Page - 123 - in Maximum Tire-Road Friction Coefficient Estimation

6 Results and conclusion are designed for dry roads, cf. Section 1.3.2, the adaptation to low-friction surfaces is especially interesting. The results in Section 6.3 support the idea that the identification of low-friction surfaces is more accurate than that of dry conditions for a given dynamic excitation of bax. It has to be mentioned that the work of Lex et al., [LKE13b], was based on the same assumptions concerning the requirements for application in an AEB and, in addition, was based on the exact same measurement data that has been used within this work. Using ANNs, see Section 2.2.2, the achieved estimation results are comparable to those achievedwith theparticlefilterpresentedwithin this chapter. For summer tiredata, the MAE was given between 0.17 and 0.22 for the four tires, [LKE13b], which is comparable to the results in Table 6.1 and Table 6.2. Nevertheless, it has to be mentioned that a direct comparison of the two results is only legitimate for a rough evaluation. Whereas the MAE shown by Lex et al., [LKE13b], count for a larger amount of measurement data, the resulting MAE shown in this work only count for individual measurements. 6.6. Discussion and outlook The results of the sensitivity analysis in Section 4 suggest that it is worth focusing on the wheel dynamics or wheel-related variables rather than the vehicle’s body reaction. This statement is directly supported by other works such as Rajamani, [RPPL12], and indirectly supported by the existence of much research that focuses on the horizontal tire forces or the longitudinal slips and the side slip angles, cf. Section 2.2.2. In this work, the wheel’s angular momentum was used to calculate the expected longitudinal tire forcesFx,i. The results presented in this chapter suggest that the proposed observer based on a particle filter can fulfill the requirements of an AEB in certain driving states. This is especially true for inner-city applications, as the requirements for an estimate of the friction potential are relatively low at these vx, cf. Figure 6.16. In particular, the particle filter with resampling and using ∆µmax within the re- initialisation strategy is promising, as the result is not too noisy yet still converges quickly. The estimate of the friction potential itself only depends on the variable par- ticles. The fixed particles are only used to detect when re-initialisation is necessary. In order toensure that theestimatedvaluegetscloser to therealvalueof the frictionpoten- tial, the limit of ∆µmax has to be optimized. Due to the nature of the proposed method, the influence of the road and the tire condition are estimated inseparably. Therefore, 123