Analysis of significance of variables in IC engine operation: an empirical methodology

Approaches for studying internal combustion engines have evolved from fundamental analytical modelling to gross phenomenological studies, then to computational approaches in the last few decades. Additionally, lack of unified models of IC engines even after almost 150 years since its introduction has made computational approaches inevitable. And yet, accurate computational solutions incur high engineering costs; and they cut out in delivering quick and pragmatic industry feasible solutions. Therefore, this study took an alternative empirical approach that addressed the comprehensivity of the system. First, approximated empirical model for experimental data using artificial neural network was developed, and it offered prediction accuracy of 0.85 +/- 0.12 (as mean +/- standard deviation) for 19 response variables. This predicted dataset was further used as a duplicate dataset for validating the results of an empirical method that accounted for the comprehensivity of the system. Analysis of dependencies between these variables along with earlier studies revealed that empirical redundancy exists in IC engine operation. Subsequent processing based on statistical principles and empirical guidelines provided for sorted engine variables in order of their significance. This was quantified through two indexes introduced in the study: representation rank, and importance rank. Identification of these variables and their significance should advance computational studies to progress empirical modelling.