Engine Cylinder Pressure Identification And Misfire Diagnosis

The main research object is the well-known and widely studied cylinder pressure and misfire diagnosis.Cylinder pressure is the most directly variable in monitoring the engine inner status.There is also a great impact in crankshaft speed when misfire fault is occurred.Crankshaft speed is not only associated with cylinder pressure,but also interrelated with engine misfire diagnosis.Then,a novel pressure identification method involving frequency-amplitude modulated Fourier series(FAMFS)and an extended-Kalman-filter-optimized engine model is proposed and a novel misfire diagnosis method based on speed model and residual series time-frequency domain transformation of extended-Kalman-filter-optimized engine model is presented.As a crucial and critical factor in monitoring the internal state of an engine,cylinder pressure is mainly used to monitor the burning efficiency,to detect engine faults,and to compute engine dynamics.Even though the invasive cylinder pressure sensor is now greatly improved,its high cost,low reliability,and short lifespan due to severe working environments are criticisms enumerated by researchers all around the world.Aimed at a low-cost,real-time,non-invasive,high-accuracy soft cylinder pressure sensor,we propose a novel pressure identification method involving frequency-amplitude modulated Fourier series(FAMFS)and an extended-Kalman-filter-optimized engine model,which is based on theories regarding explosion and burning together with Newton's laws.The extended-Kalman-filter-optimized engine model is an iterative speed model associated with the throttle opening value and the crankshaft load.The extended-Kalman-filter is then used to estimate the optimal output of this iteration model.The optimal output of the speed iteration model is utilized to separately compute the frequency and amplitude of the cylinder pressure cycle-by-cycle.A standard engine's working cycle,identified by the 24 th order Fourier series,is determined.Using frequency and amplitude obtained from the iteration model to modulate the Fourier series yields a complete pressure model.A commercial engine(EA211)provided by the China FAW Group corporate R&D center is used to verify the method.Test results show that this novel method possesses high accuracy and real-time capability Thus,the novel method's accuracy and feasibility are verified.The engine misfire fault can be divided in two kinds of fault,one is misfire fault at “steady state condition” and the other is misfire fault at “dynamic state condition”.Aiming at solving the misfire fault at “steady state condition”,a method based on cylinder pressure identified by Fourier Transform and L-M optimized BP Neural Network is proposed.Use the steady state data simulated from AMESim to train BP Neural Network for obtaining the relationship between open value of valve and frequency.Use Fourier Transform to identify cylinder pressure.By contrasting the pressure of engine identified by Fourier Transform and the pressure mapped from crank speed,thus the misfire failure is be diagnosed.By offsetting the phase and the frequency to the identified model,accuracy and generalization of the identified model are improved.Re-offset phase and frequency to identified model when misfire fault occurs for gaining high tracking ability.On the “dynamic state condition side”,in regard to the existing misfire detection strategy,the throttle opening value and load should be constant or the accuracy of the detection strategy under complex situations will be badly affected.Aiming at solving this problem,a novel engine model based on explosion and burning theory together with newton law is proposed.This model is an iteration model associated with throttle opening value and load.Kalman filter is then used to estimate the optimal output of this iteration model.Misfire fault is classified according to number of cylinder which is malfunction.Threshold is determined by transforming the changing rate of residual series between kalman observer speed optimal value and crankshaft speed value from time domain to frequency domain.Misfire detection strategy called time-frequency transform of residual is then determined.Analysis the misfire fault in frequency domain to gain the characteristic of each fault.AMESim is used to simulate the model.Results show that relating to different kinds of misfire there exist obvious differences of residual serie's time-frequency characteristics on low-frequency,fundamental-frequency and doubling-frequency under both misfire condition and normal condition.Neural Network is used to complete the fault classification due to the characteristics each fault presented.Feasibility,validity and accuracy are then been proven.