Fuzzy Neural Network Learning for Practical Intelligent Powertrain Exhaust Gas Temperature Prediction

The exhaust gas temperature plays a leading role in the thermal efficiency of the automotive powertrain system. It also determines the performance of catalytic converters for removing toxic gases. When the exhaust temperature is excessively high, it will give rise to extra heat energy loss in the exhaust systems. Meanwhile overheating leads to engine component damage, engine knock, fuel pre-ignition and emission after-treatment system failure. The inherent formation mechanism of exhaust temperature is highly complicated, which depends on multiple factors in the combustion process as well as heat and mass transfer. The maximum combustion efficiency itself requires the stoichiometric mixture (air to fuel ratio equal to 14.7). Two parameters with dominant impact on exhaust gas temperature are engine speed and engine load. Some typical delays occur in engine operations as well due to fuel atomization, air and fuel mixing, vaporization, heat transfer, and so on. It is essential to operate at the optimal exhaust temperature to maximize the performance to cost ratio and to avoid severe damage. Ambiguity and uncertainty are inevitable, which gives rise to high complexity in modeling and prediction of the exhaust gas temperature. The goal of this research instead is to design a feasible and applicable simple exhaust gas temperature model for potential optimal engine design. The intelligent hybrid fuzzy neural network learning is proposed to model the exhaust gas temperature in the powertrain system with high nonlinearity, high complexity and high uncertainty. The fuzzy system is introduced to deal with the uncertainty and ambiguity through fuzzy sets, covering fuzzification, inference engine and defuzzification subsystems. Hybridization is made when artificial neural networks involve in together with the fuzzy system. In this case, training, learning, predication and validation of the hybrid fuzzy neural network powertrain exhaust gas temperature model can be implemented. This simple model can be accomplished with ease, which can also be extended to the exhaust gas temperature model prediction across arbitrary types of automotive engines.