Evaluating Sensor Placement in Vibration-Based Engine Misfire Detection Using Artificial Neural Networks

This paper examines the effect of sensor positioning on the ability of artificial neural networks (ANN) to classify engine misfires using vibration signals. When it faces misfires that can affect engine performance, fuel economy, and emissions, these events create specific vibration patterns that can be analyzed for diagnostic purposes. However, vibration-based diagnostics only work if the sensors are placed in a way that allows for precise and representative signals to be obtained. Using a 4-cylinder engine as the test rig, vibration data are acquired at four different sensor locations with various misfire conditions. The signals were subjected to pre-processing and feature extraction, resulting in an optimized subset of features for building an ANN. Separate neural networks were then created for each sensor location, and their accuracy and classification performance were evaluated using validation and testing metrics. According to the outcomes, diagnostic accuracy varied significantly depending on sensor placement, with the highest test accuracy of 82.36% achieved by placing the sensor close to the engine's central components. This site recorded vibration signals with the most diagnostic characteristics, highlighting its importance for credible fault recognition. The results also showed that the true positive rate differed for each classification according to different sensor positions, highlighting the importance of considering the unique characteristics of each individual cylinder. This work emphasizes the need for vibration diagnostics in conjunction with machine learning as a non-intrusive, low-cost event detection process that also provides a base for necessary design guidance towards advanced real-time fault detection systems. Future work focuses on improving multi-signal integration to improve diagnostic technology. These findings support the development of more effective and reliable engine fault detection systems.