Damage location and identification of the wing structure with Probabilistic Neural Networks

The subject of Structural Health Monitoring (SHM) is emerging as an area of interest for aerospace, mechanical and automotive industry. It is defined as a system with the ability to continuously and automatically monitor the physical states of a structure and outer environment by means of embedded or attached sensors. Then, the SHM system can interpret adverse changes caused by the structure damage with the intention of improving the structure's reliability and reducing the maintenance cost. Therefore, the key of implementation of SHM is that SHM system knows what changes in the structure to look for and how to identify them. Usually, the process of the damage identification is to detect the damage, extract the features which can accurately represent the characteristic of the damage, and identify the damage. Although, there has been an amount of work performed in the area of SHM, most efforts are focusing on civil structures and the technology level has reached some modest degree of maturing. For the particular structures like the aircraft structure, due to its special characteristics, there will be a lot of problems needed for analysis. In this paper, taking the aircraft wing structure as an example, its finite element model has been constructed to simulate the fifteen damage patterns according to the actual mechanical properties of the wing structure. Combined with the dissymmetry of the geometry characteristic of the wing structure, the change of the natural frequency is extracted as the feature parameter. Then, the damage feature vector based on the feature parameters is constructed as the input vector of the Probabilistic Neural Network (PNN). Finally, taking σ =0.1, 0.05, 0.01 and 0.001 respectively, the PNN is trained with the damage training samples, and the ability of damage identification of the PNN is vivificated by the verification samples. The results of the PNN training and verification show that the damages of the upper and lower skin at the front spar are hard to distinguish correctly. In order to recognize them correctly, the PNN must increase other feature parameters which are more sensitive to the damages at the two locations. Anyway, the method of PNN has great identification effect for the wing structure, and it can be applied to the aircraft structure health monitoring.