Vibration Measurement by Machine Vision Using Up Sampled Cross Correlation and Finite Difference Algorithm

Abstract

Recently, machine vision systems play an important role in engineering measurements, especially in vibration measurement. This research proposes a new intelligent vibration measurement technique using machine vision systems for fault detection and condition monitoring. A non-reflective white paper sticker of known size, marked with one or more coloured dots, was pasted on the surface of the machine (farm tractor) to obtain error-free results in template matching. Tracking was performed on dots in the image region of interest (ROI). It allows the freedom to capture images without giving scale-factor training to the vision system and simplify the camera calibration. Vibration was calculated using up-sampled cross-correlation (UCC) and a finite difference algorithm (FDA) from images captured by a machine vision system equipped with a macro lens that can capture a specific point on the object very finely. The white sticker significantly reduces image noise in the region of interest, eliminating the need for pre-processing video frames. With images adjusted to real-world dimensions, displacement values are more accurate, allowing for precise vibration calculations at multiple points in a component. This study validates measurement uncertainty against CSIR-NPL, showing lower error percentages at higher vibration levels. These findings support their use in vibration analysis.