A data mining approach to study the significance of nonlinearity in multistation assembly processes

Abstract

Linear models of multistation manufacturing processes are commonly used for variation reduction and other quality improvement purposes. Yet the nonlinear nature of variation propagation in multistation manufacturing processes makes people inevitably wonder at what point does the linear model cease to provide a reasonable approximation of the nonlinear system. This paper presents a data mining method to study the significance of nonlinearity effects in a multistation process. The data mining method consists of two major components: (i) an aggressive factor covering design, which uses a design set of affordable size to assess the significance of nonlinearity in a multistation process with hundreds of variables; (ii) a multiple-additive-regression-tree-based predictive model, which can help identify the critical, influential factors and partial dependence relationships among the factors and the response. Using the data mining approach, insights are garnered about how these critical factors affect the significance of nonlinearity in a multistation process. Decision guidelines are provided to help users decide when a nonlinear model, instead of a linear one, should be applied.

Acknowledgements

The authors gratefully acknowledge financial support from the NSF under grants DMI-032214 and DMI-0348150, and from the State of Texas Advanced Technology Program under grant 000512-0237-2003. The authors also appreciate the editor and the referees for their valuable comments and suggestions.

Notes

*RPC stands for the reduced product construction method.

*A Gaussian correlation function is used. The best performance is presented, chosen from using a constant, a linear, or a quadratic regression model.