Approaches to Embed a Software-based Adaptive Prognostic Estimation Kernel into a PHM System-onChip

This paper presents considerations and approaches related to embedding a software-based Adaptive Prognostic Estimation kernel into a Prognostic Health Management System-on-Chip (PHM SoC). Ridgetop Group's Adaptive Remaining Useful Life EstimationT (ARULET) software suite includes a two-stage Adaptive Prognostic Estimation kernel comprising robust, accurate, and a fast-converging predictive analytics kernel to produce meaningful prognostic estimates for State-of-Health (SoH), Remaining Useful Life (RUL), and Prognostic Horizon (PH). Existing PHM Standards and practices typically require a combination of hardware, firmware, and software elements for Sensors (S), Data Acquisition (DA), Data Management (DM), State Detection (SD), Health Assessment (HA), Prognostic Assessment (PA), Advisory Generation (AG), and Health Management (HM). For many PHM applications it is impractical to have a single SoC comprising all of the described PHM elements. We believe a more feasible and practical approach is to provide a PHM SoC solution having a multiplicity of sensors, microcontrollers, and a single Adaptive Prognostic Estimation SoC. The approach separates each PHM element into four overlapped groups: (1) S, DA, and DM; (2) DM, SD, and HA; (3) HA, PA, and AG; (4) AG and HM. Group 1 produces Feature Data (FD) that is extracted from ConditionBased Data (CBD) and has a characteristic signature that is highly correlated to degradation leading to failure. Groups 2 and 3 are a realization of group 2 and 3 as a single, two- stage SoC solution. Prognostication of a commercial-off-the-shelf quadcopter is used as an example to demonstrate a design of a PHM SoC in which AG is through light-emitting diodes (LED) and HM is initiated by an operator. A summary, conclusion, and follow-on activity section ends the paper.