Fault prediction management method and system for inorganic board production line based on data analysis

Abstract

The present invention relates to the technical field of fault monitoring, and discloses a fault prediction management method and system for an inorganic board production line based on data analysis, wherein a fixed frame is installed above the discharge end of the production line, a high-definition camera is arranged at the center of the lower side of the fixed frame, and a rotating lamp stand for evenly emitting white light is rotatably connected to the outer side of the upper end of the high-definition camera, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end; when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the rotating lamp stand rotates around the high-definition camera at a uniform speed within a fixed time period, and the high-definition camera shoots the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards; timestamps and angle stamps are annotated for the multi-angle high-definition images of each inorganic board; the multi-angle high-definition images of each inorganic board are classified and sorted according to the timestamps and angle stamps to obtain a data set.

Description

Technical Field

[0001] The present invention relates to the field of fault monitoring, and more specifically, to a method and system for predicting and managing faults of an inorganic board production line based on data analysis. Background Art

[0002] At present, the problem of rough or defective board surface in the pre-coating production line of inorganic boards is still a key problem in the production process. If there is a problem in the production line, it may cause the surface roughness of the board to increase, or in more serious cases, obvious surface damage. For malignant damage, such as cracks, peeling, etc.

[0003] Existing monitoring methods, such as manual inspection or automated inspection equipment, can easily identify and eliminate malignant damage. However, the subtle roughness changes on the transparent colloid surface of inorganic boards are often difficult to detect effectively by naked eyes or conventional image analysis methods because the surface is smooth and patternless. This makes it difficult to detect and predict faults in the inorganic board production line in a timely manner, affecting production quality and economic benefits. Summary of the invention

[0004] The present invention provides an inorganic board production line fault prediction management method and system based on data analysis, which solves the technical problems raised in the background technology.

[0005] First, the fault prediction management method of inorganic board production line based on data analysis includes:

[0006] Step 101, a fixing frame is installed above the discharge end of the production line, a high-definition camera is arranged at the center of the lower side of the fixing frame, a lamp for evenly emitting white light is rotatably connected to the outer side of the upper end of the high-definition camera, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end;

[0007] Step 102, when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a constant speed within a fixed time period, and the high-definition camera photographs the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards;

[0008] Step 103: annotate the multi-angle high-definition image of each inorganic board with a timestamp and an angle stamp; classify and sort the multi-angle high-definition image of each inorganic board according to the timestamp and the angle stamp to obtain a data set. ,in Represents k time series images of the nth angle;

[0009] Step 104, preprocessing the data set to obtain a preprocessed data set; inputting the data of each angle in the preprocessed data set into a preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and determining whether the output of the detection model is credible through angle symmetry;

[0010] Step 105, during the pre-coating operation period of the M inorganic boards, collect power data of the circuit of the inorganic board production line at fixed time intervals, and monitor the power data;

[0011] Step 106, if the output of the detection model is credible, the output of the detection model is adopted, and fault prediction is performed on the inorganic board production line based on the credible output of the detection model and the monitoring results of the electric energy data; if the output of the detection model is unreliable, the detection model is reversely optimized based on the output of the detection model.

[0012] Furthermore, the dataset is preprocessed, including:

[0013] Crop and align the images in each data set in the dataset to obtain a standard image dataset;

[0014] The images in each data set in the standard image dataset are grayed out, and the grayed out value of each pixel in the image is normalized to obtain a preprocessed dataset.

[0015] Furthermore, the preset detection model includes a shadow extraction layer, a feature construction layer, and a prediction layer;

[0016] The shadow extraction layer includes N feature modules. The receptive field of the feature module increases linearly. The convolution kernel size of the feature module increases linearly. The N feature modules are processed by the preprocessed data set. Perform convolution operation on the image in, extract features layer by layer, and obtain k boundary feature matrices;

[0017] The feature construction layer calculates the Euclidean distance of adjacent boundary feature matrices to obtain k-1 Euclidean distance values, and constructs a feature sequence based on the k-1 Euclidean distance values ​​in chronological order;

[0018] The prediction layer is used to input the feature sequence, and the prediction layer outputs the hidden state. The hidden state is used to obtain the roughness of the i-th angle of the M+1-th inorganic board through the classifier.

[0019] Furthermore, the calculation formula of the prediction layer is as follows:

[0020] ;

[0021] in, represents the feature sequence of the feature construction layer input of the u-th unit of the prediction layer, represents the hidden state of the output of the u-th unit in the prediction layer, and They represent the update gate, reset gate and candidate hidden state of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the update gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the reset gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the u-th unit in the prediction layer, respectively. represents the hidden state of the output of the u-1th unit in the prediction layer, Assign a value of 0, represents point-by-point multiplication, sigmoid represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function.

[0022] Furthermore, the training samples of the detection model include:

[0023] Collect multi-angle high-definition images of P historical inorganic boards, and use the data set constructed by the multi-angle high-definition images of the 1st to the i-th inorganic boards as the training data of the training samples of the detection model. The roughness manually evaluated for the i+1th inorganic board is used as the sample label of the training sample.

[0024] Furthermore, the angular symmetry is used to determine whether the output of the detection model is credible, including:

[0025] The detection model is used to group the roughness of the n angles of the M+1th inorganic board into two groups, and the Group validation sets and label each validation set; the sum of the angles in each group is a multiple of 180 degrees;

[0026] based on The group validation set constructs a symmetric error matrix: ;in, Indicates that the first and The absolute difference in roughness of the group validation set is used to construct the Laplace matrix based on the symmetric error matrix: ;in, represents the Laplacian matrix; represents the degree matrix, ; represents the element in the i-th row and i-th column of the degree matrix, represents the sum of all elements in the i-th row of the matrix S;

[0027] The eigenvalue of the Laplace matrix is ​​obtained by QR decomposition; if the eigenvalue of the Laplace matrix is ​​judged to be ≤ a preset characteristic threshold, the output of the detection model is credible, otherwise the output of the detection model is not credible.

[0028] Furthermore, the electric energy data is monitored, including:

[0029] Electric energy data includes current data, voltage data and harmonic data;

[0030] Calculate the average current and average voltage during the pre-coating operation period of M inorganic boards;

[0031] Calculate the current fluctuation rate and the voltage fluctuation rate based on the average current and the average voltage;

[0032] Calculate the total harmonic distortion of current and voltage based on the harmonic data;

[0033] Determine whether the current fluctuation rate, voltage fluctuation rate, current total harmonic distortion and voltage total harmonic distortion are greater than a preset fluctuation rate threshold range and distortion threshold range;

[0034] If the current fluctuation rate, the voltage fluctuation rate, the current total harmonic distortion and the voltage total harmonic distortion are greater than the preset fluctuation rate threshold range and the distortion threshold range, it is determined that the electric energy data is abnormal.

[0035] Furthermore, the detection model is reversely optimized according to the output of the detection model, including:

[0036] Setting the symmetric loss function ;in, represents the symmetric loss function, represents the loss weight coefficient, and They represent the two roughnesses of the symmetric angle of the M+1th inorganic plate;

[0037] The loss function constructed by combining the sample label of the training sample and the predicted roughness of the detection model with the symmetry loss function is obtained to obtain a composite loss function, and the composite loss function is used as the loss function of the detection model.

[0038] Furthermore, based on the reliable output and power data of the detection model, fault prediction is performed on the inorganic board production line;

[0039] According to the detection model, the maximum roughness of the M+1th inorganic board among the roughnesses of the M+1th inorganic board at n angles is obtained;

[0040] If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the power data is abnormal, it is determined that there is an electrical fault in the inorganic board production line;

[0041] If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the electric energy data is normal, it is determined that there is a mechanical failure in the inorganic board production line;

[0042] If the electric energy data is abnormal and the maximum roughness of the M+1th inorganic board is less than the preset roughness threshold, it is determined that there is fluctuation in the power grid system to which the inorganic board production line is connected, and the inorganic board production line has no faults.

[0043] Secondly, the inorganic board production line fault prediction management system based on data analysis includes:

[0044] The image acquisition module has a fixed frame installed above the discharge end of the production line, and a high-definition camera is set in the center of the lower side of the fixed frame. The upper outer side of the high-definition camera is connected to a lamp for evenly emitting white light, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end; when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a uniform speed within a fixed time period, and the high-definition camera shoots the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards;

[0045] The image classification module annotates the timestamp and angle stamp of each inorganic board's multi-angle HD image; classifies and sorts the multi-angle HD image of each inorganic board according to the timestamp and angle stamp to obtain a data set ,in Represents k time series images of the nth angle;

[0046] The prediction module preprocesses the data set to obtain a preprocessed data set; the data of each angle in the preprocessed data set are respectively input into the preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and the output of the detection model is determined to be credible through angle symmetry;

[0047] The power quality monitoring module collects power data of the circuit of the inorganic board production line at fixed time intervals during the pre-coating operation period of M inorganic boards, and monitors the power data;

[0048] The processing module adopts the output of the detection model if it is credible, and predicts the fault of the inorganic board production line based on the credible output of the detection model and the monitoring results of the electric energy data; if the output of the detection model is unreliable, the detection model is reversely optimized based on the output of the detection model.

[0049] The beneficial effects of the present invention are: by installing a high-definition camera, a rotating light stand and an infrared monitoring device, accurate multi-angle lighting monitoring of the inorganic board production process is achieved, and image data marked with timestamps and angle stamps are automatically classified, sorted and preprocessed. The roughness of the inorganic board is analyzed based on the detection model, and the credibility of the model results is judged by the angle symmetry. If it is not credible, the model is automatically optimized; if it is credible, the degree of fault of the production line is judged according to the roughness; the present invention improves the accuracy of automated monitoring of the production line and the effectiveness of fault prediction, ensuring the stability and efficiency of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the inorganic board production line fault prediction management method based on data analysis of the present invention;

[0051] Figure 2 It is a module diagram of the inorganic board production line fault prediction management system based on data analysis of the present invention;

[0052] Figure 3 It is a plan view of the device for collecting data on the inorganic plate at the discharge end of the present invention.

[0053] In the figure: 201, image acquisition module; 202, image classification module; 203, prediction module; 204, power quality monitoring module; 205, processing module. DETAILED DESCRIPTION

[0054] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0055] like Figure 1 to Figure 3 As shown, the inorganic board production line fault prediction management method based on data analysis includes:

[0056] Step 101, a fixing frame is installed above the discharge end of the production line, a high-definition camera is arranged at the center of the lower side of the fixing frame, a lamp for evenly emitting white light is rotatably connected to the outer side of the upper end of the high-definition camera, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end;

[0057] Step 102, when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a constant speed within a fixed time period, and the high-definition camera photographs the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards;

[0058] Step 103: annotate the multi-angle high-definition image of each inorganic board with a timestamp and an angle stamp; classify and sort the multi-angle high-definition image of each inorganic board according to the timestamp and the angle stamp to obtain a data set. ,in Represents k time series images of the nth angle;

[0059] Step 104, preprocessing the data set to obtain a preprocessed data set; inputting the data of each angle in the preprocessed data set into a preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and determining whether the output of the detection model is credible through angle symmetry;

[0060] Step 105, during the pre-coating operation period of the M inorganic boards, collect power data of the circuit of the inorganic board production line at fixed time intervals, and monitor the power data;

[0061] Step 106, if the output of the detection model is credible, the output of the detection model is adopted, and fault prediction is performed on the inorganic board production line based on the credible output of the detection model and the monitoring results of the electric energy data; if the output of the detection model is unreliable, the detection model is reversely optimized based on the output of the detection model.

[0062] It should be noted that when the inorganic board production line is in use, the processing error will gradually increase with the increase of operating time; when the processing error exceeds the qualified standard, it means that there is a fault in the production line. This fault can be reflected from the roughness of the pre-coated surface of the inorganic board. Then, by detecting the roughness of the pre-coated surface of the inorganic board, the fault prediction of the inorganic board production line can be realized.

[0063] The infrared monitoring device includes a light emitting end and a light receiving end. When the inorganic board arrives at the discharge port, the arrival of the inorganic board is judged through the signal feedback from the optical fiber receiving end, thereby starting the motor to drive the rotating light table to rotate around the high-definition camera at a constant speed; during the period of one rotation of the rotating light table, the high-definition camera collects pictures of the discharge port at several moments, and obtains pictures of multiple angles of light shining on the top of the inorganic board.

[0064] In one embodiment of the present invention, preprocessing the data set includes:

[0065] Crop and align the images in each data set in the dataset to obtain a standard image dataset;

[0066] The images in each data set in the standard image dataset are grayed out, and the grayed out value of each pixel in the image is normalized to obtain a preprocessed dataset.

[0067] In one embodiment of the present invention, the cropping and registration of images in each data set in the data set to obtain a standard image data set can be implemented by using OpenCV tools.

[0068] In one embodiment of the present invention, the preset detection model includes a shadow extraction layer, a feature construction layer and a prediction layer;

[0069] The shadow extraction layer includes N feature modules. The receptive field of the feature module increases linearly. The convolution kernel size of the feature module increases linearly. The N feature modules are processed by the preprocessed data set. Perform convolution operation on the image in, extract features layer by layer, and obtain k boundary feature matrices;

[0070] The feature construction layer calculates the Euclidean distance of adjacent boundary feature matrices to obtain k-1 Euclidean distance values, and constructs a feature sequence based on the k-1 Euclidean distance values ​​in chronological order;

[0071] The prediction layer is used to input the feature sequence, and the prediction layer outputs the hidden state. The hidden state is used to obtain the roughness of the i-th angle of the M+1-th inorganic board through the classifier.

[0072] In one embodiment of the present invention, the calculation formula of the shadow extraction layer is as follows:

[0073] The receptive field size of the i-th feature module The formula is: ;in, represents the receptive field size of the first feature module, Indicates the increased receptive field size of each feature module;

[0074] The convolution kernel size of the i-th feature module The formula is: ;in, represents the convolution kernel size of the first feature module, Indicates the convolution kernel size added to each feature module;

[0075] The convolution operation formula of the i-th feature module is: ; ;in, Represents any image in the preprocessed dataset. represents the feature map extracted by the i-th feature module, represents the feature map extracted by the i-1th feature module, represents the convolution operation, Represents the bias term of the i-th feature module; express Activation function;

[0076] Output of the shadow extraction layer It is expressed as: ;in, Represents the weight of the feature map extracted by the i-th feature module.

[0077] In one embodiment of the present invention, the calculation formula of the prediction layer is as follows:

[0078] ;

[0079] in, represents the feature sequence of the feature construction layer input of the u-th unit of the prediction layer, represents the hidden state of the output of the u-th unit in the prediction layer, and They represent the update gate, reset gate and candidate hidden state of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the update gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the reset gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the u-th unit in the prediction layer, respectively. represents the hidden state of the output of the u-1th unit in the prediction layer, Assign a value of 0, represents point-by-point multiplication, sigmoid represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function.

[0080] In one embodiment of the present invention, the training samples of the detection model include:

[0081] Collect multi-angle high-definition images of P historical inorganic boards, and use the data set constructed by the multi-angle high-definition images of the 1st to the i-th inorganic boards as the training data of the training samples of the detection model. The roughness manually evaluated for the i+1th inorganic board is used as the sample label of the training sample.

[0082] In one embodiment of the present invention, determining whether the output of the detection model is credible through angle symmetry includes:

[0083] The detection model is used to group the roughness of the n angles of the M+1th inorganic board into two groups, and the Group validation sets and label each validation set; the sum of the angles in each group is a multiple of 180 degrees;

[0084] based on The group validation set constructs a symmetric error matrix: ;in, Indicates that the first and The absolute difference in roughness of the group validation set is used to construct the Laplace matrix based on the symmetric error matrix: ;in, represents the Laplacian matrix; represents the degree matrix, ; represents the element in the i-th row and i-th column of the degree matrix, represents the sum of all elements in the i-th row of the matrix S;

[0085] The eigenvalue of the Laplace matrix is ​​obtained by QR decomposition; if the eigenvalue of the Laplace matrix is ​​judged to be ≤ a preset characteristic threshold, the output of the detection model is credible, otherwise the output of the detection model is not credible.

[0086] In one embodiment of the present invention, monitoring electric energy data includes:

[0087] Electric energy data includes current data, voltage data and harmonic data;

[0088] Calculate the average current and average voltage during the pre-coating operation period of M inorganic boards;

[0089] Calculate the current fluctuation rate and the voltage fluctuation rate based on the average current and the average voltage;

[0090] Calculate the total harmonic distortion of current and voltage based on the harmonic data;

[0091] Determine whether the current fluctuation rate, voltage fluctuation rate, current total harmonic distortion and voltage total harmonic distortion are greater than a preset fluctuation rate threshold range and distortion threshold range;

[0092] If the current fluctuation rate, the voltage fluctuation rate, the current total harmonic distortion and the voltage total harmonic distortion are greater than the preset fluctuation rate threshold range and the distortion threshold range, it is determined that the electric energy data is abnormal.

[0093] In one embodiment of the present invention, by Get the average voltage ;in, Indicates that M inorganic boards are pre-coated at fixed time intervals during the period of operation. Sub voltage data; express The index of the sub-voltage data; Indicates Sub voltage data;

[0094] pass Get the average power ;in, Indicates that M inorganic boards are pre-coated at fixed time intervals during the period of operation. Secondary current data; express Index of secondary current data; Indicates Secondary current data;

[0095] pass Get the current fluctuation rate ;in, express The maximum value among the sub-voltage data; express The minimum value in the sub-voltage data;

[0096] pass Get the current fluctuation rate ;in, express The maximum value in the sub-current data; express The minimum value in the sub-current data;

[0097] pass Get the voltage total harmonic distortion ;

[0098] pass Get the total harmonic distortion of current ;

[0099] It should be noted that the voltage total harmonic distortion Should be kept below 5%; voltage total harmonic distortion Should be kept below 10%.

[0100] In one embodiment of the present invention, the detection model is reversely optimized according to the output of the detection model, including:

[0101] Setting the symmetric loss function ;in, represents the symmetric loss function, represents the loss weight coefficient, and They represent the two roughnesses of the symmetric angle of the M+1th inorganic plate;

[0102] The loss function constructed by combining the sample label of the training sample and the predicted roughness of the detection model with the symmetry loss function is obtained to obtain a composite loss function, and the composite loss function is used as the loss function of the detection model.

[0103] Fault prediction for inorganic board production line based on reliable output and power data of detection model;

[0104] According to the detection model, the maximum roughness of the M+1th inorganic board among the roughnesses of the M+1th inorganic board at n angles is obtained;

[0105] If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the power data is abnormal, it is determined that there is an electrical fault in the inorganic board production line;

[0106] If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the electric energy data is normal, it is determined that there is a mechanical failure in the inorganic board production line;

[0107] If the electric energy data is abnormal and the maximum roughness of the M+1th inorganic board is less than the preset roughness threshold, it is determined that there is fluctuation in the power grid system to which the inorganic board production line is connected, and the inorganic board production line has no faults.

[0108] The inorganic board production line fault prediction management system based on data analysis is applied to the method described, including:

[0109] The image acquisition module 201 has a fixed frame installed above the discharge end of the production line, a high-definition camera is arranged at the center of the lower side of the fixed frame, and a lamp for evenly emitting white light is rotatably connected to the outer side of the upper end of the high-definition camera, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end; when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a uniform speed within a fixed time period, and the high-definition camera shoots the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards;

[0110] The image classification module 202 annotates the multi-angle high-definition image of each inorganic board with a timestamp and an angle stamp; classifies and sorts the multi-angle high-definition image of each inorganic board according to the timestamp and the angle stamp to obtain a data set. ,in Represents k time series images of the nth angle;

[0111] Prediction module 203 pre-processes the data set to obtain a pre-processed data set; inputs the data of each angle in the pre-processed data set into a preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and determines whether the output of the detection model is credible through angle symmetry;

[0112] The power quality monitoring module 204 collects power data of the circuit of the inorganic board production line at fixed time intervals during the pre-coating operation period of the M inorganic boards, and monitors the power data;

[0113] Processing module 205, if the output of the detection model is credible, the output of the detection model is adopted, and fault prediction is performed on the inorganic board production line based on the credible output of the detection model and the monitoring results of the electric energy data; if the output of the detection model is unreliable, the detection model is reversely optimized based on the output of the detection model.

[0114] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A fault prediction management method for an inorganic board production line based on data analysis, characterized in that: include: Step 101, a fixing frame is installed above the discharge end of the production line, a high-definition camera is arranged at the center of the lower side of the fixing frame, a lamp for evenly emitting white light is rotatably connected to the outer side of the upper end of the high-definition camera, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end; Step 102, when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a constant speed within a fixed time period, and the high-definition camera photographs the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards; Step 103: annotate the multi-angle high-definition image of each inorganic board with a timestamp and an angle stamp; classify and sort the multi-angle high-definition image of each inorganic board according to the timestamp and the angle stamp to obtain a data set. ,in Represents k time series images of the nth angle; Step 104, preprocessing the data set to obtain a preprocessed data set; inputting the data of each angle in the preprocessed data set into a preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and determining whether the output of the detection model is credible through angle symmetry; Step 105, during the pre-coating operation period of the M inorganic boards, collect power data of the circuit of the inorganic board production line at fixed time intervals, and monitor the power data; Step 106, if the output of the detection model is credible, the output of the detection model is adopted, and fault prediction is performed on the inorganic board production line based on the credible output of the detection model and the monitoring result of the electric energy data; If the output of the detection model is unreliable, the detection model is reversely optimized according to the output of the detection model; The angular symmetry is used to determine whether the output of the detection model is credible, including: The detection model is used to group the roughness of the n angles of the M+1th inorganic board into two groups, and the Group validation sets and label each validation set; the sum of the angles in each group is a multiple of 180 degrees; based on The group validation set constructs a symmetric error matrix: ;in, Indicates that the first and The absolute difference in roughness of the group validation set is used to construct the Laplace matrix based on the symmetric error matrix: ;in, represents the Laplacian matrix; represents the degree matrix, ; represents the element in the i-th row and i-th column of the degree matrix, represents the sum of all elements in the i-th row of the matrix S; The eigenvalue of the Laplace matrix is ​​obtained by QR decomposition; if the eigenvalue of the Laplace matrix is ​​judged to be ≤ a preset characteristic threshold, the output of the detection model is credible, otherwise the output of the detection model is not credible; The preset detection model includes a shadow extraction layer, a feature construction layer, and a prediction layer; The shadow extraction layer includes N feature modules. The receptive field of the feature module increases linearly. The convolution kernel size of the feature module increases linearly. The N feature modules are processed by the preprocessed data set. Perform convolution operation on the image in, extract features layer by layer, and obtain k boundary feature matrices; The feature construction layer calculates the Euclidean distance of adjacent boundary feature matrices to obtain k-1 Euclidean distance values, and constructs a feature sequence based on the k-1 Euclidean distance values ​​in chronological order; The prediction layer is used to input feature sequences, and the prediction layer outputs hidden states. The hidden states are used by the classifier to obtain the roughness of the i-th angle of the M+1-th inorganic board; The training samples of the detection model include: Collect multi-angle high-definition images of P inorganic boards in history, and use the data set constructed by the multi-angle high-definition images of the 1st to the i-th inorganic boards as the training data of the training samples of the detection model, and use the roughness of the i+1th inorganic board manually evaluated as the sample label of the training sample; Reverse optimization of the detection model based on the output of the detection model, including: Setting the symmetric loss function ;in, represents the symmetric loss function, represents the loss weight coefficient, and They represent the two roughnesses of the symmetric angle of the M+1th inorganic plate; The loss function constructed by combining the sample label of the training sample and the predicted roughness of the detection model with the symmetry loss function is obtained to obtain a composite loss function, and the composite loss function is used as the loss function of the detection model.

2. The inorganic board production line fault prediction management method based on data analysis according to claim 1 is characterized in that: Preprocess the dataset, including: Crop and align the images in each data set in the dataset to obtain a standard image dataset; The images in each data set in the standard image dataset are grayed out, and the grayed out value of each pixel in the image is normalized to obtain a preprocessed dataset.

3. The inorganic board production line fault prediction management method based on data analysis according to claim 2 is characterized in that: The calculation formula of the prediction layer is as follows: ； in, represents the feature sequence of the feature construction layer input of the u-th unit of the prediction layer, represents the hidden state of the output of the u-th unit in the prediction layer, and They represent the update gate, reset gate and candidate hidden state of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the update gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the reset gate of the u-th unit in the prediction layer, respectively. and They represent the first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the u-th unit in the prediction layer, respectively. represents the hidden state of the output of the u-1th unit in the prediction layer, Assign a value of 0, represents point-wise multiplication, sigmoid represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function.

4. The inorganic board production line fault prediction management method based on data analysis according to claim 1 is characterized in that: Monitor power data, including: Electric energy data includes current data, voltage data and harmonic data; Calculate the average current and average voltage during the pre-coating operation period of M inorganic boards; Calculate the current fluctuation rate and the voltage fluctuation rate based on the average current and the average voltage; Calculate the total harmonic distortion of current and voltage based on the harmonic data; Determine whether the current fluctuation rate, voltage fluctuation rate, current total harmonic distortion and voltage total harmonic distortion are greater than a preset fluctuation rate threshold range and distortion threshold range; If the current fluctuation rate, the voltage fluctuation rate, the current total harmonic distortion and the voltage total harmonic distortion are greater than the preset fluctuation rate threshold range and the distortion threshold range, it is determined that the electric energy data is abnormal.

5. The inorganic board production line fault prediction management method based on data analysis according to claim 4 is characterized in that: Fault prediction for inorganic board production line based on reliable output and power data of detection model; According to the detection model, the maximum roughness of the M+1th inorganic board among the roughnesses of the M+1th inorganic board at n angles is obtained; If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the power data is abnormal, it is determined that there is an electrical fault in the inorganic board production line; If the maximum roughness of the M+1th inorganic board is greater than the preset roughness threshold and the electric energy data is normal, it is determined that there is a mechanical failure in the inorganic board production line; If the electric energy data is abnormal and the maximum roughness of the M+1th inorganic board is less than the preset roughness threshold, it is determined that there is fluctuation in the power grid system to which the inorganic board production line is connected, and the inorganic board production line has no faults.

6. An inorganic board production line fault prediction management system based on data analysis, applied to the method as described in any one of claims 1 to 5, characterized in that: include: The image acquisition module has a fixed frame installed above the discharge end of the production line, and a high-definition camera is set in the center of the lower side of the fixed frame. The upper outer side of the high-definition camera is connected to a lamp for evenly emitting white light, and the white light falls obliquely on the top of the discharge end; infrared monitoring devices for detecting inorganic boards are installed on both sides of the discharge end; when the infrared monitoring device detects that the inorganic board has arrived at the discharge port, the lamp rotates around the high-definition camera at a uniform speed within a fixed time period, and the high-definition camera shoots the inorganic board within a preset time period at a fixed time interval to obtain multi-angle high-definition images of M inorganic boards; The image classification module annotates the timestamp and angle stamp of each inorganic board's multi-angle HD image; classifies and sorts the multi-angle HD image of each inorganic board according to the timestamp and angle stamp to obtain a data set ,in Represents k time series images of the nth angle; The prediction module preprocesses the data set to obtain a preprocessed data set; the data of each angle in the preprocessed data set are respectively input into the preset detection model to obtain the roughness of the n angles of the M+1th inorganic board, and the output of the detection model is determined to be credible through angle symmetry; The power quality monitoring module collects power data of the circuit of the inorganic board production line at fixed time intervals during the pre-coating operation period of M inorganic boards, and monitors the power data; The processing module adopts the output of the detection model if it is credible, and performs fault prediction on the inorganic board production line based on the credible output of the detection model and the monitoring result of the electric energy data; If the output of the detection model is unreliable, the detection model is reversely optimized based on the output of the detection model.

Images