SIGNAL AND IMAGE PROCESSING FOR TURBOMACHINERY MONITORING

Oil&Gas industry requires reliable turbomachinery and industry 4.0 relies on sensing and data processing to make this possible. The collected data and its processing can predict failure and improve the health of turbomachines, resulting in a reduction of maintenance cost and increase in performance. This thesis includes two parts: the first one is about processing of dynamic pressure signals to detect stall and surge in centrifugal compressors; the second one is about gas turbine rotor blade temperature estimation from infrared image. Stall and surge detection Pressure signals from a model test of a centrifugal compressor stage are recorded and processed. Relation between instabilities, signal unsteadiness and spectra is investigated. A novel approach, based on deep learning, was proposed and tested. Gas turbine blade temperature estimation Accurate mesurements of the surface temperature profile inside a gas turbine from infrared thermographic images requires non trivial models of infrared emission and reflection. In this section, a novel approach for achieving this goal is presented, which includes a calibration algorithm using reference thermocouples in some positions on the stator.