A study of a rapid method for detecting the machined surface roughness

To achieve the surface roughness of the large-scale machined surface detection rapidly, a surface roughness detection method of “multi-dimensional feature parameters matrix + BP neural network algorithm + automatic acquisition of regions of interest” is proposed. A multi-dimensional feature parameters matrix containing the strong correlation coefficients related to the surface roughness is constructed by extracting various feature parameters of clean image based on the gray-level co-occurrence matrix. The BP neural network model is used to predict the surface roughness with multi-dimensional feature parameters matrix as input. The region of interest in the detected images is extracted consequently by gray value transformation, image filtering, and morphological methods, and the method of “multi-dimensional feature parameters matrix + BP neural network algorithm” is used to detect the surface roughness of the region of interest. The automatic surface roughness detection system is constructed by combining the proposed surface roughness detection methods to analyze the condition of the machined surface and give the surface roughness value rapidly. Compared with experiments, the detecting accuracy and efficiency of the developed system are evaluated regarding the different machined surface features. The results show that the relative errors between the clean image and the interfered image with chips are 6.41% and 5.46%, and the average value of a single detection time is not more than 1.15 s, which can meet the accuracy and time requirements of high-efficiency detection and provides certain technical support for the industrial automated detection.