Geração de mapas de eficiência volumétrica e razão ar-combustível por lógica fuzzy em motores de combustão interna

Abstract

Engine calibration through look-up tables is widely used in electronic control units, but it often requires many testing iterations. As an alternative, this work investigates the use of fuzzy logic to model two core mixture-management blocks: volumetric efficiency (VE) and air-fuel ratio (AFR). The objective was to build interpretable fuzzy maps and compare them with reference maps. To this end, a Mamdani system was developed in MATLAB/Simulink, with nine membership functions per variable and centroid defuzzification, generating 2-D/3-D surfaces and executing a time-based simulation with startup and cruise phases. The static evaluation indicated good adherence: for VE, the correlation was close to 0.94, with a mean absolute error of approximately 8 percentage points and a root-mean-square error of 10.27 percentage points; for AFR, the correlation was around 0.93, with a mean error of approximately 0.54 percentage points and a root-mean-square error of 0.6940. In the dynamic analysis, the estimated hourly fuel-consumption curves exhibited the same peak-and-valley structure between the fuzzy model and the real map, with small differences explained by two design choices: the smoothing inherent to fuzzy surfaces and a slightly lower rotational sensitivity in the fuzzy VE. At low load, both models maintained stoichiometric AFR; at high load, both enriched the mixture, preserving the expected behavior. It was concluded that fuzzy maps reproduce essential physical trends, provide transparency for tuning, and can replace static maps in simulation scenarios and, with local refinements, in real-time applications. As future work, it is suggested to increase rotational sensitivity in the VE model, revise the richer AFR regions, and include temperature and pressure compensations.