Braking Torque Estimation Through Machine Learning Algorithms

MotoGP class motorcycles rely on carbon braking system to cope with their incredible acceleration capability and high speed. Hence, assessing the torque generated by the front discs is a key to improve the vehicle performance. As direct measurement of the braking torque is not allowed during races, its value may be estimated through a physical model, using as inputs the brake fluid pressure (monitored on board), the braking system geometry and the friction coefficient (μ). However, the results obtained with this method are highly limited by the knowledge of the instantaneous friction coefficient between the disc rotor and the pads. Since the value of μ is a highly nonlinear function of many variables (namely temperature, pressure and angular velocity of the disc), an analytical model appears impractical to establish. This work aims to implement an innovative algorithm, based on machine learning, for determining μ from the signals regularly available in races, to enable accurate breaking torque computation. The proposed method consists of two main tools. An artificial neural network (ANN) is developed to approximate the unknown function that relates the input variables to μ, while a Kalman filter (KF) is implemented to estimate the real temperature distribution on the disc surface that constitutes one of the most important ANN inputs. The proposed algorithm has been successfully validated with real data collected from extensive tests in racetracks, with a special sensor setup.