Spark plug failure detection using Z-freq and machine learning

Abstract

Preprogrammed monitoring of engine failure due to spark plug misfire can be traced using a method called machine learning. Unluckily, a challenge to get a high-efficiency rate because of a massive volume of training data is required. During the study, these failure-generated were enhanced with a novel statistical signal-based analysis called Z-freq to improve the exploration. This study is an exploration of the time and frequency content attained from the engine after it goes under a specific situation. Throughout the trial, the misfire was formed by cutting the voltage supplied to simulate the actual outcome of the worn-out spark plug. The failure produced by fault signals from the spark plug misfire were collected using great sensitivity, space-saving and a robust piezo-based sensor named accelerometer. The achieved result and analysis indicated a significant pattern in the coefficient value and scattering of Z-freq data for spark plug misfire. Lastly, the simulation and experimental output were proved and endorsed in a series of performance metrics tests using accuracy, sensitivity, and specificity for prediction purposes. Finally, it confirmed that the proposed technique capably to make a diagnosis: fault detection, fault localization, and fault severity classification.