Value Prediction of Design Parameters Based on Weighting Least Squares

Abstract

Parametric equipments or standard parts usually have many different types of original design parameters. So when designing some new specifications, it requires a lot of estimation or trial and error to determine the value trends and intervals of other unknown design parameters. Based on a finite number of historical examples of design parameter groups, the paper gives an algorithm to fit value trend line using multivariate linear weighted least squares method, whose weights are designed by using distance-proximity coefficient and correlation coefficient. The algorithm uses a small amount of new design parameters, fits value trend lines of other unknown parameters, predicts all other design parameters, finally makes up a design parameter group for a new specification. Two test results of standard parts from home and abroad show that, the accuracy of value prediction is able to meet the requirements of engineering applications.