Intelligent troubleshooting of vertical bandsaws, Leveraging Ensemble Learning on Low-Level Data

Abstract

Bandsaw machines are the backbone of the modern wood industry. Due to immense load stresses and operational speeds various parts may fail frequently, hence constant monitoring is required. Most of the time small faults like imbalance or sturdiness, which if not corrected, progress into cracks and malfunction. This research proposes an intelligent troubleshooting mechanism to monitor the health of high-speed rotary machines and enables the operator to be informed instantly about the abnormality. The test case considered is a vertical bandsaw machine. The faults studied are unwanted looseness or tightness of the moving parts of the machine like wheel pulleys, blade, and motor belts. Other frequently occurring faults associated with non-moving parts like blade guide (looseness) and cutting bed (tilt) have also been considered. The study is performed in three stages (a) collection of 6 degrees of freedom vibration data from machine operation in both stable and faulty conditions with an IoT device. The sampling rate is 5 Hz, (b) applying various state-of-the-art machine learning algorithms to the data. The best classifier is selected based on accuracy on normal and corrupted noisy data, (c) updating the results on a semi-digital twin model of the actual machine for the operator to decide the needed action. The proposed classifier has excellent accuracy on both clean (93.22%) as well as corrupted data (87.34%) with a small difference of 5.88%. The study is important as it gives a complete framework for the bandsaw machines' operation from fault detection to fault localization and actionable process.