Resin tank pressure resistance test method and system based on deep learning

Abstract

The invention discloses a resin tank pressure resistance test method and system based on deep learning, and relates to the technical field of data processing. The method comprises the following steps: performing significant feature analysis based on a surface image of a target resin tank, determining a plurality of abnormal areas, performing defect identification, and constructing a defect distribution diagram according to an identification result; based on equipment attributes, setting retrieval constraint conditions to retrieve voltage withstanding test records of similar products, obtaining a sample data set, performing supervision training on the deep learning operator by using the sample data set, and constructing a voltage withstanding performance predictor; environment parameters of the target resin tank in a preset time zone are collected, the defect distribution diagram and the environment parameters are input into the pressure resistance predictor, and the predicted maximum pressure bearing value is output to serve as a pressure resistance test result. The technical problem that in the prior art, a resin tank pressure resistance test result is not accurate enough is solved, defect distribution and environmental parameters are analyzed in combination with deep learning, and the technical effect of improving pressure resistance prediction accuracy is achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a resin tank pressure test method and system based on deep learning. Background Art

[0002] As an important industrial container, resin tanks are widely used in many fields such as chemical industry, petroleum, and water treatment. Its pressure resistance is one of the key indicators to measure the quality and safety of resin tanks, and is directly related to the operating reliability and service life of the equipment. Traditional resin tank pressure resistance test methods usually rely on physical pressure tests or simple historical data analysis. Although these methods can provide certain test results, they have problems such as insufficient accuracy, low data utilization, and inability to fully consider the impact of product defects and environmental parameters on pressure resistance. In addition, with the diversification of resin tank application scenarios, complex environments have put forward higher requirements on their pressure resistance performance, and it is difficult to meet actual needs by relying solely on traditional testing methods. Summary of the invention

[0003] The present application provides a resin tank pressure test method and system based on deep learning, which solves the technical problem that the resin tank pressure test results are not accurate enough in the prior art.

[0004] In view of the above problems, the present application provides a resin tank pressure test method and system based on deep learning.

[0005] In a first aspect of the present application, a resin tank pressure test method based on deep learning is provided, the method comprising: Based on the surface image of the target resin tank, a significant feature analysis is performed to determine multiple abnormal areas, and defects are identified for the multiple abnormal areas, and a defect distribution map is constructed according to the identification results; the device attributes of the target resin tank are obtained, and retrieval constraints are set based on the device attributes to retrieve the pressure test records of similar products, a sample data set is obtained, and the sample data set is used to perform supervised training on a deep learning operator to construct a pressure performance predictor; the environmental parameters of the target resin tank in a predetermined time zone are collected, the defect distribution map and the environmental parameters are input into the pressure performance predictor, and the predicted maximum pressure value is output as the pressure test result.

[0006] A second aspect of the present application provides a resin tank pressure test system based on deep learning, the system comprising: Analysis and identification module: performs significant feature analysis based on the surface image of the target resin tank, determines multiple abnormal areas, identifies defects in the multiple abnormal areas, and constructs a defect distribution map based on the identification results; Model construction module: obtains the device attributes of the target resin tank, sets retrieval constraints based on the device attributes to retrieve the pressure test records of similar products, obtains a sample data set, and uses the sample data set to supervise the training of the deep learning operator to construct a pressure performance predictor; Pressure test module: collects the environmental parameters of the target resin tank in a predetermined time zone, inputs the defect distribution map and environmental parameters into the pressure performance predictor, and outputs the predicted maximum pressure value as the pressure test result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, a significant feature analysis is performed based on the surface image of the target resin tank to determine multiple abnormal areas, and defects are identified for the multiple abnormal areas, and a defect distribution map is constructed based on the identification results. Then, the device attributes of the target resin tank are obtained, and the retrieval constraints are set based on the device attributes to retrieve the pressure test records of similar products, a sample data set is obtained, and the sample data set is used to supervise the training of the deep learning operator to construct a pressure performance predictor. Finally, the environmental parameters of the target resin tank in the predetermined time zone are collected, the defect distribution map and environmental parameters are input into the pressure performance predictor, and the predicted maximum pressure value is output as the pressure test result. The technical problem that the pressure test results of the resin tank in the prior art are not accurate enough is solved, and the technical effect of improving the accuracy of the pressure performance prediction is achieved by combining deep learning to analyze the defect distribution and environmental parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic flow chart of a resin tank pressure test method based on deep learning provided in an embodiment of the present application.

[0010] Figure 2 A schematic diagram of the structure of a resin tank pressure test system based on deep learning provided in an embodiment of the present application.

[0011] Explanation of the reference numerals: analysis and identification module 11 , model building module 12 , voltage resistance testing module 13 . DETAILED DESCRIPTION

[0012] The present application solves the technical problem in the prior art that the results of the resin tank pressure resistance test are not accurate enough by providing a resin tank pressure resistance test method and system based on deep learning.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0015] Embodiment 1, as Figure 1 As shown, the present application provides a resin tank pressure test method based on deep learning, wherein the method includes: A significant feature analysis is performed based on the surface image of the target resin tank to determine a plurality of abnormal areas, and defects are identified for the plurality of abnormal areas, and a defect distribution map is constructed according to the identification results.

[0016] The surface of the target resin tank is comprehensively scanned using high-precision imaging equipment to collect surface images. The collected images are preprocessed, including denoising, contrast enhancement, and edge feature extraction, to ensure the accuracy and robustness of subsequent analysis. The preprocessed surface images are subjected to significant feature analysis to extract multiple abnormal areas, which usually manifest as texture abnormalities, color differences, or geometric shape abnormalities. Defect identification is performed on these abnormal areas to determine information such as the type, size, and location of the defects, and a defect distribution map is constructed. The defect distribution map is used to visualize the defect distribution on the surface of the target resin tank and provide data support for subsequent pressure resistance performance predictions.

[0017] Furthermore, a significant feature analysis is performed based on the surface image of the target resin tank to identify multiple abnormal areas, including: Based on low-resolution acquisition of a surface image of a target resin tank, the surface image is input into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit; center-surrounding difference calculations are performed on the multi-scale image set to generate a brightness feature map, a color feature map, and a direction feature map, and the brightness feature map, the color feature map, and the direction feature map are superimposed to generate an initial significant image; the initial significant image is extracted according to an initial significant feature threshold to obtain a plurality of significant feature regions that meet the initial significant feature threshold, and are set as the plurality of abnormal regions.

[0018] Specifically, an imaging device is used to collect the surface image of the target resin tank in a low-resolution mode to reduce the computational complexity of data processing; the collected surface image is input into a Gaussian pyramid model, and a multi-scale Gaussian downsampling operation is performed to generate a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit, which respectively decompose the brightness information, color information, and direction information, and the Gaussian pyramid blurs and downsamples the image by applying a Gaussian filter layer by layer to separate feature information of different scales. For each image in the multi-scale image set, the center-surrounding difference is calculated to capture the significant characteristics between the local area and the surrounding area, and a brightness feature map, a color feature map, and a direction feature map are generated. The brightness feature map, the color feature map, and the direction feature map represent the significant feature distribution of brightness, color, and direction, respectively, wherein the brightness feature map: extracts the brightness difference of the local area to reflect the light and dark contrast area in the image; the color feature map: calculates the color contrast difference based on the RGB or Lab color space to highlight the color mutation area; the direction feature map: extracts the edge direction difference area through gradient direction analysis (such as Sobel operator or direction filter). The brightness feature map, color feature map and direction feature map are weightedly superimposed to generate an initial salient image; the initial salient feature threshold is set to screen the highly salient areas in the initial salient image, the initial salient image is binarized according to the salient feature threshold, and the areas that meet the threshold conditions are extracted. These areas represent abnormal areas with significant features in the image; the extracted significant feature areas are marked as multiple abnormal areas to provide a target range for subsequent defect identification.

[0019] Furthermore, performing defect identification on the plurality of abnormal areas and constructing a defect distribution map according to the identification results includes: Defect recognition features are configured, wherein the defect recognition features include crack features and deformation features; surface defects of the resin tank are retrieved according to the crack features and deformation features, a sample defect area set, a sample crack feature set and a sample deformation feature set are obtained, supervised learning is performed on the convolutional neural network until convergence, and a surface defect recognition plug-in is obtained; based on high-resolution acquisition of multiple abnormal images of the multiple abnormal areas, defects are recognized on the multiple abnormal images using the surface defect recognition plug-in, multiple defect features are output, and the defect distribution map is constructed in combination with the positions of multiple abnormal areas.

[0020] Specifically, defect recognition features are configured, and crack features and deformation features are used as the main recognition indicators, where crack features include parameters such as crack length, width and edge roughness, and deformation features include information such as surface depressions and bulges’ geometric dimensions and displacement amplitude; based on crack features and deformation features, possible defects on the surface of the resin tank are retrieved, and sample defect area sets, sample crack feature sets and sample deformation feature sets are extracted. These data are used to perform supervised learning training on the convolutional neural network, and through iterative optimization until the model converges, a surface defect recognition plug-in dedicated to defect recognition is generated; high-resolution imaging equipment is used to image multiple abnormal areas Image acquisition is performed to obtain multiple abnormal images; these abnormal images are input into the aforementioned surface defect recognition plug-in for automated defect recognition analysis. The surface defect recognition plug-in accurately detects the crack and deformation features in the abnormal images based on the feature parameters extracted during model training, and outputs multiple specific defect features, including the type, location, size, severity and other information of each defect; the output multiple defect features are combined with the geographic location data of the abnormal area to generate a defect distribution map of the resin tank, which intuitively shows the spatial distribution of surface defects of the target resin tank, and provides an accurate data basis for subsequent pressure resistance performance prediction and risk assessment.

[0021] The device attributes of the target resin tank are obtained, and retrieval constraints are set based on the device attributes to retrieve the pressure test records of similar products, a sample data set is obtained, and the sample data set is used to perform supervised training on a deep learning operator to construct a pressure performance predictor.

[0022] Extract the equipment properties of the target resin tank, including material properties (such as material type, material thickness, elastic modulus and tensile strength), structural properties (such as tank size, wall thickness distribution, structural form and connection method) and working time (such as cumulative operating time or service life). Set retrieval constraints based on the extracted equipment properties, for example, select records with the same or similar material types, records with matching structural forms, and records with the same working time (such as new products, products in use or products nearing the end of their life); use these conditions to filter out test data of similar products that meet the conditions from the historical pressure test record database. Organize the retrieved pressure test records of similar products into a sample data set, which contains test data related to equipment properties (such as material properties, structural design, working time), pressure test results (such as maximum pressure value, failure pressure) and test conditions (such as test environment temperature and humidity). The data set is cleaned and normalized to ensure data consistency, and key features are annotated. The cleaned sample data set is input into the deep learning operator for supervised training. During the training process, through multiple iterative optimizations, the model learns the nonlinear mapping relationship between pressure resistance performance and equipment properties, while considering the dynamic impact of different working hours on pressure resistance performance. During the training process, the loss function is used to evaluate the difference between the predicted value and the actual test value, and the model weights are updated through the back propagation algorithm until the model converges. The completed deep learning operator is used as a pressure resistance performance predictor, which can accurately predict the maximum pressure resistance value of the target resin tank based on the input resin tank equipment properties, defect characteristics and environmental parameters.

[0023] Furthermore, based on the device attributes, a search constraint condition is set to search for withstand voltage test records of similar products to obtain a sample data set, including: The device attributes of the target resin tank are obtained, wherein the device attributes include material attributes, structural attributes and working time; the material attributes, structural attributes and working time are expanded according to the attribute tolerance range to obtain material attribute thresholds, structural attribute thresholds and working time thresholds; test constraints are configured, wherein the test constraints include pressure holding time and expected pressure drop; the material attribute thresholds, structural attribute thresholds and working time thresholds are used as device retrieval constraints, the pressure holding time and expected pressure drop are used as conditional retrieval constraints, and pressure test records of similar products are retrieved based on big data to obtain sample defect distribution sets, sample environmental parameter sets and sample maximum pressure value sets as the sample data sets.

[0024] Specifically, the equipment attributes of the target resin tank are extracted, including material attributes (such as material type, elastic modulus, tensile strength, etc.), structural attributes (such as wall thickness, tank shape, dimensional parameters, etc.) and working time (such as cumulative operating time or service life) as basic parameters for retrieval. According to the specific value of each equipment attribute, an appropriate tolerance interval is set in combination with experience or standards. For example, the compatible range of material types and physical property thresholds can be expanded for material attributes, the upper and lower limits of wall thickness and size can be expanded for structural attributes, and the allowed phased intervals can be set for working time; finally, material attribute thresholds, structural attribute thresholds and working time thresholds are formed to ensure the coverage and diversity of retrieval results. According to the specific needs of pressure test data, test constraints are configured, including but not limited to pressure holding time (such as the minimum time requirement for pressure holding) and expected pressure drop (such as the maximum allowable pressure loss). These conditions are used to screen records with consistent test standards to ensure the comparability and validity of sample data. The material property threshold, structural property threshold and working time threshold are used as equipment retrieval constraints, and the pressure holding time and expected pressure drop are used as test condition retrieval constraints to form a complete retrieval strategy; the pressure test records of similar products are retrieved based on the big data platform; through the retrieval, the sample defect distribution set (including defect location, type and characteristic distribution), sample environmental parameter set (including test environment temperature, humidity, temperature difference, humidity difference, external pressure, etc.) and sample maximum pressure value set (including maximum pressure value and failure pressure point) that meet the conditions are obtained, and these data are integrated into a sample data set to provide high-quality data support for the subsequent training of deep learning operators.

[0025] Furthermore, the sample data set is used to supervise the deep learning operator and construct a voltage resistance performance predictor, including: Q deep learning operators are configured, wherein the deep learning operators include at least BP neural network, support vector machine and random forest, and Q is an integer greater than or equal to 3; the sample data set is divided into Q equal parts, and Q parts are selected with replacement for Q times to construct a first training set, and Q training sets are selected iteratively for Q times to obtain Q training sets; with sample defect distribution and sample environmental parameters as input and sample maximum pressure bearing value as output, the Q deep learning operators are supervised and trained respectively using the Q training sets to obtain Q converged pressure resistance performance prediction units, and the pressure resistance performance predictor is integrated to construct.

[0026] Specifically, first, Q deep learning operators are configured, where the deep learning operators include at least BP neural network, support vector machine and random forest, and Q is an integer greater than or equal to 3, so as to ensure the diversity of multiple algorithms and the robustness of the results; then, the sample data set is divided into Q parts, and Q selections are made by sampling with replacement to construct the first training set; Q different training sets are constructed by iterative selection Q times to ensure the randomness of data distribution and the comprehensiveness of model training; then, the defect distribution and environmental parameters in the sample data set are used as input, and the maximum pressure value of the sample is used as output, and the Q deep learning operators are supervised and trained respectively by using the constructed Q training sets. During the training process, the parameters of the operators are adjusted by iterative optimization so that each operator achieves performance convergence on its own training set, and finally Q independent pressure resistance performance prediction units are obtained; finally, the Q converged pressure resistance performance prediction units are integrated, and the final pressure resistance performance predictor is constructed by using model fusion technology (such as weighted average or voting mechanism) to improve the accuracy and robustness of the prediction results. Through this method, the pressure resistance performance predictor can accurately predict the maximum pressure value of the target resin tank based on its defect distribution and environmental parameters, providing a reliable performance evaluation tool for practical applications.

[0027] Furthermore, the integrated construction of the pressure resistance performance predictor includes: Verify and obtain Q prediction accuracies of the Q convergent voltage withstand performance prediction units; based on the coefficient of variation method, configure Q output weights according to the Q prediction accuracies; based on the principle of ensemble learning, fuse the Q output weights and the Q convergent voltage withstand performance prediction units to generate the voltage withstand performance predictor.

[0028] Specifically, the Q voltage withstand performance prediction units that have been trained and converged are verified separately, and their prediction performance is tested on an independent verification set, and the prediction accuracy (such as mean square error or absolute error) of each prediction unit is calculated to obtain Q prediction accuracy values; then, these prediction accuracies are weighted based on the coefficient of variation method, and the output weight of each prediction unit is calculated, where the coefficient of variation method assigns weights by measuring the stability of the prediction accuracy, so that units with high prediction accuracy and small fluctuations obtain larger weights; then, according to the principle of ensemble learning, the Q converged voltage withstand performance prediction units are fused with their corresponding output weights, and the final voltage withstand performance predictor is generated by weighted averaging or other fusion strategies (such as weighted voting or stacking models).

[0029] The environmental parameters of the target resin tank in a predetermined time zone are collected, the defect distribution map and the environmental parameters are input into the pressure resistance performance predictor, and the predicted maximum pressure bearing value is output as the pressure resistance test result.

[0030] The environmental parameters of the target resin tank in the predetermined time zone are collected, including temperature, humidity, external pressure and vibration intensity, etc. These parameters can be acquired through real-time monitoring by environmental sensors; the collected environmental parameters are combined with the previously generated defect distribution map and input as input data into the constructed pressure resistance performance predictor. The pressure resistance performance predictor comprehensively analyzes the input defect distribution information and environmental parameters, predicts the maximum pressure value of the target resin tank, outputs the maximum pressure value that the resin tank may withstand, and uses the predicted maximum pressure value as the pressure resistance test result to provide data support for the safety performance evaluation of the resin tank.

[0031] Furthermore, the environmental parameters within the predetermined time zone include a mean temperature, a mean temperature difference, a mean humidity, and a mean humidity difference.

[0032] Environmental sensors arranged at the test site are used to continuously collect environmental data within a predetermined time zone, including temperature, humidity and their changes. Based on the collected raw data, the mean temperature and mean temperature difference within the predetermined time zone are calculated. The mean temperature reflects the average temperature level during the period, while the mean temperature difference quantifies the amplitude of temperature fluctuations. At the same time, the mean humidity and mean humidity difference are calculated. The mean humidity is used to characterize the overall level of air humidity during the period, while the mean humidity difference reflects the dynamic range of humidity changes. The above four environmental parameters (mean temperature, mean temperature difference, mean humidity and mean humidity difference) are combined with the previously generated defect distribution map and input into the pressure resistance performance predictor as comprehensive input data. Based on the input data, the pressure resistance performance predictor analyzes the comprehensive impact of defect characteristics on the pressure bearing capacity of the resin tank under different environmental conditions, and accurately predicts the maximum pressure bearing value of the resin tank.

[0033] In summary, the embodiments of the present application have at least the following technical effects: First, a significant feature analysis is performed based on the surface image of the target resin tank to determine multiple abnormal areas, and defects are identified for the multiple abnormal areas, and a defect distribution map is constructed based on the identification results. Then, the device attributes of the target resin tank are obtained, and the retrieval constraints are set based on the device attributes to retrieve the pressure test records of similar products, a sample data set is obtained, and the sample data set is used to supervise the training of the deep learning operator to construct a pressure performance predictor. Finally, the environmental parameters of the target resin tank in the predetermined time zone are collected, the defect distribution map and environmental parameters are input into the pressure performance predictor, and the predicted maximum pressure value is output as the pressure test result. The technical problem that the pressure test results of the resin tank in the prior art are not accurate enough is solved, and the technical effect of improving the accuracy of the pressure performance prediction is achieved by combining deep learning to analyze the defect distribution and environmental parameters.

[0034] Embodiment 2 is based on the same inventive concept as the resin tank pressure test method based on deep learning in the previous embodiment. Figure 2As shown, the present application provides a resin tank pressure test system based on deep learning, wherein the system includes: Analysis and identification module 11: Perform significant feature analysis based on the surface image of the target resin tank, determine multiple abnormal areas, and identify defects in the multiple abnormal areas, and construct a defect distribution map based on the identification results; Model construction module 12: Obtain the device attributes of the target resin tank, set retrieval constraints based on the device attributes to retrieve the pressure test records of similar products, obtain a sample data set, and use the sample data set to supervise the training of the deep learning operator to construct a pressure performance predictor; Pressure test module 13: Collect the environmental parameters of the target resin tank in a predetermined time zone, input the defect distribution map and environmental parameters into the pressure performance predictor, and output the predicted maximum pressure value as the pressure test result.

[0035] Furthermore, the analysis and identification module 11 is used to perform the following method: Based on low-resolution acquisition of a surface image of a target resin tank, the surface image is input into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit; center-surrounding difference calculations are performed on the multi-scale image set to generate a brightness feature map, a color feature map, and a direction feature map, and the brightness feature map, the color feature map, and the direction feature map are superimposed to generate an initial significant image; the initial significant image is extracted according to an initial significant feature threshold to obtain a plurality of significant feature regions that meet the initial significant feature threshold, and are set as the plurality of abnormal regions.

[0036] Furthermore, the analysis and identification module 11 is used to perform the following method: Defect recognition features are configured, wherein the defect recognition features include crack features and deformation features; surface defects of the resin tank are retrieved according to the crack features and deformation features, a sample defect area set, a sample crack feature set and a sample deformation feature set are obtained, supervised learning is performed on the convolutional neural network until convergence, and a surface defect recognition plug-in is obtained; based on high-resolution acquisition of multiple abnormal images of the multiple abnormal areas, defects are recognized on the multiple abnormal images using the surface defect recognition plug-in, multiple defect features are output, and the defect distribution map is constructed in combination with the positions of multiple abnormal areas.

[0037] Furthermore, the model building module 12 is used to execute the following method: The device attributes of the target resin tank are obtained, wherein the device attributes include material attributes, structural attributes and working time; the material attributes, structural attributes and working time are expanded according to the attribute tolerance range to obtain material attribute thresholds, structural attribute thresholds and working time thresholds; test constraints are configured, wherein the test constraints include pressure holding time and expected pressure drop; the material attribute thresholds, structural attribute thresholds and working time thresholds are used as device retrieval constraints, the pressure holding time and expected pressure drop are used as conditional retrieval constraints, and pressure test records of similar products are retrieved based on big data to obtain sample defect distribution sets, sample environmental parameter sets and sample maximum pressure value sets as the sample data sets.

[0038] Furthermore, the model building module 12 is used to execute the following method: Q deep learning operators are configured, wherein the deep learning operators include at least BP neural network, support vector machine and random forest, and Q is an integer greater than or equal to 3; the sample data set is divided into Q equal parts, and Q parts are selected with replacement for Q times to construct a first training set, and Q training sets are selected iteratively for Q times to obtain Q training sets; with sample defect distribution and sample environmental parameters as input and sample maximum pressure bearing value as output, the Q deep learning operators are supervised and trained respectively using the Q training sets to obtain Q converged pressure resistance performance prediction units, and the pressure resistance performance predictor is integrated to construct.

[0039] Furthermore, the model building module 12 is used to execute the following method: Verify and obtain Q prediction accuracies of the Q convergent voltage withstand performance prediction units; based on the coefficient of variation method, configure Q output weights according to the Q prediction accuracies; based on the principle of ensemble learning, fuse the Q output weights and the Q convergent voltage withstand performance prediction units to generate the voltage withstand performance predictor.

[0040] Furthermore, the withstand voltage test module 13 is used to perform the following method: The environmental parameters within the predetermined time zone include a mean temperature, a mean temperature difference, a mean humidity, and a mean humidity difference.

[0041] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0042] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0043] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A resin tank pressure test method based on deep learning, characterized in that: The method comprises: Performing significant feature analysis based on the surface image of the target resin tank, determining multiple abnormal areas, performing defect recognition on the multiple abnormal areas, and constructing a defect distribution map according to the recognition results; Acquire the device attributes of the target resin tank, set retrieval constraints based on the device attributes to retrieve the withstand voltage test records of similar products, acquire a sample data set, and use the sample data set to supervise the deep learning operator to build a withstand voltage performance predictor; The environmental parameters of the target resin tank in a predetermined time zone are collected, the defect distribution map and the environmental parameters are input into the pressure resistance performance predictor, and the predicted maximum pressure bearing value is output as the pressure resistance test result.

2. The resin tank pressure test method based on deep learning according to claim 1 is characterized in that: Based on the surface image of the target resin tank, significant feature analysis is performed to identify multiple abnormal areas, including: Based on low-resolution acquisition of a surface image of a target resin tank, the surface image is input into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit; Performing center-surrounding difference calculation on the multi-scale image set to generate a brightness feature map, a color feature map and a direction feature map, and performing superposition processing on the brightness feature map, the color feature map and the direction feature map to generate an initial salient image; The initial salient image is extracted according to an initial salient feature threshold, and a plurality of salient feature regions satisfying the initial salient feature threshold are obtained and set as the plurality of abnormal regions.

3. The resin tank pressure test method based on deep learning according to claim 2 is characterized in that: Defect identification is performed on the multiple abnormal areas, and a defect distribution map is constructed according to the identification results, including: configuring defect recognition features, wherein the defect recognition features include crack features and deformation features; Retrieve surface defects of the resin tank according to the crack characteristics and deformation characteristics, obtain a sample defect region set, a sample crack feature set and a sample deformation feature set, perform supervised learning on the convolutional neural network until convergence, and obtain a surface defect recognition plug-in; Based on high-resolution acquisition of multiple abnormal images of the multiple abnormal areas, the surface defect recognition plug-in is used to perform defect recognition on the multiple abnormal images, multiple defect features are output, and the defect distribution map is constructed in combination with the positions of multiple abnormal areas.

4. The resin tank pressure test method based on deep learning according to claim 1 is characterized in that: The retrieval constraint conditions are set based on the device attributes to retrieve the withstand voltage test records of similar products and obtain a sample data set, including: Acquire the device attributes of the target resin tank, wherein the device attributes include material attributes, structural attributes, and working time; Expanding the material attribute, structural attribute and working time according to the attribute tolerance interval to obtain a material attribute threshold, a structural attribute threshold and a working time threshold; Configure test constraints, wherein the test constraints include pressure holding time and expected pressure drop; The material property threshold, structural property threshold and working time threshold are used as equipment retrieval constraints, and the pressure holding time and expected pressure drop are used as conditional retrieval constraints. Pressure test records of similar products are retrieved based on big data to obtain a sample defect distribution set, a sample environmental parameter set and a sample maximum pressure value set as the sample data set.

5. The resin tank pressure test method based on deep learning according to claim 4 is characterized in that: The sample data set is used to perform supervised training on the deep learning operator to construct a voltage resistance performance predictor, including: Configure Q deep learning operators, where the deep learning operators include at least BP neural network, support vector machine and random forest, and Q is an integer greater than or equal to 3; Divide the sample data set into Q equal parts, select Q parts with replacement, construct a first training set, and iterate and select Q parts Q times to obtain Q training sets; Taking the sample defect distribution and the sample environmental parameters as input and the sample maximum pressure bearing value as output, the Q training sets are used to perform supervised training on the Q deep learning operators respectively, to obtain Q converged pressure resistance performance prediction units, and the pressure resistance performance predictor is integrated to construct.

6. The resin tank pressure test method based on deep learning according to claim 5 is characterized in that: The pressure resistance performance predictor is constructed in an integrated manner, including: Verifying and obtaining Q prediction accuracies of the Q convergent withstand voltage performance prediction units; Based on the coefficient of variation method, Q output weights are configured according to the Q prediction accuracies; Based on the principle of integrated learning, the withstand voltage performance predictor is generated by fusion according to the Q output weights and the Q converged withstand voltage performance prediction units.

7. The resin tank pressure test method based on deep learning according to claim 1, characterized in that: The environmental parameters within the predetermined time zone include a mean temperature, a mean temperature difference, a mean humidity, and a mean humidity difference.

8. The resin tank pressure test system based on deep learning is characterized by: For implementing the resin tank pressure test method based on deep learning according to any one of claims 1 to 7, the system comprises: Analysis and recognition module: performs significant feature analysis based on the surface image of the target resin tank, determines multiple abnormal areas, performs defect recognition on the multiple abnormal areas, and constructs a defect distribution map based on the recognition results; Model building module: obtaining the device attributes of the target resin tank, setting retrieval constraints based on the device attributes to retrieve the withstand voltage test records of similar products, obtaining a sample data set, and using the sample data set to supervise the deep learning operator to build a withstand voltage performance predictor; Pressure test module: collects the environmental parameters of the target resin tank in a predetermined time zone, inputs the defect distribution map and environmental parameters into the pressure performance predictor, and outputs the predicted maximum pressure value as the pressure test result.

Images