Enhancement of large engine technology through machine learning

The ongoing digitalization of today’s world provides valuable opportunities for improving existing large internal combustion engines (ICE) technology and enabling or supporting the development of new solutions. In particular, machine learning has recently opened up promising new avenues. The analysis and use of large amounts of data that has been generated either experimentally (e.g., by sensors inside ICEs) or virtually (e.g., by simulation tools) effectively provide insights into previously unknown correlations. On the one hand, this allows generation of an added value for research tools such as engine testing and simulation. On the other hand, the additional benefits can eventually be employed in series applications.
This paper presents actual applications of how machine learning approaches enhance the diverse research being conducted on modern large engines. For all applications, realistic and applicationrelated data from experiments or simulations serve for model training and validation and the outcomes are described by means of quantitative results to understand the achieved benefits. The topics covered range from fundamental research on how to enhance simulation methods to fault diagnosis on engine test beds to condition monitoring and predictive maintenance on key engine components such as cylinder liners, fuel injection valves or sliding bearings and finally to engine control
applications for combustion anomalies. Each topic is introduced by discussing the underlying task as well as the implemented machine learning approaches, which can include purely data-driven as well as hybrid methods that also take physical relationships into account. Altogether this provides a comprehensive overview of the versatile ways in which machine learning can be beneficially deployed