Magnetic signal denoising based on auxiliary sensor array and deep noise reconstruction

Abstract

Magnetic field measurement technologies are extensively applied in nondestructive testing, quality inspection, and fault diagnosis of machines and equipment. Magnetic signals are highly susceptible to noise interference, and how to effectively conduct signal denoising under strong noise disturbance remains a challenge. This study designs a novel auxiliary sensor array comprising center and distributed satellite sensors for magnetic noise reconstruction and signal denoising. The signals of the center and satellite sensors are highly correlative owing to the propagation of the magnetic noise through space. An attention-based long short term memory model is designed and trained to establish the mapping relationship among the center and satellite signals. The noise of the center sensor can then be reconstructed from those of the satellite sensors in the procedure of magnetic field measurement. Finally, the denoised signal of the center sensor is obtained by subtracting the noisy signal with the reconstructed noise. The effectiveness and superiority of the proposed method are validated by experiments and compared with the conventional methods. The designed auxiliary sensor array and algorithm are easily implemented with remarkable robustness, flexibility, and scalability, thereby showing greatly potential use for magnetic signal denoising in a variety of mechatronic systems.