Valves are an essential component of systems in many domains, including ground support equipment for space and aeronautics. Due to the large number of valves and their high usage frequency, valves are one of the most commonly failed components in such systems. The loss of a valve can cause a loss of certain system functions or result in degraded performance that may fail to meet system requirements. Therefore, determining valve health and predicting remaining useful life can provide much value to a maintenance program.
We take a physics-based approach to the problem of valve prognostics. We develop physics models of valves that incorporate damage progression models. We estimate the damage state and the parameters that define its evolution, and use this to predict remaining useful life using the physics-based model. We implement this in a robust model-based, statistical framework. The architecture is outlined in the following figure.
The system is provided with inputs uk and provides measured outputs yk. Prognostics may begin at t=0, with the damage estimation module determining estimates of the states and unknown parameters, represented as a probability distribution. The prediction module uses the joint state-parameter distribution, along with hypothesized future inputs, to compute EOL and RUL as probability distributions at desired prediction times. In parallel, a fault detection, isolation, and identification (FDII) module may be used to determine which damage mechanisms are active, represented as a fault set F. The damage estimation module may then use this result to limit the space of parameters that must be estimated. Alternatively, prognostics may begin only when diagnostics has completed.