Fault Detection And Diagnosis For Gas Turbine Based On Information Entropy Models

Nowadays; with evolving technology, humans endeavor to cope with somehindrances to efficient and accurate production. This ongoing innovation has reachedthe maintenance field whereby, gigantic and complex machines can be monitored todiscover trips existing within them and predict possible failures prior to their occurrence.Particularly, it becomes necessary for power stations that can be thought of as the coreor driving hub of several processes. Hence, the gas turbine fault detection and diagnosisbecame a recursive concern to avoid its incurred perturbations of the production andoptimally enhance its service life. The combination of damaging factors to which a gasturbine must endure so as to carry out its function, such as prolonged service overelevated temperatures, high spinning speed, soaring pressures and contagiousenvironment; make it prone to increasing faults rate. Such faults; depending on theirgravity, they may spawn security incidents, unit malfunctions and shutdowns whichimpede or halt production with accompanied economic deficiencies.Fault detection and diagnosis of a gas turbine covers a series of health monitoringapproaches. Here, we emphasized on Shannon information theory by-products. First, weapplied an early developed statistic approach, based on mutual information namely“Maximal Information Coefficient (MIC)”, to come up with a stable pattern, which is aroughly constant snapshot of interrelationships between all parametric sensors plantedin different corners of the turbine. That unaltered pattern reflects usual flawlesscondition that should be based on to decide the fault occurrence upon unconformity offurther measurements. Secondly, we conducted a survey on exhaust gas temperaturesvia information entropy under its Kernel-extended version to select a subset of moreinformative features constructed to allude fault hints and C4.5decision tree to flag datasamples according to fault classes by means of selected features’ partitioning rules.The objective of this work was two-fold; primarily, to rudimentarily introduceMIC in abnormality awareness of gas turbine system via significantly associated sensorsand secondly, to combine exhaust temperature information entropy with kernel spacesand decision tree concepts in order to analyze its extended performance on gas turbinediagnosis. Data analyses revealed that outcomes from those two Shannon informationtheory derivatives are trustworthy and deserve more furtherance focus for the gasturbine to be safely and effectively operated.