Contributions to Digital Shadows Through Virtual Sensing for Population-Based Structural Health Monitoring of Infrastructure

• Grant number: 25/05664-1
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: August 01, 2025
• End date: February 28, 2029
• Field of knowledge: Engineering - Mechanical Engineering - Mechanics of Solids
• Principal Investigator: Samuel da Silva
• Grantee: Henrique Cordeiro Novais
• Host Institution: Faculdade de Engenharia (FEIS). Universidade Estadual Paulista (UNESP). Campus de Ilha Solteira. Ilha Solteira , SP, Brazil

Abstract

Ensuring operational safety and monitoring the lifespan of infrastructures, such as bridges and wind turbines, requires continuous monitoring combined with structural inspections, primarily through structural health monitoring (SHM) methods. However, despite significant advances in machine learning algorithms and hardware, such as new sensors, finding generalized, cost-effective, and efficient solutions remains challenging. These solutions must provide a high probability of detection with low false positives and negative rates. One of the main challenges is the limited availability of adequate data for training classifiers or extracting reference information during the commissioning of healthy infrastructures. Additionally, the high cost of monitoring-especially for sensors and data acquisition systems-can be prohibitive for large populations of infrastructure. For instance, in wind farms with hundreds of turbines, budget constraints often prevent the provision of sensors for all turbines. In this context, this thesis aims to contribute by proposing using metamodels for regression and classification of structural conditions, utilizing sensor data from a population of infrastructures, and assuming that some are well monitored with high-quality data and others have limited sensor information. The focus will be on balancing data sources and targets: datasets with extensive histories and high-quality information will be combined with limited datasets characterized by sparse data, reduced historical records, or fewer sensors. To achieve this, we will employ a co-kriging strategy to balance the extracted metrics and generate a digital shadow model capable of tracking infrastructure health. We plan to utilize more than two datasets. To enhance the quality of datasets with sparse or low-quality sensors, virtual sensing will be applied, complementing the available information. The thesis will use particle filters as an estimator due to their ability to handle general conditions, such as potential nonlinearities, and their foundation in Bayesian formulation. Additionally, transfer learning tools, such as domain adaptation, will be used to transfer knowledge between digital shadow models for diagnosing structural conditions. The application examples will include strain measures and vibration data from benchmark structures such as bridges and wind turbine towers, laboratory-scale experimental data, and potentially field data provided by international collaborators with whom the thesis intends to collaborate. (AU)