Machine learning techniques applied to infrastructure nondestructive evaluation tasks

행사분류	세미나	게시일	2024-05-08
행사기간	2024-05-31 ~ 2024-05-31
사전 등록기간	2024-05-08 00:00 ~ 2024-05-28 23:00
개최장소	인천대학교 교수회관(2호관) 3층 대회의실(305호)
※ 등록자조회를 하시려면 아래 버튼을 눌러 주시기 바랍니다. 등록자 조회
Machine learning techniques applied to infrastructure nondestructive evaluation tasks John S. Popovics (Department of Civil and Environmental Engineering, University of Illinois) * 일   시 : 2024년 5월 31일(금) 오전10시 * 장   소 : 인천대학교 교수회관(2호관) 3층 대회의실(305호) * 등록방법 : 홈페이지에서 5월28일(화)까지 사전등록 * 등록인원 : 100명 (선착순) * 등  록  비 : 무 료 Machine learning (ML) technologies offer promise to improve non-destructive evaluation (NDE) capabilities, especially where large data sets are analyzed. This presentation offers an overview of two different NDE applications where ML is applied to interpret NDE data. In both cases, the particular challenges of the applications are reviewed, and then solutions to those challenges are presented. In the first case, ML techniques are employed within a data-driven approach to estimate the residual stress state of a rail structure. Impulse-driven multi-resonance rail vibration data are utilized for this purpose where we hypothesize that correlations between the resonance frequencies and rail axial stress condition can provide estimation of rail stress state without reference measurement or models and while minimizing influences from the environment and rail support conditions. Field measurements collected from an in-service rail line over a period of nearly two years show that rail vibration resonances vary with rail temperature and rail stress conditions. A set of ML techniques are deployed to identify useful resonances and to establish relations between resonance natural frequency and stress state.  Established predictions of rail stress state satisfy measurement accuracy expectations. In the second case, physics-informed neural networks (PINNs) are used to interpret ultrasonic wave data collected from concrete. PINNs offer a promising alternative that combine the strength of physics-based models with the flexibility of data-driven approaches. The effectiveness of the proposed PINN models is evaluated using mechanical wave data from a variety of test specimens, each with unique defects, material properties, and geometries that contain simulated defects. Material properties such as wave velocity, quality factor, Young’s modulus and shear modulus are then predicted over the spatial domain using PINN models. Notably, the PINN models are able to detect micrometer-scale cracks in large concrete specimens.  Throughout, the developed PINN models show superiority over traditional signal processing methods or purely data-driven methods because the models are able to predict multiple inhomogeneous properties such as wave velocity, Young’s modulus, etc. using a single measurement dataset on different types of structures.    Collectively, the presented research cases demonstrate that ML techniques provide advanced processing, analysis, and interpretation approaches to better characterize complicated civil engineering infrastructure materials and structures in situ.