Using HPC in Gas Turbines Blade Fault Diagnosis

Parallelization of two approaches used for identification of faults in blades of a gas turbine is presented. The first approach, termed as direct simulation, is aiming to populate a fault diagnosis database allowing early identification of faults. The second approach, termed as the inverse one, gives a more focused solution to fault identification, by using the direct approach in an iterative way which permits the estimation of the blade geometry alterations. By using state of the art parallel tools such as ScaLAPACK library and exploiting inherent coarse-grained parallelism in calculating the elements of the Jacobian needed in the iterative method encouraging speedups have been obtained. The test cases presented include theoretically produced fault signals as well as experimental cases, where actual measurement data are shown to produce in quasi real time the geometrical deformations which existed in the test engine.