Research Of Control-Oriented Combustion Model For DICI Engine

Nowadays,decreasing emissions and saving energy are the two main challenges for internal combustion engines.Model-based combustion comtrol seems to have a good potential in order to achieve both fuel consumption and pollutant emission reductions,also considering the increasing computational performance of modern ECUs(Engine Control Units).Developing a precise control-oriented engine model is the precondition to perform model-based combustion control.Moreover,parameter matching and performance optimization of engine also greatly depend on a precise control-oriented engine model.Therefore,it is very crucial to develop a precise control-oriented engine model.0D physics-based and direct engin model are the two best options to select as candidate control-oriented engine models.Wiebe function-based combustion laws and accumulated fuel mass approach are two of best-known 0D combustion models to build physical engine models.The main challenges for the development of control-oriented engine model are the calibration of engine model parameters and the identification of correlation between inputs and outputs in the engine model.According to the investigation of literature,how to automaticly calibrate Wiebe function-based combustion model is still an important problem.For the identification of correlation between inputs and outputs,there are two problems need to further research:how to select proper inputs for each output and which method is suitable to simulate the correlation between parameters.The investigation scope of inputs candidates for selecting proper inputs is still not comprehensive enough.Moreover,there are still no mature rules and processes to perform the selection of proper inputs.MAP-based lookup table,empirical(or semi-empirical)function and artificial intelligence algorithm represented by artificial neural network(ANN)are widely used methods in recent decades to simulate the correlation between parameters,while,the performances of each method when used to develop control-oriented engine model are still not clear.This work was mainly focused on the development of control-oriented engine models.In this work,sensitivity analysis of Wiebe function was carried out and automatic calibration algorithm of Wiebe function-based combustion model was developed for the first time;Comparison between ANN-and MAP-based predictive engine models based on triple-Wiebe function was performed;Correlation analysis and sensitivity analysis were combined for the first time to select proper independent variables for each denpendent variable in engine model,and the rules and process for the selection of independent variables using correlation and sensitivity analysis were novelly created.The contents and conclusions of this research are as follows:(1)Design of testing plan,engine test,and calibration of parameters in engines’ working process were carried out to collect and prepare experimental data for the development of control-oriented engine models.Three types of diesel engines were employed to support this research,namely a single-injection diesel engine with mechanical fuel injection system,a single-injection diesel engine with common-rail fuel injection system and a multi-injection diesel engine with common-rail fuel injection system.The parameterize method of engine working process was described.Working process parameters were calculated using experiment data to prepare for correlation analysis and development of control-oriented engine model.The parameterize includes the calculation of heat release rate from in-cylinder pressure and the estimitation of in-cylinder pressure,pumping loss,indicated mean efficient pressure,friction loss,incylinder temperature,EGR ratio and EGR mass flow.(2)Sensitivity analysis of Wiebe function was proceeded on the single-injection diesel engine with common-rail system,and an automatical calibration algorithm of Wiebe function-based combustion model was developed and the development of 0D engine predictive physical model based on Wiebe functions was carried out on the single-injection diesel engine with mechanical fuel injection system.Map-based approach and ANN were respectively used to identify the correlations between Wiebe parameters and engine operation variables.The comparison results between map-based approach and ANN shown that ANN presents better predictive performance than map-based approach to build 0D engine predictive physical model.(3)For the multi-injection diesel engine,both the physics-based and direct engine models were built using empirical approaches,respectively called as "empirical physics-based model"and "empirical direct model".The empirical physics-based engine model was based on the accumulated fuel mass approach,and power law-based emipirical function is used to fit the correlation between dependent and independent parameters.In order to select proper input parameters for engine models,Pearson and partial correlation analysis were sequentially carried out to select the candidates of input parameters,and then sensitivity analysis was proceeded to confirm the final list of input parameters.ANN was also applied to replace semi-empirical approach to identify the correlation in both the physics-based and mathematical engine models,thus "ANN physics-based model" and "ANN direct model" were built.The set and data of input variables that were used for the ANNs were the same as those used for the semi-empirical approaches.(4)The predictive performances of the four kinds of control-oriented models of multi-injection diesel engine,namely "empirical physics-based model","empirical direct model","ANN physics-based model" and "ANN direct model",were compared under steady-state and transient condition,over a WHTC(Worldwide Harmonized Heavy-duty Transient Cycle).The ANN direct model showed the best accuracy in the estimation of the combustion metrics at both steady-state and transient operation;the accuracy of the empirical physics-based model is slightly worse than that of the ANN direct model.The accuracy of the ANN physics-based model is higher than that of the empirical physics-based model at steady-state operation,but it shows a high deterioration in transient operation.The accuracy of the empirical direct models is worse than that of the empirical physics-based model and ANN direct models at steady-state and transient operation.However,the accuracy can still be considered acceptable,also considering the simple mathematical structure of this kind of model,and the low number of tuning parameters(and,therefore,of experimental data needed for calibration).Computational time of all the models(excluding the ANN physics-based model,which is the least robust approach)was estimated when they were run on the ETAS ES910 device.The results shown that the ANN direct model and the empirical direct model feature the lowest computational time,is lower than 50μs.The computational time required by the empirical physics-based model is the highest one,is of the order of 350μs.However,it is also compatible with real-time combustion control applications,since it is far lower than the typical engine cycle time.