DEVELOPMENT OF A GENERIC PROCESS PLANNING SYSTEM

Abstract

Enhancing the adaptability of CAPP systems and developing appropriate mathematical models, modelling algorithms for the process planning parts, process equipment and planning methodologies have been recognized as emerging tasks in the current CAPP research and development. This thesis contributes to these critical issues in the following aspects: To develop generic CAPP systems, the mechanisms for the construction of process planning systems must be flexible enough to incorporate new application-specific rules, standards and expertise. This requires the system to have strong capability, like a human planning engineer, to upgrade its knowledge and skills regularly from training, planning practices and creative learning. In this research, the adaptability of GCAPP system is achieved by implementing connectionist modelling or neural networks techniques. Due to the self-learning ability of neural networks, the developed models demonstrate very strong adaptability in modelling all the time-independent activities of process planning such as feature recognition, feature prioritization and operations planning. To model the hierarchical nature of process planning intelligence, an intelligent system based on the holo-informatic intelligence formula is developed to simulate the intuitive, syntactical and functional inference capabilities of human planners. The intuitive knowledge, mainly involved in the part understanding stage, is processed using 'neocognitron' mechanism for feature recognition and manufacturing environment description. The philosophy of converting the part geometry back to algebraic form and processing it with neural networks is the basis of the developed system to deal with low-level intuitive pattern recognition. Learning new features and recognizing intersected features are two outstanding features of GCAPP system due to the implementation of this concept. Most feature-based planning activities such as operations selection and tools selection, etc., are syntactical and time-independent in nature. FAM technique is employed in the GCAPP system for the processing of syntactic information. The implementation of FAM demonstrates a seamless integration of intuitive and syntactical information and strong capabilities in fast decision making, self learning and uncertainty processing. The functional information in process planning is mainly dealt with in the operations sequencing model. Feature prioritization, an intelligent way of experienced engineers to develop process sequences for complex parts, is computerized in this research by means of evaluating the manufacturabilities of features. The generation of operation sequences, based on the feature's priorities, is performed using a knowledge-based system. This sub-system integrated with the FAM-based planning models and the neuron-based part recognition models form a hierarchical intelligent structure of the GCAPP system.