A Noninvasive Machine Learning Solution for Estimating the Rotation Speed of a Heat Engine

Abstract

There are situations in practice when it is necessary to estimate the rotation speed of the heat engine of a car when its tachometer is not in a good health. In this case, a quick method of speed estimation, noninvasive if possible, even if it may not be very accurate, may be of help. The present paper proposes such a method for estimating the rotation speed of an internal combustion engine, utilizing the signals produced by the engine vibrations acquired with a mobile phone and supervised machine learning (ML) algorithms. The paper describes the complete process of the method, with details regarding the data acquisition and preprocessing, features building and ML algorithms implementation. An example of field deployment is also provided and an analysis is made about how a series of parameters influences the method and may be optimized in terms of two important criteria: accuracy and computation effort. Finally, a trade-off between the two criteria is carried out, specifying the optimal conditions for deploying the method in the field.