Using deep learning to diagnose preignition in turbocharged spark-ignited engines

Internal combustion engines of today are expected to reduce their greenhouse gas emissions to comply with global climate change mitigation targets. This can be achieved using low-carbon fuels, introducing more hybridization, and improving their efficiency. The potential of artificial intelligence in contributing to these pathways is immense. In fact, researchers have already been using machine learning (ML) techniques for better control and optimization of engines, predicting performance and emissions, and detecting faults in internal combustion engines. This work looks at different ways in which such techniques have been implemented in spark-ignited engines. Thereafter, one specific application has been detailed: use of ML to diagnose stochastic preignition events in turbocharged engines. Preignition is an abnormal combustion event, often leading to excessively high peak pressures and pressure oscillations, which may damage the engine hardware. To diagnose preignition cycles from normal cycles, two deep neural network models were used; one using principal component analysis data as input and the other using direct time-series data as input. The former model was able to better differentiate between preignition and normal cycles in the current work.