Research On Corrosion Of Metals In Nature Environment Based On Image Recognition

Corrosion morphology image is one of the most important features toevaluate the material's corrosivity. In this paper, the digital image processingmethods were used in corrosion electrochemistry research. According to thecharacters of corrosion in the natural environment, the relations between thefeatures of corrosion morphology images and the sample corrosion data werestudied. The main conclusions are as follows: The corrosion morphology image processing system were established whichwas suited for sample exposed to natural environment. The suitable parameter tocapture images and corresponding pre-processing methods were discussed anddetermined. The results showed that the median filter was an efficient method toeliminate the noise signal that produced during image capturing. While the fuzzyenhancement method was very effective to make the corrosion morphologyfeatures showed clearly. Furthermore, a new image threshold method was given. Corrosion morphology images of samples exposed to marine were capturedby using the established image processing system and then the fractal theory wasused to analyze the images. The fractal dimension, generalized dimension, areafactor and laucinarity of improved metallic corrosion images were calculated.Taking the obtained features and their corrosion modality as the knowledge base,the diagnosing system identifying corrosion modality was established according tothe theory of fuzzy pattern recognition. The morphology of corrosion images canbe identified by acquired fractal characters. Based on wavelet image analysis and modern mathematics methods, a novelanalytical method was presented to study the early stages of atmospheric corrosionof nonferrous metal. The principle of the method involves exposing samples to theatmosphere and then capturing digital images of the surface morphology of theexposed samples. The improved images were decomposed by multi-resolutionwavelet transformation. Every subimage contained different information of aspecific scale and orientation. The energy values were obtained as feature vectors.The canonical correlation analysis was used to gain the correlative coefficientbetween the features and the corrosion loss. The weighted features were analyzedto obtain the material weight loss due to corrosion by using an Artificial NeuralNetwork (ANN) model. The results indicated that the computational methods IIdeveloped for corrosion analysis could provide reasonable results for estimatingmaterial weight loss due to corrosion morphology image such as pure zinc exposedin natural environment and accelerated aluminum alloy corrosion samples. Consultation and forecast diagnosis system of metal exposed in marine andatmospheric environment had been developed by using the object-orientedprogramming language, Visual Basic respectively. The system both comprisedthree major modules: database and management system, corrosion prediction andcorrosion morphology analysis. In the first part, the common data managementfunction such as data input, data modify, data delete and data refresh werediscussed. There are two different methods used in the corrosion predictionmodule, ANN model and gray model. Using the ANN method, the corrosion depthof new materials in different marine environment can be predicted, while the graymodel method can give the corrosion rate of materials in the future. Taking thegray distribution or fractal characters of metallic samples and their corrosionmorphology as the knowledge base respectively, the diagnosing system identifyingcorrosion morphology was established according to the theory of fuzzy patternrecognition. The modules of image processing can calculate the fractal features ofimages and decompose images by wavelet analysis. A simple and convenient system to capture real-time corrosion morphologyimages in-situ of stainless steel sample exposed in NaCl solution was presented.The electroche...