Real-time monitoring method and device for working state of gearbox

Abstract

The invention provides a real-time monitoring method and device for the working state of a gearbox, and relates to the technical field of gearbox testing. According to the method, all feature information of scale spaces such as a time domain and a frequency domain is obtained through wavelet change integration, then, in the training process of an artificial intelligence model, modes such as an attention mechanism and soft thresholding are adopted, feature screening is carried out on the time domain information and the frequency domain information, denoising is achieved, key features related to fault types are obtained, and the fault types can be identified. According to the method and the system, the key features and the fault types are obtained, training is performed based on the key features and the fault types to obtain the artificial intelligence model, the gearbox is monitored in real time based on the obtained artificial intelligence model, noise in the vibration signals can be effectively weakened, the influence of the noise on gearbox fault prediction is weakened, and the accuracy and reliability of fault diagnosis of the shape-feature-integrated gearbox are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gearbox testing, and in particular to a real-time monitoring method and device for the working state of a gearbox. Background Art

[0002] In the field of intelligent tunneling equipment in coal mines, the shape-integrated gearbox is a key transmission component, and its operating status directly affects the overall performance and operating efficiency of the tunneling equipment. In order to ensure the safe and stable operation of the tunneling equipment, it is particularly important to perform real-time and accurate fault detection on the shape-integrated gearbox. However, in actual application, the fault detection process faces many technical challenges, among which the noise interference problem is particularly prominent.

[0003] The conformal integrated gearbox will generate complex vibration and noise signals during operation. These signals not only contain the operating status information of the gearbox itself, such as the interaction and friction between the components, which generates a large amount of mechanical noise, but also are mixed with various interference factors from the external environment, such as the vibration of the tunnel boring machine, the noise of the underground environment, and electromagnetic noise. In addition, changes in environmental factors such as air humidity and temperature in the well may also cause the instability of sensor performance, thereby introducing additional noise. These factors cause the collected fault signals to be often severely polluted by noise, making the extraction and identification of fault features extremely difficult.

[0004] Traditional fault detection methods mainly rely on technical means such as spectrum analysis and wavelet analysis to try to separate fault features from mixed signals. However, in practical applications, these methods are often limited by noise interference, which greatly reduces the accuracy and reliability of the detection results. Especially in complex working conditions such as tunnel boring machines, the diversity and uncertainty of noise interference make the application effect of traditional methods even more discounted.

[0005] At present, the noise interference problem in the fault detection process of the shape-integrated gearbox leads to low fault detection accuracy and reliability of the shape-integrated gearbox. Summary of the invention

[0006] The present invention provides a real-time monitoring method and device for the working state of a gear box, which can improve the accuracy and reliability of fault detection results of a form-integrated gear box.

[0007] In a first aspect, the present invention provides a real-time monitoring method for the working state of a gearbox, the method comprising: obtaining vibration signals of a morphologically integrated gearbox under various fault states, and recording the fault type of each vibration signal; performing wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image; the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal; based on the two-dimensional time-frequency image and the fault type of each vibration signal, performing neural network training to obtain an artificial intelligence model, and during the training process, using an attention mechanism and a soft thresholding method to denoise and screen the two-dimensional time-frequency image, and extracting key features to identify the fault type; based on the artificial intelligence model, the working state of the morphologically integrated gearbox is monitored in real time to obtain a fault detection result.

[0008] In a possible implementation, a vibration signal is subjected to wavelet transform processing to obtain a two-dimensional time-frequency image, including: analyzing the frequency range, time domain characteristics and noise composition of the vibration signal; determining a wavelet basis function based on the frequency range, time domain characteristics and noise composition; decomposing the vibration signal based on the wavelet basis function to obtain wavelet coefficients of multiple scales; and reconstructing the wavelet coefficients of multiple scales to obtain a two-dimensional time-frequency image.

[0009] In one possible implementation, a neural network is trained based on the two-dimensional time-frequency image and the fault type of each vibration signal to obtain an artificial intelligence model, including: generating multiple training samples based on the two-dimensional time-frequency image and the fault type of each vibration signal; and training a neural network based on the multiple training samples to obtain an artificial intelligence model.

[0010] In a possible implementation, neural network training is performed based on multiple training samples to obtain an artificial intelligence model, including: step one, building an initial model, and setting the model parameters of the initial model as the model parameters of the current iteration process; the initial model includes a deep residual shrinkage module and a convolutional neural network module, and the model parameters of the deep residual shrinkage module include attention parameters and soft threshold parameters; the model parameters of the convolutional neural network model include the number of layers, quantity and weights of neurons; step two, based on the two-dimensional time-frequency image of the current training sample, and the deep residual shrinkage module and model parameters in the current iteration process, key features are obtained; step three, based on the convolutional neural network model and model parameters in the current iteration process model parameters, and key features, and output the predicted fault type; step 4, compare the deviation between the fault type of the current training sample and the predicted fault type; step 5, if the deviation meets the preset conditions, exit the iteration and execute step 6; if the deviation does not meet the preset conditions, change the model parameters of the deep residual shrinkage module and the convolutional neural network module, and repeat steps 2 to 5 until the deviation meets the preset conditions; step 6, determine whether the training set in the training sample has been trained. If the training is completed, build an artificial intelligence model based on the model parameters of the current deep residual shrinkage module and the convolutional neural network module; if the training is not completed, change the training sample and repeat steps 2 to 6 until the training set is trained.

[0011] In a possible implementation, key features are obtained based on the two-dimensional time-frequency image of the current training sample, and the deep residual shrinkage module and model parameters in the current iteration process, including: based on the two-dimensional time-frequency image of the current training sample, and the residual block of the deep residual shrinkage module, the two-dimensional time-frequency image is preliminarily shrunk to obtain the high-order features of the current training sample; based on the attention parameters of the current iteration process, the weight coefficients of each channel feature in the high-order features of the current training sample are determined; based on the high-order features of the current training sample, and the weight coefficients of each channel feature, the attention feature is generated; based on the soft threshold parameters in the current iteration process, the attention feature is deeply shrunk to obtain the key features.

[0012] In one possible implementation, based on an artificial intelligence model, the working state of the shape-integrated gearbox is monitored in real time to obtain fault detection results, including: obtaining the real-time vibration signal of the shape-integrated gearbox in the current period; performing wavelet transformation processing based on the real-time vibration signal to obtain a two-dimensional time-frequency image of the current period; inputting the two-dimensional time-frequency image of the current period into the deep residual shrinkage module of the artificial intelligence model to obtain the key features of the current period; based on the key features of the current period and the convolutional neural network module of the artificial intelligence model, outputting the fault detection results of the current period.

[0013] In one possible implementation, based on an artificial intelligence model, the working state of the shape-integrated gearbox is monitored in real time. After the fault detection result is obtained, it also includes: if the fault detection result is a minor fault, a fault monitoring instruction is generated, and the fault monitoring instruction is used to instruct the monitored gear system to operate with a fault; minor faults include wear, pitting and scratches; if the fault detection result is a serious fault, a shutdown instruction is generated, and the shutdown instruction is used to instruct the gear system to shut down, and serious faults include broken teeth.

[0014] In one possible implementation, if the fault detection result is a minor fault, a fault monitoring instruction is generated, which then includes: recording infrared data after the gear system fault; dividing the infrared data after the gear system fault into sliding time windows to obtain infrared data of multiple sliding time windows; based on the infrared data of multiple sliding time windows and a stress analysis model, determining analysis results of multiple sliding time windows, the analysis results including the fault type, the fault probability of each fault type and the fault location; based on the analysis results of multiple sliding time windows, analyzing the fault at the fault location in the fault detection result to determine the fault trend; based on the fault detection result, the analysis results of multiple sliding time windows, and the fault trend, determining whether the gear system is shut down.

[0015] In a possible implementation, based on an artificial intelligence model, the working state of the morphologically integrated gearbox is monitored in real time. After the fault detection result is obtained, it also includes: if the fault detection result is a gearbox fault, the real-time operating parameters of the morphologically integrated gearbox in the current period are obtained; the real-time operating parameters include speed, temperature, pressure and load; based on the real-time operating parameters of the current period and the two-dimensional time-frequency image of the current period, the operating characteristics of the current period are generated; based on the operating characteristics of the current period and a preset Gaussian process regression model, stress prediction is performed to obtain the stress characteristics of the morphologically integrated gearbox; based on the stress characteristics, the fault detection results are verified to determine the fault type of the morphologically integrated gearbox.

[0016] In a possible implementation, before stress prediction is performed based on the operating condition characteristics of the current period and a preset Gaussian process regression model to obtain the stress characteristics of the morphologically integrated gearbox, it also includes: obtaining the vibration signal, operating condition parameters and stress characteristics of the morphologically integrated gearbox in a normal state and various fault states; generating multiple operating condition characteristics based on the vibration signal and the operating condition parameters; performing Gaussian regression fitting based on the multiple operating condition characteristics and the stress characteristics corresponding to each operating condition characteristic to obtain a Gaussian process regression model.

[0017] In the second aspect, an embodiment of the present invention provides a real-time monitoring device for the working status of a gearbox, the monitoring device comprising a communication module and a processing module, the communication module being used to obtain vibration signals of a morphologically integrated gearbox under various fault conditions and record the fault type of each vibration signal; the processing module being used to perform wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image; the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal; based on the two-dimensional time-frequency image and the fault type of each vibration signal, a neural network is trained to obtain an artificial intelligence model, and during the training process, an attention mechanism and a soft thresholding method are used to denoise and screen the two-dimensional time-frequency image, and key features are extracted to identify the fault type; based on the artificial intelligence model, the working status of the morphologically integrated gearbox is monitored in real time to obtain a fault detection result.

[0018] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, the memory storing a computer program, the processor being used to call and run the computer program stored in the memory to execute the steps of the method described in the first aspect and any possible implementation method of the first aspect.

[0019] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and wherein when the computer program is executed by a processor, the steps of the method described in the first aspect and any possible implementation method of the first aspect are implemented.

[0020] The present invention provides a real-time monitoring method and device for the working state of a gearbox. The present invention obtains all feature information in scale spaces such as time domain and frequency domain through wavelet transformation integration. Then, in the training process of an artificial intelligence model, attention mechanism and soft thresholding are used to perform feature screening on the time domain information and the frequency domain information to achieve denoising, and obtain key features related to the fault type. The artificial intelligence model is trained based on the key features and the fault type. The gearbox is monitored in real time based on the artificial intelligence model obtained in this way, which can effectively reduce the noise in the vibration signal, reduce the influence of the noise on the gearbox fault prediction, and improve the accuracy and reliability of the morphologically integrated gearbox fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, 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 labor.

[0022] Figure 1It is a flow chart of a method for real-time monitoring of the working state of a gear box provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a real-time monitoring device for the working state of a gear box provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0024] In the description of the present invention, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, "at least one" and "plurality" refer to two or more. The words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not limit them to be different.

[0025] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.

[0026] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include other steps or modules that are not listed, or may optionally include other steps or modules that are inherent to these processes, methods, products or devices.

[0027] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be described through specific embodiments in conjunction with the accompanying drawings of the present invention.

[0028] like Figure 1As shown, an embodiment of the present invention provides a method for real-time monitoring of the working state of a gearbox. The method includes steps S101-S104.

[0029] S101. Obtain vibration signals of the conformal integrated gearbox under various fault conditions, and record the fault type of each vibration signal.

[0030] For example, the embodiments of the present invention can use high-precision vibration sensors installed in key parts of the gearbox, such as bearings, gears, etc., to capture its vibration signals under different operating conditions. In order to obtain comprehensive fault data, different types of faults can be simulated in a laboratory environment, such as gear wear, bearing damage, shaft misalignment, etc. For each fault state, it is necessary to record the label of the vibration signal in detail, including information such as fault type, fault degree, and occurrence time, for subsequent data processing and model training.

[0031] S102, performing wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image.

[0032] In the embodiment of the present application, the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal.

[0033] For example, the embodiment of the present invention can select a suitable wavelet basis function for transformation according to the characteristics of the vibration signal, such as Morlet wavelet, Db wavelet, etc. The wavelet transform is multi-scale decomposed to obtain time-frequency information of different frequency components. The decomposed time-frequency information is represented in the form of a two-dimensional image, where the horizontal axis represents time, the vertical axis represents frequency, and the color or brightness of the image represents the strength or energy of the signal.

[0034] As a possible implementation manner, step S102 can be specifically implemented as steps S1021-S1024.

[0035] S1021. Analyze the frequency range, time domain characteristics and noise composition of the vibration signal.

[0036] Exemplarily, embodiments of the present invention may use fast Fourier transform (FFT) or other spectrum analysis methods to determine the main frequency components of the vibration signal and their distribution range. This helps to select appropriate wavelet basis functions and decomposition scales. Observe the time domain waveform of the vibration signal and analyze the periodicity, non-stationarity and other characteristics of the signal. These characteristics are crucial for understanding the time-varying behavior of the signal and subsequent wavelet transform processing. Identify the noise components in the vibration signal, including white noise, colored noise, etc. The presence of noise may affect the effect of the wavelet transform, so it needs to be properly processed in subsequent steps.

[0037] S1022. Determine a wavelet basis function based on the frequency range, time domain characteristics and noise composition.

[0038] Exemplarily, the embodiment of the present invention can select the most suitable wavelet basis function according to the frequency range, time domain characteristics and noise composition of the vibration signal. Common wavelet basis functions include Haar wavelet, Daubechies wavelet, Morlet wavelet, etc. When selecting, it is necessary to consider the orthogonality, compact support, symmetry and other characteristics of the wavelet basis function. According to the frequency resolution requirements of the vibration signal, the decomposition scale of the wavelet transform is determined. The more decomposition scales, the higher the frequency resolution, but the computational complexity will also increase. Therefore, it is necessary to make a trade-off between resolution and computational efficiency.

[0039] S1023. Decompose the vibration signal based on wavelet basis functions to obtain wavelet coefficients of multiple scales.

[0040] Exemplarily, the embodiment of the present invention can use the selected wavelet basis function and decomposition scale to perform discrete wavelet transform on the vibration signal. This step decomposes the signal into wavelet coefficients at multiple scales, and the coefficients at each scale represent the frequency components of the signal at that scale. The decomposed wavelet coefficients are stored in an appropriate data structure for subsequent reconstruction processing.

[0041] S1024. Reconstruct the wavelet coefficients of multiple scales to obtain a two-dimensional time-frequency image.

[0042] Exemplarily, embodiments of the present invention may use an inverse discrete wavelet transform (IDWT) or other reconstruction algorithms to reconstruct the stored wavelet coefficients back into a time domain signal. This step typically involves inverse transforming and superimposing the wavelet coefficients at each scale. The reconstructed signal is represented in the form of a two-dimensional image, where the horizontal axis represents time, the vertical axis represents frequency (or scale), and the color or brightness of the image represents the intensity or energy of the signal. This step can be implemented by programming, using an image processing library (such as OpenCV) to generate and display a two-dimensional time-frequency image.

[0043] S103, based on the two-dimensional time-frequency image and the fault type of each vibration signal, a neural network training is performed to obtain an artificial intelligence model.

[0044] In the embodiment of the present application, the attention mechanism and soft thresholding method are used to denoise and screen the two-dimensional time-frequency image during the training process, and key features are extracted to identify the fault type.

[0045] Exemplarily, the embodiments of the present invention can perform preprocessing operations such as normalization and cropping on the two-dimensional time-frequency image to improve the training efficiency and accuracy of the model. The attention mechanism is introduced into the neural network so that the model can automatically focus on the key areas in the image, thereby improving the accuracy of fault identification. The image is denoised using the soft thresholding method to reduce the impact of noise on model training. At the same time, by screening key features, the model input is further simplified and the generalization ability of the model is improved. The preprocessed two-dimensional time-frequency image and the corresponding fault type label are used to train the neural network. During the training process, the network parameters are continuously adjusted to minimize the loss function and improve the recognition accuracy of the model.

[0046] As a possible implementation manner, step S103 may be specifically implemented as steps S1031 - S1032 .

[0047] S1031. Generate multiple training samples based on the two-dimensional time-frequency image and the fault type of each vibration signal.

[0048] Exemplarily, an embodiment of the present invention can pair each two-dimensional time-frequency image with its corresponding vibration signal fault type to form a training sample. Ensure that each sample contains sufficient fault information so that the neural network can learn the fault characteristics from it. In order to improve the generalization ability of the model, the original training samples can be subjected to data enhancement processing. This includes operations such as rotation, scaling, translation, and adding noise to generate more variant samples. Data enhancement helps the model to be more robust when processing vibration signals under different conditions. The generated training samples are divided into a training set, a validation set, and a test set. The training set is used to train the neural network; the validation set is used to adjust the model parameters during the training process to avoid overfitting; and the test set is used to evaluate the performance of the final model.

[0049] S1032. Based on multiple training samples, neural network training is performed to obtain an artificial intelligence model.

[0050] Exemplarily, for the processing of two-dimensional time-frequency images, common network architectures include convolutional neural networks (CNNs) and their variants, such as Deep CNN, ResNet, etc. These networks can effectively extract features in the images. Select an appropriate loss function to measure the difference between the model prediction results and the true fault types. For classification problems, common loss functions include Cross-Entropy Loss, Focal Loss, etc. The choice of the loss function will directly affect the training effect and performance of the model. Adopt an appropriate optimization algorithm to update the weights and biases of the neural network to minimize the loss function. Common optimization algorithms include Stochastic Gradient Descent (SGD), Adam, RMSprop, etc. During the training process, hyperparameters such as the learning rate and momentum can be adjusted according to the performance of the validation set. During the training process, regularly monitor metrics such as the loss function values and accuracy of the training set and the validation set to evaluate the training progress and performance of the model. If it is found that the performance of the model on the validation set begins to decline (i.e., overfitting occurs), methods such as Early Stopping and Regularization can be adopted to alleviate the overfitting problem. After the training is completed, save the trained model to disk for subsequent inference or further tuning. At the same time, a function to load the model also needs to be provided to reload and use the model when needed.

[0051] Exemplarily, step S1032 can be specifically implemented as steps one to six.

[0052] Step one, build an initial model and set the model parameters of the initial model as the model parameters of the current iteration process.

[0053] In some embodiments, the initial model includes a deep residual shrinkage module and a convolutional neural network module. The model parameters of the deep residual shrinkage module include attention parameters and soft threshold parameters; the model parameters of the convolutional neural network model include the number of layers, the number, and the weights of the neurons.

[0054] Exemplarily, the deep residual shrinkage module combines the attention mechanism and the soft thresholding method to extract key features in the vibration signal. The attention parameters are used to control the attention of the model to different frequency components, and the soft threshold parameters are used to remove noise and redundant information. The convolutional neural network module consists of multiple convolutional layers, pooling layers, and fully connected layers to further extract and classify features. Parameters such as the number of layers, the number, and the weights of the neurons determine the complexity and learning ability of the network. Model parameter initialization: Use methods such as random initialization or pre-trained weights to assign values to the parameters of the initial model. These parameters will be updated through the backpropagation algorithm in the subsequent iteration process.

[0055] Step 2: Based on the two-dimensional time-frequency image of the current training sample, as well as the deep residual shrinkage module and model parameters in the current iteration process, key features are obtained.

[0056] For example, the two-dimensional time-frequency image of the current training sample is used as the input of the deep residual shrinkage module. The key features in the image are extracted through the attention mechanism and soft thresholding method of the deep residual shrinkage module. These features reflect the changes of the vibration signal at different frequencies and time scales.

[0057] Exemplarily, step 2 may be specifically implemented as A1-A4.

[0058] A1. Based on the two-dimensional time-frequency image of the current training sample and the residual block of the deep residual shrinkage module, the two-dimensional time-frequency image is preliminarily shrunk to obtain the high-order features of the current training sample.

[0059] Among them, the embodiment of the present invention can use the two-dimensional time-frequency image of the current training sample as the input of the deep residual shrinkage module. Ensure that the image data has been properly preprocessed, such as normalization, denoising, etc. The residual block in the deep residual shrinkage module includes components such as convolutional layers, batch normalization layers, and activation functions. The convolutional layer extracts local features in the image, the batch normalization layer accelerates the training process and improves the stability of the model, and the activation function (such as ReLU) introduces nonlinearity. These components work together on the input image to generate preliminary high-order features. In the residual block, a skip connection is usually included, which adds the input features directly to the output features to achieve feature fusion and direct transfer of gradients. This helps to alleviate the gradient vanishing problem in deep networks.

[0060] A2. Based on the attention parameters of the current iteration process, determine the weight coefficients of each channel feature in the high-order features of the current training sample.

[0061] For example, the attention mechanism is introduced in the deep residual shrinkage module to dynamically adjust the weights of each channel feature in the high-order feature. This is usually achieved through components such as global average pooling, fully connected layers, and sigmoid activation functions. The global average pooling layer compresses the spatial dimension of the high-order feature into a single channel to obtain a compact feature representation. Then, the correlation between the channels is learned through the fully connected layer, and the sigmoid activation function is used to map the output to the [0,1] interval as the weight coefficient of each channel feature.

[0062] A3. Generate attention features based on the high-order features of the current training sample and the weight coefficients of each channel feature.

[0063] Exemplarily, the weight coefficients calculated in step 2 are multiplied element-by-element with the high-order features to achieve feature weighting. This helps to emphasize important features and suppress irrelevant or noisy features. Attention feature generation: The weighted features are attention features. They reflect the importance of each channel in the original high-order features and provide more valuable information for subsequent feature extraction and classification.

[0064] A4. Based on the soft threshold parameters in the current iteration process, the attention features are deeply shrunk to obtain the key features.

[0065] Exemplarily, the soft thresholding operation in the deep residual shrinkage module is used to further remove noise and redundant information. It processes the attention features element by element based on the soft threshold parameter in the current iteration. The soft thresholding function is usually defined as f(x) = sign(x) * max(|x| - λ, 0), where x is an element in the attention feature, λ is the soft threshold parameter, and sign(x) is the sign function. Through the soft thresholding operation, small values ​​in the attention feature (usually corresponding to noise or redundant information) are set to zero or close to zero, while larger values ​​(corresponding to important features) are retained. In this way, more refined and robust key features are obtained for subsequent classification tasks.

[0066] Step three: Based on the convolutional neural network model and model parameters in the current iteration process, as well as the key features, output the predicted fault type.

[0067] For example, the embodiment of the present invention can input the extracted key features into a convolutional neural network module, and map the features to the fault type space through operations such as convolution, pooling, and full connection. Based on the mapped features, a classifier such as a softmax function is used to output the probability distribution of the predicted fault type.

[0068] Step 4: Compare the deviation between the fault type of the current training sample and the predicted fault type.

[0069] Exemplarily, the embodiment of the present invention can use a measurement method such as a cross entropy loss function to calculate the deviation between the actual fault type and the predicted fault type of the current training sample. According to the value of the loss function, the prediction performance of the model is evaluated. The smaller the loss value, the better the prediction performance of the model.

[0070] Step 5: If the deviation meets the preset conditions, exit the iteration and execute step 6; if the deviation does not meet the preset conditions, change the model parameters of the deep residual shrinkage module and the convolutional neural network module, and repeat steps 2 to 5 until the deviation meets the preset conditions.

[0071] Exemplarily, the preset conditions usually include the number of iterations, loss value threshold or validation set performance, etc. When these conditions are met, the iteration process will terminate. Use the back propagation algorithm and optimizer (such as Adam, SGD, etc.) to update the model parameters of the deep residual shrinkage module and the convolutional neural network module according to the gradient of the loss function. Repeat steps 2 to 5 until the deviation meets the preset conditions or the maximum number of iterations is reached.

[0072] Step six, determine whether the training set in the training sample has been trained. If the training is completed, build an artificial intelligence model based on the model parameters of the current deep residual shrinkage module and convolutional neural network module; if the training is not completed, change the training sample and repeat steps two to six until the training set is trained.

[0073] Exemplarily, ensure that all samples in the training set have been used to train the model. Use the model parameters of the trained deep residual shrinkage module and convolutional neural network module to build the final artificial intelligence model. Evaluate the performance of the model on the validation set to ensure that it has good generalization ability. If the validation set performance is poor, it may be necessary to adjust the model structure, optimizer parameters, or data enhancement strategy. If the training set is not trained, replace the new training sample and repeat steps 2 to 6 until the training set is fully trained.

[0074] S104. Based on the artificial intelligence model, the working state of the shape-integrated gearbox is monitored in real time to obtain fault detection results.

[0075] Exemplarily, the embodiment of the present invention can collect vibration signals in real time during the operation of the gearbox, and perform wavelet transform processing to obtain a two-dimensional time-frequency image. The processed two-dimensional time-frequency image is input into a trained artificial intelligence model for reasoning to obtain a fault detection result. The fault detection result is output in a visual form, such as the fault type, fault degree, location of occurrence, etc. At the same time, the result can be sent to a remote monitoring center or an alarm system can be triggered as needed. According to the fault detection results, corresponding maintenance suggestions are given, such as replacing damaged parts, adjusting operating parameters, etc., to ensure the safe and stable operation of the gearbox.

[0076] As a possible implementation manner, step S104 can be specifically implemented as steps S1041 - S1044 .

[0077] S1041. Obtain a real-time vibration signal of the shape-integrated gearbox in the current period.

[0078] Exemplarily, an embodiment of the present invention can install acceleration sensors or vibration sensors at key locations of a shape-integrated gearbox to capture the vibration signal of the gearbox in real time. The vibration signal of the gearbox is collected in real time at a certain sampling rate and resolution through a data acquisition system. Ensure that the sampling rate and resolution are high enough to capture subtle changes in the vibration of the gearbox. Preprocess the collected vibration signal, such as denoising, filtering, and normalization, to improve the quality and reliability of the signal.

[0079] S1042: Perform wavelet transform processing based on the real-time vibration signal to obtain a two-dimensional time-frequency image of the current period.

[0080] Exemplarily, the embodiments of the present invention can select appropriate wavelet basis functions and decomposition layers for wavelet transform according to the characteristics of the gearbox vibration signal. Common wavelet basis functions include Haar wavelet, Daubechies wavelet, etc. Through wavelet transform, the vibration signal is converted from the time domain to the time-frequency domain to obtain a time-frequency image of different frequency components changing with time. This helps to capture the frequency changes and fault characteristics in the gearbox vibration. The time-frequency data obtained by the wavelet transform is visualized to generate a two-dimensional time-frequency image of the current time period. Ensure that the image is clear and easy to identify.

[0081] S1043. Input the two-dimensional time-frequency image of the current period into the deep residual shrinkage module of the artificial intelligence model to obtain the key features of the current period.

[0082] Exemplarily, the embodiments of the present invention can load a trained artificial intelligence model, including a deep residual shrinkage module and a convolutional neural network module. The two-dimensional time-frequency image of the current period is used as the input of the deep residual shrinkage module. Ensure that the image format is consistent with the model input requirements. Through the residual block, attention mechanism, soft thresholding and other components in the deep residual shrinkage module, the input image is feature extracted and noise suppressed to obtain the key features of the current period. These features reflect the key information in the gearbox vibration signal, such as fault characteristics, frequency changes, etc.

[0083] S1044. Based on the key features of the current time period and the convolutional neural network module of the artificial intelligence model, output the fault detection results of the current time period.

[0084] Exemplarily, the embodiments of the present invention can input the extracted key features into a convolutional neural network module, and map the features to the fault type space through operations such as convolutional layers, pooling layers, and fully connected layers. In the output layer of the convolutional neural network module, a classifier such as a softmax function is used to classify the mapped features, and the fault detection results for the current period are output. The detection results are usually given in the form of a probability distribution, indicating the possibility of different fault types. Based on the output results, the current working status of the gearbox and the possible causes of the fault are explained. If a fault is detected, an alarm is issued in a timely manner or corresponding measures are taken for maintenance.

[0085] The present invention provides a real-time monitoring method for the working state of a gearbox. All feature information in scale spaces such as time domain and frequency domain is obtained through wavelet transformation integration. Then, in the training process of an artificial intelligence model, attention mechanism and soft thresholding are used to perform feature screening on the time domain information and the frequency domain information to achieve denoising, and key features related to the fault type are obtained. The artificial intelligence model is trained based on the key features and the fault type. The gearbox is monitored in real time based on the artificial intelligence model obtained in this way, which can effectively reduce the noise in the vibration signal, reduce the influence of the noise on the gearbox fault prediction, and improve the accuracy and reliability of the gearbox fault diagnosis with integrated form.

[0086] Optionally, the real-time monitoring method for the working status of a gearbox provided in an embodiment of the present invention further includes steps S201 - S202 after step S104 .

[0087] S201. If the fault detection result is a minor fault, a fault monitoring instruction is generated.

[0088] In some embodiments, the fault monitoring instructions are used to instruct the gear system of the monitoring profile integrated gearbox to operate with faults; minor faults include wear, pitting and scratches.

[0089] For example, if the fault type is wear, pitting and scratching, the gear system will not fail immediately, but the long-term existence may affect the performance and life of the gear system. The gear system can be monitored and operated with faults.

[0090] Exemplarily, after step S201, steps B1-B5 are also included.

[0091] B1. Record infrared data after gear system failure.

[0092] For example, after a gear system fails, the existence of the fault is first confirmed by visual inspection, vibration analysis or other diagnostic means. An infrared thermal imager is used to record infrared images or video data of the faulty gear system to ensure that the entire fault period is covered. The infrared data is converted into a digital format for subsequent processing and analysis.

[0093] B2. Divide the infrared data after the gear system failure into sliding time windows to obtain infrared data of multiple sliding time windows.

[0094] Exemplarily, an embodiment of the present invention can set the size of the sliding time window (i.e., the length of time contained in each window) according to the time range of the fault occurrence and the sampling frequency of the infrared data. The sliding time window should be small enough to capture the dynamic changes of the fault, and large enough to contain sufficient fault information. Determine the sliding step size, that is, the time overlap between two adjacent windows. Divide the recorded infrared data according to the set sliding time window to obtain multiple continuous and non-overlapping (or with a certain overlap) time window data. The infrared data in each time window should contain sufficient temperature distribution information for subsequent stress analysis and fault detection.

[0095] B3. Based on the infrared data of multiple sliding time windows and the stress analysis model, the analysis results of multiple sliding time windows are determined.

[0096] In some embodiments, the stress analysis model is used to analyze the temperature and stress changes under various conditions.

[0097] In some embodiments, the analysis results include the fault type, the fault probability of each fault type, and the fault location.

[0098] Exemplarily, the infrared data of each sliding time window is processed using the constructed stress analysis model. The model should be able to predict the stress distribution and possible fault types of the gear system based on the temperature distribution information in the infrared data. Information such as the fault type, the fault probability of each fault type, and the fault location are extracted from the output of the model. The fault type may include wear, pitting, broken teeth, etc.; the fault probability indicates the possibility of each fault type in the current time window; and the fault location indicates the specific location of the fault in the gear system.

[0099] B4. Based on the analysis results of multiple sliding time windows, analyze the faults at the fault locations in the fault detection results to determine the fault trend.

[0100] For example, based on the analysis results of multiple sliding time windows, the changing trends of fault type, fault probability and fault location over time are observed. Statistical analysis, machine learning or data mining methods are used to identify the development trend of faults, such as whether the fault is aggravated or spread to other components.

[0101] B5. Determine whether the gear system is shut down based on the fault detection results, the analysis results of multiple sliding time windows, and the fault trend.

[0102] For example, the overall state of the gear system is comprehensively evaluated by combining the original fault detection results (such as the preliminary fault judgment obtained by the stress analysis model), the analysis results of multiple sliding time windows, and the fault trend analysis. Considering factors such as the fault type, fault probability, fault location, fault development trend, the importance of the gear system, and the downtime cost, a decision on whether to shut down is made.

[0103] If the evaluation results indicate that the gear system has a serious fault or the fault trend is uncontrollable, a shutdown order is generated. The shutdown order should include information such as the reason for the shutdown, the shutdown time, the safety measures after the shutdown, and the maintenance plan. Relevant personnel are notified to perform the shutdown operation and take necessary safety measures and maintenance measures in accordance with the requirements of the shutdown order.

[0104] S202: If the fault detection result is a serious fault, a shutdown instruction is generated.

[0105] In some embodiments, the shutdown instruction is used to instruct the gear system to shut down, and serious faults include broken teeth.

[0106] For example, if the fault type is broken teeth, the gear system may fail immediately or cause serious safety hazards, and the gear system should be instructed to shut down immediately for maintenance.

[0107] In this way, the present invention can classify and process the gear system faults according to their severity, reduce the downtime of the gear system, and improve the accuracy, convenience and detection efficiency of the gear system fault detection.

[0108] Optionally, a stress analysis model may be constructed through steps C1-C9.

[0109] C1. Obtain infrared data and stress data of the gear system of the shape-integrated gearbox under various conditions.

[0110] Among them, the various conditions include normal, worn, pitted, scratched and broken teeth.

[0111] In some embodiments, the stress data includes stress at each key point at each time within a set time period under various states.

[0112] For example, the embodiments of the present invention may use a high-precision infrared thermal imager and a stress sensor. The infrared thermal imager is used to capture the temperature distribution on the gear surface, and the stress sensor is arranged at the key position of the gear to measure the stress change. According to the operating speed and fault characteristics of the gear system, an appropriate data sampling rate is set to ensure that sufficient fault information can be captured.

[0113] C2. Based on the infrared data in the gear system under various states, time windows are divided to generate multiple first input features.

[0114] C3. Generate stress characteristics based on stress data under various states.

[0115] In some embodiments, the embodiments of the present invention can organize stress data under various states to form a stress time series or a stress distribution diagram. Preprocess the stress data, such as removing outliers, smoothing, etc. Extract features of the stress data, such as stress peak value, stress mean value, stress distribution range, etc., as stress features.

[0116] Exemplarily, step C3 may be specifically implemented as steps C31-C34.

[0117] C31. Based on the pre-built 3D twin model of the gear system, determine the stress-related points of each key point.

[0118] In some embodiments, the stress-related points include direct contact nodes and indirect contact nodes.

[0119] C32. Based on the stress data under various states, extract the stress data of each key point and the stress data of stress-related points of each key point.

[0120] C33. For any state, based on the stress data of each key point at each moment in the state, and the stress data of stress-related points of each key point, the stress time series of each key point and the stress time series of stress-related points of each key point are constructed.

[0121] C34. Based on the stress time series of each key point and the stress time series of stress-related points of each key point, spatial sorting is performed to obtain the stress characteristics under this state.

[0122] C4. Based on the state type, the first input feature and the stress feature are divided to obtain the first input feature and the stress feature corresponding to each state type.

[0123] Exemplarily, an embodiment of the present invention can divide the first input characteristics and stress characteristics according to the state type of the gear system (such as normal, broken teeth, wear, pitting, scratches, etc.) to obtain the first input characteristics and stress characteristics corresponding to each state type.

[0124] C5. Taking the first input feature corresponding to each state type as input and the stress feature corresponding to each state type as output, a first training sample is obtained.

[0125] C6. Taking the stress characteristics corresponding to each state type as input and each state type as output, a second training sample is obtained.

[0126] C7. Based on the first training sample, perform neural network training to obtain a first model.

[0127] Exemplarily, the embodiment of the present invention can select a suitable neural network architecture (such as a multilayer perceptron, a convolutional neural network, etc.) and set corresponding hyperparameters (such as a learning rate, the number of iterations, etc.). The neural network is trained using the first training sample, and the network weights are adjusted by the back propagation algorithm so that the network can accurately predict the stress characteristics under a given temperature change. After the training is completed, a first model is obtained, which can receive the temperature change as an input and output the predicted stress characteristics.

[0128] C8. Based on the second training sample, perform neural network training to obtain a second model.

[0129] Exemplarily, the embodiment of the present invention can select a suitable neural network architecture and hyperparameters. The neural network is trained using the second training sample so that the network can identify the state type of the gear system according to the stress characteristics. After the training is completed, a second model is obtained, which can receive the stress characteristics as input and output the identified state type.

[0130] C9. Construct a stress analysis model based on the first model and the second model.

[0131] Exemplarily, the embodiment of the present invention can integrate the first model and the second model to construct a complete stress analysis model. The model can receive the temperature change of the gear system as input, first predict the stress characteristics through the first model, and then identify the state type through the second model. The model can also be further optimized and adjusted as needed to improve the accuracy and robustness of the model.

[0132] Optionally, the real-time monitoring method for the working status of a gearbox provided in an embodiment of the present invention further includes steps S301 - S304 after step S104 .

[0133] S301. If the fault detection result is a gearbox fault, obtain the real-time operating parameters of the morphologically integrated gearbox in the current time period.

[0134] In some embodiments, the real-time operating parameters include speed, temperature, pressure, and load.

[0135] Exemplarily, in an embodiment of the present invention, a variety of sensors, such as a speed sensor, a temperature sensor, a pressure sensor, and a load sensor, can be installed on the gearbox to capture the operating parameters of the gearbox in real time. The data of these sensors are collected in real time through a data acquisition system, and the synchronization of the data is ensured. That is, the data of all sensors are collected at the same time point for subsequent analysis and processing. The collected operating parameter data is preprocessed, including denoising, filtering, outlier processing, etc., to improve the accuracy and reliability of the data.

[0136] S302: Generate operating condition characteristics of the current period based on the real-time operating condition parameters of the current period and the two-dimensional time-frequency image of the current period.

[0137] Exemplarily, the embodiments of the present invention can fuse real-time operating parameters (such as speed, temperature, pressure and load) with features in the two-dimensional time-frequency image. This can be achieved by adding the operating parameters as additional feature dimensions to the feature vector of the time-frequency image. Using machine learning or deep learning technology, key features that can reflect the current operating conditions of the gearbox are extracted from the fused feature vector. These features may include vibration modes, temperature change trends, pressure fluctuations, etc. of the gearbox.

[0138] S303: Based on the working condition characteristics of the current period and the preset Gaussian process regression model, stress prediction is performed to obtain the stress characteristics of the shape-integrated gearbox.

[0139] Exemplarily, an embodiment of the present invention may require the use of historical data and operating condition characteristics to train a Gaussian process regression model before fault detection. This model can learn the relationship between operating condition characteristics and gearbox stress. Model input: The operating condition characteristics of the current period are used as the input of the Gaussian process regression model. Stress prediction: Use the trained Gaussian process regression model to perform stress prediction on the operating condition characteristics of the current period. The prediction results may include stress distribution and stress levels of various components of the gearbox. Stress feature extraction: Extract key features that can reflect the stress state of the gearbox from the prediction results, such as the maximum stress value, stress change trend, etc.

[0140] S304. Based on the stress characteristics, the fault detection result is verified to determine the fault type of the shape-integrated gearbox.

[0141] Exemplarily, an embodiment of the present invention may be a fault type library: establish a database containing various gearbox fault types, each fault type corresponds to a specific stress feature. Match the predicted stress features with the features in the fault type library. Determine which fault type the current fault is closest to by calculating the similarity or distance metric. Based on the matching result, determine the fault type of the morphologically integrated gearbox. If the match is successful, the corresponding fault type is output; if the match is unsuccessful, it may indicate that a new fault type has occurred or further diagnosis is required. Based on the determined fault type, take corresponding maintenance measures or repair plans to ensure the normal operation of the gearbox and extend its service life.

[0142] In this way, the embodiment of the present invention can comprehensively consider the stress characteristics, check the fault detection results, ensure the accuracy of the fault detection results, and further improve the accuracy and reliability of the fault diagnosis of the shape-integrated gearbox.

[0143] Optionally, the real-time monitoring method for the working status of a gearbox provided in an embodiment of the present invention further includes steps S401 - S403 before step S303 .

[0144] S401. Obtain vibration signals, operating parameters and stress characteristics of the shape-integrated gearbox in a normal state and various fault states.

[0145] For example, in order to obtain data of the gearbox in normal state and various fault states, a series of experiments need to be designed. These experiments should cover various operating conditions and fault types that the gearbox may encounter. During the experiment, sensors are used to collect the vibration signals (such as acceleration, displacement, etc.), operating parameters (such as speed, temperature, pressure, load, etc.) of the gearbox in real time, as well as stress characteristics obtained by other means (such as strain gauges, finite element analysis, etc.). The collected data is marked to clarify the gearbox state (normal or a certain fault type) corresponding to each set of data. This helps with subsequent data analysis and model training. The collected data is stored in a safe and reliable storage medium for subsequent data processing and model training.

[0146] S402: Generate multiple operating condition characteristics based on the vibration signal and the operating condition parameters.

[0147] Exemplarily, the embodiments of the present invention can pre-process the collected vibration signals, such as denoising, filtering, standardization, etc., to improve the quality and consistency of the signals. Signal processing techniques (such as time domain analysis, frequency domain analysis, time-frequency analysis, etc.) and machine learning algorithms (such as principal component analysis, cluster analysis, etc.) are used to extract key features that can reflect the working conditions of the gearbox from the vibration signals. The extracted vibration features are integrated with the working condition parameters (such as speed, temperature, pressure, load, etc.) to form a vector containing multiple working condition features. These feature vectors will be used for subsequent model training and stress prediction.

[0148] S403: Based on multiple operating condition characteristics and stress characteristics corresponding to each operating condition characteristic, Gaussian regression fitting is performed to obtain a Gaussian process regression model.

[0149] Exemplarily, an embodiment of the present invention can use the operating condition feature vector generated in step S402 as input and the corresponding stress feature as output to construct a training data set. Gaussian process regression (GPR) is selected as the modeling method. GPR is a non-parametric Bayesian regression method that can handle high-dimensional input data and nonlinear relationships. The Gaussian process regression model is trained using the training data set. During the training process, it is necessary to set the hyperparameters of the model (such as kernel function, kernel parameters, etc.), which can be optimized by maximizing the marginal likelihood function of the model. Use cross-validation or other validation methods to validate the trained model to ensure the accuracy and generalization ability of the model. Store the trained Gaussian process regression model in an appropriate location for subsequent use in stress prediction and fault diagnosis.

[0150] In this way, the embodiment of the present invention can use the Gaussian process regression algorithm to perform data fitting on the operating characteristics and stress characteristics before verifying the fault detection results through stress characteristics to obtain a Gaussian process regression model, thereby more comprehensively considering the fault detection results and further improving the accuracy and reliability of the morphologically integrated gearbox fault diagnosis.

[0151] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0152] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0153] Figure 2 The schematic diagram of the structure of a real-time monitoring device for the working state of a gearbox provided by an embodiment of the present invention is shown. The monitoring device 500 includes a communication module 501 and a processing module 502 .

[0154] The communication module 501 is used to obtain vibration signals of the conformal integrated gearbox under various fault conditions and record the fault type of each vibration signal.

[0155] The processing module 502 is used to perform wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image; the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal; based on the two-dimensional time-frequency image and the fault type of each vibration signal, a neural network training is performed to obtain an artificial intelligence model, and during the training process, an attention mechanism and a soft thresholding method are used to denoise and screen the two-dimensional time-frequency image, and key features are extracted to identify the fault type; based on the artificial intelligence model, the working state of the morphologically integrated gearbox is monitored in real time to obtain a fault detection result.

[0156] Figure 3Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, the electronic device 600 of this embodiment includes: a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program 603, the steps in the above-mentioned method embodiments are implemented, for example Figure 1 Alternatively, when the processor 601 executes the computer program 603, the functions of each module / unit in the above-mentioned device embodiments are realized, for example, Figure 2 The functions of the communication module 501 and the processing module 502 are shown.

[0157] Exemplarily, the computer program 603 may be divided into one or more modules / units, which are stored in the memory 602 and executed by the processor 601 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program 603 in the electronic device 600. For example, the computer program 603 may be divided into Figure 2 A communication module 501 and a processing module 502 are shown.

[0158] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0159] The memory 602 may be an internal storage unit of the electronic device 600, such as a hard disk or memory of the electronic device 600. The memory 602 may also be an external storage device of the electronic device 600, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 600. Further, the memory 602 may also include both an internal storage unit of the electronic device 600 and an external storage device. The memory 602 is used to store the computer program and other programs and data required by the terminal. The memory 602 may also be used to temporarily store data that has been output or is to be output.

[0160] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A real-time monitoring method for the working state of a gearbox, characterized in that: include: Obtain the vibration signals of the conformal integrated gearbox under various fault conditions and record the fault type of each vibration signal; Performing wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image; the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal; Based on the two-dimensional time-frequency image and the fault type of each vibration signal, a neural network training is performed to obtain an artificial intelligence model. During the training process, an attention mechanism and a soft thresholding method are used to denoise and screen the two-dimensional time-frequency image, and key features are extracted to identify the fault type. Based on the artificial intelligence model, the working state of the conformal integrated gearbox is monitored in real time to obtain fault detection results.

2. The real-time monitoring method of the working state of the gear box according to claim 1 is characterized in that: The step of performing wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image includes: Analyzing the frequency range, time domain characteristics and noise composition of the vibration signal; Determining a wavelet basis function based on the frequency range, time domain characteristics and noise composition; Decomposing the vibration signal based on the wavelet basis function to obtain wavelet coefficients of multiple scales; The wavelet coefficients of the multiple scales are reconstructed to obtain the two-dimensional time-frequency image.

3. The real-time monitoring method of the working state of the gear box according to claim 1 is characterized in that: The neural network training is performed based on the two-dimensional time-frequency image and the fault type of each vibration signal to obtain an artificial intelligence model, including: Generate multiple training samples based on the two-dimensional time-frequency image and the fault type of each vibration signal; Based on the multiple training samples, neural network training is performed to obtain the artificial intelligence model.

4. The real-time monitoring method of the working state of the gear box according to claim 3 is characterized in that: The performing of neural network training based on the plurality of training samples to obtain the artificial intelligence model comprises: Step 1: build an initial model and set the model parameters of the initial model as the model parameters of the current iteration process; the initial model includes a deep residual shrinkage module and a convolutional neural network module, and the model parameters of the deep residual shrinkage module include attention parameters and soft threshold parameters; the model parameters of the convolutional neural network model include the number of layers, quantity and weights of neurons; Step 2: Based on the two-dimensional time-frequency image of the current training sample, as well as the deep residual shrinkage module and model parameters in the current iteration process, key features are obtained; Step 3: output the predicted fault type based on the convolutional neural network model and model parameters in the current iteration process and the key features; Step 4: compare the deviation between the fault type of the current training sample and the predicted fault type; Step 5: If the deviation meets the preset conditions, exit the iteration and execute step 6; if the deviation does not meet the preset conditions, change the model parameters of the deep residual shrinkage module and the convolutional neural network module, and repeat steps 2 to 5 until the deviation meets the preset conditions; Step six, determine whether the training set in the training sample has been trained. If the training is completed, construct the artificial intelligence model based on the model parameters of the current deep residual shrinkage module and the convolutional neural network module; if the training is not completed, change the training sample and repeat steps two to six until the training set is trained.

5. The real-time monitoring method of the working state of the gear box according to claim 4, characterized in that: The two-dimensional time-frequency image based on the current training sample, as well as the deep residual shrinkage module and model parameters in the current iteration process, obtains key features, including: Based on the two-dimensional time-frequency image of the current training sample and the residual block of the deep residual shrinkage module, the two-dimensional time-frequency image is preliminarily shrunk to obtain high-order features of the current training sample; Based on the attention parameters of the current iteration process, the weight coefficients of each channel feature in the high-order features of the current training sample are determined; Generate attention features based on the high-order features of the current training sample and the weight coefficients of each channel feature; Based on the soft threshold parameter in the current iteration process, the attention feature is deeply shrunk to obtain the key feature.

6. The real-time monitoring method of the working state of a gear box according to claim 1, characterized in that: Based on the artificial intelligence model, the working state of the shape-integrated gearbox is monitored in real time to obtain fault detection results, including: Obtain the real-time vibration signal of the shape-integrated gearbox in the current period; Based on the real-time vibration signal, wavelet transform processing is performed to obtain a two-dimensional time-frequency image of the current period; Input the two-dimensional time-frequency image of the current period into the deep residual shrinkage module of the artificial intelligence model to obtain the key features of the current period; Based on the key features of the current period and the convolutional neural network module of the artificial intelligence model, the fault detection results of the current period are output.

7. The real-time monitoring method of the working state of a gear box according to claim 1, characterized in that: After the working state of the shape-integrated gearbox is monitored in real time based on the artificial intelligence model and the fault detection result is obtained, the method further includes: If the fault detection result is a minor fault, a fault monitoring instruction is generated, wherein the fault monitoring instruction is used to instruct the gear system of the monitoring shape-integrated gearbox to operate with a fault; the minor fault includes wear, pitting and scratches; If the fault detection result is a serious fault, a shutdown instruction is generated, and the shutdown instruction is used to instruct the gear system to shut down. The serious fault includes a broken tooth.

8. The method for real-time monitoring of the working state of a gear box according to claim 7, characterized in that: If the fault detection result is a minor fault, a fault monitoring instruction is generated, and then the following steps are further included: Record infrared data after gear system failure; Divide the infrared data after the gear system failure into sliding time windows to obtain infrared data of multiple sliding time windows; Determine analysis results of the multiple sliding time windows based on the infrared data of the multiple sliding time windows and the stress analysis model, wherein the analysis results include fault types, fault probabilities of each fault type, and fault locations; Based on the analysis results of the multiple sliding time windows, analyzing the faults at the fault locations in the fault detection results to determine the fault trends; Based on the fault detection result, the analysis results of the multiple sliding time windows, and the fault trend, it is determined whether the gear system is shut down.

9. The method for real-time monitoring of the working state of a gearbox according to claim 1, characterized in that: After the working state of the shape-integrated gearbox is monitored in real time based on the artificial intelligence model and the fault detection result is obtained, the method further includes: If the fault detection result is a gearbox fault, then the real-time operating parameters of the morphologically integrated gearbox in the current time period are obtained; the real-time operating parameters include speed, temperature, pressure and load; Based on the real-time working condition parameters of the current period and the two-dimensional time-frequency image of the current period, the working condition characteristics of the current period are generated; Based on the working condition characteristics of the current period and a preset Gaussian process regression model, stress prediction is performed to obtain the stress characteristics of the shape-integrated gearbox; Based on the stress characteristics, the fault detection result is checked to determine the fault type of the form-fitting integrated gearbox.

10. A real-time monitoring device for the working state of a gearbox, characterized in that: include: A communication module is used to obtain vibration signals of the conformal integrated gearbox under various fault conditions and record the fault type of each vibration signal; A processing module is used to perform wavelet transform processing on the vibration signal to obtain a two-dimensional time-frequency image; the two-dimensional time-frequency image is used to reflect the time domain information and frequency domain information of the vibration signal; based on the two-dimensional time-frequency image and the fault type of each vibration signal, a neural network training is performed to obtain an artificial intelligence model, and during the training process, an attention mechanism and a soft thresholding method are used to perform denoising and screening on the two-dimensional time-frequency image, and key features are extracted to identify the fault type; Based on the artificial intelligence model, the working state of the conformal integrated gearbox is monitored in real time to obtain fault detection results.

Images