DNN-CNN-based analysis of autoignition and combustion characteristics of DME/NH3 homogeneous mixtures under marine engine relevant conditions

Dimethyl ether (DME)/ammonia (NH 3 ) blended-fuel combustion has significant potential for low-carbon emissions and high-efficiency energy utilization. Zero-dimensional homogeneous combustion models coupled with chemical mechanisms can predict the autoignition characteristics of homogeneous mixtures. However, they cannot effectively describe autoignition combustion behavior under transient temperature and pressure variations. In this study, a computational program was developed using a self-established homogeneous combustion simulation model to investigate the autoignition combustion behavior of DME/NH 3 homogeneous mixtures during transient temperature and pressure variations under marine engine relevant conditions. A homogeneous reactor model was used to describe the in-cylinder process, while gas exchange and heat transfer were considered. Intake temperature, intake pressure, fuel composition, intake valve closing timing (IVC), and exhaust gas recirculation (EGR) rate were used as basic operating parameters to generate the pressure-curve dataset. A dual surrogate model consisting of a deep neural network (DNN) and a one-dimensional convolutional neural network (1D-CNN) was constructed. The DNN was used to predict peak pressure and maximum pressure rise rate, and its outputs were combined with the basic operating parameters as inputs to the 1D-CNN to reconstruct the pressure curve. By coupling the surrogate models with particle swarm optimization, using the basic operating parameters as decision variables and thermal efficiency as the screening objective, the intrinsic autoignition and combustion characteristics of the mixture and their effects on thermal efficiency were analyzed under four representative loads. The results show that, with increasing load, the operating conditions with the highest thermal efficiency shift toward higher intake states and lower DME energy fractions, while CA50 is gradually delayed and combustion duration increases.