A Fault Diagnosis System for Chemical Equipment

Abstract

The present invention relates to the field of fault diagnosis technology, specifically a chemical equipment fault diagnosis system, the system includes: an image acquisition module, a chemical equipment monitoring camera is installed, the real-time image signal of each chemical equipment in the monitoring area is acquired, the acquired image signal is sampled and partitioned, the grayscale distribution, edge intensity and texture details of the image are extracted, each extracted parameter is arranged on a time axis, and image timing information is acquired. In the present invention, more accurate chemical equipment fault diagnosis is achieved through multi-level data acquisition, image processing and feature extraction processes. By acquiring the monitoring area image in real time, and sampling the image by partition, the grayscale distribution, edge intensity and texture details are extracted, these parameters are integrated into the image timing information arranged on the time axis, and the image data is calibrated and the brightness is adjusted to ensure data consistency under various lighting environments.

Description

Technical Field

[0001] The invention relates to the technical field of fault diagnosis, and in particular to a chemical equipment fault diagnosis system. Background Art

[0002] The field of fault diagnosis technology is dedicated to identifying potential or existing fault characteristics in various types of equipment such as mechanical, electrical, and chemical through a variety of monitoring, analysis, and prediction methods. The core of this technical field is to accurately capture abnormal signals during the operation of the equipment, such as changes in characteristic values ​​such as vibration, temperature, and pressure, and to deeply explore and analyze the causes, patterns, and development trends of equipment failures through real-time monitoring and data analysis models. The chemical equipment fault diagnosis system is designed to intelligently monitor the health status of chemical equipment and identify faults, capture abnormal signals of equipment, and perform fault mode analysis.

[0003] However, existing fault diagnosis technologies are prone to false alarms or missed alarms under conditions such as low light and high noise, which affects the reliability of diagnosis. They also mainly rely on a single or a small number of physical parameters, such as vibration and temperature changes during equipment operation, and are prone to ignoring key information such as rich surface texture and curvature changes in image data, and are unable to fully present the operating status of the equipment. In addition, current fault detection is often based on static data pattern matching, which makes it difficult to achieve trend prediction and difficult to capture dynamically changing potential fault characteristics in a timely manner. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a chemical equipment fault diagnosis system.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme: A chemical equipment fault diagnosis system comprises:

[0006] The image acquisition module is equipped with a chemical equipment monitoring camera to obtain real-time image signals of each chemical equipment in the monitoring area, sample and partition the collected image signals, extract the grayscale distribution, edge intensity and texture details of the image, arrange each extracted parameter on a time axis, obtain image timing information, calibrate and adjust the pixel values ​​of the image timing information, and generate calibration image data;

[0007] The image enhancement module normalizes the grayscale distribution parameters contained in the calibration image data, increases the brightness of the grayscale pixel values ​​collected in a low-light environment, performs denoising calculations on the pixel signal intensity in the interference area, obtains image denoising data, performs edge convolution operations on the key areas of the chemical equipment in the image denoising data, fuses the convolution results with the denoised grayscale parameters, and obtains a high-definition key area image;

[0008] The feature extraction module divides the surface features of the chemical equipment in the image into regions based on the high-definition key area image, calculates the texture difference value, morphological curvature change value and grayscale gradient value in each region, performs clustering operation on the calculation results, obtains key morphological features, compares the key morphological features with the equipment failure data in the historical database, and establishes morphological feature comparison results;

[0009] The fault diagnosis module extracts the regional features with high matching degree based on the morphological feature comparison results, calculates the time trend of morphological changes in the matching area, dynamically predicts the offset of the trend curve, and then dynamically marks the area with abnormal offset to establish equipment fault warning information.

[0010] Preferably, the steps of acquiring the image timing information are:

[0011] The real-time image signal of each chemical equipment in the monitoring area is sampled and partitioned, and the grayscale distribution parameters, edge intensity parameters and texture detail parameters of the image are extracted to obtain the initial parameter set of the image signal;

[0012] Arrange each parameter in the image signal initial parameter set according to the time axis to generate an image signal parameter sequence arranged on the time axis, and compare the pixel values ​​in the image signal parameter sequence arranged on the time axis in layers to establish image time sequence arrangement parameters;

[0013] The pixel values ​​in the image timing parameters are calibrated and dynamically adjusted using the formula:

[0014] Generate calibration image timing information; where, Represents the adjusted image timing information, is the pixel intensity value within the partition, is the weight factor, which is adjusted according to the feature weights of grayscale, edge and texture. is the pixel timing value at the previous time point, is the offset coefficient.

[0015] Preferably, the step of acquiring the calibration image data is:

[0016] Based on the calibration image time sequence information, the grayscale distribution parameters in the image are normalized to obtain normalized grayscale distribution parameters;

[0017] Performing brightness adjustment on the normalized grayscale distribution parameters collected in a low-light environment, and dynamically correcting the grayscale distribution parameters using the pixel intensity in the image collection to generate grayscale distribution parameters after brightness adjustment;

[0018] The grayscale distribution parameters after brightness adjustment are subjected to denoising calculation of the pixel signal intensity in the interference area, using the formula:

[0019] Generate denoised calibration image data; where, represents the denoised calibration image data, is the grayscale pixel value after the brightness in the image is increased. is the noise intensity in the interference area, is the brightness parameter of the pixel area, is the noise attenuation coefficient.

[0020] Preferably, the steps of acquiring the image denoising data are:

[0021] Based on the calibration image data, the grayscale distribution parameters of the key areas of the chemical equipment are extracted, and the grayscale distribution parameters are subdivided and layered to obtain the grayscale parameters of the layered key areas;

[0022] Performing edge convolution operation on the grayscale parameters of the layered key areas, combining edge information of the layered grayscale parameters with texture features, and generating convolution edge feature parameters;

[0023] The convolution edge feature parameters are fused with the grayscale parameters in the calibration image data using the formula:

[0024] Generate image denoising data; where, represents image denoising data, is the grayscale distribution value in the convolution edge feature, is the calibration grayscale parameter after denoising, is the total number of grayscale features in the fusion area.

[0025] Preferably, the steps of acquiring the high-definition key area image are:

[0026] Based on the image denoising data, the grayscale distribution and edge convolution information of the chemical equipment area are extracted, and the image edge features of the key area are enhanced to obtain enhanced edge feature parameters;

[0027] The enhanced edge feature parameters are fused with the key area texture features in the image denoising data, and the areas with high matching degree are screened by feature matching degree to generate fused matching feature parameters;

[0028] Based on the fusion matching feature parameters, texture smoothing is performed using the formula:

[0029] Generate high-definition key area images; where For high-definition key area images, represents the matching texture smoothness parameter, To enhance the grayscale convolution result of edge features, is the total number of matching regions, Indicates the number of pixels in the key area.

[0030] Preferably, the steps of acquiring the key morphological features are:

[0031] Based on the high-definition key area image, the surface of the chemical equipment is divided into regions, and the texture difference value, morphological curvature change value and gray gradient value of each region are identified and extracted to form a feature parameter set;

[0032] Performing cluster analysis on the texture difference values, morphological curvature change values ​​and grayscale gradient values ​​in the feature parameter set to generate classification feature clusters;

[0033] The classification feature clusters are compared and analyzed using the formula:

[0034] Generate key morphological features; among them, Represents key morphological features, is the texture difference value in the clustering feature, is the change value of the morphological curvature, is the total number of features involved in clustering, is the weight adjustment coefficient, which reflects the importance of each feature to the key morphological features.

[0035] Preferably, the steps of obtaining the morphological feature comparison result are:

[0036] Matching the key morphological features with the equipment failure data in the historical database one by one, comparing each key morphological feature with the corresponding feature in the failure data, and generating preliminary comparison and matching parameters;

[0037] Based on the preliminary comparison and matching parameters, the morphological features of the matching area are screened, and the area with a matching degree higher than a preset threshold is extracted to obtain a high matching area feature set;

[0038] Perform dynamic matching evaluation on the high matching area feature set, using the formula:

[0039] Generate morphological feature comparison results; among them, is the result of morphological feature comparison. is the key feature value in the high matching area, is the corresponding value of the historical fault feature, is the total number of features in the high matching area, is the importance weight of the matching feature, It is a dynamic adjustment coefficient.

[0040] Preferably, the steps of obtaining the equipment failure warning information are:

[0041] Based on the morphological feature comparison results, a difference analysis is performed between the regional feature set with a high matching degree and the feature parameters of the equipment in a normal state, the estimated fault features are identified, and preliminary fault warning parameters are generated;

[0042] Performing dynamic trend analysis on the preliminary fault warning parameters, screening features with abnormal trends by evaluating the rate of change of feature difference values ​​over time, and forming a dynamic warning feature set;

[0043] Based on the dynamic warning feature set, the warning parameters are predicted in time series using the formula:

[0044] Generate equipment failure warning information; among which, Indicates equipment failure warning information. is the difference parameter in the dynamic warning feature, is the characteristic change rate in the time series, is the total number of warning features, is the time decay coefficient.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, a more accurate chemical equipment fault diagnosis is achieved through multi-level data acquisition, image processing and feature extraction processes. By acquiring the monitoring area image in real time and performing partition sampling on the image, the grayscale distribution, edge intensity and texture details are extracted, and these parameters are integrated into the image timing information arranged on the time axis, and the image data is calibrated and the brightness is adjusted to ensure data consistency under various lighting environments. The use of denoising and convolution operation fusion significantly improves the clarity of the key area image, optimizes the image detail resolution, and ensures the complete display of the equipment surface features. By comparing with historical fault data, the high-matching regional features are further located, and on this basis, trend prediction and abnormal annotation are carried out to make timely warnings of the fault change trend of the equipment. It provides effective support for improving the safety of chemical equipment operation and the sensitivity of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0049] See also Figure 1 The present invention provides a technical solution: a chemical equipment fault diagnosis system comprising:

[0050] The image acquisition module is equipped with a chemical equipment monitoring camera to obtain the real-time image signal of each chemical equipment in the monitoring area, sample and partition the acquired image signal, extract the grayscale distribution, edge intensity and texture details of the image, arrange each extracted parameter on the time axis, obtain the image timing information, calibrate and adjust the pixel value of the image timing information, and generate calibration image data;

[0051] The image enhancement module normalizes the grayscale distribution parameters contained in the calibration image data, increases the brightness of the grayscale pixel values ​​collected in the low-light environment, performs denoising calculations on the pixel signal intensity in the interference area, obtains image denoising data, performs edge convolution operations on the key areas of the chemical equipment in the image denoising data, and fuses the convolution results with the denoised grayscale parameters to obtain high-definition key area images;

[0052] The feature extraction module divides the surface features of chemical equipment in the image into regions based on high-definition key area images, calculates the texture difference value, morphological curvature change value and grayscale gradient value in each region, performs clustering operations on the calculation results, obtains key morphological features, compares the key morphological features with the equipment failure data in the historical database, and establishes morphological feature comparison results;

[0053] The fault diagnosis module extracts the regional features with high matching degree based on the morphological feature comparison results, calculates the time trend of morphological changes in the matching area, dynamically predicts the offset of the trend curve, and then dynamically marks the area with abnormal offset to establish equipment fault warning information.

[0054] The steps to obtain image timing information are as follows:

[0055] The real-time image signal of each chemical equipment in the monitoring area is sampled and partitioned, and the grayscale distribution parameters, edge intensity parameters and texture detail parameters of the image are extracted to obtain the initial parameter set of the image signal;

[0056] Arrange each parameter in the initial parameter set of the image signal according to the time axis to generate a time axis arranged image signal parameter sequence, and compare the pixel values ​​in the time axis arranged image signal parameter sequence in layers to establish image time sequence arrangement parameters;

[0057] The pixel values ​​in the image timing parameters are calibrated and dynamically adjusted using the formula:

[0058] Generate calibration image timing information; where, Represents the adjusted image timing information, is the pixel intensity value within the partition, is the weight factor, which is adjusted according to the feature weights of grayscale, edge and texture. is the pixel timing value at the previous time point, is the offset coefficient.

[0059] Specifically, the real-time image signal of the chemical equipment in the monitoring area is sampled and partitioned. The partitioning process is carried out according to the distribution characteristics of the image signal, and the image is divided into multiple monitoring sub-areas. The image signal of each area is further extracted with grayscale distribution parameters, edge intensity parameters and texture detail parameters. The image light and dark distribution range and brightness difference value are extracted from the grayscale distribution parameters, and the edge intensity changes and texture detail differences are recorded to generate a preliminary image signal initial parameter set. The grayscale parameters are classified into light and dark levels, and the edge parameters are classified into boundary contour layers. The image signal initial parameter set is generated in the layered order, and the parameters of each sub-area are recorded for subsequent analysis.

[0060] Arrange each parameter in the initial parameter set of the image signal according to the time axis, record the grayscale distribution and edge strength changes of each sub-region at different time points in layers, extract important structural contour changes by monitoring the fluctuation trend of edge strength, and extract the change trend of texture detail parameters at the same time, generate an image signal parameter sequence arranged on the time axis, and then set layered comparison standards for grayscale distribution and texture detail parameters respectively, set lower and higher comparison thresholds, form an image signal parameter sequence for layered comparison of pixel values, and establish image time sequence arrangement parameters;

[0061] formula: The calculation process:

[0062] The pixel values ​​in the image timing parameters are calibrated and dynamically adjusted based on this formula. Represents the adjusted image timing information. The calculation requires calling the pixel intensity value in each pixel partition , weight factor , according to the different grayscale, edge and texture feature parameters, the pixel time value at the previous time point , and a small offset coefficient Used to stabilize the value to obtain a reasonable parameter value range. The value is set as follows: For example, the intensity values ​​of grayscale, edge, and texture are 12, 10, and 15 respectively, and the pixel timing value at the previous time point is taken. , is 0.1.

[0063] Substituting into the formula:

[0064]

[0065] Calculate each term to get:

[0066]

[0067] The result is 4.10, indicating the adjusted image timing information value for subsequent image processing steps.

[0068] The steps to obtain the calibration image data are:

[0069] Based on the calibration image time sequence information, the grayscale distribution parameters in the image are normalized to obtain normalized grayscale distribution parameters;

[0070] The normalized grayscale distribution parameters collected in a low-light environment are brightness adjusted, and the pixel intensity in the image acquisition is used for dynamic correction to generate the grayscale distribution parameters after brightness adjustment;

[0071] The grayscale distribution parameters after brightness adjustment are used to perform denoising calculation on the pixel signal intensity in the interference area, using the formula:

[0072] Generate denoised calibration image data; where, represents the denoised calibration image data, is the grayscale pixel value after the brightness in the image is increased. is the noise intensity in the interference area, is the brightness parameter of the pixel area, is the noise attenuation coefficient.

[0073] Specifically, the real-time image signals of chemical equipment in the monitoring area are sampled and partitioned, and the parameters such as grayscale distribution, edge strength and texture details of the image signals are extracted and partitioned one by one. The extracted parameter sets are integrated and classified to obtain the initial parameter sets of each partition, and then each parameter is arranged along the time axis. The abnormal data and drift trends in the time series are smoothed and adjusted to ensure the accuracy and consistency of signal acquisition. The adjusted sequence data are dynamically adjusted, and the pixel values ​​are layered and compared based on the grayscale distribution, edge strength and texture features to reduce noise interference, optimize the temporal arrangement of the image signal, and establish the image temporal arrangement parameters.

[0074] Based on the image timing arrangement parameters generated above, the pixel values ​​in the parameter set are further calibrated and dynamically adjusted, the pixel intensities in different areas are normalized to form a standardized data format, the image brightness in a low-light environment is gradually adjusted, and the average value of the image pixel intensity in the brightness distribution is used for smoothing correction. Through dynamic updating and noise processing in specific areas, the impact of uneven lighting is minimized. During the denoising process, the pixel intensity in a specific area is adjusted according to a preset coefficient to ensure that the pixel signal maintains a high clarity after normalization, thereby generating grayscale distribution parameters after brightness adjustment.

[0075] formula The calculation process:

[0076] For denoising, the parameters Indicates the grayscale pixel value of each pixel after the brightness is increased, obtained through real-time acquisition equipment; Indicates the intensity level of the noise area, which is detected by the noise monitoring module; Indicates the brightness distribution of the area where each pixel is located, and the brightness parameter value is calculated by the grayscale analysis module; parameter Represents the noise attenuation coefficient, which is dynamically set according to the pixel area distance and brightness distribution characteristics.

[0077] Get , , , , substitute it into the formula and get:

[0078] The numerical result 1.31 represents the clarity index of this specific pixel in the denoised image data, reflecting the signal strength after denoising. The calculation of this value shows that under given noise conditions, the pixel signal in this area is significantly improved after the brightness is increased, which means that this area has higher reliability and accuracy in subsequent analysis. If 1.31 is greater than a preset threshold (such as 1.0), it indicates that the image signal quality in this area is good and suitable for subsequent feature extraction; conversely, if it is less than the threshold, it may need further processing or be marked as a potential problem area.

[0079] The steps to obtain image denoising data are:

[0080] Based on the calibration image data, the grayscale distribution parameters of the key areas of the chemical equipment are extracted, and the grayscale distribution parameters are subdivided and layered to obtain the grayscale parameters of the layered key areas;

[0081] Perform edge convolution operation on the grayscale parameters of the layered key area, combine the edge information of the layered grayscale parameters with the texture features, and generate convolution edge feature parameters;

[0082] The convolution edge feature parameters are fused with the grayscale parameters in the calibration image data using the formula:

[0083] Generate image denoising data; where, represents image denoising data, is the grayscale distribution value in the convolution edge feature, is the calibration grayscale parameter after denoising, is the total number of grayscale features in the fusion area.

[0084] Specifically, based on the denoised calibration image data, the grayscale distribution of key areas of the chemical equipment is refined, and the grayscale values ​​are used to segment the subtle features of the key areas layer by layer. The grayscale parameters are divided into multiple subsets and grayscale distribution models of different levels are established. The grayscale value distribution in the key area is analyzed layer by layer through the hierarchical grayscale model, and the corresponding edge feature parameters are generated. The edge convolution operation is called to calculate the edge features and texture information of the area, and the edge convolution parameters are compared with the grayscale hierarchical distribution model layer by layer. The edge changes of the grayscale parameters of each level are gradually calculated, and the texture features are combined for item-by-item operation to generate the convolution edge feature parameters. The convolution edge feature parameters comprehensively reflect the grayscale change trend and texture characteristics of the key areas in the denoised calibration image data.

[0085] formula The calculation process:

[0086] This formula is used to fuse the convolution edge feature parameters with the grayscale parameters in the denoised calibration image data to generate image denoising data. The parameters in the formula are obtained in different ways. is the grayscale distribution value in the convolution edge feature. Based on the extracted edge features, the edge pixels and background are refined through convolution operation to obtain the grayscale distribution. Represents the grayscale parameter after denoising, which is the grayscale denoising value obtained by adjusting the brightness of the original grayscale data under low light conditions. is the total number of grayscale features in the fusion area, which is obtained by counting the number of sub-areas extracted according to the grayscale distribution of the key area. During the calculation process, based on the set sample setting , and calculate:

[0087] Conclusion The value is 0.592, which indicates the denoised grayscale distribution of the key area of ​​the image after fusion. If the value is close to 1, it means the fusion result is good, and if the value is too small, it means the fusion effect is poor.

[0088] The steps to obtain high-definition key area images are:

[0089] Based on the image denoising data, the grayscale distribution and edge convolution information of the chemical equipment area are extracted, and the image edge features of the key area are enhanced to obtain the enhanced edge feature parameters;

[0090] The enhanced edge feature parameters are fused with the key area texture features in the image denoising data, and the high matching area is screened by feature matching to generate fused matching feature parameters;

[0091] On the basis of fusion matching feature parameters, texture smoothing is performed using the formula:

[0092] Generate high-definition key area images; where For high-definition key area images, represents the matching texture smoothness parameter, To enhance the grayscale convolution result of edge features, is the total number of matching regions, Indicates the number of pixels in the key area.

[0093] Specifically, based on the grayscale distribution and edge convolution information obtained from image denoising data, the image grayscale information is first extracted from the key areas of the chemical equipment and refined into edge features. The edge distribution is identified by detecting the local extreme values ​​of the grayscale. The texture difference and edge curvature change feature values ​​are further extracted in the layered image areas. The local grayscale extreme value parameters are integrated with the edge information parameters obtained by the convolution calculation to form an enhanced edge feature parameter. The convolution operation is applied to the grayscale changes in adjacent areas for edge detection to further extract the high-order edge features of the image. The gradient information in the enhanced edge feature is integrated with the texture difference to obtain the enhanced edge feature parameter.

[0094] The obtained enhanced edge feature parameters are fused with the texture features of the key areas of chemical equipment in the image denoising data. The areas with higher structural similarity are screened out by detecting the feature matching degree. The comprehensive features of grayscale, edge and texture are threshold matched and screened. A dynamic screening threshold is set for the matching degree between features. The areas above the threshold are clustered. The feature areas that meet the clustering criteria are integrated into the fused matching feature parameters, and secondary feature enhancement is performed on them to generate fused matching features.

[0095] formula The calculation process:

[0096] In this formula, represents the final high-definition key area image, is the matching texture smoothness parameter, and the specific value can be obtained by detecting the texture features of the high matching feature area. To enhance the grayscale convolution result of edge features, the value is obtained by convolution calculation. Represents the total number of all high-matching fusion regions, is the total number of pixels in the key area. The high-definition key area image is calculated step by step by calculating the sum of the square values ​​of the grayscale convolution of the smoothness of the fused area and the enhanced edge features and dividing it by the square root of the number of pixels in the key area, e.g. The value of is 0.75, The average value is 1.2. is 30, If it is 1000, then substitute it into the formula:

[0097] First, substitute the parameters into the formula:

[0098]

[0099] Calculate the parameters:

[0100] .

[0101] calculate .

[0102] because , in the summation part .

[0103] calculate .

[0104] Substitute the formula to get the high-definition key area image value :

[0105] This value represents the overall smoothness and edge sharpness of the generated high-definition key area image. If the result is greater than 1, it means that the smoothness and edge features are good and meet the high-definition image standard; if it is less than 1, there may be insufficient edge features or smoothing processing deviation, and further adjustment of edge features or smoothing parameters is required to ensure image quality.

[0106] The steps to obtain key morphological features are:

[0107] Based on high-definition key area images, the surface of chemical equipment is divided into regions, and the texture difference value, morphological curvature change value and grayscale gradient value of each region are identified and extracted to form a feature parameter set;

[0108] Cluster analysis is performed on the texture difference value, morphological curvature change value and gray gradient value in the feature parameter set to generate classification feature clusters;

[0109] Compare and analyze the classification feature clusters using the formula:

[0110] Generate key morphological features; among them, Represents key morphological features, is the texture difference value in the clustering feature, is the change value of the morphological curvature, is the total number of features involved in clustering, is the weight adjustment coefficient, which reflects the importance of each feature to the key morphological features.

[0111] Specifically, based on high-definition key area images, the surface images of chemical equipment are divided into multiple regions, and the edge detection method is used to accurately locate the texture difference values, morphological curvature change values ​​and grayscale gradient values ​​of the equipment surface. The texture feature parameters of each region are extracted in layers according to the specific differences. After extraction, the morphological curvature change value is used as the main parameter of the equipment surface concave and convex conditions. The morphological curvature change value is calculated based on the edge contour change points in the region, and the smoothness of the equipment is analyzed based on the grayscale gradient value. The texture feature differences of different regions are represented by generating texture difference value data, and the texture difference values, morphological curvature change values ​​and grayscale gradient values ​​of all regions are integrated into a preliminary feature parameter set.

[0112] The texture difference values, morphological curvature change values ​​and grayscale gradient values ​​of each region in the preliminary feature parameter set are clustered according to their similarity. The mean and variance of the difference values ​​of each region are statistically analyzed and classified into specific clusters in ascending order of variance. Each feature cluster represents a certain type of morphological feature on the surface of the device. At the same time, the difference value range within each cluster is compared, and the mean square error between the features is calculated. According to the mean square error of the difference values ​​within the cluster and the similarity of the morphological features, they are divided into different categories to generate classification feature clusters.

[0113] formula The calculation process:

[0114] Represents the key morphological features, which are obtained by calculating the weighted average of the texture difference value and the morphological curvature change value of each cluster. Represents the texture difference value of each region in the cluster feature, which is obtained by the local extreme value method in the difference value distribution. is the morphological curvature change value, which indicates the convexity or concavity of the surface morphology. It is calculated by the incremental value of the curvature change parameter. Indicates the total number of feature clusters involved in clustering, weight coefficient Based on the degree of influence of each morphological change on the key features of the device surface, it is adjusted as the curvature change value increases. It is obtained by calculating the weighted average of the texture difference value and the morphological curvature change value in the cluster. The specific calculation process is as follows:

[0115] In the feature clustering process, the total number of clustered feature clusters is , the texture difference value in each cluster and the morphological curvature change value The observation data are as follows:

[0116] Cluster 1: ,

[0117] Cluster 2: ,

[0118] Cluster 3: ,

[0119] Cluster 4: ,

[0120] Cluster 5: ,

[0121] Set the weight factor To enhance the influence of the change in morphological curvature. Substitute the formula for calculation:

[0122]

[0123] Compute each weighted product step by step:

[0124] Weighted product for cluster 1:

[0125] Weighted product for cluster 2:

[0126] Weighted product for cluster 3:

[0127] Weighted product for cluster 4:

[0128] Weighted product for cluster 5:

[0129] Add the results and divide by :

[0130]

[0131] Final calculation results Indicates the value of the key morphological feature, which reflects the overall morphological characteristics of the device surface under the current clustering parameters. A value greater than 50 indicates that the device surface is relatively rough and has significant morphological features, while a value less than 40 indicates that the surface is relatively flat and has weak morphological features.

[0132] The steps for obtaining the morphological feature comparison results are as follows:

[0133] Match the key morphological features with the equipment failure data in the historical database one by one, compare each key morphological feature with the corresponding feature in the failure data, and generate preliminary comparison and matching parameters;

[0134] Based on the preliminary comparison and matching parameters, the morphological features of the matching area are screened, and the areas with matching degrees higher than the preset threshold are extracted to obtain a high matching area feature set;

[0135] Perform dynamic matching evaluation on the high matching area feature set using the formula:

[0136] Generate morphological feature comparison results; among them, is the result of morphological feature comparison. is the key feature value in the high matching area, is the corresponding value of the historical fault feature, is the total number of features in the high matching area, is the importance weight of the matching feature, It is a dynamic adjustment coefficient.

[0137] Specifically, the key morphological features are matched with the equipment failure data in the historical database one by one. Based on each equipment failure feature parameter in the historical database, the texture difference value, morphological curvature change value and grayscale gradient value in the key morphological features are matched with the failure data in the database one by one, and the matching degree is compared. The difference and mean square error between each key morphological feature and the database feature are calculated in the item-by-item comparison process, and the cumulative sum of the mean square error of each difference value is averaged to generate preliminary comparison and matching parameters. Then, the overall parameters of all feature comparison and matching degrees are calibrated and corrected to obtain preliminary comparison and matching parameters.

[0138] Based on the preliminary comparison and matching parameters, the morphological features are screened, and the areas with a comparison matching degree higher than the set threshold are extracted. The preliminary comparison and matching parameters are compared with the fault database in the historical database one by one. Based on the threshold, the matching areas with a degree higher than the set value are classified as high-matching areas, and the areas below the threshold are eliminated to ensure that the feature set of the screened high-matching areas has complete data consistency. The invalid feature value items of the eliminated low-matching areas are eliminated during the comparison, and the feature set of the high-matching areas is obtained by integration.

[0139] formula The calculation process:

[0140] In the high matching area feature set, set the key feature value in the high matching area Corresponding value of historical fault characteristics The importance weights are 70 and 65 respectively. Set to 0.5, adjustment coefficient is 0.8, and each parameter is substituted into the formula for calculation. The formula first calculates the characteristic value of the high matching area and historical fault characteristic values The products are accumulated and then divided by the weight adjustment coefficient.

[0141] The operation process is:

[0142]

[0143] After calculation, we get Indicates the overall matching degree of the morphological feature comparison results.

[0144] The steps to obtain equipment failure warning information are as follows:

[0145] Based on the morphological feature comparison results, the high-matching regional feature set is analyzed for differences with the feature parameters of the equipment in normal state, the estimated fault features are identified, and preliminary fault warning parameters are generated;

[0146] Perform dynamic trend analysis on the preliminary fault warning parameters, evaluate the rate of change of feature difference values ​​over time, screen features with abnormal trends, and form a dynamic warning feature set;

[0147] Based on the dynamic warning feature set, the warning parameters are predicted in time series using the formula:

[0148] Generate equipment failure warning information; among which, Indicates equipment failure warning information. is the difference parameter in the dynamic warning feature, is the characteristic change rate in the time series, is the total number of warning features, is the time decay coefficient.

[0149] Specifically, based on the morphological feature comparison results, each high-matching regional feature set is extracted, and the regional features are analyzed item by item with the comparison parameters under the normal operating state of the equipment, so as to calculate the difference value feature parameters with abnormal trends. The abnormal features in the equipment are preliminarily identified through high-matching feature comparison analysis. After completing the difference analysis, a preliminary fault warning parameter set is set according to the comparison results, including difference value, change rate parameter, abnormal frequency and other specific contents. The difference value is calculated using the average difference method, comparing the average value of the normal state with the abnormal change value, and the average difference is normalized to form the preliminary warning parameters. The warning parameters are used to dynamically analyze the changing trend of fault characteristics.

[0150] A dynamic trend analysis is performed on the preliminary fault warning parameters, with the change rate of the difference parameters in a specific time interval as the core. During the dynamic trend analysis, the change rate of each time period is compared and recorded, and the change values ​​of the difference parameters at multiple time points are collected into a sequence. The trend fitting method is used to compare the slope of the change curve in each time series, and the features with a growth rate higher than the preset standard are screened out to form a dynamic warning feature set. The dynamic trend calculation method is the cumulative difference average method, and the trend growth rate screening threshold is determined according to the change amplitude of historical data. Features above this value are marked as abnormal.

[0151] formula The calculation process:

[0152] parameter : represents the difference parameter in the dynamic warning feature. To ensure its accuracy, it needs to be extracted from the real-time monitoring data and normalized. For example, in a certain period of time, if the difference parameter values ​​are monitored as 0.8, 1.2, and 1.0, the normalized difference parameter can be unified as 0.9.

[0153] parameter : Indicates the rate of change of features in a time series. It is calculated based on the increase in the difference value of consecutive time points and the time interval. For example, if the time interval is 10 minutes and the difference parameter value increases from 0.8 to 1.2, the change rate is .

[0154] parameter : represents the time decay coefficient, which is used to weight the features with significant recent changes. Based on historical trend data, it can be taken , for example, the time range is 1 hour to cover the dynamic changes of the features.

[0155] parameter : Indicates the total number of warning features, which is set according to the total number of features in real-time monitoring, for example .

[0156] Substitute the above values ​​into the formula for calculation, the steps are as follows:

[0157] Substitute the values ​​and calculate the exponential weighted value of each feature item:

[0158]

[0159]

[0160]

[0161] Adding the above calculation results together gives The final value of:

[0162]

[0163] This result indicates the warning information value of equipment failure. When this value is greater than a certain threshold, it can be determined that the risk of equipment failure is high and further investigation is required.

Claims

1. A chemical equipment fault diagnosis system, characterized in that: The system comprises: The image acquisition module is equipped with a chemical equipment monitoring camera to obtain real-time image signals of each chemical equipment in the monitoring area, sample and partition the collected image signals, extract the grayscale distribution, edge intensity and texture details of the image, arrange each extracted parameter on a time axis, obtain image timing information, calibrate and adjust the pixel values ​​of the image timing information, and generate calibration image data; The image enhancement module normalizes the grayscale distribution parameters contained in the calibration image data, increases the brightness of the grayscale pixel values ​​collected in a low-light environment, performs denoising calculations on the pixel signal intensity in the interference area, obtains image denoising data, performs edge convolution operations on the key areas of the chemical equipment in the image denoising data, fuses the convolution results with the denoised grayscale parameters, and obtains a high-definition key area image; The feature extraction module divides the surface features of the chemical equipment in the image into regions based on the high-definition key area image, calculates the texture difference value, morphological curvature change value and grayscale gradient value in each region, performs clustering operation on the calculation results, obtains key morphological features, compares the key morphological features with the equipment failure data in the historical database, and establishes morphological feature comparison results; The fault diagnosis module extracts the regional features with high matching degree based on the morphological feature comparison results, calculates the time trend of morphological changes in the matching area, dynamically predicts the offset of the trend curve, and then dynamically marks the area with abnormal offset to establish equipment fault warning information.

2. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the image timing information are as follows: The real-time image signal of each chemical equipment in the monitoring area is sampled and partitioned, and the grayscale distribution parameters, edge intensity parameters and texture detail parameters of the image are extracted to obtain the initial parameter set of the image signal; Arrange each parameter in the image signal initial parameter set according to the time axis to generate an image signal parameter sequence arranged on the time axis, and compare the pixel values ​​in the image signal parameter sequence arranged on the time axis in layers to establish image time sequence arrangement parameters; The pixel values ​​in the image timing parameters are calibrated and dynamically adjusted using the formula: ； Generate calibration image timing information; where, Represents the adjusted image timing information, is the pixel intensity value within the partition, is the weight factor, which is adjusted according to the feature weights of grayscale, edge and texture. is the pixel timing value at the previous time point, is the offset coefficient.

3. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the calibration image data are as follows: Based on the calibration image time sequence information, the grayscale distribution parameters in the image are normalized to obtain normalized grayscale distribution parameters; Performing brightness adjustment on the normalized grayscale distribution parameters collected in a low-light environment, and dynamically correcting the grayscale distribution parameters using the pixel intensity in the image collection to generate grayscale distribution parameters after brightness adjustment; The grayscale distribution parameters after brightness adjustment are subjected to denoising calculation of the pixel signal intensity in the interference area, using the formula: ； Generate denoised calibration image data; where, represents the denoised calibration image data, is the grayscale pixel value after the brightness in the image is increased. is the noise intensity in the interference area, is the brightness parameter of the pixel area, is the noise attenuation coefficient.

4. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the image denoising data are as follows: Based on the calibration image data, the grayscale distribution parameters of the key areas of the chemical equipment are extracted, and the grayscale distribution parameters are subdivided and layered to obtain the grayscale parameters of the layered key areas; Performing edge convolution operation on the grayscale parameters of the layered key areas, combining edge information of the layered grayscale parameters with texture features, and generating convolution edge feature parameters; The convolution edge feature parameters are fused with the grayscale parameters in the calibration image data using the formula: ； Generate image denoising data; where, represents image denoising data, is the grayscale distribution value in the convolution edge feature, is the calibration grayscale parameter after denoising, is the total number of grayscale features in the fusion area.

5. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps of acquiring the high-definition key area image are as follows: Based on the image denoising data, the grayscale distribution and edge convolution information of the chemical equipment area are extracted, and the image edge features of the key area are enhanced to obtain enhanced edge feature parameters; The enhanced edge feature parameters are fused with the key area texture features in the image denoising data, and the areas with high matching degree are screened by feature matching degree to generate fused matching feature parameters; Based on the fusion matching feature parameters, texture smoothing is performed using the formula: ； Generate high-definition key area images; where For high-definition key area images, represents the matching texture smoothness parameter, To enhance the grayscale convolution results of edge features, is the total number of matching regions, Indicates the number of pixels in the key area.

6. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the key morphological features are: Based on the high-definition key area image, the surface of the chemical equipment is divided into regions, and the texture difference value, morphological curvature change value and gray gradient value of each region are identified and extracted to form a feature parameter set; Performing cluster analysis on the texture difference values, morphological curvature change values ​​and grayscale gradient values ​​in the feature parameter set to generate classification feature clusters; The classification feature clusters are compared and analyzed using the formula: ； Generate key morphological features; among them, Represents key morphological features, is the texture difference value in the clustering feature, is the change value of the morphological curvature, is the total number of features involved in clustering, is the weight adjustment coefficient, which reflects the importance of each feature to the key morphological features.

7. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the morphological feature comparison result are: Matching the key morphological features with the equipment failure data in the historical database one by one, comparing each key morphological feature with the corresponding feature in the failure data, and generating preliminary comparison and matching parameters; Based on the preliminary comparison and matching parameters, the morphological features of the matching area are screened, and the area with a matching degree higher than a preset threshold is extracted to obtain a high matching area feature set; Perform dynamic matching evaluation on the high matching area feature set, using the formula: ； Generate morphological feature comparison results; among them, is the result of morphological feature comparison. is the key feature value in the high matching area, is the corresponding value of the historical fault feature, is the total number of features in the high matching area, is the importance weight of the matching feature, It is a dynamic adjustment coefficient.

8. The chemical equipment fault diagnosis system according to claim 1, characterized in that: The steps for obtaining the equipment failure warning information are as follows: Based on the morphological feature comparison results, a difference analysis is performed between the regional feature set with a high matching degree and the feature parameters of the equipment in a normal state, the estimated fault features are identified, and preliminary fault warning parameters are generated; Performing dynamic trend analysis on the preliminary fault warning parameters, screening features with abnormal trends by evaluating the rate of change of feature difference values ​​over time, and forming a dynamic warning feature set; Based on the dynamic warning feature set, the warning parameters are predicted in time series using the formula: ； Generate equipment failure warning information; among which, Indicates equipment failure warning information. is the difference parameter in the dynamic warning feature, is the characteristic change rate in the time series, is the total number of warning features, is the time decay coefficient.

Images