Linear Workshop Process Scheduling using Neural Nets

The paper describes the development and application of a neural network based approach to predicting the process time in a linear construction workshop. Specifically, the workshop situation considered was the production of molded timber strips. The objective was to establish an accurate means of estimating how long a molding order would take to process, thus to enable more efficient sequencing of multiple jobs and to provide customers with accurate estimates of delivery dates. The neural network model was developed to predict the production rate of molding (linear feet per minute) as a function of a range of independent variables (including molding profile, type of timber, and thickness of timber). A number of alternative neural network architectures were studied and their performances compared, and a sensitivity analysis was undertaken to identify the importance of the various independent variables. Training and testing of the neural networks was undertaken using a set of 3,000 sample jobs collected during 1998 and 1999. The neural network proved to be a viable approach to modeling this type of linear process. Suggests are provided for further improving performance of the model by use of alternative neural systems.