Enhanced signal processing for MFL-based steel pipe defect detection combining ICEEMDAN with multi-feature-GA-WTD method

Abstract

The magnetic flux leakage (MFL) method has been widely utilized for steel pipe inspection, demonstrating its effectiveness in examining continuous ferromagnetic structures. However, MFL signals are often contaminated by various noise sources in field environments, complicating defect detection and analysis. To address this challenge, this study proposes an advanced signal processing method combining Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and a multi-feature-based Intrinsic Mode Function (IMF) selection strategy. Key features such as energy entropy, permutation entropy, and harmonic ratio were used to evaluate IMFs, with their relative importance optimized using a Genetic Algorithm (GA). Selected IMFs were further refined through Wavelet Threshold Denoising (WTD) to suppress residual noise and enhance signal fidelity. The proposed method was experimentally validated using a portable MFL sensor device designed for high usability in field applications. Achieving a signal-to-noise ratio (SNR) of 35.52 dB and a correlation coefficient (CC) of 0.94, the method demonstrated precise detection of small-scale defects in steel pipes, even under significant noise interference. These results highlight the method’s ability to enhance defect detection accuracy by suppressing noise while preserving critical defect-related signals, paving the way for more efficient and reliable steel pipe inspections in real-site applications.