SETUP PLANNING USING HOPFIELD NEURAL NETWORK AND SIMULATED ANNEALING

Abstract

This thesis reports a new approach to setup planning of prismatic parts using Hopfield neural network coupled with simulated annealing. The approach aims at solving two main problems which have not been satisfactorily solved yet, i.e., (1) the conflict between feature precedence relationships and the sequence of final setups, and (2) the re-setup along certain tool approach direction(s). The approach deals with setup planning in two stages, i.e., (I) sequence all the features of a workpiece according to its geometric and technological constraints, and (2) identify setups from the sequenced features. In the first stage, the task of feature sequencing is converted to a constraint optimization problem (COP) which is similar to the travelling salesman problem (TSP). The setup time due to setup and tool changes is incorporated into the 'distance' between features. The energy concept in Hopfield network is adopted to model the TSP problem by attaching the constraints of precedence and critical tolerance relationships between features as penalty functions. Simulated annealing is incorporated to solve the COP and the local minima problem. The feature sequence obtained aims at minimizing the number of setups and tool changes while ensuring that no feature precedence relationship is violated and critical tolerance violation is kept to a minimum. In the second stage, setups are generated from the sequenced features using a vector intersection approach based on common tool approach directions. The performance of this approach is demonstrated by the simulation algorithm and a few case studies. The results show that this approach is able to perform process planning tasks by considering various constraints concurrently.