An industrial production and manufacturing intelligent supervision system based on an internet of things

Abstract

The application relates to the field of production and manufacturing supervision, and particularly relates to an industrial production and manufacturing intelligent supervision system based on the Internet of Things, which comprises a data collection module, a feature extraction module, a finite element simulation module and a generative adversarial network model, wherein the data collection module is used for collecting vibration information of equipment, the feature extraction module is used for obtaining a trajectory of a dominant frequency changing over time, the finite element simulation module is used for simulating a stress nephogram according to input energy distribution and the trajectory of the dominant frequency changing over time, and the generative adversarial network model is used for generating a crack distribution diagram, and the authenticity of an image is judged by comparing the crack distribution diagram with the simulated stress nephogram. The application can prepare in advance before equipment damage by collecting vibration information of equipment and performing frequency band analysis on the vibration information, so that economic losses caused by failure to handle in time when a fault occurs can be avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of production supervision, and particularly relates to an industrial production supervision system based on the Internet of Things. BACKGROUND

[0002] With the continuous development of the Internet of Things, the association between devices is becoming more and more complex, and the systematized and automated production line gradually replaces the traditional manual production. With the continuous popularization of systematization and automation, more and more enterprises begin to use intelligent supervision to supervise the production line. Unlike manual supervision, the intelligent supervision system has higher advantages in supervision efficiency and supervision cost, and faster feedback efficiency in responding to faulty devices.

[0003] The existing supervision system used in industrial production usually judges whether the device in the Internet of Things is in an abnormal working state according to the service life, current, voltage and other working conditions of the device. Although this supervision method can accurately know the location of the abnormal state, it cannot prevent the device from failing in advance, because the aging of the device does not cause the device to be unable to use. However, when the load of the aging device is too large, it may cause direct failure problems, which cannot be handled by detecting current, voltage and other conditions. SUMMARY

[0004] The present application provides the following technical solutions in view of the deficiencies of the prior art.

[0005] An industrial production supervision system based on the Internet of Things, comprising a data collection module, a feature extraction module, a finite element simulation module and a generative adversarial network model.

[0006] The data collection module is used to collect vibration information of the device, the feature extraction module is used to perform wavelet packet decomposition on the vibration information to obtain energy distribution of each frequency band, and obtain a trajectory of dominant frequency changing with time from an abnormal frequency band, the finite element simulation module is used to simulate a stress cloud map according to the input energy distribution and the trajectory of the dominant frequency changing with time, the generative adversarial network model is used to generate a crack distribution map according to the trajectory of the dominant frequency changing with time and random noise, compare the crack distribution map with the simulated stress cloud map to judge the authenticity of the image, and analyze the stress cloud map and the crack distribution map to judge the fault condition of the device.

[0007] As an improvement of the above technical solution, the data collection module at least comprises a laser Doppler vibration measurement module and an acoustic emission signal processing module.

[0008] The vibration information at least includes vibration signals of the device surface and stress wave information, the laser Doppler vibration testing module is used to test the vibration signals of the device surface, and the acoustic emission signal processing module is used to capture the stress wave information released by the internal crack of the material.

[0009] As an improvement of the above technical solution, the feature extraction module at least includes a time domain convolution network module.

[0010] The method for obtaining the trajectory of the dominant frequency changing over time from the abnormal frequency band includes the following steps:

[0011] The input vibration information is decomposed into several sub-bands by the time domain convolution module, the energy entropy of each frequency band is calculated, and the abnormal frequency band is located according to the energy entropy.

[0012] The trajectory of the dominant frequency changing over time is extracted from the abnormal frequency band, and the instantaneous impact feature is captured.

[0013] As an improvement of the above technical solution, the wavelet packet decomposition adopts a multi-Bessel wavelet decomposition method, decomposes the vibration information into a fourth-order wavelet, generates 16 sub-bands, and calculates the energy entropy of each sub-band.

[0014] As an improvement of the above technical solution, the trajectory of the dominant frequency changing over time is extracted from the abnormal frequency band, including the following steps:

[0015] The vibration information is transformed into sub-bands by the multi-Bessel wavelet decomposition method and the time-frequency distribution is calculated to generate a time-frequency matrix.

[0016] The maximum energy frequency of each time point is found, the time-frequency coordinates of the energy frequency are recorded, and the time-frequency ridge line of the energy frequency is output as the extracted trajectory of the dominant frequency changing over time.

[0017] As an improvement of the above technical solution, the generation of the time-frequency matrix depends on the following formula:

[0018]

[0019] Wherein, represents the input original vibration information, is a scale parameter, is a translation parameter, is a wavelet base function, is a scaling factor.

[0020] As an improvement of the above technical solution, the stress cloud simulation method of the finite element simulation module includes the following steps:

[0021] The trajectory of the main frequency changing over time is input to a finite element simulation module to obtain a corresponding dynamic load, frequency and energy characteristics of the dynamic load are extracted, and transient load boundary conditions are obtained according to the frequency and energy characteristics.

[0022] The time-frequency ridge line slope is obtained according to the trajectory of the main frequency changing over time, and the crack growth rate is estimated according to the ridge line slope.

[0023] The stress contour is simulated according to the transient load boundary conditions and the crack growth rate, if the matching degree of the simulated stress contour and the measured time-frequency characteristics is higher than a preset value, it is judged that the characteristics are effective, and if it is lower than the preset value, the feature extraction is re-performed through the time domain convolution module.

[0024] As an improvement of the above technical solution, the generative adversarial network model comprises a generator module and a discriminator module.

[0025] The generator module is used to generate a crack distribution map according to the trajectory of the main frequency changing over time and random noise, and the discriminator module is used to compare the input crack distribution map with the simulated stress contour to judge the authenticity probability of the simulated stress contour image.

[0026] As an improvement of the above technical solution, the judging of the authenticity of the image comprises the following steps:

[0027] If the authenticity probability of the image is greater than or equal to a preset threshold after comparing the crack distribution map with the simulated stress contour, it is directly output, and if the authenticity probability of the image is lower than the preset threshold after comparing the crack distribution map with the simulated stress contour, it is considered that there is an error in the simulation result.

[0028] The beneficial effects of the present application are:

[0029] By collecting the vibration information of the equipment and performing frequency band analysis on the vibration information, the equipment material can be simulated and simulated, the crack and other conditions can be judged, and the use of the equipment can be analyzed. According to this information, preparations can be made in advance before the equipment is damaged, and economic losses caused by failure to handle in time can be avoided. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The present application is a principle block diagram. DETAILED DESCRIPTION

[0031] Following, specific, concrete examples will be used to illustrate the embodiments of the application. Those skilled in the art will easily understand other advantages and purposes of the application from the description of the application. The application can also be implemented or applied in other different specific embodiments, and various modifications or changes can be made to the details in the description based on different views and applications without departing from the spirit of the application.

[0032] The existing supervision system used in industrial production manufacturing usually judges whether the device in the Internet of Things is in an abnormal working state according to the service life, current, voltage and other working conditions of the device. Although this supervision method can accurately know the position of the abnormal state, it cannot prevent the failure of the device in advance, because the aging of the device does not cause the device to be unusable, but when the load of the aging device is too large, direct failure problems may occur. Such failure problems cannot be handled by detecting current, voltage and other conditions.

[0033] In order to solve the above problems, please refer to the figure, provide a kind of based on the Internet of Things industrial production manufacturing intelligent supervision system, including: data collection module, feature extraction module, finite element simulation module and generative adversarial network model.

[0034] The data collection module is used to collect the vibration information of the device, the feature extraction module is used to perform wavelet packet decomposition on the vibration information to obtain the energy distribution of each frequency band, and the trajectory of the dominant frequency changing with time is obtained from the abnormal frequency band. The finite element simulation module is used to simulate the stress nephogram according to the input energy distribution and the trajectory of the dominant frequency changing with time. The generative adversarial network model is used to generate a crack distribution map according to the trajectory of the dominant frequency changing with time and random noise, compare the crack distribution map with the simulated stress nephogram to judge the authenticity of the image, and analyze the stress nephogram and the crack distribution map to judge the failure condition of the device.

[0035] First, the data collection module collects the vibration information of the device, which includes at least the vibration signal and stress wave information on the surface of the device. The vibration signal on the surface of the device refers to the vibration state generated by the device during operation. The related information of these vibration states is usually collected by the laser Doppler vibration measurement module under the data collection module. The vibration speed on the surface of the device is measured by laser interference. The stress wave information refers to the internal crack condition of the manufacturing device. The internal crack condition cannot be measured by laser interference, so the acoustic emission signal processing method is used to measure, i.e. AE detection device, which detects the stress wave emitted from the internal structure of the stressed material to judge the damage degree of the internal structure. In this way, the internal crack condition of the material can be detected.

[0036] After the collection of vibration information is completed, feature extraction needs to be performed on the vibration information. In this step, we usually use a feature extraction module. In this embodiment, a time domain convolution network module is used, which is based on a TCN model. By adjusting the dilation coefficient, the receptive field of the convolution kernel can be expanded, thereby more effectively processing long-distance dependent features. Therefore, by using the TCN model to extract features from the vibration information, long-time sequence data can be effectively processed while maintaining the time sequence.

[0037] In consideration of the possibility that the generated output image of the crack condition is a grayscale image, when performing feature extraction, wavelet packet decomposition can be used to process the vibration information. The wavelet packet decomposition uses a Morlet wavelet decomposition method to decompose the vibration information into a fourth-order wavelet, generate 16 sub-bands, and calculate the energy entropy of each sub-band.

[0038] Based on the above analysis, it can be known that the method of obtaining the trajectory of the dominant frequency changing with time from the abnormal frequency band can adopt the following steps:

[0039] The input vibration information is decomposed into 16 sub-bands by the time domain convolution module, and the energy entropy of each frequency band is calculated. The abnormal frequency band is located according to the energy entropy.

[0040] The trajectory of the dominant frequency changing with time is extracted from the abnormal frequency band, and the instantaneous impact feature is captured.

[0041] For the TCN model, a neural network module needs to be established first, and the input vibration signal is extracted by the neural network module. After the input vibration information is decomposed into 16 sub-bands and the energy entropy is calculated, it is input into the neural network module. The structure of the neural network module is as follows:

[0042] class DilatedConv1d(nn.Module):

[0043] def __init__(self, in_channels=64, out_channels=64, dilation=1):

[0044] super().__init__()

[0045] self.conv = nn.Conv1d(

[0046] in_channels,

[0047] out_channels,

[0048] kernel_size=3,

[0049] dilation = dilation, # dilation factor (2^i)

[0050] padding = dilation # keep the timing length unchanged )

[0052] self.relu = nn.ReLU()

[0053] def forward(self, x):

[0054] return self.relu(self.conv(x))

[0055] The line class DilatedConv1d(nn.Module) defines a new class DilatedConv1d that inherits from nn.Module, indicating that this is a custom PyTorch neural network module. nn.Module is the base class for all neural network layers and models in PyTorch.

[0056] def __init__(self, in_channels=64, out_channels=64, dilation=1), which represents the constructor of the class and initializes the parameters of this convolutional module.

[0057] super().__init__(), which calls the constructor of the parent class nn.Module to ensure that the class inherited from nn.Module is correctly initialized.

[0058] self.conv = nn.Conv1d(...), which defines a 1D convolutional layer, namely nn.Conv1d.

[0059] self.relu = nn.ReLU(), which defines a ReLU activation function (nn.ReLU()). ReLU is a commonly used activation function that sets negative values to zero and keeps positive values unchanged. ReLU is commonly used to improve the non-linear representation ability of neural networks.

[0060] def forward(self, x): This is the forward propagation function of the model. The forward function in PyTorch defines how data flows through the network module.

[0061] return self.relu(self.conv(x)), which means that data is processed through the following two steps:

[0062] 1. First, the input x is passed to the convolutional layer self.conv(x), which performs a 1D dilated convolution operation and returns the output after convolution.

[0063] 2. Then, the output of the convolution is processed non-linearly using the ReLU activation function `self.relu()`. Finally, the result after convolution and activation is returned.

[0064] The data output by the above architecture is usually an important feature of the vibration signal in the time domain. In this embodiment, after transforming the vibration information into sub-bands using multi-Besch wavelet decomposition, it is necessary to calculate the time-frequency distribution information during the transformation process. After integrating this time-frequency distribution information, a time-frequency matrix is ​​generated, which depends on the following formula:

[0065]

[0066] in, This represents the original vibration information input. For scale parameters, For translation parameters, For wavelet basis functions, This is the scaling factor.

[0067] After generating the time-frequency matrix, the temporal convolutional neural network module iterates through the time points to find the maximum energy frequency at each time point and records the time-frequency coordinates of that energy frequency. The connection of these time-frequency coordinates is the time-frequency ridge, which is the trajectory of the dominant frequency changing over time.

[0068] Extracting time-frequency ridges typically relies on the following code:

[0069] def extract_ridge(tf_matrix):

[0070] ridges = []

[0071] for t in range(tf_matrix.shape[1]):

[0072] # Traverse time points

[0073] freq_idx = np.argmax(tf_matrix[:, t]) # Find the maximum energy frequency at the current moment.

[0074] ridges.append((t, freq_idx)) # Record time-frequency coordinates

[0075] return ridges

[0076] ridges = [], represents the initialization of the ridge line list, through the "for t in range(tf_matrix.shape[1]):" code to traverse the time points, and through "freq_idx = np.argmax(tf_matrix[:, t])" to find the maximum energy frequency at this time point, and through "ridges.append((t, freq_idx))" to record the time-frequency coordinates of the current time point, and the record position is the ridge line list ridges, when all the time points are recorded, the connection of the obtained time-frequency coordinates is the time-frequency ridge line.

[0077] After obtaining the time-frequency ridge line, the stress contour can be simulated and generated through finite element simulation analysis, specifically, the stress contour simulation method of the finite element simulation module includes the following steps:

[0078] Input the time-frequency ridge line into the finite element simulation module to obtain the corresponding dynamic load, extract the frequency and energy characteristics of the dynamic load, and obtain the transient load boundary condition according to the frequency and energy characteristics.

[0079] The above steps depend on the following code:

[0080] def prepare_fem_input(ridges):

[0081] freq = [f for (t, f) in ridges]

[0082] energy = compute_energy(ridges)

[0083] load_params = {

[0084] 'frequency': np.mean(freq),

[0085] 'amplitude': energy_to_force(energy),

[0086] 'duration': len(ridges) * time_step

[0087] }

[0088] return load_params

[0089] The corresponding data parameters are extracted from the ridge line list, which usually contains frequency and energy characteristics. Specifically, freq = [f for (t, f) in ridges] represents extracting the frequency characteristics of each ridge line and generating a frequency list freq, where each element in the table represents the maximum energy frequency corresponding to a certain time point. energy = compute_energy(ridges) represents calculating the energy characteristics of the vibration signal by calling the compute_energy function, and calculating the overall energy according to the distribution of ridges in time and frequency. 'frequency': np.mean(freq) represents calculating the average of all frequencies in the frequency list to represent the frequency of the applied load. 'amplitude': energy_to_force(energy) represents converting the calculated energy into the amplitude of the applied force. 'duration': len(ridges) * time_step represents estimating the duration of the applied load by multiplying the number of ridges by the time step. Finally, these three parameters are integrated as the return of the transient load boundary conditions.

[0090] On the basis of the above steps, the length of crack growth is further estimated, and the length of crack growth needs to consider the relationship between frequency and dynamic load, so it is also necessary to obtain the time-frequency ridge slope according to the trajectory of the dominant frequency changing with time, and then estimate the crack growth rate according to the ridge slope.

[0091] Specifically, the relationship between frequency and dynamic load is as follows:

[0092] wherein, represents the value of dynamic load at time t, represents the amplitude of dynamic load, which is obtained by back calculation from frequency band energy entropy, represents the dominant frequency extracted from the time-frequency ridge, which represents the most significant frequency component in the signal, is a time variable, represents a sinusoidal function, which represents the dynamic load changing with time.

[0093] The calculation of crack propagation rate depends on the following formula:

[0094]

[0095] wherein, represents the rate of crack propagation, C is a material constant, which is determined by experiment and is usually related to the fatigue characteristics of the material, K is the stress intensity factor amplitude, which is the stress level at the crack tip, and is related to the external load and the crack size. When the load increases, consequently, driving the crack propagation, C is a material parameter, usually an exponent, indicating the sensitivity of the crack propagation rate to the stress intensity factor.

[0096] After obtaining the transient load boundary conditions and the crack growth rate, the stress cloud map can be simulated according to these parameters. If the matching degree of the simulated stress cloud map and the measured time-frequency characteristics is higher than the preset value, the characteristics are considered valid. If it is lower than the preset value, the feature extraction is re-performed through the time domain convolution module.

[0097] The preset value can usually be set to 5%-10%. When the error is less than the preset value, it is considered that the simulation result generated according to the characteristics is accurate, that is, the characteristics are valid. If the error is higher than the preset value, it is considered that the simulation result generated according to the characteristics has a large error, that is, the validity of the characteristics is questionable. At this time, the feature extraction can be re-performed through the time domain convolution module, and when re-extracting, the number of layers of wavelet packet decomposition needs to be increased, that is, the decomposition wavelet order can be set to 6, and then the calculation in the above steps is performed again.

[0098] Of course, after the simulation image is generated, the visualization of stress and crack also needs to be performed. In this embodiment, a generative adversarial network model is used for processing, wherein the generative adversarial network model includes a generator module and a discriminator module. The generator module is used to generate a crack distribution map according to the trajectory of the dominant frequency changing over time and random noise, and the discriminator module is used to compare the input crack distribution map and the simulated stress cloud map to judge the authenticity probability of the simulated stress cloud map image.

[0099] Specifically, the authenticity of the image includes the following steps:

[0100] If the authenticity probability of the image is greater than or equal to the preset threshold value after comparing the crack distribution map with the simulated stress cloud map, it is directly output. If the authenticity probability of the image is lower than the preset threshold value after comparing the crack distribution map with the simulated stress cloud map, it is considered that the simulation result has an error.

[0101] The generator module and the discriminator module in the above scheme are further implemented as follows:

[0102] In order to enable the generator to receive the generated characteristics, the extracted feature information also needs to be converted into a 128-dimensional time sequence feature vector through the following code, as follows:

[0103] class TCN(nn.Module):

[0104] def forward(self, x):

[0105] # input x: (batch_size, seq_len, input_dim)

[0106] # output: (batch_size, 128)

[0107] return self.tcn_layers(x)[:, -1, :]# take the last state of the sequence

[0108] The structure of the generator module is set up as follows:

[0109] class Generator(nn.Module):

[0110] def __init__(self):

[0111] super().__init__()

[0112] self.fc = nn.Linear(160, 256*8*8)

[0113] self.upconv = nn.Sequential(

[0114] nn.ConvTranspose2d(256, 128, 4, stride=2, padding=1),

[0115] nn.BatchNorm2d(128),

[0116] nn.ReLU(),

[0117] nn.ConvTranspose2d(128, 64, 4, stride=2, padding=1),

[0118] nn.BatchNorm2d(64),

[0119] nn.ReLU(),

[0120] nn.ConvTranspose2d(64, 1, 4, stride=2, padding=1),

[0121] nn.Tanh() )

[0123] def forward(self, z, features):

[0124] x = torch.cat([z, features], dim=1)

[0125] x = self.fc(x).view(-1, 256, 8, 8)

[0126] return self.upconv(x)

[0127] The function principle of the above code is as follows:

[0128] Define a class named Generator, which inherits from nn.Module, which is the basic class of all neural network modules in PyTorch.

[0129] In the __init__ method, call the constructor of the parent class super().__init__().

[0130] self.fc=nn.Linear(160,256*8*8), which represents using a fully connected layer Linear to combine the input feature vector and random noise (128-dimensional feature vector + 32-dimensional noise), with a total of 160 dimensions, and the output size is 256x8x8. The role of this layer is to expand the input into a higher-dimensional feature map for subsequent deconvolution processing.

[0131] self.upconv=nn.Sequential(……), which represents using nn.Sequential to define a series of deconvolution layers (transposed convolution layers) to upsample smaller feature maps to generate larger images. The role of these layers is to gradually expand the feature map from 8x8 to 64x64.

[0132] The forward method defines the forward propagation process of the generator:

[0133] Input: Receive a random noise vector z and a feature vector features.

[0134] Concatenate: Use torch.cat to concatenate the noise z and the feature vector in the channel dimension (dim=1) to form a new input tensor.

[0135] Fully connected layer processing: Process the concatenated tensor through the fully connected layer self.fc, and then use.view(-1, 256, 8, 8) to reshape it into 256 feature maps with a size of 8x8, preparing for the subsequent deconvolution process.

[0136] Output crack distribution image: input the feature map into self.upconv to generate the final crack distribution image.

[0137] The structure of the discriminator module is as follows:

[0138] class Generator(nn.Module):

[0139] def __init__(self):

[0140] super().__init__()

[0141] self.fc = nn.Linear(160, 256*8*8)

[0142] self.upconv = nn.Sequential(

[0143] nn.ConvTranspose2d(256, 128, 4, stride=2, padding=1),

[0144] nn.BatchNorm2d(128),

[0145] nn.ReLU(),

[0146] nn.ConvTranspose2d(128, 64, 4, stride=2, padding=1),

[0147] nn.BatchNorm2d(64),

[0148] nn.ReLU(),

[0149] nn.ConvTranspose2d(64, 1, 4, stride=2, padding=1),

[0150] nn.Tanh() )

[0152] def forward(self, z, features):

[0153] x = torch.cat([z, features], dim=1)

[0154] x = self.fc(x).view(-1, 256, 8, 8)

[0155] return self.upconv(x)

[0156] The functional principle of the above code is as follows:

[0157] Define a class named Discriminator, which inherits from nn.Module, which is the base class of all neural network modules in PyTorch. In the __init__ method, initialize and call the constructor method of the parent class super().__init__().

[0158] A series of convolutional layers, activation functions, batch normalization, and fully connected layers are defined using nn.Sequential to form a sequential network structure. Specifically, convolutional layers (Conv2d), activation functions (LeakyReLU), batch normalization (BatchNorm2d), flattening (Flatten), and fully connected layers (Linear).

[0159] The forward method defines how to perform forward propagation:

[0160] Accept two inputs: img (crack distribution map) and stress_map (stress cloud map).

[0161] Use torch.cat to concatenate the two images in the channel dimension to form a tensor with shape (B, 2, H, W), where B is the batch size, H and W are the height and width of the image.

[0162] Input the concatenated tensor into the defined convolutional network downconv, and limit the output to between 0 and 1 using the torch.sigmoid function, representing the probability of the input image being real. Usually, a threshold a is set, when the output result is between a-1, it is considered that the crack distribution map is close to the real situation, if the output result is between 0-a, it is considered that the image has a large error with the real situation.

[0163] Finally, the generated crack distribution map is the image with strong authenticity identified, combined with the simulation results of the stress cloud map to judge the fault condition of the equipment. When the equipment is at the critical point of failure, an alarm can be sent in advance to remind the construction personnel to shut down the corresponding equipment and replace it. This monitoring scheme can achieve early warning action before the equipment is damaged, and compared with traditional current, voltage, and heat monitoring, it can play a better preventive role. Of course, in this system, traditional detection systems can still be integrated for unified coordination and scheduling to achieve a more perfect intelligent monitoring system for production and manufacturing.

[0164] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the present application. Any person skilled in the art can make modifications or changes to the above examples without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed by the present application should be covered by the claims of the present application.

Claims

1. An industrial production manufacturing intelligent supervision system based on Internet of Things, characterized in that, The method comprises the following steps: A data collection module is configured to collect vibration information of the equipment, wherein the vibration information comprises at least vibration signals and stress wave information of the surface of the equipment; A feature extraction module is configured to perform wavelet packet decomposition on the vibration information to obtain energy distribution of each frequency band, and to obtain a trajectory of a dominant frequency varying with time from an abnormal frequency band; A finite element simulation module is configured to simulate a stress cloud map according to the input energy distribution and the trajectory of the dominant frequency varying with time; A generative adversarial network model is configured to generate a crack distribution map according to the trajectory of the dominant frequency varying with time and random noise, to compare the crack distribution map with the simulated stress cloud map to determine the authenticity of the image, and to analyze the stress cloud map and the crack distribution map to determine the fault condition of the equipment; The feature extraction module comprises at least a time domain convolution network module; The method of obtaining the trajectory of the dominant frequency varying with time from the abnormal frequency band comprises the following steps: The input vibration information is decomposed into 16 sub-bands by the time domain convolution module, and the energy entropy of each frequency band is calculated, and the abnormal frequency band is located according to the energy entropy; The trajectory of the dominant frequency varying with time is extracted from the abnormal frequency band to capture the instantaneous impact characteristics; The wavelet packet decomposition adopts a multi-Bézier wavelet decomposition mode to decompose the vibration information into a fourth-order wavelet to generate 16 sub-bands, and to calculate the energy entropy of each sub-band; The method of extracting the trajectory of the dominant frequency varying with time from the abnormal frequency band comprises the following steps: The vibration information is transformed into sub-bands by the multi-Bézier wavelet decomposition mode, and the time-frequency distribution is calculated to generate a time-frequency matrix; The maximum energy frequency at each time point is found, the time-frequency coordinates of the energy frequency are recorded, and the time-frequency ridge line of the energy frequency is output as the extracted trajectory of the dominant frequency varying with time; The generation of the time-frequency matrix depends on the following formula: wherein, represents the input raw vibration information, is a scale parameter, is a translation parameter, is a wavelet basis function, is a scaling factor; The method of simulating the stress cloud map by the finite element simulation module comprises the following steps: The trajectory of the dominant frequency varying with time is input into the finite element simulation module to obtain corresponding dynamic load, the frequency and energy characteristics of the dynamic load are extracted, and the transient load boundary condition is obtained according to the frequency and energy characteristics; The time-frequency ridge line slope is obtained according to the trajectory of the dominant frequency varying with time, and the crack growth rate is estimated according to the ridge line slope; The stress cloud map is simulated according to the transient load boundary condition and the crack growth rate, if the matching degree of the simulated stress cloud map and the measured time-frequency characteristics is higher than a preset value, it is determined that the feature is valid, if the matching degree is lower than the preset value, the feature extraction is performed again by the time domain convolution module; The relationship between the frequency and the dynamic load is as follows: wherein represents the value of the dynamic load at time t, represents the amplitude of the dynamic load, represents the dominant frequency extracted from the time-frequency ridge, is a time variable, represents a sinusoidal function; The calculation of the crack propagation rate depends on the following formula: wherein, K is the rate of crack propagation, C is a material constant determined experimentally and is usually related to the fatigue properties of the material, K is the stress intensity factor amplitude, which is the stress level at the crack tip and is related to the external load and the crack size, C is a material parameter. 2.The industrial production and manufacturing intelligent supervision system based on the Internet of Things according to claim 1, characterized in that: The data collection module comprises at least a laser Doppler vibration measurement module and an acoustic emission signal processing module; The laser Doppler vibration measurement module is configured to test the vibration signals of the surface of the equipment, and the acoustic emission signal processing module is configured to capture the stress wave information released by the internal cracks of the material. 3.The industrial production and manufacturing intelligent supervision system based on the Internet of Things according to claim 1, characterized in that: The generative adversarial network model comprises a generator module and a discriminator module; The generator module is used for generating a crack distribution map according to a time-varying trajectory of a main frequency and random noise, and the discriminator module is used for comparing the input crack distribution map with a simulated stress cloud map to judge a probability of authenticity of the simulated stress cloud map image.

4. The intelligent monitoring system for industrial production and manufacturing based on the Internet of Things according to claim 3, characterized in that: The judging of the authenticity of the image comprises the following steps: If the authenticity probability of the image is greater than or equal to a preset threshold after the comparison between the crack distribution map and the simulated stress cloud map, the image is directly output; if the authenticity probability of the image is lower than the preset threshold after the comparison between the crack distribution map and the simulated stress cloud map, it is considered that there is an error in the simulation result.

Images