The initial conception of the neural network model for the driving active safety estimation

K Dobaj
2018 IOP Conference Series: Materials Science and Engineering  
The article describes the prototype neural network diagnostics model application to the driving active safety determination. The neural network fitting performance was used as an diagnostic indicator. The model used the two-layer-feed-forward network, including 10 sigmoid neurons in the hidden layer and linear output layer neurons. The model application example was involved with driving and braking with working ABS and ESP systems on snowy road. The 3axis vehicle body acceleration, the speeds
more » ... wheels and the vehicle linear velocity were measured. Different kinds of tyres with various pumping pressures were applied. The measurements were obtained during the repeatable test runs. The diagnostic reference characteristics was chosen. The reference characteristics was involved with application the winter tyres with the proper pumping pressure level. The neural network was reproducing the data involved with the reference characteristics from the data involved with the test runs with inadequate tyre kinds or pumping pressure levels. The more weather-adequate the tyre kind and pumping pressure are, the smaller reproduction errors occur. Thus, the neural network reproduction error was used as an diagnostics indicator. The conducted analysis show the approach drawbacks, resulting from the vibrations and limited repeatability. The further steps, dealing with these problems, including the signal approximation methods, were proposed as the conclusions.
doi:10.1088/1757-899x/421/3/032007 fatcat:soqowp4vpje67jjhwafrcni6aq