Cost Effective Computer Vision Based Structural Health Monitoring using Adaptive LMS Filters
Structural health monitoring (SHM) algorithms based on Adaptive Least Mean Squares (LMS) filtering theory can directly identify time-varying changes in structural stiffness in real time in a computationally efficient fashion. However, the best metrics of seismic structural damage are related to permanent and plastic deformations. The recent work done by the authors uses LMS-based SHM methods with a baseline non-linear Bouc-Wen structural model to directly identify changes in stiffness (modelling or construction error), as well as plastic or permanent deflections, in real-time. The algorithm validated, in silico, on a non-linear sheartype concrete structure using noise-free simulation-derived structural responses. In this paper, efficiency of the proposed SHM algorithm in identifying stiffness changes and plastic/permanent deflections under different ground motions is assessed using a suite of 20 different ground acceleration records. The results show that even with a fixed filter tuning parameters, the proposed LMS SHM algorithm identifies stiffness changes to within 10% of true value in 2.0 seconds. Permanent deflection is identified to within 14% of the actual as-modelled value using noisefree simulation-derived structural responses. Accuracy of the proposed SHM algorithm mainly relies on providing high-speed structural responses. However, due to a variety of practical constraints, direct high frequency measurement of displacement and velocity is not typically possible. This study explores the idea that emerging high speed line scan cameras can offer a robust and high speed displacement measure required for the modified LMS-based SHM algorithm proposed for non-linear yielding structures undergoing seismic excitation, and can be used for more precise estimation of the velocity using measured acceleration and displacement data. The displacement measurement method is tested to capture displacements of a computer-controlled cart under 20 different displacement records. The method is capable of capturing displacements of the cart with less than 2.2% error.