Industrial equipment fault diagnosis and model construction method and device

Abstract

The invention discloses an industrial equipment fault diagnosis model construction method and device. The method comprises the following steps: acquiring target industrial equipment fault diagnosis training data; preprocessing the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; inputting the target feature vector matrix data into a target mixed deep learning model for training, and outputting target industrial equipment fault diagnosis result data; judging whether the target industrial equipment fault diagnosis result data meets a target preset diagnosis result data requirement or not; if yes, recording target mixed deep learning model parameters corresponding to the current target mixed deep learning model, and obtaining a target industrial equipment fault diagnosis teacher model; and performing knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model. The method has the advantages of being high in diagnosis precision, excellent in calculation efficiency, high in interpretability and high in data fusion capacity.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial Internet, and particularly to a method and device for industrial equipment fault diagnosis and model construction, which are applicable to industrial equipment status monitoring, fault prediction and anomaly detection. Background Art

[0002] With the rapid development of industrial Internet, intelligent monitoring and maintenance of industrial equipment have become a key link to ensure stable production operation. The existing industrial equipment fault diagnosis methods mainly include the following: First is rule-based diagnosis, that is, relying on manually formulated fault rules, which are difficult to adapt to complex industrial environments and have high rule update costs; Second is machine learning-based diagnosis, that is, traditional machine learning methods (such as random forest and SVM, etc.) rely on manual feature engineering and have weak adaptability to data distribution changes; Third is deep learning-based diagnosis, that is, although convolutional neural network (CNN) or recurrent neural network (RNN) can automatically extract features, there are problems such as poor generalization and high computational overhead when dealing with long-time series and multi-modal industrial equipment operation data.

[0003] Currently, large models (such as Transformer, GPT, and BERT) have made breakthrough progress in fields such as natural language processing and computer vision. Their powerful data understanding and feature learning capabilities make them have great application potential in industrial equipment fault diagnosis. However, directly applying large models in industrial scenarios still faces the following challenges: First, it is the challenge of high computational cost, that is, industrial equipment usually requires low-latency fault diagnosis, while large model inference has high overhead and is difficult to meet real-time requirements; Second, it is the challenge of poor data adaptability, that is, industrial data has the characteristics of high noise and multi-modal, and existing large models are mainly for text and images, lacking specially optimized architectures; Second, it is the challenge of lack of interpretability, that is, industrial equipment fault diagnosis needs to provide clear root cause analysis of faults, while large models are usually "black boxes" and it is difficult to explain their diagnostic decisions. Summary of the Invention

[0004] Based on this, it is necessary to address the above problems and propose a device for constructing an industrial equipment fault diagnosis model to solve the following problems of the prior art: large models for industrial equipment fault diagnosis are difficult to meet real-time requirements, have poor data adaptability, and lack interpretability of diagnostic decisions.

[0005] The first technical solution of the embodiment of the present invention is as follows:

[0006] A method for constructing a fault diagnosis model of industrial equipment, comprising: obtaining target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multi-modal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multi-modal industrial equipment operation data; preprocessing the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; inputting the target feature vector matrix data into a target hybrid deep learning model to be trained for training, and outputting target industrial equipment fault diagnosis result data; where the target hybrid deep learning model consists of a target transformer neural network and a target temporal convolutional neural network; determining whether the target industrial equipment fault diagnosis result data meets the requirements of target preset diagnosis result data; if so, recording the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model; performing knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model.

[0007] The second technical solution of the embodiment of the present invention is:

[0008] A method for diagnosing faults of industrial equipment, which is implemented based on the industrial equipment fault diagnosis model described in any one of the above, comprising: obtaining target industrial equipment operation data to be diagnosed, where the target industrial equipment operation data includes target multi-modal industrial equipment operation data; preprocessing the target industrial equipment operation data to obtain target feature vector matrix operation data; inputting the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing, and outputting target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; diagnosing the operating condition of the target industrial equipment corresponding to the target industrial equipment operation data according to the target industrial equipment fault prediction result data.

[0009] The third technical solution of the embodiment of the present invention is:

[0010] A computer device, comprising a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps:

[0011] Obtain the target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multi-modal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multi-modal industrial equipment operation data; preprocess the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; input the target feature vector matrix data into a target hybrid deep learning model to be trained and output target industrial equipment fault diagnosis result data; wherein, the target hybrid deep learning model consists of a target transformer neural network and a target temporal convolutional neural network; determine whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data; if so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model; perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model; or perform the following steps:

[0012] Obtain the target industrial equipment operation data to be diagnosed, where the target industrial equipment operation data includes target multi-modal industrial equipment operation data; preprocess the target industrial equipment operation data to obtain target feature vector matrix operation data; input the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing and output target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; diagnose the operating condition of the target industrial equipment corresponding to the target industrial equipment operation data according to the target industrial equipment fault prediction result data.

[0013] The fourth technical solution of the embodiment of the present invention is:

[0014] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to perform the following steps:

[0015] Obtain the target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multi-modal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multi-modal industrial equipment operation data; preprocess the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; input the target feature vector matrix data into a target hybrid deep learning model to be trained and output target industrial equipment fault diagnosis result data; where the composition of the target hybrid deep learning model includes a target transformer neural network and a target temporal convolutional neural network; determine whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data; if so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model; perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model; or perform the following steps:

[0016] Obtain the target industrial equipment operation data to be diagnosed, where the target industrial equipment operation data includes target multi-modal industrial equipment operation data; preprocess the target industrial equipment operation data to obtain target feature vector matrix operation data; input the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing and output the target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; diagnose the operating condition of the target industrial equipment corresponding to the target industrial equipment operation data according to the target industrial equipment fault prediction result data.

[0017] The present invention first obtains the target industrial equipment fault diagnosis training data, secondly preprocesses the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data, secondly inputs the target feature vector matrix data into a target hybrid deep learning model to be trained and outputs target industrial equipment fault diagnosis result data, and then determines whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data. If so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model, and finally perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model. Compared with the prior art, the present invention has the following beneficial effects:

[0018] First, it can optimize the large model architecture and improve the adaptability of industrial equipment fault diagnosis in the scenario.

[0019] Second, it can reduce the inference latency and enable the industrial equipment fault diagnosis to meet the real-time requirements of the industrial site.

[0020] Third, it can enhance the interpretability and improve the diagnostic credibility and usability of the large model.

[0021] Specifically, compared with the prior art, the present invention has the following significant advantages:

[0022] 1. High diagnostic accuracy. The prior art mainly relies on a single model structure of CNN (Convolutional Neural Network) or RNN (Recurrent Neural Network), and it is difficult to handle long-term temporal dependencies and local features simultaneously. The present invention innovatively combines a hybrid architecture of Transformer (i.e., the target transformer neural network of the present invention) and TCN (i.e., the target temporal convolutional neural network of the present invention), which can capture long-range dependencies through the self-attention mechanism and extract local temporal features using causal convolution, thereby improving the accuracy of fault identification.

[0023] 2. Excellent computational efficiency. Traditional deep learning models have high computational overhead and are difficult to deploy in industrial sites. The present invention realizes lightweight model design through knowledge distillation and model compression technologies, significantly reducing the computational complexity and making the system more suitable for industrial real-time diagnosis requirements.

[0024] 3. Strong interpretability. Existing deep learning methods are often "black box" models, and the diagnostic results lack interpretability. The present invention combines attention mechanism visualization and knowledge graph technology to make the diagnostic process more transparent, facilitating engineers to understand and verify the diagnostic results.

[0025] 4. Strong data fusion ability. Traditional methods can usually only process single-type data, while the feature fusion module designed in the present invention can process multi-source heterogeneous data simultaneously, improving the system's ability to identify complex faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0027] Among them:

[0028] Figure 1 is the flowchart of an implementation manner of a method for constructing an industrial equipment fault diagnosis model in an embodiment;

[0029] Figure 2 is the flowchart of an implementation manner of a method for diagnosing industrial equipment faults in an embodiment;

[0030] Figure 3 It is a structural block diagram of an implementation manner of a computer device in an embodiment. Specific implementation manner

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] Please refer to Figure 1 Combined with Figure 1 It can be obtained that a method for constructing an industrial equipment fault diagnosis model in an embodiment of the present invention includes the following steps:

[0033] Step S101: Obtain target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multimodal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multimodal industrial equipment operation data.

[0034] Among them, the target historical multimodal industrial equipment operation data includes target historical sensor time series data, target historical equipment log data, and target historical vibration image data corresponding to the target industrial equipment, and the target historical industrial equipment fault data includes target historical signal loss data and target historical parameter anomaly data.

[0035] Among them, the types of industrial equipment faults vary greatly according to different industries and actual scenarios. This step does not limit the specific type of fault. For example: used in an electrical scenario, it can be faults such as short circuits and open circuits. Used in a data acquisition scenario, it can be sensor faults such as data anomalies and signal losses.

[0036] Step S102: Preprocess the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data.

[0037] Step S103: Input the target feature vector matrix data into a target hybrid deep learning model to be trained for training, and output target industrial equipment fault diagnosis result data; among them, the target hybrid deep learning model includes a target transformer neural network and a target temporal convolutional neural network.

[0038] Among them, the target transformer neural network in this step can be optionally Transformer, and the target temporal convolutional neural network in this step can be optionally TCN.

[0039] Among them, Transformer is a powerful deep learning model architecture, especially suitable for processing sequential data such as text, and has achieved remarkable results in the field of natural language processing. Its self-attention mechanism enables the model to capture global dependencies in the input sequence, thereby improving processing efficiency and performance. TCN (Temporal Convolutional Network) is a convolutional neural network used to process time series data. Compared with traditional recurrent neural networks, TCN can better capture temporal correlations and is not prone to the problem of gradient vanishing. TCN has a wide range of applications in fields such as audio, video, language, and finance.

[0040] Step S104: Determine whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data.

[0041] Among them, the target industrial equipment fault diagnosis result data includes a correspondence table between equipment operation data and fault prediction results. The equipment operation data corresponds to the target historical multi-modal industrial equipment operation data, and the fault prediction results correspond to the target historical industrial equipment fault data. Specifically, at each specific time node, there are corresponding various operation data of the industrial equipment, and each operation data corresponds one-to-one with the historical fault data. If at each specific time node, the fault prediction results match (are the same or differ slightly) the target historical industrial equipment fault data, it means that the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data.

[0042] Step S105: If the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model.

[0043] Among them, if the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data, it means that the training of the target hybrid deep learning model has been successfully completed and needs to be recorded.

[0044] Step S106: Perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model.

[0045] Among them, in order to achieve model lightweight, this step needs to use knowledge distillation and model compression methods to optimize the model of the target industrial equipment fault diagnosis teacher model.

[0046] In this embodiment, optionally, the preprocessing of the target industrial equipment fault diagnosis training data to obtain the target feature vector matrix data includes:

[0047] First, perform data cleaning, data standardization, and data alignment on the target industrial equipment fault diagnosis training data in sequence to obtain the target standardized training data.

[0048] Second, perform time-domain feature extraction, frequency-domain feature extraction, feature fusion, and feature selection on the target standardized training data in sequence to obtain the target feature vector matrix data.

[0049] Among them, the data cleaning process includes removing outliers, handling missing values, and eliminating noise from the target industrial equipment fault diagnosis training data. The data standardization process includes normalizing the target industrial equipment fault diagnosis training data after data cleaning using the Z-score or Min-Max method. The data alignment process includes time alignment processing of the multi-source heterogeneous data of the target industrial equipment fault diagnosis training data after normalization processing.

[0050] Among them, Z-Score standardization, also known as standard score or Z-score, is a statistical measurement method used to represent the relative position of a value relative to the mean and standard deviation of the entire data set.

[0051] Among them, Min-max standardization, also known as deviation standardization, is a common data preprocessing technique mainly used to scale data to the range [0,1].

[0052] Among them, time-domain feature extraction and frequency-domain feature extraction are two commonly used methods in data processing, which respectively focus on the characteristics of data in the time domain and frequency domain. Time-domain feature extraction refers to the features directly extracted from time series data, which describe the characteristics of data in time. Time-domain features are very important for analyzing and understanding the dynamic behavior of data, and are widely used in fields such as data processing, audio analysis, and biomedical signal processing. Frequency-domain features are the features extracted from the frequency components of data, which describe the distribution of data at different frequencies. Frequency-domain features help to understand the frequency characteristics and energy distribution of data.

[0053] In short, by extracting time-domain and frequency-domain features, the characteristics of the signal can be more comprehensively understood, thereby providing useful information for subsequent data processing and analysis.

[0054] In this embodiment, optionally, the inputting the target feature vector matrix data into the target hybrid deep learning model to be trained and outputting the target industrial equipment fault diagnosis result data includes:

[0055] First, input the target feature vector matrix data into the target hybrid deep learning model to be trained for training.

[0056] Among them, in this step, the deep learning model is used to complete the deep characterization of the fault features and the fault mode recognition of the target feature vector matrix data.

[0057] Second, extract the time-domain features and frequency-domain features of the target feature vector matrix data through the target transformer neural network, and process the time-series features of the target feature vector matrix data through the target temporal convolutional neural network to complete the deep characterization of the fault features and the fault mode recognition corresponding to the target feature vector matrix data.

[0058] Third, output the target industrial equipment fault diagnosis result data corresponding to the time-series features of the target feature vector matrix data.

[0059] Among them, the extraction of the time-domain features and frequency-domain features of the target feature vector matrix data through the target transformer neural network includes:

[0060] First, capture the long-range dependencies in the sequence data corresponding to the target feature vector matrix data through the target multi-head self-attention mechanism of the target transformer neural network.

[0061] Second, retain the time-series information of the sequence corresponding to the target feature vector matrix data through the target position encoding module of the target transformer neural network.

[0062] Third, improve the feature expression ability of the target feature vector matrix data through the feed-forward neural network layer of the target transformer neural network, and use layer normalization technology to ensure the stability of the training process of the target hybrid deep learning model and the improvement of the convergence speed.

[0063] Specifically, first, for the target feature vector matrix data, Transformer (the target transformer neural network) is used as the backbone network for feature extraction. Transformer captures long-range dependencies in sequence data through the multi-head self-attention mechanism, where the position encoding module is used to retain the temporal information of the sequence. The feed-forward neural network layer in the network further enhances the feature expression ability, and at the same time, the Layer Normalization technology is used to ensure the stability of the training process and effectively improve the convergence speed of the model. Among them, Layer Normalization is a technology used to regularize input features in neural networks, mainly used to improve the stability and training speed of the model. Second, in terms of temporal modeling, the TCN (i.e., the target temporal convolutional neural network) structure is introduced to process temporal features. TCN uses causal convolutional layers to ensure that only historical information is used during prediction, and dilated convolutions are used to expand the receptive field to achieve effective modeling of long sequence data. The residual connection structure in the network alleviates the vanishing gradient problem of deep networks, and the multi-layer stacking design further enhances the expression ability of the model.

[0064] Since the self-attention mechanism itself does not contain the order information of the elements in the sequence, Transformer provides a unique encoding for each input position through position encoding, enabling the model to understand the relative positions of the elements in the sequence. In addition, Transformer uses residual connections between the input and output of each sub-layer to avoid the vanishing gradient problem in deep networks and improve the training efficiency of the model.

[0065] In this embodiment, optionally, the knowledge distillation and model compression processing of the target industrial equipment fault diagnosis teacher model to obtain the target industrial equipment fault diagnosis model includes:

[0066] First, perform knowledge distillation processing on the target industrial equipment fault diagnosis teacher model to obtain the target industrial equipment fault diagnosis student model; wherein, the distillation loss of converting the target industrial equipment fault diagnosis teacher model into the target industrial equipment fault diagnosis student model through knowledge distillation is less than the preset distillation loss.

[0067] Among them, in this step, through a pre-designed distillation loss function, the knowledge of the target industrial equipment fault diagnosis teacher model is transferred to the target industrial equipment fault diagnosis student model. During the knowledge distillation process, a joint optimization strategy is adopted, taking into account both the supervision information of the true label and the soft label guidance of the target industrial equipment fault diagnosis teacher model to ensure that the target industrial equipment fault diagnosis student model is lightweight while maintaining high performance.

[0068] Among them, Knowledge Distillation is a machine learning technique used to transfer the knowledge contained in a large and complex model (referred to as the "teacher model") to a small and lightweight model (referred to as the "student model"). In this way, the student model can inherit the performance of the teacher model as much as possible while maintaining a small model size.

[0069] The core idea of knowledge distillation is to use the output probability distribution (soft target) of the teacher model to guide the training of the student model, rather than relying solely on the true labels (hard targets). This method not only replicates the output results of the teacher model, but more importantly, mimics its "thinking process", enabling the student model to inherit its generalization ability and reasoning logic.

[0070] Second, optimize and compress the model of the target industrial equipment fault diagnosis student model to obtain a target industrial equipment fault diagnosis model.

[0071] In this embodiment, optionally, the optimizing and compressing the model of the target industrial equipment fault diagnosis student model to obtain a target industrial equipment fault diagnosis model includes:

[0072] First, reduce the precision of the model parameters corresponding to the target industrial equipment fault diagnosis student model.

[0073] Second, use model pruning technology to remove redundant connections of the target industrial equipment fault diagnosis student model and simplify the structure of the target industrial equipment fault diagnosis student model.

[0074] Specifically, in this step, the precision of the model parameters of the target industrial equipment fault diagnosis student model is reduced from 32 bits to 8 bits through 8-bit quantization, significantly reducing the model storage space. At the same time, use model pruning technology to remove redundant connections and appropriately simplify the model structure. These optimization measures greatly improve the inference efficiency while maintaining the model accuracy.

[0075] After the above processing steps, finally output an optimized target industrial equipment fault diagnosis model that not only maintains high diagnostic accuracy but also has good practicality. The target industrial equipment fault diagnosis model can effectively balance accuracy and efficiency and is suitable for deployment and use in actual industrial environments.

[0076] Please refer to Figure 2 , combined with Figure 2 It can be obtained that a method for diagnosing industrial equipment faults according to an embodiment of the present invention is implemented based on the above industrial equipment fault diagnosis model, and it includes:

[0077] Step S201: Obtain the operation data of the target industrial equipment to be diagnosed, where the operation data of the target industrial equipment includes target multi-modal industrial equipment operation data.

[0078] Step S202: Preprocess the operation data of the target industrial equipment to obtain the operation data of the target feature vector matrix.

[0079] Step S203: Input the operation data of the target feature vector matrix into the industrial equipment fault diagnosis model for processing, and output the target industrial equipment fault prediction result data corresponding to the operation data of the target industrial equipment.

[0080] Step S204: According to the target industrial equipment fault prediction result data, diagnose the operation status of the target industrial equipment corresponding to the operation data of the target industrial equipment.

[0081] In this embodiment, optionally, the preprocessing of the operation data of the target industrial equipment to obtain the operation data of the target feature vector matrix includes:

[0082] First, perform data cleaning, data standardization, and data alignment processing on the operation data of the target industrial equipment in sequence to obtain the target standardized operation data.

[0083] Second, perform time-domain feature extraction, frequency-domain feature extraction, feature fusion, and feature selection processing on the target standardized operation data in sequence to obtain the operation data of the target feature vector matrix.

[0084] Among them, the data cleaning process includes removing outliers, processing missing values, and eliminating noise from the operation data of the target industrial equipment. The data standardization process includes normalizing the operation data of the target industrial equipment after data cleaning using the Z-score or Min-Max method. The data alignment process includes time alignment processing of multi-source heterogeneous data of the operation data of the target industrial equipment after normalization processing.

[0085] In this embodiment, optionally, the diagnosing the operation status of the target industrial equipment corresponding to the operation data of the target industrial equipment according to the target industrial equipment fault prediction result data includes:

[0086] First, calculate the target industrial equipment fault prediction confidence corresponding to the target industrial equipment fault prediction result data.

[0087] Among them, in this step, the confidence is calculated through prediction variance analysis, that is, multiple predictions are made for the same input, and the dispersion degree of the prediction results is calculated. That is, the more stable the prediction, the higher the confidence.

[0088] Second, according to the confidence level of the target industrial equipment fault prediction, generate a target industrial equipment fault diagnosis report for the target industrial equipment operation data corresponding to the target industrial equipment.

[0089] Among them, in this step, by receiving the real-time operation data of the industrial equipment in real time and performing online fault identification. First, feature extraction and standardization preprocessing are performed on the real-time operation data to ensure that the data format is consistent with the training model stage. Subsequently, the optimized industrial equipment fault diagnosis model performs inference on the processed data and outputs the target industrial equipment fault prediction result data. At the same time, calculate the confidence level of the target industrial equipment fault prediction result data to provide a reliable reference for subsequent decision-making. Finally, integrate the fault type and confidence level information to generate a standardized diagnosis report.

[0090] Figure 3 The internal structure diagram of a computer device in an embodiment is shown. This computer device can specifically be a terminal or a server. As Figure 3 shown, this computer device includes a processor, a memory, and a network interface connected through a device bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of this computer device stores an operating device and can also store a computer program. When the computer program is executed by the processor, the processor can implement the above-mentioned method for constructing an industrial equipment fault diagnosis model. The internal memory can also store a computer program. When the computer program is executed by the processor, the processor can execute the above-mentioned method for constructing an industrial equipment fault diagnosis model or execute the above-mentioned method for diagnosing an industrial equipment fault in any one of the above. Those skilled in the art can understand, Figure 3 the structure shown in

[0091] In another embodiment, a computer device is proposed, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the following steps:

[0092] Obtain target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multi-modal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multi-modal industrial equipment operation data; preprocess the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; input the target feature vector matrix data into a target hybrid deep learning model to be trained and output target industrial equipment fault diagnosis result data; where the composition of the target hybrid deep learning model includes a target transformer neural network and a target temporal convolutional neural network; determine whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data; if so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model; perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model; or perform the following steps:

[0093] Obtain target industrial equipment operation data to be diagnosed, where the target industrial equipment operation data includes target multi-modal industrial equipment operation data; preprocess the target industrial equipment operation data to obtain target feature vector matrix operation data; input the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing and output target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; diagnose the operating condition of the target industrial equipment corresponding to the target industrial equipment operation data according to the target industrial equipment fault prediction result data.

[0094] In another embodiment, a computer-readable storage medium is proposed, storing a computer program, which when executed by a processor causes the processor to perform the following steps:

[0095] Obtain target industrial equipment fault diagnosis training data, where the target industrial equipment fault diagnosis training data includes target historical multi-modal industrial equipment operation data and target historical industrial equipment fault data corresponding and matched to the target historical multi-modal industrial equipment operation data; preprocess the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; input the target feature vector matrix data into a target hybrid deep learning model to be trained and output target industrial equipment fault diagnosis result data; where the composition of the target hybrid deep learning model includes a target transformer neural network and a target temporal convolutional neural network; determine whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data; if so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain a target industrial equipment fault diagnosis teacher model; perform knowledge distillation and model compression processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model; or perform the following steps:

[0096] Obtain target industrial equipment operation data to be diagnosed, where the target industrial equipment operation data includes target multi-modal industrial equipment operation data; preprocess the target industrial equipment operation data to obtain target feature vector matrix operation data; input the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing and output target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; diagnose the operating condition of the target industrial equipment corresponding to the target industrial equipment operation data according to the target industrial equipment fault prediction result data.

[0097] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0098] In the embodiment of the present invention, first, target industrial equipment fault diagnosis training data is obtained. Secondly, the target industrial equipment fault diagnosis training data is preprocessed to obtain target feature vector matrix data. Secondly, the target feature vector matrix data is input into a target hybrid deep learning model to be trained for training, and target industrial equipment fault diagnosis result data is output. Then, it is judged whether the target industrial equipment fault diagnosis result data already meets the requirements of the target preset diagnosis result data. If so, the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model are recorded to obtain a target industrial equipment fault diagnosis teacher model. Finally, knowledge distillation and model compression processing are performed on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis model. Compared with the prior art, the present invention has the following beneficial effects:

[0099] First, the large model architecture can be optimized to improve the adaptability of industrial equipment fault diagnosis in scenarios.

[0100] Second, the inference latency can be reduced to enable industrial equipment fault diagnosis to meet the real-time requirements of the industrial field.

[0101] Third, the interpretability can be enhanced to improve the diagnostic credibility and usability of the large model.

[0102] Specifically, compared with the prior art, the present invention has the following significant advantages:

[0103] 1. High diagnostic accuracy. Existing technologies mainly rely on a single model structure such as CNN (Convolutional Neural Network) or RNN (Recurrent Neural Network), and it is difficult to handle long-term temporal dependencies and local features simultaneously. The present invention innovatively combines a hybrid architecture of Transformer (i.e., the target transformer neural network of the present invention) and TCN (i.e., the target temporal convolutional neural network of the present invention). It can capture long-range dependencies through the self-attention mechanism and extract local temporal features using causal convolution, thereby improving the accuracy of fault identification.

[0104] 2. Excellent computational efficiency. Traditional deep learning models have high computational overhead and are difficult to be deployed in industrial fields. The present invention realizes lightweight model design through knowledge distillation and model compression technologies, significantly reducing the computational complexity and making the system more suitable for industrial real-time diagnosis requirements.

[0105] 3. Strong interpretability. Existing deep learning methods are often "black box" models, and the diagnostic results lack interpretability. The present invention combines attention mechanism visualization and knowledge graph technology to make the diagnostic process more transparent, facilitating engineers to understand and verify the diagnostic results.

[0106] 4. Strong data fusion ability. Traditional methods can usually only process single-type data, while the feature fusion module designed in the present invention can process multi-source heterogeneous data simultaneously, improving the system's ability to identify complex faults.

[0107] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features do not conflict, they should be considered as the scope described in this specification.

[0108] The above-described embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for constructing an industrial equipment fault diagnosis model, characterized in that: include: Acquire target industrial equipment fault diagnosis training data, wherein the target industrial equipment fault diagnosis training data includes target historical multimodal industrial equipment operation data and target historical industrial equipment fault data corresponding to and matching the target historical multimodal industrial equipment operation data; Preprocessing the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data; Inputting the target feature vector matrix data into the target hybrid deep learning model to be trained for training, and outputting the target industrial equipment fault diagnosis result data; wherein the target hybrid deep learning model comprises a target converter neural network and a target time series convolutional neural network; Determine whether the target industrial equipment fault diagnosis result data meets the target preset diagnosis result data requirements; If so, record the target hybrid deep learning model parameters corresponding to the current target hybrid deep learning model to obtain the target industrial equipment fault diagnosis teacher model; The target industrial equipment fault diagnosis teacher model is subjected to knowledge distillation and model compression processing to obtain a target industrial equipment fault diagnosis model.

2. The method for constructing an industrial equipment fault diagnosis model according to claim 1, characterized in that: The preprocessing of the target industrial equipment fault diagnosis training data to obtain target feature vector matrix data includes: The target industrial equipment fault diagnosis training data is sequentially processed by data cleaning, data standardization and data alignment to obtain target standardized training data; The target standardized training data is sequentially subjected to time domain feature extraction, frequency domain feature extraction, feature fusion and feature selection to obtain the target feature vector matrix data; Among them, the data cleaning process includes removing outliers, processing missing values ​​and eliminating noise on the target industrial equipment fault diagnosis training data, the data standardization process includes using Z-score or Min-Max method to normalize the target industrial equipment fault diagnosis training data after data cleaning, and the data alignment process includes time alignment of multi-source heterogeneous data of the target industrial equipment fault diagnosis training data after normalization.

3. The method for constructing an industrial equipment fault diagnosis model according to claim 1, characterized in that: The step of inputting the target feature vector matrix data into the target hybrid deep learning model to be trained for training, and outputting target industrial equipment fault diagnosis result data, comprises: Inputting the target feature vector matrix data into the target hybrid deep learning model to be trained for training; Extracting the time domain features and frequency domain features of the target feature vector matrix data through the target converter neural network, and processing the time series features of the target feature vector matrix data through the target time series convolutional neural network, to complete the deep characterization of the fault features and fault mode recognition corresponding to the target feature vector matrix data; Outputting the target industrial equipment fault diagnosis result data corresponding to the time series characteristics of the target feature vector matrix data; The step of extracting the time domain features and frequency domain features of the target feature vector matrix data through the target converter neural network includes: Capturing the long-range dependencies in the sequence data corresponding to the target feature vector matrix data through the target multi-head self-attention mechanism of the target converter neural network; retaining the timing information of the target feature vector matrix data corresponding sequence through the target position encoding module of the target converter neural network; The feature expression capability of the target feature vector matrix data is improved through the feedforward neural network layer of the target converter neural network, and the layer normalization technology is used to ensure the stability and convergence speed of the target hybrid deep learning model training process.

4. The method for constructing an industrial equipment fault diagnosis model according to claim 1, characterized in that: The target industrial equipment fault diagnosis teacher model is subjected to knowledge distillation and model compression processing to obtain a target industrial equipment fault diagnosis model, including: Performing knowledge distillation processing on the target industrial equipment fault diagnosis teacher model to obtain a target industrial equipment fault diagnosis student model; wherein the distillation loss of the target industrial equipment fault diagnosis teacher model converted into the target industrial equipment fault diagnosis student model through knowledge distillation is less than a preset distillation loss; The target industrial equipment fault diagnosis student model is optimized and model compressed to obtain a target industrial equipment fault diagnosis model.

5. The method for constructing an industrial equipment fault diagnosis model according to claim 4, characterized in that: The target industrial equipment fault diagnosis student model is optimized and model compressed to obtain the target industrial equipment fault diagnosis model, including: Reducing the accuracy of model parameters corresponding to the target industrial equipment fault diagnosis student model; The model pruning technology is used to remove redundant connections of the target industrial equipment fault diagnosis student model, so as to simplify the structure of the target industrial equipment fault diagnosis student model.

6. A method for diagnosing industrial equipment faults, which is implemented based on the industrial equipment fault diagnosis model according to any one of claims 1 to 5, characterized in that: include: Acquiring target industrial equipment operation data to be diagnosed, wherein the target industrial equipment operation data includes target multimodal industrial equipment operation data; Preprocessing the target industrial equipment operation data to obtain target feature vector matrix operation data; Inputting the target feature vector matrix operation data into the industrial equipment fault diagnosis model for processing, and outputting target industrial equipment fault prediction result data corresponding to the target industrial equipment operation data; According to the target industrial equipment fault prediction result data, the operating status of the target industrial equipment corresponding to the target industrial equipment operating data is diagnosed.

7. The method for diagnosing industrial equipment failure according to claim 6, characterized in that: The preprocessing of the target industrial equipment operation data to obtain target feature vector matrix operation data includes: The target industrial equipment operation data is sequentially cleaned, standardized and aligned to obtain target standardized operation data; The target standardized operation data is sequentially subjected to time domain feature extraction, frequency domain feature extraction, feature fusion and feature selection to obtain the target feature vector matrix operation data; Among them, the data cleaning process includes removing outliers, processing missing values ​​and eliminating noise on the target industrial equipment operation data, the data standardization process includes normalizing the target industrial equipment operation data after data cleaning using the Z-score or Min-Max method, and the data alignment process includes time alignment of multi-source heterogeneous data of the target industrial equipment operation data after normalization.

8. The method for diagnosing industrial equipment failure according to claim 6, characterized in that: The step of diagnosing the operating status of the target industrial equipment corresponding to the target industrial equipment operating data according to the target industrial equipment fault prediction result data includes: Calculating the target industrial equipment failure prediction confidence corresponding to the target industrial equipment failure prediction result data; A target industrial equipment fault diagnosis report of the target industrial equipment corresponding to the target industrial equipment operation data is generated according to the target industrial equipment fault prediction confidence level.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the method for constructing an industrial equipment fault diagnosis model as described in any one of claims 1 to 5, or executes the method for diagnosing industrial equipment faults as described in any one of claims 6 to 8.

10. A computer device, characterized in that: It includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method for constructing an industrial equipment fault diagnosis model as described in any one of claims 1 to 5, or executes the method for diagnosing industrial equipment faults as described in any one of claims 6 to 8.

Images