Surface Roughness Measurement Based On Laser Spackle Image

The expectation to the stability, accuracy and long lives of the mechanical product become increasingly demanding, and these largely depend on the quality of their parts. The traditional part shape, size and other parameters can be more precise control with the development and popularization of computer aided design and manufacture. The surface roughness was not widely appreciated enough in the past, which has significant implication to the fitting property, wear resistance, fatigue strength, corrosion resistance and sealing. Therefore, more and more attention focused on the measurement and control of the part surface roughness. The existing methods for surface roughness measuring including contact measurement, measurement based on machine vision, and based on optical, all of which can hardly measure roughness on-line real time. Some just explore of some rules preliminary, and haven’t formed a complete measurement scheme.The paper built a system for laser speckle image acquisition, and proposes a solution for recognition of surface craftsmanship and surface roughness measurement. The paper compares different solutions for no uniform illumination, and uses homomorphic filtering based on Butterworth filter to solve this problem. Through the improving of the local binary pattern, the M-LBP(magnitude considered LBP) is proposed. Combining with K-Nearest Neighbor algorithm, the M-LBP can achieve the goal of surface craftsmanship recognition, and the recognition accuracy rate is improved significantly. With the Tamura texture theory, the paper extracts texture coarseness Fcrs, contrast Fcon and direction Fdir of the speckle image. Using the monotonic relationship between Fcrs and surface roughness Ra, the function is fitted between them. Finally, the effectiveness of the proposed scheme is verified by the experiment. The accuracy rate of surface craftsmanship recognition reach 98%, and the fractional error of the measurement for plain grinding parts is not bigger than 7%. The paper has built measuring models for different kinds of parts, and proposed a new ideal for surface roughness measurement.