Automated measurement of Vickers hardness using image segmentation with neural networks

Vickers hardness is widely used in material performance testing due to its versatility and simple calculation. Traditional automated methods of measuring it are often easily affected by the unsatisfactory surface of the material, and yield large errors. For a more robust measurement of indentations that can yield results similar to those of manual measurements within an allowable range of error, this study proposes an automatic method to measure the Vickers indentation. It uses a convolutional neural network to segment the Vickers indentation from the background in images, and uses a bounding box to measure the length of the diagonal of the indentation. To train and verify the neural network, the authors constructed a dataset of images featuring indentations in samples composed of a variety of materials. Experiments were performed on standard hardness blocks, TiO2, copper, and nylon to compare the performance of the proposed method with the results of manual measurements. The results verify the robust performance of the proposed method on complex surfaces with deformable indentations. The code is available at https://github.com/SimonLeeCHN/IndentMes_code_ref. •The proposed method can accommodate images of Vickers indentation of any size.•It avoids interference by the background in its indentation measurements.•Its results are close to those of manual measurement in some scenarios.•It can help intuitively assess the surface properties of the material.•It accelerates the assembly line in industrial production and improves efficiency.