Engineering Data Analysis

Solutions to two major data defects (autocorrelation and multicolinearity) commonly encountered in the use of multivariable linear regression analysis when applied to engineering data are described. A computer program, incorporating a new method for the determination of the ridge parameter, is introduced for their implementation. The methods described are then applied to a complex data set involving a demonstration of the use of densified refuse-derived fuel burned in an industrial spreader stoker boiler. The analysis shows that, had only the more conventional regression analysis been applied, the fits to some response variables would have been overvalued while others would have been undervalued, and that the relationships between many of the response and predictor variables would have been determined incorrectly.