classifier network through the dot product operation of the modulation parameters, and the expression is as follows: , in, Z is the working condition embedding vector, represents the modulated characteristics, ω g is the modulation factor, B(x) represents the network before modulation, ⊙ is the Hadamard product, and B(x) and the modulated characteristics Perform element-by-element multiplication operation, and B(x)′ represents the modulated network.

[0011] Optionally, also include: In the inner loop of meta-learning, according to the model-independent meta-learning algorithm, the initial parameters By supporting The gradient descent step on the training task The expression for a single gradient update step is as follows: , in, is the inner loop learning rate, is the support set loss gradient; Using cross entropy as the loss function of the base classifier network, the expression is as follows: , in, ( ) represents the loss function, represents the true label of the i-th sample, Indicates i The input data of samples, log represents the logarithmic function, and ∑ represents the sum of all samples; By optimizing the objective function, the parameters of each meta-learning base classifier network are updated, and the expression is as follows: , Among them, min means minimizing the objective function, Indicates that all samples p(T) Distributed tasks To sum, Indicates the task The loss function is α, which is a hyperparameter and controls the step size of the gradient update. Represents the new loss function calculated after updating the model parameters using the gradient descent method; In the outer loop of meta-learning, the initial parameters are refined by reducing the relevant loss, which consists of the loss associated with the parameters of a specific task on the query set, expressed as follows: , Among them, θ represents the parameters of the model, is the learning rate for updating model parameters, Represents the model parameters θ Calculate the gradient, Represents the model, Indicates that the query data set The loss function on .

[0012] The present invention has the following beneficial effects: The present invention proposes a small sample process monitoring method suitable for a multi-operating condition environment, which is used to solve the problem of small sample fault detection under different working conditions. The present invention embeds the prior knowledge of the working condition into the model-agnostic meta-learning (MAML) framework to improve the learning ability. Specifically, a plurality of small sample tasks are first constructed using the meta-learning framework. Subsequently, the convolutional neural network (CNN) is used to extract features from the samples and construct an embedding vector representing the working condition mode. The sample set contains multiple categories, and the normal samples and the fault samples have different contributions to the representation of the working condition mode. In view of the above problems, the present invention introduces a sample attention mechanism to ensure that in the process of generating the embedding vector, the samples containing more working condition mode information have a higher weight. Afterwards, multiple conditional layers are introduced into the basic classifier of the meta-learning to flexibly introduce the prior information of the working condition. The conditional layer uses the feature-wise linear modulation (FiLM) method to generate modulation parameters by calculating the embedding vector, and then adjusts the parameters of the basic classifier network point by point, and references the prior information of the working condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of a small sample process monitoring method applicable to a multi-operating environment provided in Example 1 of the present invention.

[0014] Figure 2 This is a confusion matrix diagram for small sample fault monitoring under four working conditions provided in Example 1 of the present invention.

[0015] Figure 3 A visual comparison chart of the classification results of various methods provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the technical solution of the present invention, the small sample process monitoring method applicable to a multi-operating environment provided by the present invention is described in detail below with reference to the accompanying drawings.

[0017] Embodiment 1 In response to the actual needs of small sample fault monitoring under multiple working conditions, this embodiment proposes an embedded fault monitoring method that combines meta-learning and prior knowledge of working conditions, aiming to solve the problems of poor adaptability and insufficient generalization ability of traditional methods under changing working conditions. This embodiment introduces a sample attention mechanism and a dynamic embedding network to extract common knowledge between task-specific features and working conditions from a small number of samples, thereby enhancing the robustness and flexibility of the model. The meta-training phase optimizes model parameters to obtain initial knowledge across tasks, and the meta-testing phase quickly tunes and adapts to new working conditions to achieve efficient migration and fault detection from known working conditions to unknown working conditions.

[0018] This embodiment includes: obtaining data and performing data preprocessing, embedding prior knowledge of working conditions, constructing a meta-learning base classifier network, and multi-working condition fault monitoring based on meta-learning.

[0019] This embodiment achieves efficient migration and generalization from known working conditions to unknown working conditions through sample attention mechanism and dynamic feature embedding. The meta-training phase learns common knowledge to provide robust initial parameters for the model, and the meta-test phase completes fault detection of unknown working conditions by quickly adapting the support set. This process ensures the high accuracy and high adaptability of the model under small sample and multi-working conditions.

[0020] To solve the problem of unbalanced working condition information in different samples, the meta-learning method based on working condition prior knowledge introduces a sample attention mechanism. This mechanism dynamically assigns weights to each sample to highlight the information contribution of key samples. Specifically, the attention score is generated by randomly initialized weights and normalized using softmax. Subsequently, the input features of the sample are weighted according to the attention score to generate a weighted sample representation. The collection of these weighted samples is further used as the input for feature extraction, which strengthens the model's ability to focus on key information.

[0021] After sample attention weighting, the model extracts features through a convolutional neural network. CNN consists of four layers, each of which includes convolution operations, batch normalization (BN), and ReLU activation functions, extracting spatial features from low-order to high-order layer by layer. In the last layer, the model uses an adaptive average pooling layer to compress high-dimensional features into a vector of fixed size. This process provides sufficient feature support for the generation of embedded vectors.

[0022] Based on the weighted sample set, the model generates working condition embedding vectors through a mapping function. These vectors not only contain the characteristic information of the samples, but also reflect the commonalities and differences between working conditions. In addition, the model designs a dynamic embedding layer, which dynamically generates embedding vectors by traversing each preset dimension, and finally forms a complete embedding representation, providing a global feature expression for subsequent fault classification.

[0023] In the meta-training phase, the model learns the task characteristics and rules under known working conditions through the support set, and optimizes the network parameters to obtain a good initial state. First, the working condition embedding vector is input into the conditional layer, and the scaling parameters and bias parameters related to the working condition are generated through the gating mechanism, and the feature representation of the basic learner is dynamically modulated. Subsequently, the model calculates the gradient on the support set and iteratively updates the parameters of the embedding network and the base classifier. Finally, the modulated features are input into the classifier to complete the fault category prediction, and the common knowledge between working conditions is gradually learned. The goal of the meta-testing phase is to generalize the model to unknown working condition tasks. With the initial parameters learned in the meta-training phase, the model can quickly adapt to new tasks. When encountering unknown working conditions, the model uses a small number of samples in the support set to extract features, and quickly tunes and adjusts the parameters to adapt to the feature distribution of the new working condition. Subsequently, the model completes fault detection on the query set and evaluates its generalization performance under new tasks.

[0024] This embodiment proposes a new meta-learning method based on prior knowledge to solve the problem of small sample industrial data fault monitoring under different working conditions.

[0025] First, the industrial multi-condition data are collected through sensors. Then, these data are standardized. The specific formula for standardization is shown in formula (1): , in, is the original value, represents the standardized value, and are the mean and standard deviation of the bearing data, respectively.

[0026] Then, the standardized data is used to construct small sample tasks in the meta-learning framework. These tasks need to be constructed according to the N-way, K-shot settings. N-way means that in each meta-learning task, the model needs to classify from N different categories. K-shot means that for each category, the model only randomly selects K samples of each category during training.

[0027] The samples of each meta-learning task are encoded into an embedding vector representing each operating condition. At the same time, recognizing that the sample set contains both normal and abnormal samples, considering the differences between these samples, an attention mechanism is applied to the entire sample set to more accurately represent the conditional pattern information. The sample weights are learned through the attention mechanism and optimized using the fault classification loss to ensure that samples containing more operating condition information have a greater impact on the model. The specific format of the support set samples in the meta-learning task is as follows: , Next, the attention scores are calculated and softmax normalized over the samples of the support batch size: , Among them, is the attention score of the i-th sample, is the corresponding attention weight. Next, the attention weight is assigned to the sample: , in, is the input tensor, is the attention-weighted sample.

[0028] Convolutional neural network (CNN) is used as the main tool to extract sample working condition patterns after weighted processing by attention mechanism. Through the CNN embedding network model, the key information of working condition information is effectively captured. Then, with the support set samples as input, the vector representation of working condition information is generated through the embedding network: , in, are the embedded network parameters, Feature extraction is performed through CNN. In addition, an embedding vector layer is designed to dynamically add a linear layer to each preset dimension in the embedding list. When each embedding layer learns different weights and biases, the resulting embedding vectors show corresponding diversity. The resulting embedding vector as follows: , in, is the number of dimensions, It is input vector, It is the vector after the modulation operation of the working condition information.

[0029] In order to improve the ability of the base classifier network to capture parameters and patterns under different conditions, this study introduced a conditional layer in the base learner structure. The conditional layer is a module that dynamically adjusts the network behavior based on the conditional vector. This design allows the model to consider additional conditional information when processing input data, thereby improving its adaptability and flexibility. The network model of the meta-learning base classifier network has three components: feature extractor, conditional layer, and classifier: The base classifier network adopts a strategy in the feature extractor that combines a wide convolution kernel in the initial layer of the classifier network with multiple narrow convolution kernels in subsequent layers, aiming to address the noise problem that is prevalent in bearing fault monitoring. The wide convolution is used to capture the basic fluctuation and noise patterns in the signal. The design of the wide convolution kernel ensures the coverage of the fluctuation noise that may occur during a bearing fault, thereby preventing the omission of important information due to the convolution kernel being too small.

[0030] At the same time, the conditional layer of the base classifier adopts the method of feature linear modulation, and then calculates the embedding vector through the conditional layer Modulation parameters The parameters of the basic classifier network are modified by the dot product operation of the modulation parameters, as shown in formulas (7) and (8).

[0031] , In this way, the feature extractor focuses on extracting meta-knowledge containing prior information about the working conditions from known working conditions. Based on this learned meta-knowledge, the system can generalize more effectively when faced with new tasks under unknown conditions, thereby achieving accurate fault monitoring under different working conditions.

[0032] The Model-Agnostic Meta-Learning (MAML) algorithm strictly follows a two-layer optimization approach, where the inner loop handles the adaptive update of model parameters and the outer loop focuses on the fine-tuning of meta-parameters.

[0033] Specifically, during the meta-training process, the base network can be viewed as a network parameterized as When faced with a new task , MAML exploits the process of learning a set of initial parameters These parameters are adaptively adjusted to form new parameters In the inner loop, the initial parameters By supporting to adjust to individual training tasks by taking several gradient descent steps on When using a single gradient update step, the flow is as follows: , Among them, the step length is defined as the inner loop learning rate, is the loss gradient. This paper uses cross entropy as the loss function of the basic classifier network: , By optimizing the objective function, the parameters of each meta-learning base classifier network are updated: , In the meta-learning outer loop, the initial parameters are refined by reducing the loss with respect to the task-specific parameters on the query set: , in, The learning rate used to update the model parameters.

[0034] In order to solve the problem of small samples caused by the scarcity of fault samples, this embodiment proposes a meta-learning method based on prior working condition knowledge to realize fault monitoring under different working conditions. This embodiment also proposes a working condition information embedding network framework to convert meta-learning task samples into embedding vectors. This framework uses working condition patterns as prior knowledge to improve the performance of fault monitoring. In addition, this embodiment also proposes a new meta-learning-based basic classifier network, which aims to combine prior knowledge of working conditions for fault monitoring. When faced with unknown working conditions, the basic classifier can cleverly use previously learned prior knowledge to adjust parameters, thereby achieving fast and effective generalization. Due to its strong generalization ability, this method has shown a wide range of practical value in practical applications and has broad market prospects.

[0035] This embodiment proposes a small sample process monitoring method suitable for a multi-operating environment, which is used to solve the problem of small sample fault detection under different working conditions. This embodiment embeds the prior knowledge of the operating condition into a model-independent meta-learning framework to improve the learning ability. Specifically, a plurality of small sample tasks are first constructed using the meta-learning framework. Subsequently, a convolutional neural network is used to extract features from the samples and construct an embedding vector representing the operating mode. The sample set contains multiple categories, and normal samples and fault samples have different contributions to the representation of the operating mode. In response to the above problems, this embodiment introduces a sample attention mechanism to ensure that samples containing more operating mode information have higher weights in the process of generating the embedding vector. Afterwards, multiple conditional layers are introduced into the basic classifier of meta-learning to flexibly introduce the prior information of the operating condition. The conditional layer uses a feature linear modulation method to generate modulation parameters by calculating the embedding vector, and then adjusts the parameters of the basic classifier network point by point, referring to the prior information of the operating condition.

[0036] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A small sample process monitoring method suitable for multiple working conditions, characterized in that: include: Step S1, using sensors to collect industrial multi-operating condition data; Step S2, standardizing the industrial multi-operating condition data; Step S3, using the standardized industrial multi-condition data, constructing a small sample task in the meta-learning framework according to the N-way and K-shot settings, wherein the N-way means that in each meta-learning task, the model classifies from N different categories, and the K-shot means that for each category, the model randomly selects K samples of each category during training; Step S4: A sample attention mechanism is introduced into the meta-learning method based on prior knowledge of working conditions, and the sample attention mechanism is used to dynamically assign a weight to each sample and highlight the information contribution of key samples; Step S5, using a convolutional neural network to extract features from the sample set after weighted processing by the sample attention mechanism, so as to provide preset features to support the generation of a working condition embedding vector; Step S6: Based on the sample set after weighted processing by the sample attention mechanism, a mapping function is used to generate a working condition embedding vector, where the working condition embedding vector includes feature information of the sample and is used to reflect the commonalities and differences between working conditions; Step S7, a preset number of conditional layers are introduced into the meta-learning-based basic classifier, wherein the conditional layer uses a feature linear modulation method, generates modulation parameters according to the working condition embedding vector, adjusts the parameters of the basic classifier network point by point, and introduces working condition prior information.

2. The small sample process monitoring method applicable to multiple working conditions according to claim 1 is characterized in that: The step S2 further comprises: The expression of the standardization process is as follows: ， in, is the original value, is the value after normalization. is the mean of the bearing data, is the standard deviation of the bearing data.

3. The small sample process monitoring method applicable to multiple working conditions according to claim 2 is characterized in that: The step S4 further comprises: The format expression of the support set samples in the meta-learning method is as follows: ， in, Indicates input, Indicates the corresponding label, k For each sample in a batch; Calculate the attention score and perform softmax normalization on the samples of the support set batch size. The expression is as follows: ， Among them, S i is the attention score of the i-th sample, is the corresponding attention weight, exp represents the exponential function, batch_size Indicates the total number of samples in the batch; The attention weight is assigned to the sample as follows: ， in, is the input tensor, is the attention-weighted sample.

4. The small sample process monitoring method applicable to multiple working conditions according to claim 3 is characterized in that: The step S5 further comprises: Taking the support set samples as input, the vector representation of the working condition information is generated by embedding the network model. The expression is as follows: ， in, are the embedded network parameters, To use the convolutional neural network for feature extraction, is the attention-weighted sample, y is the corresponding label, v is a vector representation of working condition information.

5. The small sample process monitoring method applicable to multiple working conditions according to claim 4 is characterized in that: The step S6 further comprises: Get the working condition embedding vector The expression is as follows: ， in, is the number of dimensions, It is input vector, is the load case embedding vector, is the weight matrix, is the bias parameter, is a linear transformation function or a nonlinear transformation function.

6. The small sample process monitoring method applicable to multiple working conditions according to claim 5 is characterized in that: The step S7 further comprises: The conditional layer modifies the parameters of the base classifier network through the dot product operation of the modulation parameters, and the expression is as follows: ， in, Z is the working condition embedding vector, represents the modulated characteristics, ω g is the modulation factor, B(x) represents the network before modulation, ⊙ is the Hadamard product, and B(x) and the modulated characteristics Perform element-by-element multiplication operation, and B(x)′ represents the modulated network.

7. The small sample process monitoring method applicable to multiple working conditions according to claim 6 is characterized in that: Also includes: In the inner loop of meta-learning, according to the model-independent meta-learning algorithm, the initial parameters By supporting The gradient descent step on the training task The expression for a single gradient update step is as follows: ， in, is the inner loop learning rate, is the support set loss gradient; Using cross entropy as the loss function of the base classifier network, the expression is as follows: ， in, ( ) represents the loss function, represents the true label of the i-th sample, Indicates i The input data of samples, log represents the logarithmic function, and ∑ represents the sum of all samples; By optimizing the objective function, the parameters of each meta-learning base classifier network are updated, and the expression is as follows: ， Among them, min means minimizing the objective function, Indicates that all samples p(T) Distributed tasks To sum, Indicates the task The loss function is α, which is a hyperparameter and controls the step size of the gradient update. Represents the new loss function calculated after updating the model parameters using the gradient descent method; In the outer loop of meta-learning, the initial parameters are refined by reducing the relevant loss, which consists of the loss associated with the parameters of a specific task on the query set, expressed as follows: ， Among them, θ represents the parameters of the model, is the learning rate for updating model parameters, Represents the model parameters θ Calculate the gradient, Represents the model, Indicates that the query data set The loss function on .

Images