Quantification of crack in-pipe detection signals in steel pipelines based on differential eddy current detection technique with incremental permeability extraction

Quantifying internal crack detection signals in steel pipelines requires a high level of expertise from researchers. The traditional signal features cannot fully characterize the actual signals. This paper uses steel plate specimens to simulate steel pipelines for crack detection signal quantification research. Firstly, a differential eddy current test platform based on incremental permeability extraction is built, and the self-developed eddy current detection technology is used to detect the crack defects and form a quantitative database of crack defects; then, end-to-end crack detection signal quantification models DRSN1d and DRSN2d are established; finally, noise is added to the database to compare the traditional model with DRSN2d. The results show that the constructed deep learning model achieves an average crack depth inversion accuracy of ±0.018mm and an average crack width inversion accuracy of ±0.015mm, which meets the industrial requirements. The deep learning quantization model outperforms the traditional machine learning model on high-noise data, and the features formed from the end-to-end model training are also better than the traditional features.