Micro Gas Turbine fault detection and isolation with a combination of Artificial Neural Network and off-design performance analysis

Recently Micro Gas Turbines deployment in smart grids is growing, which increases engine load change during its lifecycle; consequently, lifetime reduces faster, and diagnostics is more highlighted. Engine complex dynamic limits studies to only system-level diagnostics at the full-load operation, whereas measurements’ uncertainties and gradual degradation are often neglected. This study proposes a diagnostics scheme to detect and isolate faults in a wide range of part loads and degradation in the presence of uncertainties. An off-design model of Micro Gas Turbine is developed, and uncertainties are considered for preparing a comprehensive training database. An artificial Neural Network is employed to understand the nonlinear correlation between measurements and components’ health state. Different sets of measurements are tested to minimize the number of required measurements. It demonstrates power, and shaft speed measuring is necessary for accurate detection. Moreover, to present appropriate fault isolation using power, shaft speed, exhaust temperature, compressor discharge pressure, and temperature are required. The study indicates diagnostics performance is not sensitive to load variety that exists in the database but shows considerable sensitivity to degradation severities variety. Noise level effects on diagnostics performance are investigated to evaluate the importance of sensors’ uncertainty considerations. It reveals that detection is not so sensitive to the noise level. However, isolation shows more sensitivity. The result demonstrates the high capability of the proposed approach for establishing system level and component level diagnostics in a broad operating range and dealing with measurements’ uncertainties engine high complexity and nonlinearity.