关键词: 小管径弯头, 缺陷量化, 漏磁检测, 粒子群优化, 神经网络模型, 多模态特征

Abstract: When the magnetic flux leakage (MFL) internal detector passes through the elbow of the pipe with small diameter, the lifting-off value of the detector changes and the detected MFL signal is quite different from that of the straight pipe. Therefore, it is necessary to study the applicability of the traditional quantization method. Based on the MFL simulation model of straight pipe and elbow, a set of three-axis MFL signals of defects with different specifications are obtained. Then, a multiple hybrid feature dataset is constructed and used as input to fuse the PSO-ELMAN model. After that, the quantization methods for elbows and straight pipes are constructed separately to establish the three-dimensional mapping relationship between magnetic flux leakage signal and defect size. The research results indicate that the quantization accuracy of PSO-ELMAN model in length, depth and width of elbow defect reached 98.6%, 98.3% and 97.9% respectively, which is better than the results obtained from the adoption of BP neural network. This study can provide some basis for quantification of elbow defects of pipes with small diameter.

Key words: pipe elbow with small diameter, defect quantification, magnetic flux leakage detection, particle swarm optimization, neural network model, multimodal features