Many Just-In-Time (JIT) manufacturing environments generate operational data reflecting both efficient and inefficient factory performance. Frequently data for inefficient performance is lost or discarded for fear of replicating poor performance. The purpose of this paper is two fold. First, historical JIT shop data is analyzed using a genetic algorithm (GA) to determine which shop factors are important determinants offactory performance. Second, subsequent to these important factors being identified by a GA, an artificial neural network (ANN) is used to learn the relationships between these factors and factory performance. The ANN can then be used to predict factory performance for future shop conditions and enhance shop performance. While ANN learning techniques have previously been applied to JIT production systems (Wray, Rakes, and Rees, 1997) (Markham, Mathieu, and Wray, 2000), these techniques have only been trained on data sets that reflect an efficient factory. Mathieu, Wray, and Markham (2002) investigated inefficient and efficient JIT factory performance but did not deploy either ANNs or a GA. In this paper an example application is presented using a GA to specify important shop factors and to predict saturated, starved or efficient factory performance based on dynamic shop floor data.
Wray, Barry A.; Markham, Ina S.; and Mathieu, Richard G.
"An Artificial Neural Network Approach to Learning from Factory Performance in a Kanban-Based System,"
Journal of International Information Management: Vol. 12
, Article 7.
Available at: https://scholarworks.lib.csusb.edu/jiim/vol12/iss2/7