Impact Monitoring of Large and Complex Structures Based on Transfer Learning

Abstract

Aircraft structure impact monitoring is important to the safe operation of aircraft. However, aircraft structures are often structurally complex, increasing the uncertainty of the signal during transmission. Traditional impact monitoring methods need to obtain sufficient structural change signals through dense sensor arrays to obtain good monitoring results. But too many sensors can increase the cost of operating an aircraft. Therefore, this paper adopts sparse sensor array arrangement, proposes a two-step impact monitoring strategy from region to point location, and adopts deep learning and traditional methods to monitor impact events. Firstly, the test structure is divided into several regions of a certain size, and a model capable of accurate regional location is trained by convolutional neural network. In this process, in view of the large size of the aircraft structure and the difficulty in obtaining training data, the transfer learning strategy of model fine-tuning is adopted to transfer the trained feature knowledge of the source domain model to the target domain model, reducing the cost required for data acquisition and training model. Then in the second step, on the basis of accurate regional positioning, weighted centroid method is used to estimate the impact location.