Research On Analytical Redundancy For Aero-engine Based On Onboard Model

As a measuring element, sensor plays an important role in aero-engine control system. Theaccuracy of sensor’s sign directly influences the safety and stability of aero-engine control system.Thus, in order to promote reliability of aero-engine control system it is necessary to establish a falsedetection, isolation and reconstruction system for sensors work in aero-engine control system.This thesis achieves a series of researches which, takes the support vector regression(SVR)owning statistical learning foundation and excellent generalization performance, focus on senor faultdetecting and signal reconstructing.Firstly, recursive reduced least squares support vector regression(RR-LSSVR) has been put intoapplication in aero-engine senor faults detecting and signal reconstructing. And simulations of bothsingle sensor fault and bi-sensors faults had been conducted and generated desirable results.Secondly, the senor fault detecting and signal reconstructing technique has applied onaero-engine in degeneration condition. Since RR-LSSVR is an offline study algorithm that can onlywork with offline data, author, taking advantage of similarity principle, designs a simplified modelbased on propulsion system matrix(PSM) to solve the signal deviation when aero-engine begindegeneration. The signal deviation is used to offset the estimation of RR-LSSVR signal.Thirdly, according to the complexity of the method above, author developed a new studyalgorithm-online sliding window parsimonious least squares support vector regression (OSP-LSSVR),and realized it in VC to confirm feasibility. OSP-LSSVR is an online study algorithm that can updatesupport vector with respect to aero-engine state. This method overcomes the false of offline studyalgorithm, which cannot update model.Finally, based on OSP-LSSVR, author designed aero-engine fault detecting and signalreconstructing system. This system, utilizing the most valid data to update predict model and estimateaero-engine’s dynamic outputs, accomplished single sensor fault detecting and signal reconstructing.