Geometric neuro-fuzzy transfer learning for in-cylinder pressure modelling of a diesel engine fuelled with raw microalgae oil

Bioenergy attracts more attention owing to the reduction of both air pollution and greenhouse gas emissions in a whole life cycle compared to fossil fuels. As a third-generation biofuel, Microalgae Oil (MAO) can utilise carbon dioxide and light energy at an increased photosynthetic efficiency compared to energy crops for biomass. Due to the wide variety of MAO and their blends with diesel in different ratios, characterization of these biofuels’ engine performance is difficult to be standardized, e.g., in-cylinder pressure. This paper proposes a novel approach of geometric neuro-fuzzy transfer learning (GNFTL) for in-cylinder pressure modelling of a diesel engine fuelled with MAO. Inspired by computational geometry, this approach only utilizes limited experimental data obtained by geometric screening to learn a high-precise transfer model of the in-cylinder pressure with different MAO blending ratios. Followed by the process of MAO extraction and test cell description, the proposed approach of GNFTL is presented which comprises geometric transfer domain segmentation and neuro-fuzzy transfer learning. By a comprehensive study, the results demonstrate that the proposed approach can achieve a competitive prediction accuracy whilst significantly reducing experimental efforts on used biofuel by 47.8% and operation time by 41.5%, compared to the conventional manual design of experiment.