Research Of Laser Speckle Surface Roughness Measurement Method Based On MRF In Wavelet Domain

The surface roughness is the most commonly used parameter to describe surface morphology for machined surface. The surface roughness can be used to assess various micro machined surfaces, and the value of the surface roughness can affect performance and life period of machines and instruments. It is crucial to research how to measure and evaluate the surface roughness. A new surface roughness measurement method is presented based on texture analysis by laser speckle in this paper. Based on statistical properties of laser speckle and computer texture analysis theories, texture features represented the surface roughness are extracted by modeling the laser speckle patterns of machined surface. The relationships between texture features and surface roughness are researched. Experimental measurement system is set up. The measurement results of different machined surfaces based on texture models are computed and compared. The setup configuration’s parameters changing to what degree can influence the experimental results is researched, therefore, the system robustness is discussed. This method can realize surface roughness measurement by single laser speckle texture pattern.The main research works are as follows:1. Based on the texture analysis method, the relationships between characteristics of laser speckle and surface roughness are researched. Laser speckle texture analysis models which can characterize the surface roughness are proposed. The monogamous relationship between surface roughness Ra and a texture feature is established by texture analysis of laser speckle patterns of machined surface and extracting the texture features.2. Markov Random Field models (MRF) of model methods are researched. Surface roughness extraction methods based on Gauss-MRF model and Gibbs-MRF model of laser speckle are proposed respectively. For Gauss-MRF model, texture features from second order to sixth order neighbor models are extracted according to neighbor pixels’ magnitude and direction. For Gibbs-MRF model, the neighbor pixels are defined by first order and second order clique. The parameters β2～β9, are extracted for texture clique, which can be used as texture features charactering for Gibbs-MRF model.3. Wavelet model of signal processing method is researched. Surface roughness extraction method based on wavelet model of laser speckle is proposed. Laser speckle patterns are decomposed by wavelet transform. Coefficient co-occurrence matrixes containing lots of surface roughness information are configured based on dependencies among high frequency sub-band coefficients. The values of mean, variance, energy, and entropy of coefficient co-occurrence matrix are extracted as texture features.4. Combining the above mentioned two models Markov in wavelet domain model is proposed. In the other words, surface roughness information can be extracted from laser speckle patterns based on Markov in wavelet domain model. The model is inspired by multi-resolution and highly discriminative nature of the wavelet representation and the statistical regularities capability of the MRF model. The model captures significant intra-scale and inter-scale statistical correlation of wavelet coefficients. The model texture features are extracted to investigate the relationships between texture features and surface roughness Ra.5. The experimental system is set up. The relationships between surface roughness Ra and texture features of turning, vertical milling and horizontal milling surface are researched based on the system using above mentioned three models. The results validate that Markov in wavelet domain model is the most suitable model to characterize surface roughness of turning, vertical milling and horizontal milling surface, and the model can characterize their surface roughness in some larger range also. By this experimental system, the changes of texture features are also investigated when the setup configuration’s parameters are varied. The parameters variances include laser work distance between specimens and CCD, surface texture direction related to the laser orientation, laser beam diameter, and laser light power. The result is that Markov in wavelet domain model is the best model to characterize surface roughness.Three kinds of texture analysis methods are applied in this paper. Texture features which can characterize surface roughness are extracted. According to the results, three texture models are analyzed and compared. The best texture model which is suitable to surface roughness measurement is summarized. The surface roughness measurement system consists of laser, CCD without lens and computer. It has some advantages such as simple system structure, easy operation, non-contact and nondestructive testing. The method presented in this paper has potential and magnificent reference value for online surface roughness measurement. If this measuring system is calibrated with the standard specimen surface beforehand, the surface roughness actual value Ra can be deduced in case of the same material surfaces at the same manufacture conditions.