Engine valve clearance diagnostics based on vibration signals and machine learning methods

A dynamic advancement of the design of combustion engines generates a necessity of introduction of strategies of operation based on the information related to their technical condition. The paper analyzes problems related to vibration based diagnostics of valve clearance of a piston combustion engine, significant in terms of its efficiency and durability. Methods of classification have been proposed for the assessment of the valve clearance. Experiments have been performed and described that aimed at providing information necessary to develop and validate the proposed methods. In the performed investigations, the vibration signals were obtained from a triaxial accelerometer located in the engine cylinder head. A parameterization of the obtained vibration signal has been carried out for the engine operating under different engine loads, rotation speeds and valve clearance settings. The parameterization pertained to the specific features of the vibration signals, the derivative of the vibration signal as a function of time as well as the envelope of this derivative. In the first approach, the authors developed a classifier in the form of a set of binary trees that additionally allowed distinguishing the features significant in terms of the identification of adopted classes. For comparison, the authors also developed classifiers in the form of a neural network as well as a k-nearest neighbors algorithm using the Euclidean metric. Based on the performed investigations and analyses a method of valve clearance assessment has been proposed.