Andreas Poncet

Probabilistic Diagnosis of Thermal Plants Condition

A thermal plant (such as a gas turbine) is a system subject to considerable stress, which causes creep and fatigue. The condition of each component-and therefore the overall plant performance-invariably degrades with operation time, until it is partially restored by some maintenance action. We present a pro-active solution for localizing and tracking the origin of the plant faults (compressor fouling, turbine erosion, etc.). The diagnosis is obtained by analyzing the observed symptoms (performance degradation in efficiency, flow capacity, etc.). The solution uses a probabilistic approach to compute on-line the probabilities of the potential origins of the faults given the latest process information. The updated probabilities encapsulate all the condition information and quantify how likely each fault origin is, thereby providing crucial input to risk-based planning of machine operation and maintenance