Machine Learning Integration With Combustion Physics to Develop a Composite Predictive Model for Reactivity Controlled Compression Ignition Engine

Abstract

Phasing of combustion metrics close to the optimum values across operation range is necessary to avail benefits of reactivity controlled compression ignition (RCCI) engines. Parameters like start of combustion occurrence crank angle (θsoc), occurrence of burn rate fraction reaching 50% (θ50), mean effective pressure from indicator diagram (IMEP) etc. are described as combustion metrics. These metrics act as markers for macroscopic state of combustion. Control of these metrics in RCCI engine is relatively complex due to the nature of ignition. As direct combustion control is challenging, alternative methods like combustion physics derived models are a subject of research interest. In this work, a composite predictive model was proposed by integrating trained random forest (RF) machine learning and artificial neural networks (ANN) to combustion physics derived modified Livengood-Wu integral, parametrized double-Wiebe function, autoignition front propagation speed based correlations and residual gas fraction model. The RF machine learning established a correlative relationship between physics based model coefficients and engine operating condition. The ANN developed a similar correlation between residual gas fraction parameters and engine operating condition. The composite model was deployed for the predictions of θsoc, θ50 and IMEP as RCCI engine combustion metrics. Experimental validation showed an error standard deviation (θ68.3,err) of 0.67 °CA, 1.19°CA, 0.223 bar and symmetric mean absolute percentage error of 6.92%, 7.87% and 4.01% for the predictions of θsoc, θ50 and IMEP respectively on cycle to cycle basis. Wide range applicability, lesser experiments for model calibration, low computational costs and utility for control applications were the benefits of the proposed predictive model.

A Hybrid Physics-Based and Stochastic Neural Network Model Structure for Diesel Engine Combustion Events

“…As such, the performance of the three CA50 and IMEP model frameworks was also evaluated in a stochastic environment. The models were perturbed with a Gaussian distributed disturbance (as shown in the block diagrams in Figures 4,11,17,and 19) to assess their robustness.…”

“…A linear parameter-varying support-vector machine (LPV-SVM) with multivariate inputs was used, and a maximum CA50 average tracking error of 0.7 CAD was achieved with an LPV-model-based controller. In [19], machine learning and combustion physics were integrated to predict CA50 and other combustion metrics for an RCCI engine. The random forests technique was implemented to develop correlative mapping between physicsbased model coefficients and the engine's operating condition, and an ANN was utilized for model coefficient determination.…”

“…The literature demonstrates that leveraging physics to inform ANN models could enhance predictions of engine metrics and, as such, improve real-time combustion phasing control methods without the strict need for in-cylinder sensors. While prior work using either direct ANNs or integrated ANN and physics-based approaches to predict critical engine metrics has shown promise [1,[5][6][7]13,15,16,[18][19][20][21][22][23][24], some challenges with uncertainty and complexity still need to be addressed. Some existing methods assume linear relationships and, as such, are simpler to use for control, but have a more limited ability to extrapolate, and encounter greater inaccuracy due to the nonlinearity of engine processes [7,13,18,22].…”

A misfire-integrated Gaussian process (MInt-GP) emulator for energy-assisted compression ignition (EACI) engines with varying cetane number jet fuels

“…Machine Learning (ML)-based surrogate modeling of internal combustion engines (ICE) has been widely used for a broad range of applications. 25–29 Data-driven ML approaches, in particular, are popular for building ICE surrogate models; such approaches include neural networks (NN), 30–45 Support Vector Machines (SVM), 46–49 Gaussian Processes (GPs, 50–60 also known as kriging 61 ), and other learning models. 62–68 In surrogate modeling applications with limited training runs from expensive simulators, GPs (and its recent non-stationary extensions) have several key advantages over alternate deep learning models.…”

The Start of Combustion Prediction for Methane-Fueled HCCI Engines: Traditional vs. Machine Learning Methods

“…ese techniques have been recently used by researchers to develop control models in the engine industry [27][28][29]. Using ML, it is possible to provide the numerical models to predict the SOC timing employing the engine performance dataset.…”