A nonlinear probabilistic method and contribution analysis for machine condition monitoring

Health degradation assessment from normal to failure condition of machine part or system is a key element in condition-based maintenance (CBM) system. This paper proposes a generative topographic mapping (GTM) and contribution analysis-based method to perform machine health degradation assessment and monitoring. GTM-based negative likelihood probability (NLLP) is developed to offer a comprehensible indication for quantifying machine health states. A Bayesian-inference-based probability (BIP) calculation method is further developed to analyze the failure probability of the monitored machine or component. A variable replacing-based contribution analysis method is developed to discover potential features that are effective for the detection and assessment of machine health degradation in its whole life. The experimental results on a turbine engine simulation system and a bearing testbed illustrate plausibility and effectiveness of the proposed methods.

► A generative topographic mapping-based method is proposed for machine health monitoring.
► Negative likelihood probability is capable of quantifying machine health states.
► A probabilistic indication is developed to analyze machine failure.
► A contribution analysis method can discover effective features.
► The results on machine testbed illustrate the effectiveness of the proposed method.