关键词: 模拟电路, 实时可靠度预测, 噪声辅助技术, 现场数据, 粒子滤波, 信息融合

Abstract: Signal processing and information fusion technologies are the key to real-time reliability prediction. The traditional real-time reliability analysis methods are based on specific random process and pro- bability distribution. A new real-time reliability prediction method is proposed. The proposed method uses a Kalman filter-based noise-assisted technique to calculate the fault indicators that characterize system performance degradation trend. On this basis, the particle filter technology is used to extrapolate the pseudo-failure performance of circuit system, and then the Bayesian inference as an information fusion method is introduced to update the time-varying parameters of performance distribution, thus predicting the real-time reliability of the circuit. The failure physical model of an embedded planar capacitor based on the real acceleration degradation experiment is introduced, and the effectiveness of this real-time prediction method combined with noise-assisted technology and on-site data is verified by using real data instead of ideal simulation hypothetical data. The result shows that the more the on-site data information is， the higher the prediction accuracy of circuit reliability is.Key

Key words: analogcircuit, realtimeprediction, noise-assistedtechnique, on-sitedata, particlefilter, informationfusion