Застосування нечіткої нейронної мережі для визначення інформативності факторів впливу на реалізацію зчіпної здатності дороги і шини

Abstract

The task of assessing the traction of the road and tires is one of the most relevant in the technical examination of road accidents, as the results of its solution directly affect the assessment of the effectiveness of braking of wheeled vehicles, as the main method of preventing road accidents on motor transport. In the presence of such damage to the vehicle, which makes it impossible to conduct road tests, the expert has to use outdated calculation methods. This contributes to errors and increases the uncertainty of the data on which expert opinions are formed. The paper proposes ways to improve existing approaches to estimating the adhesion factor and indicators of braking efficiency of vehicles in the automotive examination of accidents in the presence of compositional (stochastic and fuzzy) uncertainty. Analysis of the use of mathematical methods in the practice of accident investigation showed that in the absence of the possibility of using traditional mathematical methods based on the detection of accurate quantitative relationships, to study accidents in uncertainty, it is advisable to use approximate modeling methods based on fuzzy (continuous) logics. The results of the study of the braking efficiency of vehicles of category M1 in operating conditions and the results of the evaluation of the informativeness of the factors influencing the adhesion factor, using the Fuzzy Logic Toolbox of the Matlab computing environment were used to the choice and substantiation of the method of estimating the coupling qualities of car tires in the study of road accidents in conditions of uncertainty. As a result, it was found that the improvement of existing approaches to estimating the adhesion factor and efficiency of vehicle braking in autotechnical examination of road accidents in the presence of compositional uncertainty can be achieved through the use of simple ANFIS models that provide better generalizing properties.