Filtering And Classification Of 3D Engineering Surface Using High Definition Metrology

The surface morphology of machined parts not only affects the performance of the parts directly, but also can provide feedback on the manufacturing process. Therefore it is necessary to accurately analyze and evaluate the surface morphology of machined parts. With the rapid development of three-dimensional measuring techniques especially high definition metrology, it is possible to obtain cloud data from engineering surfaces in quite a short time. Based on high definition metrology, this paper focuses on the research of three-dimensional engineering surfaces, involving two aspects including surface filtering and classification.This paper analyzes the existing surface filtering methods and notices that wavelet filter has remarkable advantages and is the current research focus. However, the existing wavelet filtering methods focus on two-dimensional surfaces, and are not entirely applicable to the case of three-dimensional surfaces. In light of this, a new multi-scale geometric analysis tool called shearlets is introduced to the filtering of three-dimensional engineering surfaces. The framework of shearlet filtering is established and its metrological characteristics are analyzed. The performance of the proposed method is validated by both simulated surface and several real-world 3D engineering surfaces. Also the results are compared with the current ISO 11562 standardized Gaussian filter. The experimental results demonstrate that 1) shearlet filtering has many nice properties such as multi-resolution ability, highly directional sensitivity, shift invariance and high calculation efficiency; 2) shearlet filtering has zero phase and ideal amplitude transmission characteristics; 3) To different kind of engineering surfaces, shearlet filtering is effective for the separation of surface roughness, waviness and form; 4) The differences between the surface parameters through ISO 11562 standardized Gaussian filter and shearlet filter are within a reasonable range(5%-25%).Classification of engineering surfaces is an effective way to identify the surface quality of machined parts. This paper proposes a multiclass surface classification method called MPO-SVME. The proposed method takes high definition surface data as input, and then constructs basic classifiers using Support Vector Machine(SVM). This method can improve the classification accuracy of engine block top surfaces by using a selective ensemble strategy to select optimal classifiers. The performance of the proposed method is validated in case study and the experimental results demonstrate that the classification accuracy of using a single multiclass SVM classifier is 89.58%, while the MPO-SVME algorithm can increase the average classification accuracy to 92.22% with only selecting 2-3 basic classifiers. Therefore, the proposed classification method can increase the classification accuracy of engineering surfaces with only a handful of selected classifiers.