Experimental Validation of Artificial Neural Network Based Road Condition Classifier and its Complementation

Our previous study focused on the development of artificial neural network (ANN) classifier estimating the maximum static road friction coefficient for each wheel of the vehicle based on the Carsim/Simulink co-simulation environment by utilizing only existing sensor value without additional sensors. As follow-up to the previous study, this paper investigates the effectiveness of the proposed ANN classifier using the field test data in Software-In-Loop-Simulation (SILS) environment. Furthermore, we have extended the input of ANN classifier to improve an estimation performance. Specifically, the braking pressure and pressure gradient of each wheel were additionally set as the new inputs of the classifier (cf. the original inputs are the initial speed at braking, the vehicle deceleration, the wheel slip ratio, and the vehicle mass). Hence, the benefits of additional inputs have been clearly explored here. Moreover, the proposed scheme was challenged to other, more complicate road surface conditions (including jump and split friction roads). The original classifier guarantees a fairly accurate estimate, but we found that including brake pressure information into the classifier yielded estimates in better quality and the estimation results capture 86~95% accuracy for normal braking regardless of roads and 70~84% accuracy for extreme road conditions. This work will be a valuable asset for those who wish to develop the practical estimation methods for road friction coefficient via an ANN classifier using only inertial sensor information already available in most standard cars.