Modeling The Stable Operating Envelope For Partially Stable Combustion Engines Using Class Imbalance Learning
This work addresses a domain-specific problem for combustion engine diagnostics and control, with incremental improvements in applying existing machine learning methods to handle class imbalance in engine data.
The paper tackled the problem of modeling the stable operating envelope for homogeneous charge compression ignition (HCCI) engines by using machine learning to identify the boundary from experimental data, with results showing that cost-sensitive versions of extreme learning machines and support vector machines are well-suited for this task and have potential for predicting instability based on sensor history.
Advanced combustion technologies such as homogeneous charge compression ignition (HCCI) engines have a narrow stable operating region defined by complex control strategies such as exhaust gas recirculation (EGR) and variable valve timing among others. For such systems, it is important to identify the operating envelope or the boundary of stable operation for diagnostics and control purposes. Obtaining a good model of the operating envelope using physics becomes intractable owing to engine transient effects. In this paper, a machine learning based approach is employed to identify the stable operating boundary of HCCI combustion directly from experimental data. Owing to imbalance in class proportions in the data, two approaches are considered. A re-sampling (under-sampling, over-sampling) based approach is used to develop models using existing algorithms while a cost-sensitive approach is used to modify the learning algorithm without modifying the data set. Support vector machines and recently developed extreme learning machines are used for model development and results compared against linear classification methods show that cost-sensitive versions of ELM and SVM algorithms are well suited to model the HCCI operating envelope. The prediction results indicate that the models have the potential to be used for predicting HCCI instability based on sensor measurement history.