Deep residual shrinkage network with multichannel VMD inputs for noise reduction of micro-thrust measurement

Abstract

Micro-newton thrusters are widely utilized in the field of astronautics. Typically, the precision of micro-newton thrust measurement fundamentally depends on the background noise level. In this research, we introduce the Multichannel Variational Mode Decomposition Input Deep Residual Shrinkage Network (MV-DRSN) to identify the effective signals merged in the background noise. Experimental studies in vacuum were conducted to investigate the effect of noise reduction on MV-DRSN. It is shown that a steady-state signal with 0.1 μN as the minimum change unit can be recovered from the noises with an amplitude of 0.8 μN with an accuracy of 96.7% using MV-DRSN. In addition, the superiority of MV-DRSN over conventional ResNet has been validated, and its effectiveness in practical scenarios is verified. The proposed method has potential for noise reduction of steady-state sensor signals.