Comparing Rotational Speed and Acceleration Signals for Normal Operating State Detection in an Internal Combustion Engine

Condition monitoring (CM) relies on accurate state information. While vibration analysis using accelerometers is key to monitoring rotating machines, the high cost often limits its continuous industrial application. Building monitoring systems upon readily available quantities, such as rotational speed (RPM), significantly enhances cost-efficiency by eliminating the need for additional hardware. We studied the use of high-resolution RPM measurements as a basis for classifying different normal operating states of an internal combustion engine (ICE). To ensure computational efficiency, we leveraged state-of-the-art time-series feature extraction libraries alongside logistic regression. We built and compared classifiers based on high-resolution RPM and acceleration measurements and studied the effect of feature reduction on model accuracy. We show that, regardless of the feature extraction method, basing the classification on RPM instead of acceleration yields significantly more accurate models. These results promote RPM as a promising base for CM of ICEs.