Multimodal Data Fusion for Structural Health Monitoring of Composite Materials

• Grant number: 25/11451-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2025
• End date: August 31, 2026
• Field of knowledge: Engineering - Mechanical Engineering - Mechanics of Solids
• Principal Investigator: Kayc Wayhs Lopes
• Grantee: Tiago Yuzo Yoshida Simao
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Composite materials, widely used in sectors such as aerospace, aeronautics, and automotive, stand out due to their superior strength-to-weight ratio compared to isotropic materials like aluminum and steel. However, damage in these materials can evolve rapidly and in complex ways, increasing the risk associated with operations involving this type of material. Therefore, it is essential to develop methods that facilitate the detection and monitoring of such damage to enhance operational reliability. For structural inspection, traditional methods such as visual analysis or non-destructive testing may require specialized labor and also involve a high dependence on human interpretation, which makes real-time monitoring more difficult and increases the chance of human error during the process. Advances in computing and statistics have enabled the development of various computational techniques to enhance structural health monitoring processes by using signals collected from the structure itself over time. However, many of these techniques investigate the presence of damage based solely on signals from a single sensor, such as vibration signals, which can limit the efficiency and accuracy of diagnostics. In this context, the objective of this research project is to develop a computational model based on deep learning, combining data from different sensors, a technique known as multimodal data fusion, for the structural health monitoring of composite materials. For this study, a dataset provided by NASA on CFRP (Carbon Fiber Reinforced Polymer) plates will be used. The specimens were subjected to fatigue tests, during which piezoelectric actuators and sensors were employed to capture ultrasonic wave signals at various frequencies, and X-ray images were acquired periodically throughout the fatigue cycles. The available data will be used to train neural network models whose modalities will be combined through multimodal fusion to detect and quantify delamination and estimate the remaining useful life of the structure. It is expected that this approach will enable earlier and more accurate diagnostics of the structural condition, increasing reliability and efficiency in the maintenance of critical components.