Estimation of road surface type from brake pressure pulses of ABS

The ABS braking system uses the coefficient of friction over the slip rate to vary the brake pressure. For this, it references the friction coefficient-slip graph embedded in the ABS control unit. This method is insufficient to determine the friction coefficient suitable for real conditions on slippery surfaces where the peak is uncertain. In this study, machine learning methods such as SVM, MLP, SGD, GNB, and Extremely Randomized Trees that provide road surface type estimation regardless of road type have been developed in order to increase the braking performance of ABS by eliminating this deficiency. In this study, the compatibility between brake pressure and wheel acceleration change is taken as reference. For this, ABS tests are carried out at different speeds on both wet and slippery roads. The results are very promising and Extremely Randomized Trees reached 98% accuracy, 99% precision, 98% recall, and 97% F1-score, especially in test conditions at 30 km/h. Also, Extremely Randomized Trees reached 95% accuracy, 99% precision, 96% recall, and 97% F1-score at 60 km/h.