Study On Normal Behavior Model For Aero-engine Fuel Distributor Under Variable Operating Conditions

With the rapid development of the aviation industry,the safety and reliability of aircraft has received unprecedented concern.Aero-engine,the power source of aircraft,is one of their core components,whose condition is crucial to the safety and reliability of aircraft flight.As an important part of aero-engine control,the performance of fuel system has a decisive impact the safety and reliability of aircraft flight.Fuel distributor is a core component of aero-engine fuel system.Its failure may cause an anomaly of the entire fuel system and even affect the flight safety of aircraft.Hence,this thesis conducts research as follows:(1)Focusing on the 1000L/h flow grade fuel distributor for a certain type of aero-engine,this thesis analyzes its structure and working principle,constructs component models of the aero-engine fuel distributor,and builds the experiment table.Based on the component models,the functional module models of the fuel distributor are established.On account of the constructed component models and functional module models,a complete fuel distributor model is built and analyzed.(2)A normal behavior model of aero-engine fuel distributor based on Convolutional Neural Networks-Gated Recurrent Unit(CNN-GRU)is proposed.Taking multiple sensor data as input,the model extracts data features by Convolutional Neural Networks(CNN),and models time series data using Gated Recurrent Unit(GRU)model.Regarding model parameter settings,the thesis explores the effects of different convolutional kernel sizes,GRU unit numbers and optimization parameters on model performance.According to the experimental results,compared with CNN,Long and Short-Term Memory(LSTM),Attention-LSTM and GRU network models,the CNN-GRU model shows greater accuracy and stability in the normal behavior of the aero-engine fuel distributor.The research has practical value for the analysis of normal behavior model of aero-engine fuel distributors.(3)Considering variable operating conditions,a normal behavior model of fuel distributors based on the combination of transfer learning and CNN-GRU is proposed.By using pre-training models for transfer learning and fine-tuning it on target tasks,the model’s generalization ability is improved.As the experimental results suggest,the model performs well under variable operating conditions,providing an effective solution for research on normal behavior model of fuel distributors.With remarkable practicality and usability,it can better meet engineering needs in practical applications,reduce maintenance costs and improve safety performance.An analytical software for normal behavior model is then designed to facilitate research on normal behavior model of aero-engine fuel distributors.This study provides an effective scheme for the normal behavior model of fuel dispenser,which is helpful to improve the safety and reliability of aircraft flight.