Application of Machine Learning Methods for the Development of Internal Combustion Engines – An Overview

Abstract

Machine Learning (ML) has a strong potential to improve the performance and effectiveness of several technologies and processes. In recent years, ML has gained in importance, primarily due to its matchless success in image recognition and computer games. These ML accomplishments have motivated to transfer and adapt its algorithms and modeling methods to most scientific disciplines. For instance, in mechanical engineering, ML is coming to hold a crucial position ranging from value chain optimization (production) to substitution of complex simulation models (research and development). In the case of traditional research and development approach, the analysis and optimization of a process are implemented according to the understanding of the governing mechanisms described by physical and mathematical rules. On the contrary, the intelligence of the ML method originates from the extraction of trends and laws based on data patterns, which produces surprisingly good results in many cases. However, it is not entirely evident why it performs so well. One of the most challenging mechanical engineering topics is the improvement of the Internal Combustion Engine (ICE) towards higher efficiency and lower negative impact on the environment. ICEs are very complex systems, which involve high-speed reciprocating motions, transient gas flow and combustion chemistry. Thus, the application of ML methods for ICEs opens new perspectives regarding the modelling, control and maintenance. These topics are addressed in detail in the course of this paper, based on the most relevant published results found in the literature, to provide an overview to the actual research and development of ICE using ML methods.