Research And Realization Of Aero-engine Fault Diagnosis Expert System

As a high-precision complex mechanical thermal system,aero-engines often work under harsh environments such as high temperature,high pressure,high vibration,etc.And long-term exposure to such environments will easily reduce the service time of the components or even damage the components,resulting in severe aviation accidents.Therefore,the research on the health management and fault diagnosis of aero engines is of great significance to the development of our country’s aviation industry.This thesis is supported by the project of "Aero Engine Data Comprehensive Management,Performance Evaluation and Fault Diagnosis Software System",which is jointly investigated with AVIC Shenyang Engine Design Institute.Considering the disadvantages of traditional expert systems,an expert system for aero-engine fault diagnosis is designed and realized with data-driven algorithms.Specific issues of aero-engine data preprocessing,expert knowledge base,flight parameter prediction and software platform implementation are investigated in detail as follows.In Chapter 2,aiming at the problem of abnormal data and noise in aero-engine data,a heuristic Garrote wavelet threshold denoising data preprocessing method is proposed.It firstly analyzes the characteristics of the aero-engine data,and then the selections of wavelet basis,wavelet threshold function,and wavelet critical threshold are discussed based on wavelet threshold denoising principle.Finally,comparison experiments in simulation are conducted to verify the effectiveness of the proposed method.In Chapter 3,aiming at the difficulty of acquiring knowledge from the knowledge base of traditional expert systems,a semi-automatic knowledge acquisition method based on rough set is proposed.By analyzing the characteristics of aero-engine knowledge and related work of knowledge representation,a knowledge base architecture suitable for the proposed system is designed.Then the reduction process of extracting knowledge rules from fault data using the rough set principle is described.Finally,the process of semi-automatic knowledge extraction is illustrated considering practical applications.In Chapter 4,aiming at the problem of aero-engine parameter time-series prediction,an LSTM based prediction method is proposed,which is implemented and verified with Tensor Flow framework.The results demonstrated that the LSTM neural network is good at predicting aero-engine parameters,which lays a foundation for preventive diagnosis of faults.Finally,considering the above contents and project requirements,the software and database architecture are designed in Chapter 5.Specifically,detailed descriptions and comparisons of the main technologies are elaborated.Finally,a software system for aero-engine fault diagnosis is established with C++ in the Visual Studio 2015,and the test results are demonstrated.