Prediction And Control Of Surface Roughness In Cylindrical Traverse Grinding Process

Prediction and Control of Surface Roughness in Cylindrical Traverse Grinding ProcessAs the chief machining method of shaft parts, cylindrical traverse grinding is one of the most important machining methods in machining field. In order to improve efficiency and reduce labor cost, grinding technology is developing toward automation and intelligent. The method to predict and control the state during grinding is important to realize the automation and intelligent of grinding. The grinding result directly affects the final quality and performance of product because the grinding is usually the ultimate machining procedure. Because the cylindrical traverse grinding process is non-linearity and randomcity with too many influence factors, it's difficult to predict and control the surface roughness accurately by traditional method. The most applied BP network model was selected in this paper, and the genetic algorithm was utilized to solve the congenital limitation of BP ANN. Then the GA-BP neural network was built in this paper to predict the surface roughness. The adaptive fuzzy neural network controller for the workpiece surface roughness for cylindrical traverse grinding was put forward in this dissertation. According the prediction of GA-BP prediction model, the workpiece surface roughness was controlled used the adaptive fuzzy controller.Based on the requirement of experiment and testing, the MM 1320 grinder was reformed during the research. The construction of cylindrical traverse grinding open controlled system was established in this paper, and the man-machine interface wrote by Visual Basic 6.0 programme language. The automatic feed control of transverse and vertical worktables as well as the rotational speed control of grinding wheel and workpiece were realized in this research.Based on cylindrical traverse grinding open controlled system, the orthogonal experiment was designed with three factors of grinding depth, rotational speed and transverse feed rate. The effect of three factors on the workpiece surface roughness was obtained by means of pole difference analysis and variance analysis. According to the results of orthogonal experiment, the overall experiment is designed in order to provide training data and test data for prediction and control model.Artificial neural network (ANN), especially BP network, including its basic theory, application, advantages and shortcomings were introduced in detail in this paper. Since the genetic algorithm is strong in global search, in order to improve the performance of traditional BP network, the genetic algorithm was utilized to optimize the original weights and threshold of BP network. The prediction model of cylindrical traverse grinding surface roughness was established based on GA-BP network, and then the prediction results of GA-BP model were compared with that of BP network. The result indicated that the prediction effect of GA-BP network was better than traditional BP network model. The GA-BP prediction model is more accurate and efficient, which can satisfy the requirement.The basic theory and concept of fuzzy inference system (FIS) is introduced in this paper. The structure and the learning algorithm of Adaptive-Network-Based Fuzzy Inference System (ANFIS) were introduced in detail. To solve the problem that membership functions and fuzzy rules are difficult to defined, a adaptive fuzzy control model was established based on the ANFIS. According to the results of orthogonal experiment, the transverse feed rate was selected as control variable of adaptive fuzzy controller when surface roughness was the target. Experiment results verified that the proposed surface roughness adaptive fuzzy controller in the cylindrical traverse grinding can obtain the satisfactory roughness accuracy.