A neural network degradation model for computing and updating residual life distributions

The ability to accurately estimate the residual life of partially degraded components is arguably the most challenging problem in prognostic condition monitoring. This paper focuses on the development of a neural network-based degradation model that utilizes condition-based sensory signals to compute and continuously update residual life distributions of partially degraded components. Initial predicted failure times are estimated through trained neural networks using real-time sensory signals. These estimates are used to derive a prior failure time distribution for the component that is being monitored. Subsequent failure time estimates are then utilized to update the prior distributions using a Bayesian approach. The proposed methodology is tested using real world vibration-based degradation signals from rolling contact thrust bearings. The proposed methodology performed favorably when compared to other reliability-based and statistical-based benchmarks. Note to Practitioners-We propose a neural-network-based degradation model that estimates utilizes real-time sensory signals to estimate the failure time of partially degraded components. The proposed model has been tested and validated on rolling element bearings by using real-time vibration signatures to estimate their failure times. In order to implement, one must first identify the sensory information that is correlated with the underlying degradation process. Next, a sample of components, similar to the one being monitored is tested. Degradation-based sensory information are acquired and stored along with the corresponding operating times of each acquisition. A group of neural networks are trained using supervisory training protocols. Each neural network is trained to identify the degradation pattern of one component in the sample. This is achieved by training the network to identify the operating time corresponding to each sensory signature that is input to the network. Sensory signals from similar components operating in the field are then input to the network model and used to predict the failure time based on the latest degradation state of the component being monitored. Steps 1-4 outline the details of the implementation. The sensory-updating methodology is used to continuously update the failure time predictions as subsequent real-time signals are acquired.