This full-day workshop will be a practical guide for nonlinear regression modeling. Although theoretical analysis behind techniques will be revealed, the takeaway will be the participant’s ability to:

• Choose appropriate concepts for defining the regression objective

• Choose an optimization approach and criteria for convergence

• Apply both data-based and logical criteria for model validation and model discrimination

• Design experiments for data generation that support model validation

• Select an appropriate model design considering both order/complexity and utility in use

• Estimate model uncertainty based on data variability

Participants will receive a copy of a new textbook, which will be used as the workshop notes (or, the author’s manuscript, if the publication is delayed). Exercises and code can be implemented in any environment, but Excel/VBA will used as in-workshop examples and exercises. The author’s software will be provided (Leapfrogging as an optimizer, steady-state as stopping criteria, bootstrapping for estimating model uncertainty) along with several case-study data sets for revealing course concepts. Participants are invited to bring a laptop with Excel version 2010 or higher for in-class applications.

Models based on data are often central for model-based control, forecasting, training simulators, analysis and diagnosis, mechanism validation, design scale-up, and supervisory optimization. For many of these applications nonlinear models are preferred in order to capture the process/device behavior. Regression is the procedure of fitting models to data, and nonlinear regression means that the adjustable model coefficients do not appear linearly within the model; and, even for seemingly linear models, a variable delay introduces a nonlinear model coefficient, and which is also constrained to integer values. Workshop topics will include equation structures, optimization of parameter values in the presence of constraints and local traps, choosing optimization convergence criteria based on model properties, data preprocessing and post-processing, data-based model validation, discrimination between models, design of experiments that support validation outcomes, propagation of uncertainty, and model utility evaluation. This is not the standard linear regression approach to develop response surface model structures, and classic experimental designs such as Latin Square, Box, and Star plans. This workshop will focus on techniques for nonlinear regression and enabling design of experiments.