Data Driven Fault Diagnosis For Fuel Cell Power Plants
Publication Date: 2004-May-25
The IP.com Prior Art Database
Data Driven Fault Diagnosis For Fuel Cell Power Plants, Data driven sensor and actuator fault diagnosis for fuel cell power plants.
Data-Driven Fault Diagnosis for Fuel Cell Power Plants
power plants are subject to a number of types of operational faults, including
sensor and actuator failures, catalyst degradation, heat exchanger fouling,
deviations in fuel composition or load profile. These faults degrade power
plant performance and often result in unscheduled power plant shutdowns.
As a solution, data from previous occurrences of a given fault is used to diagnose new faults using multivariate statistical techniques. This is feasible because most faults change the multivariate distribution of power plant operating data (see first figure). One statistical "model" is generated for each type of fault as manifested in a given power plant sequential control state (for example, "P" or "R" state). New data is compared to each model and the best fit is chosen using the Fisher Discriminant Analysis technique.
One type of presently existing control software has fault diagnosis capability only in the case of gross sensor faults (when the electrical signal is out of range). Field support engineers rely on expert knowledge to manually diagnose faults. The described technique could be implemented as an onboard or remote automatic diagnostic algorithm. The use of Fisher Discriminant Analysis dramatically reduces computational requirements and reduces the effects of noise. Building one fault model for each power plant operating state permits separation of the effects of the fault on the data distribution from effects rising from changes in op...