A neuro-fuzzy based combustion sensor for the control of optimal engine combustion efficiency

Modern and advanced control systems for internal combustion engines require accurate feedback information from the combustion chamber. Whereas the in-cylinder pressure sensor provides this information through its close thermodynamic ties with the combustion process, drawbacks in its implementation push research towards other nonintrusive sensing methods. This paper suggests alternative methods of combustion phasing detection relying on measured angular crankshaft speed. Method developed, achieves sensing of angular position of the 50% of mass fraction burned (MFB50) through two steps: calculation of, so called, synthetic torque and its nonlinear transformation to a combustion feature estimator through local linear Neuro-fuzzy based model (LLNFM). In order to calibrate both parts of this virtual combustion sensor, parameters of a high-fidelity crankshaft dynamic model are identified, and LLNF model is trained with extensive experimentally collected data set. Created virtual MFB50 sensor, demonstrated its performance, on a large test data set comprised of 70% of gathered data.