Research On The Application Of Granular Computing Theory In CAPP Intelligentization

As a key link of manufacturing informatization, Computer Aided Process Planning (CAPP) plays an important role in guaranteeing the quality of process planning, improving the effciency of process planning and reducing the labor intensity of process staff. Intelligence is one of important directions of CAPP technology. The keys of developing the intelligent CAPP are process knowledge acquisition and process decision model construction. Efficient acquisition of large amount of valuable process knowledge can enrich the knowledge database of CAPP system and provide knowledge guarantee for knowledge based intelligent process plan. Effective construction of highly accurate process decision model and ulitization of the model for designing processing scheme and planning process route are the concrete embodiment of intelligent CAPP.As a emerging computing paradigm of simulating the human cognition, Granular Computing (GrC) is one of the hot fields in artificial intelligence. To explore the way of enhancing intelligent CAPP, this paper makes a deep study on the application method of GrC in the two key technologies of intelligent CAPP. In the aspect of process knowledge acquisition, a typical process route acquisition approach based on the theory of fuzzy quotient space is proposed, and the technological architecture of the process knowledge acquisition in GrC environment is designed. In the aspect of process decision model construction, a construction method of process decision model based on the theory of fuzzy quotient space and Radial Basis Function (RBF) neural network is proposed, and the optimum design of RBF neural network is realized. Throughout the paper, the main research results and characteristics are as follows:1. For the first time, the ideology and methodology of GrC is applied to process planning domain, based on which the application scope of GrC is broadened. Granulation of process information makes the characteristics of different information granules more obvious, and reduces the solving difficulty of problem space at the same time. Measurement for the granularity of information granules is used to evaluate the effect of granulation. Computing with information granules is applied to acquire process knowledge and construct process decision model.2. In the study on acquisition of typical process routes, an evaluating model for process routes similarity is established based on features of parts. This model well reflects the essence of the process route similarity. Aiming at the shortcoming in subjective determination of the cluster number based on traditional clustering analysis, which influences the quality of acquired typical process routes, the idea of using the theory of fuzzy quotient space, which is one of theoretical models of GrC, to acquire typical process route is proposed. Firstly, process information granules are formed by granulation of the process routes, and a series of hierarchical granular spaces are constructed. Then the granular spaces with hierarchical structures are measured by information entropy for judging the effect of granulation. Finally, through applying Longest Common Subsequence (LCS) algorithm, some typical process routes are obtained from process information granules that are contained in the best granular space. This method improves the quality of acquired process knowledge.3. In the study on construction of process decision model, aiming at the defects of BP neural network in slow convergence and easily falling into local optimum, RBF neural network that can quickly converge to the global optimal solution is used to replace BP neural network for constructing the process decision model. Aiming at the defect of determining the number of hidden neurons based on traditional methods which cannot make full use of the information provided by the samples, the idea of integrating the theory of fuzzy quotient space into the RBF neural network is proposed. Firstly, based on input information and output information of process rule samples, weighted Euclidean distance is adopted for granulation of process rule samples, and information entropy is used to measure the granulation effect. Then information gain is applied to determine the best granular space. Finally, those process information granules contained in the best granular space are used to compute the number of hidden neurons and their corresponding centers and deviations. Through this method, optimization design of the hidden layer structure is realized, and the accuracy of process decision model is enhanced.