A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Advanced Technologies

Thesis (M.Sc.) - Asian Institute of Technology, 1998

This thesis reports a case study of production system of Metals Workshop of PT INKA, a rolling stock manufacturer in Indonesia. The objective of the study is to suggest ways to improve the performance of the existing production system and layout based on the application of Binary Adaptive Resonance Theory (ARTl) neural network and cell formation. The ARTl algorithm is used to block-diagonalize the binary part-machine incidence matrix. ARTl algorithm is applied twice. ARTl Column is used to classify the machines and then ARTl Row is used to classify the parts. Since the solution by ARTl is not practical for this data from a case study, then the extended procedure is generated to improve the ARTl algorithm. The result is not only block-diagonalization of part-machine matrix, but also the machine cell formation. Two modifications for ARTl have been incorporated for improving the algorithm. The presentation of input vectors in decreasing number of ls enables the ARTl to classify the input vectors more consistently. Representing the bottom-up weight matrix in a vector form in computer program has also been found to the capability of ART 1 by: 1. Reducing the computer memory requirements 2. Faster computation Based on the part-machine grouping, manufacturing cells have been designed to improved the machine utilization an reduction in machine and manpower requirements. Arranging the existing machines into cells has led to substantial reduction in number of machines required and corresponding reduction in manpower. This also has resulted into uniform utilization of the manufacturing resources and avoids unnecessary duplication of manufacturing resources.