Research On The Observation Method Of Fuel Injection Performance Of High Pressure Common Rail System

The fuel injection performance of a common rail system has a huge impact on the overall performance of a diesel engine.Small changes in parameters such as injection pressure curve,opening delay time,closing delay time,injection duration and injection volume can have a huge impact on the combustion reactions in the combustion chamber,which in turn affects the diesel engine performance.Therefore,real-time monitoring of the parameters related to injection performance is of great significance to grasp the operating status of the diesel engine.When testing the common rail system on the test bench,the injector injection pressure curve,injection volume and other parameters can be obtained through pressure sensors and other equipments to adjust the parameters of the common rail system and monitor the performance of the common rail system in real time.However,in test bench such as combustion or in practical engineering applications,the injectors are enclosed in the combustion chamber and it is not possible to obtain the injection pressure curve,injection volume and other performance parameters directly through traditional methods.In this case,it is particularly necessary to obtain the injection-related parameters by various methods.This paper uses AMESim16.0 software to establish a one-dimensional simulation model of the hydraulic process of the high pressure common rail system according to the constraints of the test bench and some structural parameters,and adjusts the relevant parameters to make the system performance meet the requirements of the injection performance index.Two methods are proposed to observe the injection performance of the high pressure common rail system using different algorithms to address the needs of injection performance and the limitations of traditional injection performance observation:（1）when the available samples are insufficient,the injection volume equation can be derived and the corresponding parameter correction equation can be solved by genetic algorithm to finally obtain the injection volume;（2）when the available samples are sufficient,Gate Recurrent Unit（GRU）variant of the Long Short-Term Memory（LSTM）can be used to accurately observe the injection pressure during the entire injection cycle.For the case of small sample size,since the injection volume is a numerical value rather than a high-dimensional series,the optimization of the relevant parameters in the model can be completed with only a small amount of training data in combination with the genetic algorithm.In this paper,a fuel injection volume observation model is established with the fuel injection volume of injector as the target,and three sets of data,one for the normal state and one for the injector early end injection state in the normal operating condition and one for the injector early end injection state in the 75%operating condition,are selected as the training set and input into the genetic algorithm for multi-objective optimization.6)₁,6)₂ and6)₃ are the optimization parameters.On this basis,the model is validated for different operating conditions,different operating states and different training states of the common rail system and compared with the observations of the injected fuel volume obtained from the simulation model.Using the model with a small sample size,fuel injection was observed for three operating conditions-normal,75%and idle-and three operating states-injector prematurely ending injection and high pressure pump plunger wear-for a total of nine operating cases,with an average error of 1.6853%and a maximum error of 4.4396%over 90 test samples.In the untrained cases,the average error over the 30 test samples can be 1.4952%,0.8175%and1.8555%,with maximum errors of 2.7209%,2.0042%and 1.8555%respectively.With a small amount and variety of data,the model is able to observe the injection volume of the common rail system more accurately for any operating state and training condition except for the fuel leakage condition.For the case of a larger sample size with more relevant training data,a corresponding observation model can be built by other algorithms for this time-series sequence of the injection pressure curve of the high pressure common rail system.In this paper,a variant of the LSTM,GRU,is chosen to build an observation model of the injection pressure curve of the common rail system,and five operating cases under normal operating conditions are used as examples to build an observation model of the injection pressure of the common rail system.The rail pressure,injection pressure and control pulse width of 10 injection cycles for each operating cases were selected to form the training set,and the rail pressure,injection pressure and control pulse width of 3 injection cycles for each operating condition were selected to form the test set.The discrete descent method is used to vary the learning rate and Adam（Adaptive Momentum Estimation）is used for optimization.The rail pressure signal and other relevant timing signals are used to build an observed model of the injection pressure curve of the high pressure common rail system.The actual injection curves of different injectors under different operating conditions are compared with the observed injection curves obtained using the observed injection pressure model,and the difference between the injection start time,the maximum injection pressure time,the end time,the injection duration,the maximum injection pressure and the EMD（Empirical Mode Decomposition）are used to carry out the corresponding error analysis,and the results are compared with the 1D simulation model based on bench tests to verify the correctness and accuracy of this injection pressure observation model.Using the injection pressure curve observation model with a large sample size,the average observation error at the start of injection was 0.0013 ms and the maximum error was 0.1 ms for the 75 test samples for the five states mentioned above;the average observation error at the time of maximum injection pressure was 0.0227 ms and the maximum error was 0.2 ms;the average observation error at the end of injection was 0.0253 ms and the maximum error was0.1 ms;The average observation error at the moment of maximum injection pressure was0.0227 ms,with a maximum error of 0.2 ms;the average observation error at the moment of end of injection was 0.0253 ms,with a maximum error of 0.1 ms;the average observation error at the duration of injection was 0.0267 ms,with a maximum error of 0.2 ms;the average maximum error of injection pressure was 2.06 MPa,with a maximum error of 6.51 MPa;the average EMD was 37.88,with a maximum error of 108.32.The simulation model was able to observe the injection pressure profile of the common rail system under different operating conditions.Compared with the commonly used 1D simulation model,the average EMD of the two models is similar,37.88（1D simulation model）and 33.62（observation model of injection pressure curve of high pressure common rail system）,and the overall difference of injection pressure is similar,53.62 MPa（1D simulation model）and 54.23 MPa（observation model of injection pressure curve of high pressure common rail system）.The overall accuracy of the two models is comparable.The injection pressure curves for different operating conditions of the common rail system can be observed more accurately.The overall accuracy of the two models is generally comparable.Compared to the injection pressure curve obtained from the calibrated1D simulation model of the common rail system,the overall accuracy is comparable,but the accuracy is better than the simulation model in the higher pressure region（near the injection pressure peak）,and the calculation speed is faster and the equipment requirements are lower.