Session: 21-01: Poster Session

Paper Number: 98347

98347 - Development of Lamination Layer Signal Cancellation Technique for Cfrp Composite Using Autoencoder 

Recently, low-carbon green energy developing units are getting much attention due to the exhaustion of energy and environmental issues. The wind power generator, a typical low-carbon green energy developing unit, has less influence on the weather. Also, its generation quantity compared to the ground contact area has a point. Moreover, the generation quantity can be expanded through enlarging the surface area of wind power generator’s wind blade. Before, GFRP (Glass Fiber Reinforced Plastic) was the most popularly used material for wind blade of wind power generator. However, the usage of carbon-fiber in lieu of glass fiber continues to increase regarding the efforts to improve the efficiency of the wind power system. The spinning radius of the wings has increased by 30% every year and now the blade is as big as 120m, and the set-up (installation) cost of wind power generator is increasing while the efficiency is reducing. Before, the CFRP used for wind blade has been made with the prepreg sheet aligned in single direction. However, currently in 3MW class high-capacity wind turbine, instead of the regular Prepreg CFRP, the CFRP made of pultrusion method is used. Pultrusion CFRP has more durable and lighter characteristic compared to the Prepreg CFRP. However, the lamination layer of pultrusion CFRP tends to separate and break that causes various types of defects when stress occurs in transverse direction. Moreover, in CFRP structural integrity evaluation using ultrasonic testing, the echo signal of each layer of pultrusion CFRP makes the evaluation of defect signal very difficult. Therefore, in this study, a new technique implementing Autoencoder, a type of deep learning neural network, for pultrusion CFRP structural integrity evaluation is used to effectively eliminate the echo signal of layers in pultrusion CFRP.

Presenting Author: Yun-Taek Yeom SungKyunKwan University