Sparse Bayesian Learning for Gas Path Diagnostics

Abstract

A gas path diagnostic method based on sparse Bayesian learning is presented. Most gas path diagnostic problems present the case where there are fewer measurements than health parameters. In addition, the measurement readings can be faulty themselves and need to be determined, which further increases the number of unknown variables. The number of unknown variables exceeds the number of measurements in gas path diagnostics, making the estimation problem underdetermined. For gradual deterioration, it is common to apply a weighted-least-square algorithm to estimate the component health parameters at the same time sensor errors are being determined. However, this algorithm may underestimate the real problem and attribute parts of it to other component faults for accidental single fault events. The accidental single fault events impact at most one or two component(s). This translates mathematically into the search for a sparse solution. In this paper, we proposed a new gas path diagnostic method based on sparse Bayesian learning favoring sparse solutions for accidental single fault events. The sparse Bayesian learning algorithm is applied to a heavy-duty gas turbine considering component faults and sensor biases to demonstrate its capability and improved performance in gas path diagnostics.