Most scheduling solutions fail when uncertainty or unscheduled situations arise that were not accounted for. Also, neural network solutions have one weakness: these networks may solve the problems well but give no clear indications of how or why a particular problem is solved. One must try to decipher the network weights to understand the results. This ability to obtain knowledge from the weights is very much a research issue. In this project we:

1. Updated the current simulation model to incorporate additional variables not accounted for in the earlier models. These included:
   1. supervisor preference;
   2. storage/holding tanks for intermediate material handling;
   3. shortages of labor and/or equipment, including maintenance downtime.
2. Used decision tree induction and our current data to automatically create a rule base to solve this same scheduling problem.
3. Examined possible hybrid neural network / decision tree solutions to this problem. The methods when combined resulted in an improved scheduling system.