Ensemble fault detection based on magnetic flux leakage images with noise robustness for steel wire ropes

Abstract

Localization of local flaws is critical to the magnetic flux leakage inspection of steel wire ropes. Existing studies mainly focus on improving the denoising ability to improve LF localization accuracy, but encounter limitations caused by inconsistencies in the noise features. In contrast, the localization technique still lacks enough attention. Second stage noise (SSN), which refers to noise after the denoising process, is the primary cause of inaccurate localization. To address this challenge, this paper adopted two ideas: reducing the SSN through a more effective denoising process and decreasing the false detection of SSN by using ensemble detection in the localization process. The proposed denoising method applies dual-dimensional template matching to enhance the LF signal and suppress the resulting SSN regardless of the noise feature. In the localization stage, two localization methods were developed and combined to obtain united results, bringing improved robustness against the SSN. The results demonstrate that the proposed denoising method achieves a significant reduction in SSN. The proposed ensemble detection achieves an F1 score of 0.9575, which is much higher than existing methods.