Laser cutting machine rack fault diagnosis method based on deep learning

Abstract

The invention relates to the technical field of intelligent manufacturing and fault diagnosis, and discloses a laser cutting machine rack fault diagnosis method based on deep learning. Through constructing a rack image diagnosis model and a rack operation diagnosis model, automatic, rapid and accurate diagnosis of rack faults is realized. Firstly, a model is constructed according to historical data; secondly, rack picture data are collected and preprocessed when the laser cutting machine is started, and faults are judged through the image diagnosis model; if no fault exists, rack operation data are collected and preprocessed, and the fault is judged through an operation diagnosis model; when any model judges that a fault exists, a corresponding maintenance strategy is triggered; and finally, recording related data to a database, and sending a replacement prompt when the rack reaches the service life. According to the method, the diagnosis efficiency and accuracy are improved, the manual intervention cost is reduced, and powerful support is provided for intelligent maintenance of the laser cutting machine.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing and fault diagnosis technology, and in particular to a laser cutting machine rack fault diagnosis method based on deep learning. Background Art

[0002] As an indispensable equipment in modern industry, the operation stability and reliability of laser cutting machines are crucial to production efficiency and product quality. As a key transmission component of laser cutting machines, the state of the rack directly affects the overall performance of the equipment. However, traditional rack fault diagnosis methods often rely on manual experience and regular inspections, and have problems such as low diagnostic efficiency and insufficient accuracy. With the rapid development of artificial intelligence technology, the application of deep learning in the field of fault diagnosis is becoming more and more extensive. The present invention proposes a laser cutting machine rack fault diagnosis method based on deep learning, which aims to realize automatic, fast and accurate diagnosis of rack faults by constructing a rack image diagnosis model and a rack operation diagnosis model. This method not only improves the diagnostic efficiency, but also reduces the cost of manual intervention, providing strong support for the intelligent maintenance of laser cutting machines. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a laser cutting machine rack fault diagnosis method based on deep learning to solve the problems of low diagnostic efficiency, insufficient accuracy and reliance on manual experience in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a laser cutting machine rack fault diagnosis method based on deep learning, the method comprising: Step S100: constructing a rack image diagnosis model based on historical rack image data and historical fault data of the laser cutting machine, and constructing a rack operation diagnosis model based on historical rack vibration data, historical sound signal data and historical fault data; Step S200: when the laser cutting machine is started, rack image data is collected, the rack image data is subjected to a first data preprocessing, and a rack image diagnosis model is used to determine whether there is a rack fault and the type of the fault; Step S300: When the rack image diagnosis model determines that the rack has no fault, the laser cutting machine starts to run, collects rack operation data in real time, the rack operation data includes rack vibration data and sound signal data, performs second data preprocessing on the rack operation data, and uses the rack operation diagnosis model to determine whether the rack has a fault and the type of fault; Step S400: When the rack image diagnosis model or the rack operation diagnosis model determines that the rack has a fault, a rack fault maintenance strategy is triggered according to the judgment result; Step S500: storing rack image data, rack vibration data, sound signal data, fault data and adopted rack fault maintenance strategy in a database, and issuing a replacement reminder when the rack reaches a preset service life.

[0005] Preferably, in a possible implementation manner of the first aspect, the rack image diagnosis model is constructed using a Swin Transformer-based method, and the structure includes: A first input layer, for receiving rack image data that has been preprocessed by first data; Position encoding module, used to add position information when inputting rack image data; The feature extraction module extracts image features from the input rack image data through the hierarchical structure of Swin Transformer; The first output layer receives the feature map output by the feature extraction module, classifies the features using the fully connected layer and the softmax function, and outputs whether there is a rack fault and the specific fault type.

[0006] Preferably, in a possible implementation of the first aspect, the rack operation diagnosis model is constructed using a method based on a multi-level attention mechanism, and the structure includes: The second input layer is used to receive the pre-processed rack vibration data and sound signal data; The feature fusion module uses a convolutional neural network to extract the features of the rack vibration data and the sound signal data respectively, and fuses these features through an attention mechanism to form a joint feature representation; Multi-level attention layer, including multiple attention mechanism units, each unit further extracts features and assigns weights to the joint feature representation to focus on feature information in different time scales and frequency components; The second output layer receives the feature vector output by the multi-level attention layer and uses the fully connected layer and classifier to determine whether the rack has a fault and the specific fault type.

[0007] Preferably, in a possible implementation manner of the first aspect, the first data preprocessing includes: De-noising the rack image data to eliminate noise and interference in the image; Grayscale the rack image data to reduce the data volume and computational complexity; The rack image data is normalized to make the image data have a uniform scale range.

[0008] Preferably, in a possible implementation manner of the first aspect, the second data preprocessing includes: De-noising the rack vibration data and sound signal data to improve the signal-to-noise ratio of the data; Filter the rack vibration data and sound signal data to remove high-frequency noise and low-frequency interference; Perform time-frequency domain conversion and signal decomposition on the rack vibration data and sound signal data.

[0009] Preferably, in a possible implementation manner of the first aspect, the frequency domain conversion and signal decomposition include: Use Fast Fourier Transform to convert time domain signals into frequency domain signals; The frequency domain signal is decomposed by the variational mode decomposition method to obtain signals with different frequency components.

[0010] Preferably, in a possible implementation manner of the first aspect, the rack fault maintenance strategy includes: When it is judged as a minor fault, a warning signal is issued to remind the operator to pay attention and take preventive measures; When a serious fault is detected, the laser cutting machine should be stopped immediately to prevent the fault from further deteriorating and causing greater losses. Provide repair or replacement suggestions based on the fault type and severity, and guide operators or maintenance personnel to take maintenance measures.

[0011] Preferably, in a possible implementation manner of the first aspect, the method further comprises regularly updating and optimizing the rack image diagnosis model and the rack operation diagnosis model to improve the accuracy and adaptability of the models.

[0012] The beneficial effects of the present invention are as follows: by constructing a rack image diagnosis model and a rack operation diagnosis model, all-round monitoring and diagnosis of rack faults are achieved, effectively improving the accuracy and reliability of diagnosis; using deep learning technology, it is able to automatically learn features from a large amount of historical data, avoiding the limitations of traditional methods that rely on manual experience; real-time collection and processing of rack operation data, timely detection of potential faults, reducing equipment failure rates, and extending equipment life; regular updating and optimization of the diagnosis model ensures the accuracy and adaptability of the model, providing reliable guarantees for the intelligent maintenance of laser cutting machines. In summary, the present invention is of great significance for improving the operating stability and reliability of laser cutting machines, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A flow chart of a method for deep learning-based laser cutting machine rack fault diagnosis is provided for this application. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Embodiment 1: Figure 1 As shown, the present invention provides a laser cutting machine rack fault diagnosis method based on deep learning, the method comprising: Step S100: construct a rack image diagnosis model based on the historical rack image data and historical fault data of the laser cutting machine, and construct a rack operation diagnosis model based on the historical rack vibration data, historical sound signal data and historical fault data.

[0017] Specifically, a high-resolution industrial camera is installed above the rack of the laser cutting machine to take pictures of the rack to obtain rack image data. An acceleration sensor and a sound sensor are installed near the rack to collect vibration signals and sound signals when the rack is running, and the above data is stored in a database.

[0018] The historical rack image data, historical rack vibration data, historical sound signal data and historical fault data are obtained from the database. The rack image diagnosis model is constructed using the historical rack image data and historical fault data, and the rack operation diagnosis model is constructed using the historical rack vibration data, historical sound signal data and historical fault data.

[0019] The rack image diagnosis model is constructed using a Swin Transformer-based method, and its structure includes the first input layer, a feature extraction module, a position encoding module, and the first output layer.

[0020] The first input layer is used to receive the preprocessed rack image data. In this embodiment, the data of the first input layer is passed into the model through the pixel_values ​​parameter, which is a four-dimensional tensor with a shape of [batch_size, num_channels, height, width], which represents the batch size, number of channels, image height, and image width, respectively. Through the permute and view operations, the shape of the input image is converted from [batch_size, num_channels, height, width] to [batch_size, num_patches, num_channels], where num_patches is the number of blocks (or patches) into which the image is divided.

[0021] The position encoding module is used to add position information when the rack image data is input to enhance the model's understanding of the spatial relationship in the rack image. In this embodiment, a trainable position embedding matrix self.position_embeddings is created with a shape of [1, num_patches, hidden_size]. According to the batch size of the input image, the position embedding matrix is ​​copied to match the batch size, and the position embedding is added to the image features to add the position information.

[0022] The feature extraction module is used to extract multi-scale features from the input rack image data through the hierarchical structure of Swin Transformer. In this embodiment, SwinTransformerModel is used as the core part of the Transformer model, and the image features with added position information are input into the Transformer model. The Transformer model outputs a series of hidden states, where the last hidden state is used for subsequent classification tasks; the hidden state of the last token is extracted from the output of the Transformer model; a fully connected layer (self.fc_layer) is used to convert the feature representation into a score of the fault type, and the softmax function is applied to convert the score into a probability distribution.

[0023] The first output layer receives the feature map output by the feature extraction module, classifies the features using the fully connected layer and the softmax function, and determines whether there is a rack fault and the specific fault type.

[0024] The rack operation diagnosis model is constructed using a method based on a multi-level attention mechanism, and its structure includes a second input layer, a feature fusion module, a multi-level attention layer, and a second output layer.

[0025] The second input layer is used to receive the preprocessed rack vibration data and sound signal data. In this embodiment, the data of the second input layer is passed into the model through the signal_data parameter. This parameter is a multidimensional tensor with a shape of [batch_size, num_signals, signal_length], which represents the batch size, the number of signal types, and the length of each signal. The input time domain signal is processed by fast Fourier transform (FFT) and variational modal decomposition (VMD) operations. First, the time domain signal is converted into a frequency domain signal using FFT to obtain a tensor represented in the frequency domain with a shape of [batch_size, num_signals, num_frequencies], where num_frequencies represents the number of frequency points in the frequency domain. Then, the frequency domain signal is decomposed using the VMD method to obtain modal components of different frequency components. Each modal component can be regarded as a two-dimensional tensor with a shape of [batch_size, modal_length], where modal_length represents the length of the modal component. Finally, these modal components and the statistical features of the original time domain signal are concatenated or fused to form a fused feature tensor.

[0026] The feature fusion module uses a convolutional neural network to extract the features of vibration data and sound signal data respectively, and fuses these features through an attention mechanism to form a joint feature representation. In this embodiment, the feature fusion module is implemented by the VibrationSoundFusionModule class, which is used to fuse the features of the vibration signal and the sound signal. The vibration signal is first processed by a one-dimensional convolutional neural network (CNN), which consists of nn.Conv1d, nn.ReLU and nn.MaxPool1d layers to extract the features of the vibration signal; the sound signal is also processed by a one-dimensional CNN, and its structure is the same as that of the vibration CNN. The extracted features are fused through a multi-head attention mechanism (nn.MultiheadAttention), allowing the model to pay attention to the importance of different features during the fusion process. Finally, the fused features are projected through a fully connected layer (nn.Linear) to obtain a fused feature vector.

[0027] The multi-level attention layer includes multiple attention mechanism units, each of which performs further feature extraction and weight assignment on the joint feature representation to focus on important information in different time scales and frequency components. In this embodiment, the multi-level attention layer is implemented by the MultiLevelAttentionLayer class, which is used to further process the fused feature vector. This layer contains multiple levels of attention mechanisms, each of which consists of a linear layer, a ReLU activation function, another linear layer, and a Softmax function. This structure allows the model to capture the importance of features at different levels. At each level, the model calculates the attention weights and uses these weights to perform a weighted summation of the input features to obtain a weighted feature vector. The output feature vectors of all levels are spliced ​​together to form the final output feature vector.

[0028] The second output layer receives the feature vector output by the multi-level attention layer and uses the fully connected layer and classifier to determine whether the rack has a fault and the specific fault type.

[0029] Step S200: when the laser cutting machine is started, rack image data is collected, a first data preprocessing is performed on the rack image data, and a rack image diagnosis model is used to determine whether there is a rack fault and the type of fault.

[0030] Specifically, first, follow the normal operation process of the laser cutting machine, start the machine and ensure its stable operation. At the same time, start the high-resolution industrial camera installed above the rack, adjust the shooting angle and distance between the camera and the rack to ensure that a clear and complete rack image can be captured. The camera is set to continuously collect 10 rack image data at a rate of 1 frame per second, which will serve as an important basis for subsequent fault diagnosis.

[0031] Next, the collected rack image data is subjected to the first data preprocessing. The purpose of this step is to ensure that the rack image diagnosis model can accurately and efficiently process the image data. The preprocessing operations include denoising, grayscale processing and normalization. Denoising uses median filtering, Gaussian filtering and mean filtering denoising algorithms to eliminate noise and interference in the image and improve image quality. Grayscale processing converts color images into grayscale images to reduce the amount of data and reduce computational complexity. Finally, normalization processing gives the image data a uniform scale range, thereby enhancing the stability and accuracy of the model.

[0032] In this embodiment, the cv2.imread function is used to read the rack image file stored in the specified path. This function is provided in the OpenCV library for reading image files. It accepts an image file path as input and returns a matrix containing image data. If the image file cannot be read (for example, the file does not exist or the path is wrong), a ValueError exception is thrown and an error message is output.

[0033] After reading the image, denoising is performed to eliminate noise and interference in the image. The Gaussian filtering algorithm is used for denoising, and the standard deviation sigma of the Gaussian kernel is set to 1.

[0034] In order to reduce the amount of image data while retaining the main feature information of the image, grayscale processing is performed. The color image is converted to a grayscale image using the cv2.cvtColor function. This function is a function for color space conversion provided in the OpenCV library. It accepts a matrix containing image data and a color space conversion code as input and returns the converted image data. In this embodiment, the color image is converted from the BGR color space to the grayscale color space.

[0035] Finally, normalization is performed to map the pixel values ​​of the grayscale image to the range of [0, 1]. Normalization is a commonly used technique in image processing, which is achieved by dividing the pixel values ​​of the image by a maximum value (255.0 in this embodiment). The image data after normalization has a uniform scale range, thereby enhancing the stability and accuracy of subsequent image processing.

[0036] The preprocessed rack image data will be input into the rack image diagnosis model for fault diagnosis. The model receives the preprocessed image data through the first input layer, and then uses the feature extraction module (such as the hierarchical structure of SwinTransformer) to extract multi-scale features in the image. The position encoding module adds position information to these features to enhance the model's understanding of spatial relationships. The extracted features are then sent to the output layer for classification through the fully connected layer and the softmax function. The first output layer generates a probability distribution that clearly shows the possibility of various fault types in the rack. Based on the maximum value of the probability distribution, the model can accurately determine whether there is a fault in the rack and the specific fault type.

[0037] If the judgment result is that there is a fault, the model will output detailed information such as the fault type and fault severity, and can trigger the alarm system to promptly notify relevant personnel to handle it, ensuring that the fault is responded to quickly.

[0038] Step S300: When the rack image diagnosis model determines that the rack has no fault, the laser cutting machine starts to run and collects rack operation data in real time. The rack operation data includes rack vibration data and sound signal data. The rack operation data is subjected to a second data preprocessing and the rack operation diagnosis model is used to determine whether there is a rack fault and the type of fault.

[0039] Specifically, during the operation of the laser cutting machine, the system captures the vibration data and sound signal data of the rack in real time. These data are collected by acceleration sensors and sound sensors installed around the rack, and are recorded and stored in the form of continuous data streams. In order to efficiently utilize these data and achieve accurate fault diagnosis, a series of preprocessing steps need to be performed. First, denoising is performed to improve the signal-to-noise ratio of the data. For vibration data, adaptive filtering or Kalman filtering technology is used. These technologies can automatically adjust the filter parameters according to the statistical characteristics of the signal and effectively filter out background noise. For sound signals, denoising methods based on spectral subtraction or Wiener filtering are used to separate and remove noise components by analyzing the spectral characteristics of the sound signal.

[0040] Next, filtering is performed to eliminate high-frequency noise and low-frequency interference and highlight the effective information in the signal. High-frequency noise is mostly caused by tiny vibrations inside the equipment or the noise of the sensor itself. A low-pass filter is used to filter out these high-frequency components and retain the low-frequency and baseband parts of the signal. Low-frequency interference may be caused by the overall vibration of the machine or power supply fluctuations. These low-frequency components can be removed by a high-pass filter to reduce their interference with fault diagnosis.

[0041] In order to deeply analyze the running status of the rack, it is also necessary to convert the time domain signal into a frequency domain signal and perform signal decomposition. The fast Fourier transform (FFT) algorithm is used to efficiently convert the time domain signal into a frequency domain signal to obtain the signal spectrum, that is, the energy distribution of the signal at different frequencies, which helps to identify abnormal frequency components. At the same time, the variational mode decomposition (VMD) method is used to adaptively decompose the complex signal into a series of modal components with different center frequencies and bandwidths. These modal components reflect the characteristics of the signal at different frequency components, which helps to reveal the fault information hidden in the complex signal.

[0042] In this embodiment, two denoising functions are used. The denoise_shake_data() function uses the Kalman filter algorithm to denoise the vibration data to improve the purity and accuracy of the data. The denoise_audio_data() function first uses the pre-emphasis technology to enhance the high-frequency components of the sound signal, and then applies the discrete cosine transform (DCT) denoising method to further remove the noise in the sound signal.

[0043] In addition, a function called bandpass_filter() is used to filter the input data using a Butterworth bandpass filter. This function receives multiple parameters, including the data to be filtered, the low cutoff frequency, the high cutoff frequency, the sampling frequency, and the order of the filter, and returns the filtered data.

[0044] Finally, a main function named preprocess_data() is used to take charge of the entire data preprocessing process. It first calls the above denoising function to denoise the vibration data and sound signal data, and then uses a bandpass filter to filter the denoised data. In particular, for vibration data, a filter range of 0-50Hz is set; for sound signal data, a filter range of 50-1000Hz is set. After completing the filtering process, the function further performs a fast Fourier transform (FFT) on the vibration data to obtain its representation in the frequency domain. At the same time, the function performs variational mode decomposition (VMD) on the filtered sound signal data to extract its intrinsic modal components, providing key information for subsequent fault diagnosis.

[0045] After the second stage of preprocessing, the vibration data and sound signal data of the rack are converted into a clearer and easier to analyze form. These data are then input into the rack operation diagnosis model for fault diagnosis. In the second input layer, the data is formatted into a multi-dimensional tensor. The feature fusion module uses a convolutional neural network to extract the features of the vibration data and sound signal data respectively, and fuses these features into a joint feature representation through an attention mechanism. Then, the multi-level attention layer further analyzes the joint feature representation, and each attention mechanism unit assigns weights to the features, focusing on important information in different time scales and frequency components. Finally, the second output layer receives the feature vector output by the multi-level attention layer, and uses a fully connected layer and a classifier to determine whether the rack has a fault and the specific fault type. The model generates a probability distribution showing the possibility of various fault types in the rack. Based on the maximum value of the probability distribution, the model is able to make the final diagnostic decision.

[0046] Step S400: When the rack image diagnosis model or the rack operation diagnosis model determines that there is a rack fault, a rack fault maintenance strategy is triggered according to the determination result.

[0047] Specifically, when the rack image diagnosis model or the rack operation diagnosis model determines that there is a fault in the rack, the type and severity of the fault will be determined first. The fault types include rack wear, breakage, looseness, poor lubrication, etc., and the fault severity is divided into three levels: minor, moderate, and severe.

[0048] Once the fault type and severity are determined, the system will take appropriate measures based on the specific situation. For minor faults, the system will send out warning signals through sound, light, etc. to remind the operator to pay attention to the rack status. At the same time, the system will automatically record the time, location, type and other information of the minor fault for subsequent analysis and processing. After receiving the warning, the operator will immediately check the rack and take preventive measures, such as strengthening lubrication, adjusting the rack position, etc., to prevent the fault from further deteriorating.

[0049] If it is judged as a serious fault, the system will immediately stop the operation of the laser cutting machine to prevent the fault from causing greater damage to the laser cutting machine. At the same time, the system will also issue an emergency stop signal and display the fault type and location so that the operator can quickly locate and handle the fault. At this time, the operator should immediately notify the maintenance personnel and inform them of the fault type and location. The maintenance personnel will take appropriate maintenance or replacement measures based on the type and severity of the fault to ensure that the rack can return to normal working condition.

[0050] In the process of troubleshooting, the system will also provide specific repair or replacement suggestions based on the type and severity of the fault. These suggestions will include information such as repair steps, required tools, model and specifications of replacement parts, etc., to guide operators or maintenance personnel to take appropriate maintenance measures. Operators or maintenance personnel perform maintenance operations based on the suggestions provided by the system to ensure that the rack can be effectively repaired and restored to normal operation.

[0051] In addition, the system will automatically record the time, location, type, severity, and treatment measures of each fault. This information will be used for subsequent fault analysis to help identify the patterns and causes of faults. By analyzing fault records, fault diagnosis models and maintenance strategies can be continuously optimized to improve the accuracy of fault diagnosis and maintenance efficiency.

[0052] Step S500: storing rack image data, rack vibration data, sound signal data, fault data and adopted rack fault maintenance strategy in a database, and issuing a replacement reminder when the rack reaches a preset service life.

[0053] Specifically, the rack image data, vibration data and sound signal data, as the original input of fault diagnosis, are sent to the fault diagnosis model for in-depth analysis after a series of preprocessing. These original data intuitively reflect the appearance characteristics and operating conditions of the rack under different operating conditions, which is of great significance for understanding and grasping the performance changes and failure modes of the rack.

[0054] At the same time, the fault data includes key information such as the fault type, fault severity, and time of fault occurrence output by the fault diagnosis model. These data directly reflect the actual fault situation of the rack and are an important basis for formulating targeted maintenance strategies and adjusting equipment operating parameters.

[0055] The maintenance strategy data records in detail the maintenance measures taken for specific faults, such as repairs, replacement of parts, adjustment of operating parameters, etc. These data record the entire process of fault handling and reflect the results of fault handling, which helps to evaluate the effectiveness of maintenance measures and further optimize the maintenance process.

[0056] In the process of recording this data into the database, the system will also track and accurately calculate the service life of the rack according to the actual use of the rack and the preset service life standard. Once the use time or number of operations of the rack reaches the preset replacement threshold, the system will immediately and automatically issue a replacement reminder to notify the operator or maintenance personnel to replace the rack in time.

[0057] In one embodiment, the method further includes regularly updating and optimizing the rack image diagnosis model and the rack operation diagnosis model to improve the accuracy and adaptability of the models.

[0058] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A laser cutting machine rack fault diagnosis method based on deep learning, characterized in that: The method comprises: Step S100: constructing a rack image diagnosis model based on historical rack image data and historical fault data of the laser cutting machine, and constructing a rack operation diagnosis model based on historical rack vibration data, historical sound signal data and historical fault data; Step S200: when the laser cutting machine is started, rack image data is collected, the rack image data is subjected to a first data preprocessing, and a rack image diagnosis model is used to determine whether there is a rack fault and the type of the fault; Step S300: When the rack image diagnosis model determines that the rack has no fault, the laser cutting machine starts to run, collects rack operation data in real time, the rack operation data includes rack vibration data and sound signal data, performs second data preprocessing on the rack operation data, and uses the rack operation diagnosis model to determine whether the rack has a fault and the type of fault; Step S400: When the rack image diagnosis model or the rack operation diagnosis model determines that the rack has a fault, a rack fault maintenance strategy is triggered according to the judgment result; Step S500: storing rack image data, rack vibration data, sound signal data, fault data and adopted rack fault maintenance strategy in a database, and issuing a replacement reminder when the rack reaches a preset service life.

2. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The rack image diagnosis model is constructed using a Swin Transformer-based method, and the structure includes: A first input layer, for receiving rack image data that has been preprocessed by first data; Position encoding module, used to add position information when inputting rack image data; The feature extraction module extracts image features from the input rack image data through the hierarchical structure of Swin Transformer; The first output layer receives the feature map output by the feature extraction module, classifies the features using the fully connected layer and the softmax function, and outputs whether there is a rack fault and the specific fault type.

3. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The rack operation diagnosis model is constructed using a method based on a multi-level attention mechanism, and its structure includes: The second input layer is used to receive the pre-processed rack vibration data and sound signal data; The feature fusion module uses a convolutional neural network to extract the features of the rack vibration data and the sound signal data respectively, and fuses these features through an attention mechanism to form a joint feature representation; Multi-level attention layer, including multiple attention mechanism units, each unit further extracts features and assigns weights to the joint feature representation to focus on feature information in different time scales and frequency components; The second output layer receives the feature vector output by the multi-level attention layer and uses the fully connected layer and classifier to determine whether the rack has a fault and the specific fault type.

4. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The first data preprocessing includes: De-noising the rack image data to eliminate noise and interference in the image; Grayscale the rack image data to reduce the data volume and computational complexity; The rack image data is normalized to make the image data have a uniform scale range.

5. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The second data preprocessing includes: De-noising the rack vibration data and sound signal data to improve the signal-to-noise ratio of the data; Filter the rack vibration data and sound signal data to remove high-frequency noise and low-frequency interference; Perform time-frequency domain conversion and signal decomposition on the rack vibration data and sound signal data.

6. The laser cutting machine rack fault diagnosis method according to claim 5, characterized in that: The frequency domain conversion and signal decomposition include: Use Fast Fourier Transform to convert time domain signals into frequency domain signals; The frequency domain signal is decomposed by the variational mode decomposition method to obtain signals with different frequency components.

7. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The rack failure maintenance strategy includes: When it is judged as a minor fault, a warning signal is issued to remind the operator to pay attention and take preventive measures; When a serious fault is detected, the laser cutting machine should be stopped immediately to prevent the fault from further deteriorating and causing greater losses. Provide repair or replacement suggestions based on the fault type and severity, and guide operators or maintenance personnel to take maintenance measures.

8. The laser cutting machine rack fault diagnosis method according to claim 1, characterized in that: The method also includes regularly updating and optimizing the rack image diagnosis model and the rack operation diagnosis model to improve the accuracy and adaptability of the models.

Images