Investigation On In-Cylinder Pressure Reconstruction Technology For Diesel Engines

The cylinder pressure signal of a diesel engine is one of the important indicators of its operating status,which reflects the efficiency,stability,and performance of the engine during the combustion process.However,obtaining the cylinder pressure signal is an expensive and difficult process.Therefore,when evaluating the performance of a diesel engine,a cost-effective,accurate,and reliable way to obtain cylinder pressure information is needed.Diesel engine cylinder pressure reconstruction technology measures other low-cost signals related to cylinder pressure and then reconstructs the cylinder pressure signal using algorithms.The reconstructed cylinder pressure under a wide range of operating conditions should meet the accuracy requirements and the commonly used combustion parameters should meet the requirements while retaining the shape characteristics of the actual cylinder pressure signal.Firstly,research on cylinder pressure signal filtering technology aims to remove noise throughout the entire acquisition process to enhance the stability and accuracy of subsequent modeling.Two key parameters for cylinder pressure filtering,cycle averaging and low-pass filter cutoff frequency,are determined based on entropy and time-frequency analysis.The experimental data show that a cycle averaging of around 78 achieves optimal results in terms of computational resources and noise elimination.Time-frequency analysis of the cylinder pressure signal shows that the frequency components of different signals vary with the engine speed.When the cutoff frequency is 1.5 times the speed(rpm),the best overall effect of removing high-frequency noise and preserving the cylinder pressure curve characteristics is achieved.Secondly,based on the crankshaft dynamics model,a feature vector that can be used to reconstruct the cylinder pressure signal is derived.A lightweight artificial neural network model is established to preprocess the feature vector using principal component analysis and normalization for model training.The reconstruction accuracy of the model is then validated,and the reconstruction accuracy of commonly used combustion parameters is evaluated to assess the model’s effectiveness.The results show that the MSE of reconstructed cylinder pressure is 0.44 Bar,with an R~2 of over 97%,and the MSEs of PMAX,LPP,CA50,and IMEP are 0.26,0.81,0.91,and 0.28,respectively.Furthermore,to improve the predictive ability of the reconstruction algorithm for cylinder pressure curves under different modes,an unsupervised learning-based combustion mode clustering algorithm is proposed.Based on the idea of controlling variables,all experimental data are observed to extract feature vectors for clustering by observing the influence of different operating variables on combustion.The experimental results show that the cylinder pressure curve can be classified into four combustion modes,each with different characteristics in curve shape.In terms of operating conditions,they can be categorized as high positive timing and high load,high positive timing and low load,low positive timing and high load,and low positive timing and low load.Finally,validated the cylinder pressure reconstruction model based on the combustion mode classification algorithm.The results show that the algorithm based on combustion mode classification can significantly improve the effectiveness of cylinder pressure reconstruction.In 100 sets of operating conditions points ranging from 800 rpm to 1500 rpm,the numerical values and shapes of the cylinder pressure curve are accurately reconstructed.After using the combustion mode classification algorithm,the MSE of the algorithm decreased by 73.4%(training set)and 62.7%(test set),and the R~2 increased to over 99%.