Fault diagnosis method and apparatus for large die forging press, device, and medium

Abstract

The present application relates to the technical field of industrial intelligence, and discloses a fault diagnosis method and apparatus for a large die forging press, a device, and a medium. The method comprises: acquiring multi-modal sensor data of a large die forging press; preprocessing the multi-modal sensor data; inputting the preprocessed multi-modal sensor data into a multi-modal attention convolutional neural network model to perform feature extraction so as to obtain multi-modal features; and performing fusion processing on the multi-modal features; and inputting the fused multi-modal features into a classifier to perform classification so as to obtain a fault diagnosis result.

Description

Fault diagnosis methods, devices, equipment and media for large die forging presses

[0001] This application claims priority to Chinese Patent Application No. 202411533552.9, filed on October 30, 2024, entitled "Fault Diagnosis Method, Apparatus, Equipment and Medium for Large Forging Press", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of industrial intelligent technology, and in particular to a fault diagnosis method, device, equipment and medium for a large die forging press. Background Technology

[0003] The increasing intelligence and automation of high-end industrial equipment places higher demands on its safe, reliable operation and efficient maintenance. Large forging presses, as crucial manufacturing equipment for key metal components, play a vital role in machinery, automotive, shipbuilding, and aerospace industries. However, due to the continuous influence of environmental forces such as temperature, pressure, and impact during production, forging presses experience frequent malfunctions, leading to a gradual decline in equipment efficiency and product quality, resulting in significant economic losses for enterprises. Therefore, fault diagnosis technology has become one of the important means to ensure the safe and reliable operation of large forging presses.

[0004] In some cases, fault diagnosis techniques for large forging presses can be mainly divided into two directions: one is fault diagnosis methods based on mechanistic models or human experience, and the other is fault diagnosis methods based on machine learning and deep learning. Mechanism-based methods typically rely on a deep understanding of the internal mechanisms of the equipment, using mathematical models to predict and diagnose faults. Machine learning and deep learning-based methods, on the other hand, utilize large amounts of historical data to train models, learning the characteristics and patterns of the data to achieve fault diagnosis.

[0005] While fault diagnosis technologies have improved the fault diagnosis capabilities of large forging presses to some extent, significant shortcomings remain. For mechanistic model-based methods, the inherent mechanisms of large forging presses are difficult to determine due to their numerous components and complex, highly coupled relationships between different parts. This results in high costs for traditional mechanistic model-based or experience-based fault diagnosis, making them impractical in actual production. As for machine learning and deep learning-based methods, current applications mostly rely on single sensors for fault prediction and classification, failing to consider data dependencies between multiple sensors, leading to incomplete and inaccurate diagnostic results. Summary of the Invention

[0006] The purpose of this application is to provide a fault diagnosis method, device, equipment, and medium for a large die forging press.

[0007] To achieve the above objectives, this application provides the following solution:

[0008] In a first aspect, this application provides a fault diagnosis method for a large-scale die forging press, including:

[0009] Acquire multimodal sensor data from a large die forging press;

[0010] The multimodal sensor data is preprocessed;

[0011] The preprocessed multimodal sensor data is input into a multimodal attention convolutional neural network model for feature extraction to obtain multimodal features;

[0012] The multimodal features are fused.

[0013] The fused multimodal features are input into a classifier for classification to obtain fault diagnosis results.

[0014] In one embodiment, the step of acquiring multimodal sensor data from a large forging press includes:

[0015] Multimodal sensor data is obtained by synchronously acquiring data from displacement sensors, impact force sensors, air pressure sensors, and hydraulic sensors in the large die forging press.

[0016] In one embodiment, the step of preprocessing the multimodal sensor data includes:

[0017] The multimodal sensor data is smoothed, filtered, and normalized.

[0018] In one embodiment, the step of inputting the preprocessed multimodal sensor data into a multimodal attention convolutional neural network model for feature extraction to obtain multimodal features includes:

[0019] Construct a multimodal attention convolutional neural network model that includes multiple input channels and an attention mechanism;

[0020] The preprocessed multimodal sensor data is input into each input channel of the modal attention convolutional neural network;

[0021] The weights of the channel data in each input channel are learned through an attention mechanism, and multimodal features are obtained by feature extraction.

[0022] In one embodiment, the step of fusing the multimodal features includes:

[0023] The multimodal features extracted by the multimodal attention convolutional neural network model are subjected to dimensionality reduction processing to obtain low-dimensional feature vectors;

[0024] The low-dimensional feature vectors are then fused.

[0025] In one embodiment, the step of inputting the fused multimodal features into a classifier for classification to obtain fault diagnosis results includes:

[0026] Construct a transductive support vector machine;

[0027] The fused multimodal features are input into a transduced support vector machine for training.

[0028] The trained transductive support vector machine is used to classify unknown fault samples to obtain fault diagnosis results.

[0029] In one embodiment, before inputting the fused multimodal features into the transduced support vector machine for training, the method further includes: pre-training the transduced support vector machine using labeled samples.

[0030] Secondly, this application provides a fault diagnosis device for a large die forging press, comprising:

[0031] The data preparation module is used to acquire multimodal sensor data from large forging presses;

[0032] The multimodal sensor data is preprocessed;

[0033] The feature extraction module is used to input the preprocessed multimodal sensor data into the multimodal attention convolutional neural network model to extract multimodal features.

[0034] The multimodal features are fused.

[0035] The prediction module is used to input the fused multimodal features into the classifier for classification and to obtain the fault diagnosis results.

[0036] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the fault diagnosis method for the large die forging press described in any one of the above.

[0037] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the fault diagnosis method for the large forging press described above.

[0038] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0039] This application provides a fault diagnosis method, device, equipment, and medium for large-scale die forging presses. By fusing and preprocessing multimodal sensor data, key features are automatically extracted using a multimodal attention convolutional neural network, overcoming the problem of limited information from a single sensor. Feature fusion reduces data dimensionality and computational complexity while retaining the most critical feature information. The fused features are then input into a classifier for classification, making full use of unlabeled data and significantly improving the accuracy and generalization ability of fault diagnosis. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 is a flowchart illustrating a fault diagnosis method for a large die forging press according to an embodiment of this application.

[0042] Figure 2 is a schematic diagram of the principle of a fault diagnosis method for a large die forging press provided in an embodiment of this application.

[0043] Figure 3 is a schematic diagram of the principle of a fault diagnosis method for a large die forging press provided in an embodiment of this application.

[0044] Figure 4 is a schematic diagram of the principle of a fault diagnosis method for a large die forging press provided in an embodiment of this application.

[0045] Figure 5 is a schematic diagram of the functional modules of a fault diagnosis device for a large die forging press provided in an embodiment of this application.

[0046] Figure 6 is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] As shown in Figure 1, some embodiments of this application provide a fault diagnosis method for a large die forging press, which includes the following steps 101 to 104 in this embodiment:

[0049] Step 101: Obtain multimodal sensor data from the large forging press.

[0050] In this embodiment, the large forging press, as a manufacturing equipment for key metal components, is affected by various physical parameters, such as displacement, impact force, air pressure, and hydraulic pressure. These data are referred to as multimodal sensor data because they come from different types of sensors and reflect the operating status of the equipment in different aspects.

[0051] In one embodiment, a suitable sensor type, such as a displacement sensor, pressure sensor, or temperature sensor, is selected based on the needs of fault diagnosis. Data is collected in real time or periodically via the sensor interface to ensure the timeliness and accuracy of the data. The collected data is stored in a database for subsequent processing and analysis.

[0052] In practical applications, displacement sensors include at least one of the following: laser displacement sensors and linear variable differential transformers; pressure sensors include at least one of the following: strain gauge pressure sensors and piezoresistive pressure sensors; and temperature sensors include at least one of the following: thermocouples and resistance temperature detectors (RTDs).

[0053] Step 102: Preprocess the multimodal sensor data.

[0054] In this embodiment, since data directly obtained from sensors may contain noise, outliers, and inconsistent data formats, preprocessing is required to improve data quality. Preprocessing steps include data cleaning, data smoothing, and data normalization to ensure that the data input into the model is clean, accurate, and formatted uniformly.

[0055] In one implementation, outliers and missing values ​​are removed or corrected to ensure data integrity. Noise in the data is reduced using filtering algorithms (such as moving average filtering, Kalman filtering, etc.). Data with different dimensions are converted to the same dimension for easier subsequent processing.

[0056] Step 103: Input the preprocessed multimodal sensor data into the multimodal attention convolutional neural network model to extract features and obtain multimodal features.

[0057] In this embodiment, a multi-attention convolutional neural network (MACNN) model is used to extract features from the preprocessed data. The MACNN model combines a convolutional neural network (CNN) and an attention mechanism, enabling it to automatically learn and extract features related to fault diagnosis from the data.

[0058] In one implementation, a MACNN model is first designed and built, including an input layer, convolutional layers, an attention mechanism layer, and pooling layers. Then, preprocessed data is input into the MACNN model, and multimodal features are extracted through convolutional operations and the attention mechanism. The attention mechanism can automatically adjust the weights of different sensor data, focusing on features more important for fault diagnosis.

[0059] Referring to Figure 2, the model mainly consists of two parts: the first part is a multimodal attention mechanism convolutional neural network (MACNN), and the second part is a PCA-TSVM classifier (Principal Component Analysis-Support Vector Machine). The entire model is built based on a semi-supervised learning method. However, since traditional CNN models are usually used for supervised learning, in order to enable the MACNN part to better extract sample features, it is necessary to first perform supervised learning pre-training on the MACNN part, allowing it to better learn the label sample feature representation. Then, the pre-trained MACNN model is placed into the entire network model to construct the MACNN-PTSVM model.

[0060] As shown in Figure 2, the backbone of this model can be viewed as a traditional CNN model structure. This structure serves two main purposes: pre-training and feature extraction. Feature extraction mainly consists of convolutional layers, ReLU layers, and max-pooling layers. These three layers are first stacked twice. Then, the output layer of a conventional CNN model is removed, and PCA is added for feature fusion, facilitating fault classification using TSVM. In this model, a Batch Normalization (BN) layer is used to batch normalize the input before feature extraction in the convolutional layers, aiming to reduce the distribution differences between data collected from different sensors.

[0061] Step 104: Perform fusion processing on the multimodal features.

[0062] In this embodiment, after extracting multimodal features, these features need to be fused to integrate information from different sensor data and improve the accuracy of fault diagnosis. Feature fusion can be performed using various methods, such as feature stitching and principal component analysis (PCA).

[0063] Multimodal information fusion is a technique that combines multiple modal information from different sensors to improve recognition accuracy and robustness. In fault diagnosis tasks, multimodal information fusion methods can effectively combine multiple data modalities to improve the diagnostic performance of the model.

[0064] For example, as shown in Figure 3, feature fusion is used to fuse the data from 12 sensors; and a data feature set for 12 channels is constructed, as shown in Figure 4.

[0065] The system directly concatenates features from different modalities to form a high-dimensional feature vector. Dimensionality reduction is then applied to this concatenated high-dimensional feature vector, removing redundant information and retaining the most critical features. PCA projects the data onto a new coordinate system through a linear transformation, maximizing the variance (i.e., maximizing information) of the data in a specific direction within the new coordinate system, thus preserving the most important features.

[0066] Step 105: Input the fused multimodal features into the classifier for classification to obtain the fault diagnosis result.

[0067] In this embodiment, the fused multimodal features are input into a classifier for classification to determine whether a large forging press has a fault and the type of fault. The classifier can employ various machine learning algorithms, such as Support Vector Machine (SVM) and Random Forest. The appropriate classifier can be selected based on the fault diagnosis requirements, such as Transductive Support Vector Machine (TSVM). TSVM is a semi-supervised learning algorithm that can be trained using unlabeled data, improving the model's generalization ability.

[0068] Then, the classifier is trained using fault data with known labels, enabling it to learn the mapping relationship between fault features and fault types. The fused multimodal features are then input into the trained classifier to obtain the fault diagnosis results. The classifier outputs the corresponding fault type or fault probability based on the input features.

[0069] This application embodiment overcomes the problem of limited information from a single sensor by using multimodal sensor data fusion and preprocessing, and automatically extracting key features using a multimodal attention convolutional neural network. By performing feature fusion, the data dimensionality and computational complexity are reduced, while the most critical feature information is retained. The fused features are then input into a classifier for classification, making full use of unlabeled data and significantly improving the accuracy and generalization ability of fault diagnosis.

[0070] In one embodiment, step 101 includes: synchronously acquiring data from displacement sensors, impact force sensors, air pressure sensors, and hydraulic sensors in the large forging press to obtain multimodal sensor data.

[0071] In this embodiment, the displacement sensor is used to measure the displacement changes of various components or key points of a large forging press during operation. Displacement data is an important basis for evaluating the operating status of the equipment and predicting potential failures. The impact force sensor is installed on the impact component of the forging press to monitor the force changes during the impact process. Impact force data can reflect the response of the equipment when subjected to impact, which is particularly crucial for diagnosing failures caused by excessive impact or abnormal load. The pneumatic pressure sensor monitors the pressure changes of the pneumatic system in the forging press. The stability of the pneumatic system is crucial to the normal operation of the equipment, and abnormal pneumatic pressure data may indicate a failure of the pneumatic system. The hydraulic sensor is used to detect parameters such as pressure and flow rate of the hydraulic system. As an important component of the forging press, the operating status of the hydraulic system directly affects the overall performance of the equipment.

[0072] Synchronous data acquisition refers to simultaneously acquiring data from the aforementioned different types of sensors to ensure that the acquired data remains consistent over time. This is because the operation of a forging press is a complex dynamic process, and the state changes of various components are interconnected. Through synchronous data acquisition, a complete dataset reflecting the real-time operating status of the equipment can be obtained.

[0073] Multimodal sensor data refers to the collection of data acquired by various types of sensors. This data includes operational information of the equipment in different aspects, such as displacement, impact force, air pressure, and hydraulic pressure. These different modal data together constitute a comprehensive description of the operating status of the forging press.

[0074] This application's embodiments, by synchronously acquiring data from multiple sensors, can obtain richer and more comprehensive equipment operating status information. This provides strong support for improving the accuracy and reliability of fault diagnosis. Simultaneously, the fusion of multimodal data also provides more dimensional input features for subsequent data analysis and model training, helping to improve the performance of the fault diagnosis model.

[0075] In one embodiment, step 102 includes: smoothing, filtering and normalizing the multimodal sensor data.

[0076] In the embodiments of this application, the main purpose of smoothing is to reduce noise and random fluctuations in the data, making the data trend smoother and easier for subsequent analysis and processing. Commonly used smoothing methods include moving averages and Savitzky-Golay filtering. These methods obtain smoothed data sequences by performing weighted averaging or polynomial fitting on the data within a certain window.

[0077] Filtering is used to remove high-frequency noise and interference signals from data while retaining useful low-frequency information. This is crucial for improving the signal-to-noise ratio and highlighting fault characteristics. Commonly used filtering methods include low-pass filtering and band-pass filtering. These methods filter out signal components with frequencies higher or lower than a set appropriate cutoff frequency.

[0078] Normalization scales data proportionally to fit it into a small, specific interval (e.g., [0,1] or [-1,1]). This helps eliminate the influence of different units on data analysis, improving the convergence speed and stability of algorithms. Common normalization methods include min-max normalization and Z-score standardization. Min-max normalization linearly transforms the data to the [0,1] interval, while Z-score standardization converts the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0079] This application's embodiments effectively improve data quality by performing smoothing, filtering, and normalization on multimodal sensor data, laying a solid foundation for subsequent feature extraction and fault diagnosis. Smoothing and filtering remove noise and interference signals from the data, highlighting fault characteristics; while normalization eliminates the influence of different units on data analysis, improving the algorithm's convergence speed and stability. These preprocessing steps collectively ensure the accuracy and reliability of subsequent fault diagnosis.

[0080] In one embodiment, step 103 includes:

[0081] Step 1031: Construct a multimodal attention convolutional neural network model that includes multiple input channels and an attention mechanism.

[0082] Step 1032: Input the preprocessed multimodal sensor data into each input channel of the modal attention convolutional neural network.

[0083] Step 1033: Learn the weights of the channel data in each input channel through an attention mechanism, and extract features to obtain multimodal features.

[0084] In this embodiment, a convolutional neural network model capable of processing multimodal data (such as displacement, impact force, air pressure, hydraulic pressure, and other multi-sensor data) is constructed. This model, by introducing an attention mechanism, can automatically learn the importance of different modal data for fault diagnosis, thereby improving the accuracy and robustness of fault diagnosis. The model contains multiple input channels, each corresponding to sensor data of one modality. The model also includes convolutional layers, attention mechanism layers, and pooling layers for feature extraction, weight learning, and dimensionality reduction. The attention mechanism is a technique that simulates human visual attention, automatically focusing on important information and ignoring unimportant information when processing large amounts of data. In the multimodal attention convolutional neural network, the attention mechanism is used to learn the weights of data in different input channels, i.e., the degree of contribution of different modal data to fault diagnosis.

[0085] Before inputting the data into the model, the raw sensor data needs to be preprocessed, including smoothing, filtering, and normalization, to improve data quality and model training performance. The preprocessed multimodal sensor data is then input into each input channel of the multimodal attention convolutional neural network model. Since data from different modalities may have different physical meanings and dimensions, they need to be input into different channels separately.

[0086] During model training, the attention mechanism automatically learns the weights of data in different input channels based on the characteristics and label information of the input data. A larger weight indicates a greater contribution of that channel's data to fault diagnosis; a smaller weight indicates a smaller contribution. Features are extracted from the input data using network structures such as convolutional layers and pooling layers. During feature extraction, the attention mechanism weights the data from different channels based on the learned weights, thereby obtaining more accurate and effective multimodal features.

[0087] This application embodiment constructs a multimodal attention convolutional neural network model containing multiple input channels and an attention mechanism, and inputs preprocessed multimodal sensor data into it, which can effectively extract multimodal features related to fault diagnosis. The introduction of the attention mechanism enables the model to automatically learn the importance of different modal data, thereby improving the accuracy and robustness of fault diagnosis.

[0088] In one embodiment, step 104 includes:

[0089] Step 1041: Dimensionality reduction is performed on the multimodal features extracted by the multimodal attention convolutional neural network model to obtain a low-dimensional feature vector.

[0090] Step 1042: Fuse the low-dimensional feature vectors.

[0091] In this embodiment, the multimodal attention convolutional neural network model has completed feature extraction from the preprocessed multimodal sensor data. These features include information from different sensors (such as displacement sensors, impact force sensors, air pressure sensors, hydraulic sensors, etc.), collectively constituting multimodal features. The main purpose of dimensionality reduction is to reduce the dimension of the feature vectors, thereby simplifying the subsequent calculation process, while removing redundant information and retaining the most critical features. This is of great significance for improving the model's generalization ability and reducing overfitting. Commonly used dimensionality reduction methods include Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and t-Distributed Stochastic Neighbor Embedding (t-SNE). PCA dimensionality reduction technology transforms the original data into a new coordinate system through linear transformation, maximizing the variance on the first coordinate axis of the new coordinate system (i.e., the projection of the data in this direction is most dispersed), and so on, thereby achieving data dimensionality reduction. After PCA dimensionality reduction, multimodal features are converted into low-dimensional feature vectors. These low-dimensional feature vectors still retain most of the key information in the original data, but the dimensionality is significantly reduced.

[0092] Feature fusion is the process of combining features from different sources or modalities into a unified representation, aiming to improve model performance and robustness. Through feature fusion, complementary information between different modalities can be comprehensively utilized, resulting in a more comprehensive and accurate feature representation. Various feature fusion methods exist, including simple concatenation, weighted summation, and kernel methods. A common approach is to concatenate or perform weighted summation on the dimensionality-reduced low-dimensional feature vectors to obtain the final fused features. Feature fusion enables models to comprehensively utilize information from different sensors, improving the accuracy and comprehensiveness of fault diagnosis. Furthermore, by fusing data from multiple modalities, the model's tolerance to faults from single sensors is also stronger.

[0093] This application's embodiments achieve a more concise, comprehensive, and accurate feature representation by performing dimensionality reduction and feature fusion on the multimodal features extracted by the multimodal attention convolutional neural network model. These features not only retain the key information in the original data but also eliminate redundancy and noise, providing strong support for subsequent classification or regression tasks.

[0094] In one embodiment, step 105 includes:

[0095] Step 1051: Construct a transductive support vector machine.

[0096] In this application's embodiment set, Transduced Support Vector Machine (TSVM) is a semi-supervised learning algorithm, particularly suitable for handling situations with few labeled samples and many unlabeled samples. In fault diagnosis of large forging presses, obtaining a large number of labeled fault samples is difficult; therefore, TSVM becomes an effective classification method. The construction process of TSVM typically includes steps such as selecting a kernel function, setting penalty parameters, and determining an optimization algorithm. The choice of kernel function determines how samples are mapped in the feature space; common kernel functions include linear kernels, polynomial kernels, and radial basis function kernels (RBF). Penalty parameters are used to balance the complexity and classification error rate of the classifier. The optimization algorithm is used to solve for the objective function of TSVM to obtain the optimal classification hyperplane.

[0097] Step 1052: Input the fused multimodal features into the transduced support vector machine for training.

[0098] In this embodiment, during the training phase, the fused multimodal features are provided as input data to the TSVM classifier. These features have undergone multimodal extraction, dimensionality reduction, and feature fusion in the preceding steps, comprehensively reflecting the operating status of the large forging press. The TSVM classifier is trained using the input training samples (including a small number of labeled samples and a large number of unlabeled samples). During training, the TSVM attempts to find an optimal classification hyperplane to minimize the classification error rate, while utilizing information from unlabeled samples as much as possible to improve the classifier's generalization ability.

[0099] Step 1053: Use the trained transductive support vector machine to classify the unknown fault samples and obtain the fault diagnosis results.

[0100] In this embodiment, after training, the TSVM classifier is capable of classifying unknown fault samples. When a new fault sample appears, it can be input into the trained TSVM classifier, which will classify it into the corresponding fault category based on the sample's features and previously learned classification rules. The output of the TSVM classifier is the fault diagnosis result, i.e., the fault category to which the unknown fault sample belongs. This result can be used to guide subsequent maintenance and repair work, helping to ensure the normal operation of large forging presses.

[0101] This application embodiment constructs a TSVM classifier and trains it using fused multimodal features to obtain a classification model capable of effectively identifying faults in large forging presses. This model can fully utilize limited labeled samples and a large amount of unlabeled sample information, improving the accuracy and generalization ability of fault diagnosis.

[0102] In one embodiment, before inputting the fused multimodal features into the transduced support vector machine for training, the method further includes: pre-training the transduced support vector machine using labeled samples.

[0103] In practical applications, once fault diagnosis results are obtained, precise operation and maintenance decisions and operations can be made. First, the system can automatically generate alarm information to notify maintenance personnel to locate and handle potential faults, avoiding unplanned downtime. Furthermore, based on the diagnosed fault type and severity, it provides data support for developing precise maintenance plans, enabling a shift from scheduled maintenance to predictive maintenance. In addition, the accumulated diagnostic results can be used to analyze equipment performance degradation patterns, providing a basis for optimizing process parameters and spare parts management, thereby comprehensively improving equipment reliability, production efficiency, and safety management levels.

[0104] In this embodiment, since the number of fault samples from large forging presses is limited, directly using these limited samples to train complex convolutional neural network models may lead to overfitting or insufficient model generalization ability. Therefore, pre-training the model with labeled samples before formal training can help the model learn the characteristics and patterns of the data, thereby improving model performance. Pre-training can also help the model converge to the optimal solution more quickly during formal training. Through pre-training, the model already has certain initial weights and biases, which are closer to the optimal solution. Therefore, during formal training, the model can adjust these parameters more quickly to adapt to new data.

[0105] Before pre-training, a certain number of labeled samples need to be prepared. These samples should be representative and able to reflect the data characteristics of the large forging press under different operating conditions. The quantity and quality of the labeled samples will directly affect the pre-training effect. Appropriate pre-training parameters, such as learning rate, batch size, and number of training epochs, should be set according to the model complexity and the number of labeled samples. These parameters will affect the speed and effectiveness of pre-training. The labeled samples are then input into the multimodal attention convolutional neural network model and trained according to the preset parameters. During training, the model continuously adjusts its weights and biases to minimize the loss function value. When the loss function value converges to a certain extent, pre-training can be considered complete.

[0106] This application's embodiments effectively improve the performance and generalization ability of a multimodal attention convolutional neural network model by pre-training it with labeled samples before training the model. Pre-training helps the model learn the characteristics and patterns of the data, providing a better starting point for formal training. Simultaneously, pre-training can accelerate the model's convergence speed and improve training efficiency.

[0107] Based on the same inventive concept, this application also provides a fault diagnosis device for a large forging press, which implements the fault diagnosis method for the large forging press described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the fault diagnosis device for a large forging press provided below can be found in the limitations of the fault diagnosis method for large forging presses described above, and will not be repeated here.

[0108] In an exemplary embodiment, as shown in FIG5, a fault diagnosis device 20 for a large forging press is provided, comprising:

[0109] The data preparation module 201 is used to acquire multimodal sensor data of a large die forging press.

[0110] The multimodal sensor data is preprocessed.

[0111] The feature extraction module 202 is used to input the preprocessed multimodal sensor data into the multimodal attention convolutional neural network model to extract multimodal features.

[0112] The multimodal features are then fused.

[0113] The prediction module 203 is used to input the fused multimodal features into the classifier for classification to obtain the fault diagnosis result.

[0114] In one embodiment, the data preparation module 201 is further configured to: synchronously acquire data from the displacement sensor, impact force sensor, air pressure sensor and hydraulic sensor in the large forging press to obtain multimodal sensor data.

[0115] In one embodiment, the data preparation module 201 is further configured to: perform smoothing, filtering and normalization processing on the multimodal sensor data.

[0116] In one embodiment, the feature extraction module 202 is used to: construct a multimodal attention convolutional neural network model that includes multiple input channels and an attention mechanism.

[0117] The preprocessed multimodal sensor data is input into each input channel of the modal attention convolutional neural network.

[0118] The weights of the channel data in each input channel are learned through an attention mechanism, and multimodal features are obtained by feature extraction.

[0119] In one embodiment, the feature extraction module 202 is further configured to: perform dimensionality reduction processing on the multimodal features extracted by the multimodal attention convolutional neural network model to obtain a low-dimensional feature vector.

[0120] The low-dimensional feature vectors are then fused.

[0121] In one embodiment, the prediction module 203 is further configured to:

[0122] Build a TSVM classifier.

[0123] The fused multimodal features are then input into the TSVM classifier for training.

[0124] The trained transductive support vector machine is used to classify unknown fault samples to obtain fault diagnosis results.

[0125] In one embodiment, the prediction module 203 is further configured to:

[0126] Before training the multimodal attention convolutional neural network model, it is pre-trained using labeled samples.

[0127] This application embodiment overcomes the problem of limited information from a single sensor by using multimodal sensor data fusion and preprocessing, and automatically extracting key features using a multimodal attention convolutional neural network. By performing feature fusion, the data dimensionality and computational complexity are reduced, while the most critical feature information is retained. The fused features are then input into a classifier for classification, making full use of unlabeled data and significantly improving the accuracy and generalization ability of fault diagnosis.

[0128] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram is shown in Figure 6. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores information exchange data. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a fault diagnosis method for a large forging press.

[0129] Those skilled in the art will understand that the structure shown in Figure 6 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0130] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0131] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0132] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0133] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0134] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0135] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fault diagnosis method for a large die forging press, characterized in that, The fault diagnosis method for the large die forging press includes: Acquire multimodal sensor data from a large die forging press; The multimodal sensor data is preprocessed; The preprocessed multimodal sensor data is input into a multimodal attention convolutional neural network model for feature extraction to obtain multimodal features; The multimodal features are fused. The fused multimodal features are input into a classifier for classification to obtain fault diagnosis results.

2. The fault diagnosis method for a large die forging press according to claim 1, characterized in that, The steps for acquiring multimodal sensor data from a large forging press include: Multimodal sensor data is obtained by synchronously acquiring data from displacement sensors, impact force sensors, air pressure sensors, and hydraulic sensors in the large die forging press.

3. The fault diagnosis method for a large die forging press according to claim 1, characterized in that, The step of preprocessing the multimodal sensor data includes: The multimodal sensor data is smoothed, filtered, and normalized.

4. The fault diagnosis method for a large die forging press according to claim 1, characterized in that, The step of inputting the preprocessed multimodal sensor data into a multimodal attention convolutional neural network model for feature extraction to obtain multimodal features includes: Construct a multimodal attention convolutional neural network model that includes multiple input channels and an attention mechanism; The preprocessed multimodal sensor data is input into each input channel of the modal attention convolutional neural network; The weights of the channel data in each input channel are learned through an attention mechanism, and multimodal features are obtained by feature extraction.

5. The fault diagnosis method for a large die forging press according to claim 1, characterized in that, The step of fusing the multimodal features includes: The multimodal features extracted by the multimodal attention convolutional neural network model are subjected to dimensionality reduction processing to obtain low-dimensional feature vectors; The low-dimensional feature vectors are then fused.

6. The fault diagnosis method for a large die forging press according to claim 1, characterized in that, The step of inputting the fused multimodal features into a classifier for classification to obtain fault diagnosis results includes: Construct a transductive support vector machine; The fused multimodal features are input into a transduced support vector machine for training. The trained transductive support vector machine is used to classify unknown fault samples to obtain fault diagnosis results.

7. The fault diagnosis method for a large die forging press according to claim 6, characterized in that, Before inputting the fused multimodal features into the transduced support vector machine for training, the process also includes: pre-training the transduced support vector machine using labeled samples.

8. A fault diagnosis device for a large die forging press, characterized in that, The fault diagnosis device for the large die forging press includes: The data preparation module is used to acquire multimodal sensor data from large forging presses; The multimodal sensor data is preprocessed; The feature extraction module is used to input the preprocessed multimodal sensor data into the multimodal attention convolutional neural network model to extract multimodal features. The multimodal features are fused. The prediction module is used to input the fused multimodal features into the classifier for classification and to obtain the fault diagnosis results.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the fault diagnosis method for a large forging press according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the fault diagnosis method for the large die forging press according to any one of claims 1-7.

Images