Degradation modeling and monitoring of truncated degradation signals

Advancements in condition monitoring techniques have facilitated the utilization of sensor technology for predicting failures of engineering systems. Within this context, failure is defined as the point where a sensor-based degradation signal reaches a pre-specified failure threshold. Parametric degradation models rely on complete signals to estimate the parametric functional form and do not perform well with sparse historical data. On the other hand, non-parametric models that address the challenges of data sparsity usually assume that signal observations can be made beyond the failure threshold. Unfortunately, in most applications, degradation signals can only be observed up to the failure threshold resulting in what this article refers to as truncated degradation signals. This article combines a non-parametric degradation modeling framework with a signal transformation procedure, allowing different types of truncated degradation signals to be characterized. This article considers (i) complete signals that result from constant monitoring of a system up to its failure; (ii) sparse signals resulting from sparse observations; and (iii) fragmented signals that result from dense observations over disjoint time intervals. The goal is to estimate and update the residual life distributions of partially degraded systems using in situ signal observations. It is showed that the proposed model outperforms existing models for all three signal types.