ABSTRACT
At present, the application of machine vision methods for roughness measurement in production sites is limited by its adaptability to illumination variations during the measurement. In this study, a machine vision method for roughness measurement with robustness to illumination is proposed so as to explore the functions of its color image indices in improving the mathematical expression of the vector of three primary colors. Besides, virtual images of different-roughness surfaces were analyzed, the effects of the samples' surface texture orientations on measurement indices were discussed, and the singular value ratio was derived as an index for evaluating roughness. The experimental results showed that the samples' index values remained unchanged when the illumination was increased for both vertical and horizontal surface textures, indicating that the proposed method has strong robustness to illumination. In addition, the experimental results were verified by a support vector machine (SVM)-based method using 10 different-roughness test samples, with the verification range of 0.127-2.245 µm. It was found that the measurement accuracy reached 90%, suggesting that the proposed method is reasonable and feasible, and shows certain potential to be applied in engineering.
ABSTRACT
The existing machine-vision surface roughness measurement technique extracts relevant evaluation indices from grayscale images without using the strong sensitivity of color information. In addition, most of these measurements use a micro-vision imaging method to measure a small area and cannot make an overall assessment of the workpiece's surface. To address these issues, a method of measuring surface roughness that uses an ordinary light source and a macro-vision perspective to generate a red and green color index for each pixel is proposed in the present study. A comparison test is conducted on a set of test samples before and after surface contamination using the color index and gray-level algebraic averaging, the square of the main component of the Fourier transform in the frequency domain, and the entropy. A strong correlation between the color index and the surface roughness is established; this correlation is not only higher than that of other indices but also present despite contamination and very robust. Verification using a regression model based on a support vector machine proves that the proposed method not only has a simple apparatus and makes measurement easy but also provides high precision and is suitable over a wide measurement range. The impact of the red and green color blocks, the lighting, and the direction of the surface texture on the correlation between the color index and the roughness are also assessed and discussed in this paper.
