A Data-Driven Slip Estimation Approach for Effective Braking Control under Varying Road Conditions

Abstract

The performances of braking control systems for robotic platforms, e.g., assisted and autonomous vehicles, airplanes and drones, are deeply influenced by the road-tire friction experienced during the maneuver. Therefore, the availability of accurate estimation algorithms is of major importance in the development of advanced control schemes. The focus of this paper is on the estimation problem. In particular, a novel estimation algorithm is proposed, based on a multilayer neural network. The training is based on a synthetic data set, derived from a widely used friction model. The open loop performances of the proposed algorithm are evaluated in a number of simulated scenarios. Moreover, different control schemes are used to test the closed loop scenario, where the estimated optimal slip is used as the set-point. The experimental results and the comparison with a model based baseline show that the proposed approach can provide an effective best slip estimation.

Tire-road friction estimation and uncertainty assessment to improve electric aircraft braking system

“…More sophisticated approaches could be used (see, e.g., [19]); this is left as an extension. The scheme presented in this study is an improvement of the one presented in [17], where a simpler MLP structure and dynamic model have been considered, and the problem of prediction accuracy computation was not addressed. The MLP training is based on a large number N of input vectors X i , each one comprising n pairs of (λ, µ) .…”

“…The crucial design elements of the data set X are the construction procedure of each windowed sequence X i , the windows size n, and the number N of vectors in the training dataset. The key idea is to use the Burckhardt model to generate a large number of friction curves by exploring the friction cube according to the procedure proposed in [17]. Each curve is further corrupted by AWG noise, to improve robustness and facilitate generalization, and both the original and the noisy curves have been used.…”

“…Deep neural networks have been proposed in [16] to estimate the maximum friction value as a function of a large number of input signals, that may not be available in some brake control systems. In [17], the authors proposed an approach which, starting from a reduced number of measurements, uses a multi-layer neural network (MLP) to estimate, online, the best slip as a function of a windowed sequence of slipfriction pairs. The MLP training is based on a synthetic dataset derived from the Burckhardt model.…”

“…The first contribution of the study is given by a method, based on the idea in [18], providing an online estimation of arXiv:2211.10336v1 [eess.SY] 14 Nov 2022 epistemic uncertainty (real-time computation of a confidence measure) of the nonlinear neural network model of the slipfriction curve. The second contribution is the introduction of an improved synthetic data-set, and a MLP based on a reduced size input set with respect to the previous approach in [17], together with a more accurate dynamic model of the aircraft braking phase.…”