Revolutionizing strain gauge testing with a novel neural network approach for deformation measurement

초록

Engineering applications like material testing and quality control need precise deformation measurement. Traditional approaches, such as strain gauges and digital image correlation (DIC), are expensive and complicated. This paper introduces multi-stage U-Net (MSU-Net), a neural network-based system for deformation assessment that was trained on strain gauge test videos. Speckled specimens under tensile tension were captured at 720p, 1080p, and 2160p resolutions using a smartphone. MSU-Net predicted segmentation produced deformation values, which were then compared to testing machine measurements, yielding mean absolute error (MAE) values ranging from 0.35 to 1.45, indicating high accuracy. This innovative approach not only enhances accuracy but also significantly reduces the costs associated with deformation measurement, making it accessible for a wider range of applications in various industries. Low breakpoint error percentages in most tests demonstrated the system's capacity to accurately catch crucial deformation events, making it useful for applications that need precision failure point identification.

키워드