A roughness Measurement Method Based on Genetic Algorithm and Neural Network for Microheterogeneous Surface in Deep-Hole Parts

It is difficult to measure the surface roughness of microheterogeneous surface in deep-hole parts due to the limitation of measurement space. In this paper, we propose a new method based on microscopic vision to detect the surface roughness of R-surface in the valve. First, the clear microscopic image of R-surface is obtained by the established microscopic system, which is mainly fabricated by the long working distance lenses of digital microscopic camera. Thereafter, based on genetic algorithm (GA) and feed-forward back propagation artificial neural network (BP-ANN), a hybrid method is proposed to predict the surface roughness. In this method, the microscopic image features of R-surface are taken as the inputs to the hybrid model. GA is employed to search the optimal initial weights and thresholds of BP-ANN, which resolves the problem that training methods of BP-ANN are much sensitive to initial weight values and thresholds. In addition, in virtue of a three-dimensional surface profiler, the targets of hybrid model are calibrated by the actual roughness values of R-surface in the sample valves, where the sections of sample valves over R-surface are cut. Finally, experiments on the microscopic image acquisition and roughness calibration are conducted, as well as the prediction experiments. Moreover, the analysis results indicate that the proposed measurement method based on GA and BP-ANN exhibits high precision and stability for evaluating the microcosmic surface roughness of microheterogeneous surface in deep-hole parts.