Comparison of neural classifiers for vehicles gear estimation

Nearly all mechanical systems involve rotating machinery with gearboxes used to transmit power or/and change speed. The gear position is an indication of the driver's behavior and it is also dependent on road conditions. That is why it presents an interesting problem to estimate its value from easily measurable variables. Concerning individual vehicles, there is a specific relationship between the size of the tires, vehicle speed, regime engine, and the overall gear ratio. Moreover, there are specific ranges for vehicle speed and regime engine for each gear. This paper evaluates the use of neural network classifiers to estimate the gear position in terms of two variables: vehicle velocity and regime engine. Numerous experiments were made using three different commercial vehicles in the streets of Madrid City. A comparative analysis of the classification efficiencies of different neural classifiers such as: multilayer perceptron, radial basis function, probabilistic neural network, and linear vector quantization, is presented. The best results in terms of classification efficiency were obtained using multilayer perceptron neural network (92.7%, 91.5%, and 85.9% for a Peugeot 205, a Seat Alhambra, and a Renault Laguna respectively). The maximum likelihood classifier is used as a benchmark to compare with the neural classifiers.