Design of Engineering Experiments Part 7 – The 2k.pptVIP

Montgomery DOX 5E Fitting Regression Models It is usually interesting to build a model to relate the response to process variables for prediction, process optimization, and process control A regression model is a mathematical model which fits to a set of sample data to approximate the exact appropriate relation Low-order polynomials are widely used There is a strong “interaction” between the design of experiments and regression analysis Regression analysis is often applied to unplanned experiments Linear Regression Models In general, the response variable y may be related to k regressor variables by a multiple linear (first order) regression model y = bo + b1x1 + b2x2 + ? + bkxk + e Models of more complex forms can be analyzed similarly. E.g., y = bo + b1x1 + b2x2 + b12x1x2 + e = y = bo + b1x1 + b2x2 + b3x3 + e Any regression model that is linear in the parameters (b values) is a linear regression model, regardless of the shape of the response surface The variables have to be quantitative Model Fitting – Estimating Parameters The method of least squares is typically used Assuming that the error term e are uncorrelated random variables The data can be expressed as The model equation is yi = bo + b1xi1 + b2xi2 + ? + bkxik + ei Least square method chooses the b′s so that the sum of the squares of the errors ei is minimized Model Fitting – Estimating Parameters Model Fitting – Estimating Parameters Fitting Regression Models in Designed Experiments Fitting Regression Models in Designed Experiments Fitting Regression Models in Designed Experiments Example 10-3 Fitting Regression Models in Designed Experiments Example 10-4 Fitting Regression Models in Designed Experiments Montgomery DOX 5E The least squares function is The least squares estimators ( ) must satisfy and or Or in the matrix notation y = Xb + e The least squares estimators of b are The fitted model and residuals are Example 10-1: Response: viscosity of a polymer (y) Variables: