Industrial fault detection model training method, detection method, equipment and medium
Optimize the Transformer model through wavelet transformation and multi-layer perceptron technologies, solving the industrial fault detection problem of traditional methods in environments of data scarcity and resource constrained, and achieving efficient, accurate and lightweight deployment of the industrial fault detection model , suitable for edge computing devices.
Patent Information
- Application Number
- CN202410936704.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional machine learning methods are difficult to deploy and operate efficiently in resource-constrained environments due to poor data scarcity in industrial failure detection, especially in high-risk industries. The existing Transformer model is difficult to deploy and operate efficiently in resource-constrained environments, and is difficult to directly apply to time. Fault prediction task for sequence data.
Through wavelet transformation technology, industrial operation data is converted into time-frequency domain signals and serialized, and the Transformer model is designed with a multi-layer perceptron and multi-scale convolutional fusion layer to lightweight network, establish an industrial fault detection model, and pass Fine-tuning and quantitative technology optimizes the model structure, suitable for edge computing devices.
It realizes efficient and accurate capture of fault characteristics of the industrial fault detection model, improves the ability to analyze time series data, the overall model is lightweight, simple to deploy, and efficient to run, and is suitable for real-time monitoring and diagnosis of industrial production.
Smart Images

Figure CN118940028A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of industrial technology, and in particular to a training method, a detection method, a device and a medium for an industrial fault detection model. Background Art
[0002] In the industrial field, failure is a low-probability event, but it may bring serious losses and impacts. Especially in high-risk industries such as aerospace, energy, and manufacturing, failures may threaten life safety or interrupt production processes, causing huge losses to the economy. Currently, there are applications that use artificial intelligence technology to predict equipment failures. However, due to the relatively small number of these failure events, the number of data samples available for training is limited. Traditional machine learning methods often do not work well when faced with this data scarcity, especially supervised learning and unsupervised learning methods. Supervised learning requires a large amount of labeled data to train the model, while unsupervised learning may not capture the complexity and diversity of fault behaviors. Therefore, traditional methods have certain limitations in solving industrial fault detection and prediction.
[0003] In recent years, with the development of deep learning technology, especially the rise of Transformer structure model, it has provided new possibilities for solving the challenges brought by data scarcity in the industrial field. Transformer structure model is a neural network architecture based on self-attention mechanism, which has achieved remarkable achievements in natural language processing and image processing. Compared with traditional convolutional neural networks and recurrent neural networks, Transformer structure model can better capture long-range dependencies in sequence data, and has strong parallelization ability, which makes it perform well in processing large-scale data. It is precisely because of the advantages of Transformer structure model that it has become a powerful tool to solve the problem of data scarcity in the industrial field. Compared with traditional methods, Transformer structure model has shown better results in few-sample or even zero-sample learning problems with its attention mechanism. When applied to industrial fault detection tasks, the model based on Transformer structure can learn more complex and abstract feature representations from a small amount of labeled data based on its powerful feature learning ability and generalization ability, so as to achieve accurate identification and prediction of potential fault behaviors.
[0004] Existing Transformer models, especially pre-trained models represented by BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-Trained Transformer), usually have complex model structures and huge parameter scales, and therefore require a large amount of computing resources and memory for effective training and reasoning. Due to this limitation, Transformer models are difficult to deploy and operate efficiently in resource-constrained environments in industrial sites, and their practicality is poor.
[0005] In summary, the problems existing in related technologies need to be solved urgently. Summary of the invention
[0006] The purpose of this application is to solve one of the technical problems existing in the related art to at least a certain extent.
[0007] To this end, an object of an embodiment of the present application is to provide a training method, a detection method, a device and a medium for an industrial fault detection model.
[0008] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include:
[0009] On the one hand, an embodiment of the present application provides a method for training an industrial fault detection model, the method comprising:
[0010] Obtaining first operating data and fault type information of different mechanical equipment in various industrial processes;
[0011] Convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data;
[0012] Performing randomization processing on the sequence data to extract training data from the sequence data;
[0013] Establish a Transformer model, perform lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtain an industrial fault detection model;
[0014] The fault type information and the training data are input into the industrial fault detection model for training to obtain a trained industrial fault detection model.
[0015] In addition, the training method of the industrial fault detection model according to the above embodiment of the present application may also have the following additional technical features:
[0016] Further, in one embodiment of the present application, the obtaining of first operation data of different mechanical equipment in a plurality of industrial processes includes:
[0017] Acquire the first-hand operating data of different mechanical equipment in various industrial processes in real time through sensors;
[0018] Alternatively, historical operating data of different mechanical equipment in a variety of industrial processes are acquired, and first operating data of the mechanical equipment is obtained based on the historical operating data.
[0019] Furthermore, in one embodiment of the present application, the first operation data is converted into a time-frequency domain signal by using a wavelet transform technology, and the time-frequency domain signal is serialized to obtain corresponding sequence data, including:
[0020] Processing the first operation data by wavelet transform technology to obtain first data;
[0021] Normalizing the first data to obtain second data;
[0022] The second data is segmented by non-overlapping windows to obtain the sequence data.
[0023] Furthermore, in one embodiment of the present application, the lightweight network design of the Transformer model by using a multi-layer perceptron and a multi-scale convolutional fusion layer includes:
[0024] Constructing a feature extraction layer of the Transformer model through the multi-scale convolutional fusion layer;
[0025] Replace the multi-head attention layer in the Transformer model with the multi-layer perceptron;
[0026] The fully connected layer at the output of the Transformer model is replaced with a global average pooling layer.
[0027] Furthermore, in one embodiment of the present application, the method further includes:
[0028] The structure of the industrial fault detection model is optimized through fine-tuning technology and quantization technology.
[0029] Furthermore, in one embodiment of the present application, the method further includes:
[0030] Deploy the trained industrial fault detection model to an edge computing device.
[0031] On the other hand, an embodiment of the present application provides an industrial fault detection method, the method comprising:
[0032] Acquiring second operation data of the target mechanical equipment;
[0033] Convert the second operation data into target sequence data by wavelet transform technology and serialization processing;
[0034] The target sequence data is input into the industrial fault detection model trained by the aforementioned industrial fault detection model training method for prediction to obtain a fault detection result.
[0035] On the other hand, an embodiment of the present application provides a training system for an industrial fault detection model, the system comprising:
[0036] An acquisition unit, used to acquire first operation data and fault type information of different mechanical equipment in various industrial processes;
[0037] A conversion unit, configured to convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data;
[0038] A processing unit, used for performing randomization processing on the sequence data to extract training data from the sequence data;
[0039] An optimization unit, used for establishing a Transformer model, performing a lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtaining an industrial fault detection model;
[0040] The updating unit is used to input the fault type information and the training data into the industrial fault detection model for training to obtain a trained industrial fault detection model.
[0041] On the other hand, an embodiment of the present application provides an electronic device, including:
[0042] at least one processor;
[0043] at least one memory for storing at least one program;
[0044] When the at least one program is executed by the at least one processor, the at least one processor is enabled to implement the above method.
[0045] On the other hand, an embodiment of the present application further provides a computer-readable storage medium, which stores a program executable by a processor, and the above-mentioned processor-executable program is used to implement the above-mentioned method when executed by the processor.
[0046] The advantages and benefits of the present application will be partially given in the following description, and partially become apparent from the following description, or be understood through the practice of the present application:
[0047] The training method, detection method, equipment and medium of the industrial fault detection model disclosed in the embodiment of the present application obtain the first operation data and fault type information of different mechanical equipment in various industrial processes; convert the first operation data into time-frequency domain signals through wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data; perform randomization processing on the sequence data, and extract training data from the sequence data; establish a Transformer model, and perform lightweight network design on the Transformer model through multi-layer perceptron and multi-scale convolution fusion layer to obtain an industrial fault detection model; input the fault type information and the training data into the industrial fault detection model for training to obtain a trained industrial fault detection model. This method enables the industrial fault detection model to efficiently and accurately capture fault characteristics, improve the analysis ability of time series data, and the overall model is lightweight, simple to deploy, and efficient to operate. It is suitable for edge computing devices and can effectively meet the needs of real-time monitoring and diagnosis of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present application or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present application. For technical personnel in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0049] Figure 1 A schematic diagram of an implementation environment of a training method for an industrial fault detection model provided in an embodiment of the present application;
[0050] Figure 2 A flowchart of a method for training an industrial fault detection model provided in an embodiment of the present application;
[0051] Figure 3 A schematic diagram of the structure of a multi-scale convolutional fusion layer provided in an embodiment of the present application;
[0052] Figure 4 A schematic diagram of the architecture of an industrial fault detection model provided in an embodiment of the present application;
[0053] Figure 5 A schematic diagram of the overall process of a training method for an industrial fault detection model provided in an embodiment of the present application;
[0054] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0055] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.
[0056] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0058] 1) Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0059] 2) Computer Vision (CV): It is a science that studies how to make machines "see". To be more specific, it refers to machine vision such as using cameras and computers to replace human eyes to identify and measure targets, and further perform graphic processing so that the computer processing becomes an image that is more suitable for human eye observation or transmission to instrument detection. As a scientific discipline, computer vision studies related theories and technologies, and attempts to establish an artificial intelligence system that can obtain information from images or multi-dimensional data. Computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous positioning and map construction, and other technologies, as well as common biometric recognition technologies such as face recognition and fingerprint recognition.
[0060] 3) Machine Learning (ML): A multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies.
[0061] 4) Deep Learning (DL): Deep learning is a new research direction in the field of machine learning. It is used to learn the inherent laws and representation levels of sample data. The information obtained in the learning process is very helpful for interpreting data such as text, images and sounds. Its ultimate goal is to enable machines to have analytical learning capabilities like humans and to recognize data such as text, images and sounds.
[0062] 5) Transformer, a widely used model architecture in the field of deep learning, uses the self-attention mechanism to replace the sequence dependency processing in traditional recurrent neural networks (RNNs) and convolutional neural networks (CNNs). Self-attention allows the model to calculate in parallel when processing sequence data, rather than processing element by element like RNN, which greatly improves computational efficiency and can capture longer context information.
[0063] In the industrial field, failure is a low-probability event, but it may bring serious losses and impacts. Especially in high-risk industries such as aerospace, energy, and manufacturing, failures may threaten life safety or interrupt production processes, causing huge losses to the economy. Currently, there are applications that use artificial intelligence technology to predict equipment failures. However, due to the relatively small number of these failure events, the number of data samples available for training is limited. Traditional machine learning methods often do not work well when faced with this data scarcity, especially supervised learning and unsupervised learning methods. Supervised learning requires a large amount of labeled data to train the model, while unsupervised learning may not capture the complexity and diversity of fault behaviors. Therefore, traditional methods have certain limitations in solving industrial fault detection and prediction.
[0064] In recent years, with the development of deep learning technology, especially the rise of Transformer structure model, it has provided new possibilities for solving the challenges brought by data scarcity in the industrial field. Transformer structure model is a neural network architecture based on self-attention mechanism, which has achieved remarkable achievements in natural language processing and image processing. Compared with traditional convolutional neural networks and recurrent neural networks, Transformer structure model can better capture long-range dependencies in sequence data, and has strong parallelization ability, which makes it perform well in processing large-scale data. It is precisely because of the advantages of Transformer structure model that it has become a powerful tool to solve the problem of data scarcity in the industrial field. Compared with traditional methods, Transformer structure model has shown better results in few-sample or even zero-sample learning problems with its attention mechanism. When applied to industrial fault detection tasks, the model based on Transformer structure can learn more complex and abstract feature representations from a small amount of labeled data based on its powerful feature learning ability and generalization ability, so as to achieve accurate identification and prediction of potential fault behaviors.
[0065] In the related art, the existing Transformer models, especially the pre-trained models represented by BERT and GPT, usually have complex model structures and huge parameter scales, and thus require a large amount of computing resources and memory for effective training and reasoning. Due to this limitation, the Transformer model is difficult to deploy and run efficiently in resource-constrained environments in industrial sites, and its practicality is poor. In order to deploy the Transformer model in a resource-constrained environment, these models need to be optimized and streamlined. However, the traditional lightweight methods represented by model pruning are difficult to achieve good results for the Transformer architecture, and are more likely to cause performance degradation and cannot be directly applied. In addition, the Transformer structure is mainly designed for tasks such as natural language processing, and its model input is usually a word vector obtained by encoding natural language text. For industrial fault detection or prediction scenarios, its input is usually a time series containing binary data, which has a smaller granularity than natural language, and the order of data points within the sequence has a strict order. Therefore, the Transformer structure model applied to natural language processing is difficult to be directly applied to fault prediction tasks based on time series. In summary, as the demand for fault detection and prediction in the industrial field continues to increase, traditional methods have shown obvious shortcomings when facing the problem of data scarcity.
[0066] In view of this, the present application provides a training method, detection method, equipment and medium for an industrial fault detection model, obtains the first operation data and fault type information of different mechanical equipment in various industrial processes; converts the first operation data into a time-frequency domain signal through wavelet transform technology, and performs serialization processing on the time-frequency domain signal to obtain corresponding sequence data; performs randomization processing on the sequence data, and extracts training data from the sequence data; establishes a Transformer model, and performs lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer to obtain an industrial fault detection model; inputs the fault type information and the training data into the industrial fault detection model for training, and obtains a trained industrial fault detection model. This method enables the industrial fault detection model to efficiently and accurately capture fault characteristics, improve the ability to analyze time series data, and the overall model is lightweight, simple to deploy, and efficient to operate. It is suitable for edge computing devices and can effectively meet the needs of real-time monitoring and diagnosis of industrial production.
[0067] Please refer to Figure 1 , Figure 1The schematic diagram of the implementation environment of a training method for an industrial fault detection model provided in an embodiment of the present application is shown. In the implementation environment, the main hardware and software entities involved include a terminal device 110 and a background server 120. The terminal device 110 and the background server 120 are connected in communication.
[0068] Specifically, the training method of the industrial fault detection model provided in the embodiment of the present application can be executed separately on the terminal device 110 side, or can be executed separately on the background server 120 side, or can be executed based on data interaction between the terminal device 110 and the background server 120.
[0069] The terminal device 110 of the above embodiment may include a mobile phone, a computer, a PDA device, an intelligent voice interaction device, a vehicle-mounted terminal, etc., but is not limited thereto. The backend server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.
[0070] The terminal device 110 and the backend server 120 may establish a communication connection via a wireless network or a wired network. The wireless network or wired network uses standard communication technology and / or protocols, and the network may be set to the Internet or any other network, such as but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or any combination of a virtual private network.
[0071] Of course, it is understandable that Figure 1 The implementation environment in the present application embodiment is only provided in the industrial fault detection model training method of some optional application scenarios, the actual application is not fixed Figure 1 The hardware and software environment shown.
[0072] Below, in combination with the introduction of the aforementioned implementation environment, a training method for an industrial fault detection model provided in an embodiment of the present application is introduced and explained.
[0073] Please refer to Figure 2 , Figure 2 is a schematic diagram of a training method for an industrial fault detection model provided in an embodiment of the present application. The training method for the industrial fault detection model includes but is not limited to:
[0074] Step 210: Acquire first operation data and fault type information of different mechanical equipment in various industrial processes;
[0075] Step 220: convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data;
[0076] Step 230: performing random processing on the sequence data to extract training data from the sequence data;
[0077] Step 240: Establish a Transformer model, perform lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtain an industrial fault detection model;
[0078] Step 250: input the fault type information and the training data into the industrial fault detection model for training to obtain a trained industrial fault detection model.
[0079] In an embodiment of the present application, a training method for an industrial fault detection model is provided. The method enables the industrial fault detection model to efficiently and accurately capture fault characteristics and improve the analysis capability of time series data. Moreover, the model is lightweight as a whole, easy to deploy, and efficient to run. It is suitable for edge computing devices and can effectively meet the needs of real-time monitoring and diagnosis of industrial production.
[0080] Specifically, in an embodiment of the present application, the operating data of different mechanical equipment in the process of industrial production and manufacturing can be obtained and recorded as the first operating data. Here, the operating data of the mechanical equipment can be a vibration signal, or it can be temperature data, voltage data, or current data, etc., and the present application does not limit this. It can be understood that the operating data of the mechanical equipment is closely related to the working conditions in which it is located. When the mechanical equipment fails, its corresponding operating data will also have certain abnormalities. In an embodiment of the present application, the fault type information corresponding to these mechanical equipment can be obtained synchronously. Here, the fault type information can be used to characterize the type of fault existing in the mechanical equipment. This information can correspond to the operating data. In an embodiment of the present application, there is no limitation on the types of faults that may exist in various types of mechanical equipment.
[0081] Exemplarily, in some embodiments, the obtaining of first operation data of different mechanical equipment in multiple industrial processes includes:
[0082] Acquire the first-hand operating data of different mechanical equipment in various industrial processes in real time through sensors;
[0083] Alternatively, historical operating data of different mechanical equipment in a variety of industrial processes are acquired, and first operating data of the mechanical equipment is obtained based on the historical operating data.
[0084] In the embodiments of the present application, when obtaining the first operating data, in some embodiments, relevant sensors installed on or around the mechanical equipment, such as vibration sensors, temperature sensors, etc., can be used to collect the corresponding first operating data in real time; in some embodiments, historical working data of different mechanical equipment can also be obtained, and then the first operating data of the mechanical equipment can be analyzed based on the historical working data.
[0085] After obtaining the first operation data, the first operation data can be converted into a time-frequency domain signal using wavelet transform technology, and the time-frequency domain signal can be serialized to obtain the corresponding sequence data. Here, wavelet transform is a mathematical tool that can analyze the signal at different frequencies and time scales. Unlike the traditional Fourier transform that can only analyze the frequency domain characteristics of the signal, the wavelet transform can simultaneously obtain the time and frequency information of the signal, so it has great advantages in processing non-stationary signals (i.e., signals whose frequency changes with time). Specifically, the wavelet transform operates through a special function called a wavelet function, which has a finite duration and a finite bandwidth. By performing a wavelet transform on the signal, a series of local representations of the signal at different time scales and frequencies can be obtained, and these representations are called wavelet coefficients. The wavelet coefficients reflect the characteristics of the signal at the corresponding time scale and frequency, and can thus be used for applications such as signal analysis, compression, denoising and fault detection.
[0086] Specifically, in the embodiment of the present application, the first operation data is converted into a time-frequency domain signal by using the wavelet transform technology, and the time-frequency domain signal is serialized to obtain the corresponding sequence data, including:
[0087] Processing the first operation data by wavelet transform technology to obtain first data;
[0088] Normalizing the first data to obtain second data;
[0089] The second data is segmented by non-overlapping windows to obtain the sequence data.
[0090] In the embodiment of the present application, when the corresponding sequence data is obtained through the first operation data, the original first operation data can be first converted into a time-frequency domain signal through wavelet transform, and then serialized to obtain the sequence data. Specifically, the complex first operation data of the industrial site can be converted into wavelet coefficients of different scales and frequencies through wavelet transform, and then the wavelet coefficients can be reorganized into a sequence form. In this way, the local characteristics of the signal are captured, providing a basis for fault feature extraction.
[0091] In the embodiment of the present application, the specific formula used for wavelet transform is as follows:
[0092]
[0093] Wherein, a is a scale factor, b is a translation factor, ψ(t) is a mother wavelet function, and x(t) is an original signal. Continuous wavelet transform can obtain wavelet basis functions of different scales and positions for different a and b values, thereby providing multi-scale analysis. In the embodiment of the present application, the data obtained after wavelet transform can be recorded as the first data.
[0094] In the embodiment of the present application, the first data obtained after the wavelet transform can be normalized, so as to ensure the stability and generalization ability of the model training, and the data obtained after the normalization is recorded as the second data. Specifically, in the embodiment of the present application, the specific formula used for the normalization process is as follows:
[0095]
[0096] Among them, CWTn is the normalized second data, CWT is the size of the original first data, and CWTmax and CWTmin are the maximum value and minimum value of the original first data respectively.
[0097] In an embodiment of the present application, the normalized second data may be segmented into a plurality of non-overlapping windows, thereby maintaining the continuity and relevance of the time series data, and thus, the corresponding sequence data may be obtained as a model input.
[0098] It should be noted that in the embodiment of the present application, the size of the non-overlapping window and the sliding step size can be adjusted according to the specific problem. For example, a smaller window can capture more detailed changes, but may increase the amount of calculation. The moving window can be used to extract the local features of the signal, and the features such as the mean, variance, and spectrum in each window can be calculated, and then these features can be used in the fault diagnosis model, which is not limited in the embodiment of the present application.
[0099] In the embodiment of the present application, the obtained sequence data can be used to train the industrial fault detection model. Specifically, it can be understood that the amount of sequence data obtained may be relatively large. In the embodiment of the present application, a part (or all) of the sequence data can be extracted as training data. Exemplarily, in the embodiment of the present application, the sequence data can be first randomized to eliminate the order of the sequence data, thereby ensuring that the divided training set and test set are random samples. Then, a stratified sampling method can be used to ensure that the sample ratios of each category in the training set and the test set are similar, and the sampled data are divided, 70% of the data is used for training, and 30% of the data is used for testing. In this way, the data in the training set is the training data, and these data can be used for training later, and then the data in the test set can be used to evaluate the effect of the model. Of course, it can be understood that the data division ratio of the training set and the test set can be flexibly adjusted as needed, and is not fixed to the above example.
[0100] In the embodiment of the present application, the industrial fault detection model can be established based on the Transformer model. Specifically, the lightweight network design of the Transformer model is performed through a multi-layer perceptron and a multi-scale convolutional fusion layer, including:
[0101] Constructing a feature extraction layer of the Transformer model through the multi-scale convolutional fusion layer;
[0102] Replace the multi-head attention layer in the Transformer model with the multi-layer perceptron;
[0103] The fully connected layer at the output of the Transformer model is replaced with a global average pooling layer.
[0104] In an embodiment of the present application, in order to automatically and effectively enrich multi-scale features from the collected operating data, a multi-scale convolutional fusion (MCF) layer can be designed as a feature extraction layer of the industrial fault detection model, so that fault features at multiple time scales can be captured.
[0105] For details, please refer to Figure 3 , Figure 3 The schematic diagram of the structure of a multi-scale convolution fusion layer provided in the embodiment of the present application is shown. In the embodiment of the present application, the multi-scale convolution fusion layer includes four large convolution kernels of different scales, so as to obtain complementary and rich diagnostic information to meet the complex characteristics of fault diagnosis information in the industrial field. In order to significantly increase the nonlinear characteristics, a small 1x1 convolution kernel is connected to each convolution kernel, and then a nonlinear activation function is connected.
[0106] Reference Figure 4 , Figure 4A schematic diagram of the architecture of an industrial fault detection model provided in an embodiment of the present application is shown. In an embodiment of the present application, MLP (multi-layer perceptron) can be used to replace the multi-head attention mechanism in the traditional Transformer model, which can directly eliminate the residual connection in the Transformer and reduce the resource consumption of the attention mechanism. Specifically, the fitting ability of the original Transformer model can be transferred to the lightweight MLP layer through the knowledge distillation technology, and the Encoder structure of the Transformer can be optimized so that the model can process multiple features in parallel and improve the analysis ability of time series data.
[0107] In the embodiment of the present application, the MLP layers are fully connected, the bottom layer of the MLP is the input layer, the middle layer is the hidden layer, and the last layer is the output layer. The hidden layer is fully connected to the input layer. Assuming that the input layer is represented by a vector X, the output of the hidden layer is f(W1X+b1), W1 is the weight, b1 is the bias, and the function f can select a sigmoid function or a tanh function, which is not limited in the present application. In addition, in the embodiment of the present application, a global average pooling layer can be used to replace the fully connected layer at the output end, which can reduce the amount of calculation on the basis of maintaining global information. The global average pooling layer does not require parameters, which can avoid overfitting in this layer. Compared with the fully connected layer, the global average pooling layer is a simpler convolution structure, which is more effective in establishing the relationship between feature maps and categories. Moreover, in the embodiment of the present application, multiple fault diagnosis techniques can also be used to design a classifier to achieve classification and clustering of fault diagnosis, and multiple fault diagnosis can be performed based on symptom decomposition and category distinction methods to solve the problem that outliers cannot be diagnosed.
[0108] In some embodiments, the method further comprises:
[0109] The structure of the industrial fault detection model is optimized through fine-tuning technology and quantization technology.
[0110] In the embodiments of the present application, the model structure can also be optimized through fine-tuning and quantization techniques to reduce the computational complexity of the model and achieve lightweightness, making it suitable for edge computing devices.
[0111] Specifically, here, fine-tuning technology is a transfer learning technology used to adjust a pre-trained model to adapt to a specific task or data set. During the fine-tuning process, the model is usually trained on a task that is similar but not exactly the same as the pre-training task to refine the weights and parameters of the model. Fine-tuning the model using Prompt Tuning technology can make the model better adapted to industrial fault tasks. Quantization technology is a post-processing step that reduces the accuracy of the model. It converts the weights and activations of the model from a high-precision format to a low-precision format. The purpose of quantization is to reduce the size and inference latency of the model, making it more suitable for running in a resource-constrained environment. The model is quantized using the quantization method of post-quantization training. By reducing the size of the model, the model can be deployed in the cloud and edge, that is, the trained industrial fault detection model is deployed to the edge computing device.
[0112] It is understandable that an improved transformer network is proposed in the embodiment of the present application as an industrial fault detection model, which can use the transformer to learn long-term time-related information, thereby improving diagnostic accuracy and noise resistance. Compared with traditional networks, it is expected to achieve lightweight and better stability.
[0113] Reference Figure 5 , Figure 5 The overall flow diagram of a training method for an industrial fault detection model provided in an embodiment of the present application is shown. In an embodiment of the present application, an industrial fault detection model can be used to predict the training set to obtain a prediction result, which includes the prediction result of the model for each sample, which can be a probability distribution or a category label. According to the fault type information corresponding to the training data, evaluation indicators such as accuracy, precision, recall, and F1 score can be determined to facilitate the evaluation of the model. Understand the overall prediction performance of the model and its performance in different categories. For example, in an embodiment of the present application, the accuracy, recall, F1 score and other indicators of each category can be calculated to analyze the prediction of each category by the model, and understand the performance of the model in different categories; analyze the samples with prediction errors to understand the reasons why the model makes mistakes. Consider viewing the feature distribution of the error sample, the confidence score of the model and other information to determine the direction in which the model needs to be improved. According to the results of the analysis, the model is improved and optimized. Specifically, in an embodiment of the present application, the improvement method that can be used includes adjusting model parameters, increasing training data, improving feature engineering, etc., so that the results analysis and model improvement are continuously performed, the performance and generalization ability of the model can be continuously improved.
[0114] In an embodiment of the present application, an industrial fault detection method is also provided, comprising:
[0115] Acquiring second operation data of the target mechanical equipment;
[0116] Convert the second operation data into target sequence data by wavelet transform technology and serialization processing;
[0117] The target sequence data is input into the industrial fault detection model trained by the aforementioned industrial fault detection model training method for prediction to obtain a fault detection result.
[0118] In the embodiment of the present application, after the industrial fault detection model is trained, it can be put into actual application scenarios. For example, for the mechanical equipment to be detected, it can be recorded as the target mechanical equipment, and its corresponding second operation data can be collected. The corresponding target sequence data can be obtained through relevant processing, and the corresponding fault detection result can be predicted based on the trained industrial fault detection model.
[0119] It can be understood that the method in the embodiment of the present application has at least the following advantages:
[0120] 1. This application proposes an innovative solution to the problem of data scarcity in the industrial field. In traditional methods, the complexity of manually designed features makes it difficult to effectively capture the characteristics of the system when data is scarce. In order to overcome this problem, this application proposes a new Transformer model (MLP-Transformer), which can automatically learn important features in the data and avoid the tedious process of manually designing features. Through the Transformer model, patterns and trends in industrial data can be analyzed more accurately, thereby improving the ability to adapt to data scarcity.
[0121] 2. This application combines wavelet transform and MLP mechanism to further enhance the model's ability to capture the characteristics of time series data. Wavelet transform can effectively process non-stationary signals and convert them into frequency domain features, thereby more comprehensively describing the characteristics of time series data. Through the MLP mechanism, complex relationships and nonlinear features in the data can be learned, further enhancing the model's representation ability.
[0122] 3. In order to improve the stability and generalization ability of the model, this application introduces overlapping window technology and global average pooling technology. Overlapping window technology can introduce time overlap in time series data, increase the richness and diversity of data, and thus improve the stability and robustness of the model. Global average pooling technology helps to reduce the overfitting of the model and improve the generalization ability of the model, so that it can achieve good performance in different industrial scenarios.
[0123] 4. This application achieves lightweight deployment of the model through fine-tuning and quantization techniques. Fine-tuning technology can further optimize model parameters while maintaining model performance, making it suitable for embedded devices and real-time application scenarios. Quantization technology can reduce the storage space and computational complexity of the model, thereby improving the deployment efficiency of the model in resource-constrained environments, enabling it to be better applied in industrial production practices.
[0124] In an embodiment of the present application, a training system for an industrial fault detection model is also provided, including:
[0125] An acquisition unit, used to acquire first operation data and fault type information of different mechanical equipment in various industrial processes;
[0126] A conversion unit, configured to convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data;
[0127] A processing unit, used for performing randomization processing on the sequence data to extract training data from the sequence data;
[0128] An optimization unit, used for establishing a Transformer model, performing a lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtaining an industrial fault detection model;
[0129] The updating unit is used to input the fault type information and the training data into the industrial fault detection model for training to obtain a trained industrial fault detection model.
[0130] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0131] Reference Figure 6 , an embodiment of the present application provides an electronic device, including:
[0132] at least one processor 610;
[0133] At least one memory 620, used to store at least one program;
[0134] When at least one program is executed by at least one processor 610 , at least one processor 610 implements the above-mentioned training method of the industrial fault detection model.
[0135] Similarly, the contents of the above method embodiments are all applicable to the electronic device embodiments. The functions specifically implemented by the electronic device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0136] The embodiment of the present application also provides a computer-readable storage medium, which stores a program executable by the processor 610. The program executable by the processor 610 is used to execute the above-mentioned industrial fault detection model training method when executed by the processor 610.
[0137] Similarly, the contents of the above method embodiments are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0138] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the application is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part described as a larger operation is performed independently.
[0139] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present application. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional techniques of the engineer. Therefore, those skilled in the art can implement the present application set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the attached claims and their equivalents.
[0140] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.
[0141] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0142] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.
[0143] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0144] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0145] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.
[0146] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.
Claims
1. A training method for an industrial fault detection model, characterized in that: The method comprises: Obtaining first operating data and fault type information of different mechanical equipment in various industrial processes; Convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data; Performing randomization processing on the sequence data to extract training data from the sequence data; Establish a Transformer model, perform lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtain an industrial fault detection model; The fault type information and the training data are input into the industrial fault detection model for training to obtain a trained industrial fault detection model.
2. The training method of an industrial fault detection model according to claim 1, characterized in that: The obtaining of first operation data of different mechanical equipment in a plurality of industrial processes includes: Acquire the first-hand operating data of different mechanical equipment in various industrial processes in real time through sensors; Alternatively, historical operating data of different mechanical equipment in a variety of industrial processes are acquired, and first operating data of the mechanical equipment is obtained based on the historical operating data.
3. The training method of an industrial fault detection model according to claim 1, characterized in that: The first operation data is converted into a time-frequency domain signal by using a wavelet transform technology, and the time-frequency domain signal is serialized to obtain corresponding sequence data, including: Processing the first operation data by wavelet transform technology to obtain first data; Normalizing the first data to obtain second data; The second data is segmented by non-overlapping windows to obtain the sequence data.
4. The method for training an industrial fault detection model according to claim 1, characterized in that: The lightweight network design of the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer includes: Constructing a feature extraction layer of the Transformer model through the multi-scale convolutional fusion layer; Replace the multi-head attention layer in the Transformer model with the multi-layer perceptron; The fully connected layer at the output of the Transformer model is replaced with a global average pooling layer.
5. The method for training an industrial fault detection model according to claim 4, characterized in that: The method further comprises: The structure of the industrial fault detection model is optimized through fine-tuning technology and quantization technology.
6. A method for training an industrial fault detection model according to any one of claims 1 to 5, characterized in that: The method further comprises: Deploy the trained industrial fault detection model to an edge computing device.
7. An industrial fault detection method, characterized in that: The method comprises: Acquiring second operation data of the target mechanical equipment; Convert the second operation data into target sequence data by wavelet transform technology and serialization processing; The target sequence data is input into the industrial fault detection model trained by the training method of the industrial fault detection model according to any one of claims 1 to 6 for prediction to obtain a fault detection result.
8. A training system for an industrial fault detection model, characterized in that: The system comprises: An acquisition unit, used to acquire first operation data and fault type information of different mechanical equipment in various industrial processes; A conversion unit, configured to convert the first operation data into time-frequency domain signals by wavelet transform technology, and perform serialization processing on the time-frequency domain signals to obtain corresponding sequence data; A processing unit, used for performing randomization processing on the sequence data to extract training data from the sequence data; An optimization unit, used for establishing a Transformer model, performing a lightweight network design on the Transformer model through a multi-layer perceptron and a multi-scale convolutional fusion layer, and obtaining an industrial fault detection model; The updating unit is used to input the fault type information and the training data into the industrial fault detection model for training to obtain a trained industrial fault detection model.
9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to implement the method according to any one of claims 1 to 7 when executed by the processor.
Citation Information
Patent Citations
Online and offline hybrid recommendation method and system
CN110472152A
Rolling bearing fault diagnosis method based on multi-branch multi-scale convolutional neural network
CN110595775A
Deep learning model protection method based on confidential calculation
CN115146237A
Industrial streaming data online fault diagnosis method based on deep knowledge distillation network
CN117113169A
Cloud operation and maintenance knowledge graph construction method based on lightweight pre-training model
CN117556052A
