Abstract: In-service structural load monitoring at critical locations based on flight data is a key technique for structural prognostic and health management system of the aircraft, and a robust and precise load model is important for load monitoring and fatigue life prediction of the aircraft. Based on multi-linear regression analysis, an approach for synthetically screening the optimal combination of input variables is presented and introduced in detail. First, the multi-collinearity diagnosis should be conducted to attenuate the multi-collinearity between independent variables. Then, residual analysis is carried out to remove abnormal observations. Finally, stepwise regression is used to find the best combination of independent variables. Traditional multi-collinearity diagnosis methods have some defects, so two more feasible methods are put forward to reduce multi-collinearity based on partial correlation coefficient method and auxiliary regression equation method, respectively. As a case study, the load and stress data at a critical wing attachment bulkhead location of a typical fighter are used to illustrate the procedures of screening the optimal combination of input variables. It is proved that the approach can not only ensure the accuracy of aircraft structural load model, but also improve the stability and robustness of the model.

Key words: optimal combination of input variables, regression analysis, multi-collinearity, residual analysis, stepwise regression