Search

CN-121978099-A - Copper pipe atmospheric corrosion traceability visual diagnosis method based on interpretable AI

CN121978099ACN 121978099 ACN121978099 ACN 121978099ACN-121978099-A

Abstract

The invention relates to the technical field of material corrosion intelligent monitoring and artificial intelligence intersection, and particularly provides an atmospheric corrosion traceability visual diagnosis method for a copper pipe based on an interpretable AI, which realizes rapid, in-situ and high-precision regional classification of corrosion morphology of the copper pipe in different atmospheric environments and further reveals physical and chemical association between visual characteristics and microscopic corrosion mechanisms on which classification results depend, thereby providing a set of credible and interpretable scientific tool for intelligent corrosion diagnosis and traceability of engineering structures. The method has high precision and strong generalization capability, overcomes the dependence of the traditional method on laboratory standard images, and improves the credibility of the technology. The key visual features identified by machine learning are quantitatively associated with microscopic mechanisms such as chemical components, distribution heterogeneity and the like of corrosion products for the first time systematically, so that a new paradigm of reverse deducing corrosion environment and process through macroscopic images is realized. The method is suitable for rapid in-situ corrosion state diagnosis and regional tracing of engineering sites.

Inventors

  • WANG CHANGGANG

Assignees

  • 中国科学院金属研究所

Dates

Publication Date
20260505
Application Date
20251230

Claims (9)

  1. 1. The copper pipe atmospheric corrosion traceability visual diagnosis method based on the interpretable AI is characterized by comprising the following steps of: S1, acquiring a multi-scene real image, namely placing a copper pipe sample in at least two different typical atmospheric environments for outdoor exposure, and acquiring macroscopic corrosion images of the copper pipe surface after different exposure periods on site by using mobile terminal equipment to form an original image data set; s2, preprocessing the original image to eliminate background interference and accurately position a copper pipe main body area, and outputting a standardized target image; S3, extracting multi-dimensional visual features, namely extracting color statistical features and texture features from the standardized target image to form initial feature vectors; S4, feature aggregation and model construction, namely dividing the standardized target image into super pixels, calculating the initial feature vector in each super pixel, and carrying out feature statistics aggregation to generate a global feature vector; And S5, associating decision interpretation with a mechanism, namely carrying out interpretability analysis on the trained machine learning classification model, identifying key visual features with high contribution to regional classification, associating the key visual features with a microscopic mechanism obtained through a corrosion product characterization technology, and constructing an interpretable evidence chain from macroscopic image features to a microscopic corrosion mechanism.
  2. 2. The visual diagnosis method for tracing atmospheric corrosion of copper pipe based on interpretable AI of claim 1, wherein the robust image preprocessing of step S2 specifically comprises: S2.1, uniformly collecting the background, namely placing a copper pipe sample on a uniform light background for shooting to obtain an original RGB image; S2.2, primarily segmenting a foreground target, namely converting an original RGB image into a gray image, obtaining a binary image by adopting a self-adaptive threshold segmentation algorithm, and primarily separating a foreground from a background; S2.3, morphological optimization and accurate positioning, namely performing morphological operation on the binary image to optimize a target area, and positioning a communication area with the largest area by using a contour detection algorithm as an accurate mask of the copper pipe main body; s2.4, target extraction and size standardization, namely cutting out a copper pipe region from an original RGB image by applying the accurate mask, and scaling the region image to a preset size to obtain the standardized target image.
  3. 3. The visual diagnostic method for copper pipe atmospheric corrosion traceability based on interpretable AI according to claim 1, wherein the color statistics in step S3 comprise pixel value means, standard deviation and skewness of three color channels of red (R), green (G) and blue (B), and the texture features comprise image entropy, texture descriptors extracted based on gray level co-occurrence matrix and texture statistics histograms extracted based on local binary pattern.
  4. 4. The visual diagnostic method for tracing atmospheric corrosion of copper pipe based on interpretable AI of claim 1, wherein said feature statistics are aggregated in step S4 by computing a mean, standard deviation, minimum and maximum of said initial feature vector along a feature dimension and stitching these statistics to generate said global feature vector.
  5. 5. The visual diagnostic method for copper pipe atmospheric corrosion traceability based on interpretable AI according to claim 1, wherein the interpretable machine learning classification model in step S4 is one or more of a support vector machine, a random forest, a decision tree, a K-nearest neighbor algorithm, a logistic regression or a multi-layer perceptron.
  6. 6. The visual diagnostic method for the traceable atmospheric corrosion of copper pipes based on the interpretable AI of claim 1, wherein the interpretable analysis of step S5 comprises at least one of the following techniques: s5.1, ranking the importance of the features, namely quantifying and ranking the contribution degree of each input feature to model classification decisions; S5.2 visual attention visualization, namely generating and displaying a thermodynamic diagram of the image region concerned by the characterization model decision; s5.3, feature correlation analysis, namely calculating and analyzing statistical correlation among different visual features.
  7. 7. The AI-based visual inspection and characterization method of atmospheric corrosion of copper tubing as set forth in claim 1, wherein the corrosion product characterization technique of step S5 includes one or more of X-ray diffraction analysis, scanning electron microscopy and spectroscopy, optical microscopy.
  8. 8. The AI-based copper pipe atmospheric corrosion traceable visual diagnosis method according to any one of claims 1-7, wherein the mobile terminal device is a commercial smart phone.
  9. 9. The interpretable AI-based copper pipe atmospheric corrosion traceable visual diagnostic method of any of claims 1-7, wherein the typical atmospheric environment includes a marine atmospheric environment, an industrial-marine atmospheric environment, and a city atmospheric environment.

Description

Copper pipe atmospheric corrosion traceability visual diagnosis method based on interpretable AI Technical Field The invention relates to the technical field of material corrosion intelligent monitoring and artificial intelligence crossing, in particular to a copper pipe atmospheric corrosion traceability visual diagnosis method based on an interpretable AI. Background Copper and copper alloys are used as essential basic materials in the fields of construction, electronics, energy sources and the like, and corrosion of the copper and copper alloys in the atmospheric environment can directly influence structural safety and service life. Traditional corrosion assessment methods rely mainly on corrosion coupon weightlessness, electrochemical testing techniques, and microstructure characterization means in laboratory environments. Although the method has higher precision, the method has obvious limitations that the corrosion film-hanging method has long period and cannot reflect real-time state, the electrochemical test is often separated from the actual atmosphere environment and is difficult to reproduce a complex atmosphere-corrosion interface process, and microscopic analysis means such as a scanning electron microscope, X-ray diffraction and the like need complex sample preparation and expensive equipment and do not have the capability of on-site rapid detection. Therefore, the existing methods are difficult to meet the urgent need for rapid, in-situ, non-destructive identification of corrosion conditions in an engineering site. In recent years, with the rapid development of artificial intelligence technology, especially the successful application of computer vision and machine learning in the field of image recognition, a new technical path is provided for corrosion monitoring. Researchers try to convert corrosion morphology diagnosis into classification and identification problems based on macroscopic images, and high-efficiency and objective intelligent assessment is realized by establishing a mapping relation between image features and corrosion states. The method mainly forms two types of technical paradigms in the field at present, namely firstly, quantitative feature extraction is carried out on corrosion morphology in a controlled laboratory environment based on a traditional image processing algorithm, for example, a statistical model between corrosion degree and image features is constructed by adopting methods such as gray level co-occurrence matrix, local binary pattern, image entropy and the like, and secondly, the application of deep learning, especially convolutional neural network in corrosion image analysis is explored, and tasks such as corrosion grade classification, thickness prediction and the like are realized by utilizing an end-to-end model. Although these studies show good application prospects, there are significant drawbacks when oriented to real complex industrial scenes. Firstly, most of the existing models are constructed based on laboratory accelerated corrosion tests or standardized imaging conditions, the difference between image data and corrosion morphology formed by real outdoor atmospheric exposure is obvious, and interference factors such as uneven illumination, changeable angles, disordered background and the like in field shooting are not considered, so that the models are weak in generalization capability and poor in robustness in practical application. In particular, most of researches are based on planar samples, imaging conditions are relatively standard, and three-dimensional structures (such as pipelines and sectional materials) widely existing in engineering are difficult to directly migrate and adapt due to interaction between curved surface forms and complex illumination, and the imaging difficulty and deformation are high. Secondly, the existing method, particularly the deep learning model, is often in a 'black box' state, and can realize higher classification accuracy, but cannot explain the physical basis of model decision, and the lack of effective association from macroscopic image features to microscopic corrosion mechanisms leads to the lack of reliable scientific explanation of model prediction results, so that popularization and application of the technology in high-reliability requirement scenes such as corrosion tracing, life prediction and the like are restricted. Therefore, development of an intelligent analysis method for corrosion images based on real scene data, oriented to three-dimensional structures, and having high classification performance and strong interpretability is urgent. The method can be suitable for complex field imaging conditions, stable and robust visual characteristics are extracted, and an internal correlation between model decision and corrosion product chemical composition, spatial distribution and morphology complexity is revealed by means of interpretable machine learning technology, so that a trusted bridge from macro