Effective Independence-Based Sensor Network Design for Health Assessment of Engineered Systems

Abstract

The failure of an engineered system not only results in an enormous property loss, but also causes a substantial societal loss. The discipline of prognostics and health management (PHM) recently has received great attention as a solution to prevent unexpected failures of engineered systems. The goal of PHM is to detect anomaly states, to predict potential failures of a system, and to plan an optimal management schedule. PHM is composed of five essential functions: 1) sensing, 2) reasoning, 3) diagnostics, 4) prognostics, and 5) management. The sensing function, in which sensory data is acquired from the system of interest, is a core element needed for cost-effective execution of PHM. The success of the remaining functions in PHM highly depends on the quality of the data obtained by the sensing function. The research described herein describes the investigation of two original ideas of optimal sensor placement (OSP) for the PHM sensing function. These ideas are aimed to enable cost-effective and robust sensor data acquisition from the system. The first idea is a stochastic effective independence (EFI) method, referred to as an energy-based stochastic EFI method

the proposed method overcomes the drawbacks of existing OSP methods in the sensing function. In Research Thrust 1, the stochastic sensor network design is proposed. It takes the uncertainty of the system into consideration to give more accurate representation of the system than the deterministic sensor network design in the mean sense. Also, the explicit form of the proposed method has the benefit of lower computational requirements, as compared to the sampling-based stochastic approach. In Research Thrust 2, a robust sensor network design that considers the latent failure of the sensor is introduced. The proposed robust sensor network is designed to tolerate the partial failure of the sensor

thus, it contributes to the safety of the sensor network. The proposed method is validated to have accuracy that is comparable to the optimal sensor network design in normal conditions, and higher accuracy for situations in which there is a partial failure of the given sensor network.