The Research On Wing Load Analysis Based On Data Mining

During the flight, the aircraft’s wing will be affected by load, and the measurement of the load on the wing works through the whole design process of the plane. In order to measure the wing load, an load calibration test is conducted for establishing the load model in which the selection of the strain bridges is key. Therefore it is essential to have a research on how to choose the strain bridges effectively and establishing the best load model to improve the predictive ability for the unknown load.There are some shortcomings in the analysis methods of load model at present stage:first, there’s still room for improve the efficiency of strain bridges selection; second, the irreversible problems of the matrix is not considered in solving parameters of the load model by normal equation; last, without taking the generalization ability into account in the train stage, the established load model is weak for predict for the unknown load.The thesis studies the technology of the wing load analysis based on data mining, and proposes a model of wing load analysis based on data mining for solving the above shortcomings. In the thesis, the work we have done is as follows:1. We adopt a hybrid feature selection algorithm to select strain bridge that improves the efficiency of the strain bridge subset selection. First of all, according to a feature selection algorithm based on the correlation measurement standard for filtering strain bridges set initially,we remove the strain bridges not enough related to the target load to downscale the strain bridges. Second, we adopt a feature selection algorithm based on genetic algorithm to select bridge set further and eliminate redundancy between the strain bridges.We pick out the optimal combination of bridges and establish the load model.2. During the period of establishing the load model, we solve the model parameters by batch gradient descent algorithm and incremental gradient descent algorithm, and compare the efficiency of the two algorithms. By the iteration algorithm, we avoid the irreversible problems of the matrix in solving parameters by normal equation.3. For the potential overfitting problem in establishing the load model, we propose a regularization solution which reduces the risk of overfitting and improves the generalization ability of the model.4. We establish the load model in the real wing load data set, and test the model by cross validation. The experiment shows that the model improves the efficiency of strain bridge selection, reduces the risk of overfitting and has better predictive effect for unknown load.