Research On Weld Defect Recognition Based On Fuzzy Set And Neural Networks

Weld defect recognition is a key technology in the process of welding quality detection.Detecting the weld defect effectively and guaranteeing the quality of weld are of greatsignificance for oilfield safely production. The recognition rate of weld defect is directlyrelated to the quality of weldment and has a direct affection to the safety of oilfieldconstruction.Nowadays the problem in weld defect defection of oil pipeline mainly includes thefollowing three aspects. Firstly, the weld defect detection mode of pipeline mainly relys onthe mannul detection, which results in that the automation degree and the detection efficiencyare not high; Secondly, the detection technology mainly relys on the experience of theprofessional technician. When weld film ratings are refered, the subjectivity and factitiousexperience of the professional technician must be accompanied in the process of weldevaluation, which is non-standardized, fuzzy and easy to cause the evaluation resultno-objectivity. Thirdly, the proportion of the defect images in the film is very small, which iseasy to lead to that some defects are omited. Aiming at that the weld defect recognition isnon-standard and no-normalized, we need more reasonable method for weld defectrecognition.For the problems of the conventional weld film rating evaluation, and according to theactual situation of oil fields, this thesis integrated the fuzzy theory and the artificial neuralnetworks with the weld defect recognition process, and conducted an in-depth study about theweld defect recognition. The main research contents are as follows.1. The pre-processing of weld image is the foundation of defect recognition. Accordingto the characteristics of weld image, this thesis analyzed the noise of the image and proposedan improved median filtering method for image noise reduction. Further more, the fuzzyenhancement was used for enhance the useful information in the images.2. As the foundation of feature extraction, the characteristics of weld defect image werestudied. By analyzing the distribution of gray curve, the positions of the weld defects werelocated at first. Then extract the effective features of the defects and apply fuzzy theory todescribe the category of the feature parameters, which would be taken as the condition ofdefect recognition.3. In view of the inherent ambiguity of weld defect characteristics, the fuzzy theorybased neural network pattern recognition algorithm was studied. The fuzzy theory was applied to set up the defect training sample data for the network, instead of the traditional pattern ofinput variables, in order to improve the mapping ability of the network. Combining with theactual situation of oilfields, the method proposed was applied to the weld defect recognition.4. According to the research contents above, the weld defect recognition system wasdeveloped. The actual application showed that the weld defect recognition based on fuzzy setand neural network was easy to operate and fast in execution. As the final results anacceptable recognition rate and efficiency were obtained.