关键词: 管道, 金属磁记忆, 模糊核函数, 支持向量机, 缺陷, 识别

Abstract:

In order to solve the limitation problem in identifying pipeline defects by the metal magnetic memory (MMM) method, the present study brought forward a method for defect recognition by Support Vector Machine (SVM) based on the fuzzy kernel function via constructing maximum-minimum similarities of trapezium fuzzy data. According to the construction of recognition functions, the pipeline status could be classified into three levels, including stress concentration,micro defect and macro defect. In the comparison of features between the stress concentration area and the micro-cracking defect, the input vectors for SVM were formed by five feature vectors, including the energy increment of wavelet packet frequency bands, the modified Fourier coefficient, the peak-to-peak value of area signals, the tangential gradient and normal gradient of signal detection. After de-noising of MMM signals, the feature extraction and defect recognition were completed by virtue of experiments. Compared with recognition results of traditional liner kernel and polynomial kernel functions, the new approach can effectively identify cracking defects of pipelines, providing a new method for the pipeline defect recognition with MMM testing.

Key words: pipeline, metal magnetic memory, fuzzy kernel function, Support Vector Machine (SVM), defect, recognition