Detecting loose particle signals in multichannel recordings with transductive confidence predictor

Abstract

Loose particles left inside sealed electronic devices are one of the main factors affecting the reliability of the entire military and aerospace system. The main purpose of loose particle detection is to discriminate between the particle signals and inherent mechanical noise (component signals). To address this problem, a multichannel system for loose particle detection is presented in this paper. A synchronization-based method is proposed, which makes good use of multichannel information. The synchronization method takes variance of correlation coefficients of multichannel synchronized pulses as a detection feature and then employs a transductive confidence predictor to produce confidence measures. The experimental results show the effectiveness of the proposed method.