A new Baseline-Free method using Gaussian process for damage quantification of composite structures

• Grant number: 19/11755-9
• Support type: Scholarships abroad - Research Internship - Master's degree
• Effective date (Start): August 15, 2019
• Effective date (End): November 14, 2019
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Samuel da Silva
• Grantee: Jessé Augusto dos Santos Paixão
• Supervisor abroad: Gyuhae Park
• Home Institution: Faculdade de Engenharia (FEIS). Universidade Estadual Paulista (UNESP). Campus de Ilha Solteira. Ilha Solteira , SP, Brazil
• Local de pesquisa : Chonnam National University, Gwangju (CNU), South Korea
• Associated to the scholarship: 18/15671-1 - Damage quantification in composite material structures using extrapolation of autoregressive models coefficients, BP.MS

Abstract

Baseline-free algorithms are already familiar for applications in structural health monitoring(SHM) exploring detection and location of damages in composite structures, mainly based on measurements of active and sensing piezoelectric patches receiving guided waves where baseline data is unavailable. However, to reach a high level of SHM's hierarchy, for example, to quantify the level of damage or for forecasting this extension using only the current status, typically demands a previous model. Thus, the popular baseline-free strategies are limited to do it because most of them apply only time reversal or indices extracted directly of the output signals without these required associated models. Therefore, this project intends to provide a new contribution to this field by identifying an Auto-Regressive with an eXogenous input (ARX) model for directly monitoring the model output-error. First, this black-box model would be identified with a focus in multi-step-ahead to be able to capture the wave propagation dynamics in each structural state for a long horizon of prediction. It is essential to have a multi-step-ahead prediction model rather than a one-step-ahead, as employed by most of the works using autoregressive models for SHM purposes because the signal extrapolation to a future state can be inferred before the measures are collected. So, this is a surrogate model that can also be used to establish a correlation between damage indices extracted from output-error model and the existence of damage utilizing a trend curve regression. Next, the trend curve should be extrapolated here using a combination with Gaussian process to inspect how damage indices change are following a possible damage evolution. This work plan describes the motivation, goals, expected results, and scheduled to be followed during this internship at Chonnam National University in South Korea.