Vickers Hardness Value Test via Multi-Task Learning Convolutional Neural Networks and Image Augmentation

Hardness testing is an essential test in the metal manufacturing industry, and Vickers hardness one of most widely used measurements today. The computer-assisted requires manually generating indentations for measurement, but process tedious measured results may depend on operator’s experience. In light this, this paper proposes a data-driven approach based convolutional neural networks to measure value directly from image specimen get rid aforementioned limitations. Multi-task learning introduced proposed network improve accuracy measurement. material medium-carbon chromium-molybdenum alloy steel (SCM 440), which commonly utilized automotive industries because its corrosion resistance, high temperature, tensile strength. However, limited samples SCM 440 manual measurement procedure represent main challenge collect sufficient data training evaluation methods. regard, study introduces new mixing method augment dataset. experimental show that mean absolute error between output by architecture can be 10.2 further improved 7.6 if multi-task applied. Furthermore, robustness confirmed evaluating developed models with additional 59 unseen images provided specialists testing, provide evidence support reliability usability