Toy Model: A Naïve ML Approach to Hydrogen Combustion Anomalies 2024-01-2608

Predicting and preventing combustion anomalies leads to safe and efficient operation of the hydrogen internal combustion engine. This research presents the application of three machine learning (ML) models – K-Nearest Neighbors (KNN), Random Forest (RF) and Logistic Regression (LR) – for the prediction of combustion anomalies in a hydrogen internal combustion engine. A small experimental dataset was used to train the models and posterior experiments were used to evaluate their performance and predicting capabilities (both in operating points -speed and load- within the training dataset and operating points in other areas of the engine map).

KNN and RF exhibit superior accuracy in classifying combustion anomalies in the training and testing data, particularly in minimizing false negatives, which could have detrimental effects on the engine. The findings suggest that these naïve models are effective in identifying and flagging operating conditions with high potential for an anomaly occurring and thereby enabling timely intervention and preventive measures. The generalization of the model to conditions outside the training dataset showed sufficiently high prediction capabilities at the early stage of the development, and the new set of tested data can be included into future model training sets to improve the robustness for forthcoming testing at high-anomaly conditions before including the results into control strategies and systems.