Machine Vision-Based Application to Structural Analysis in Seismic Testing by Shaking Table

Introduction

The use of machine vision-based data for application to structural dynamics is both promising and challenging. With respect to the state-of-the-art, further developments for on-the-field applications are still needed, but, at the present stage, such techniques are already feasible for laboratory dynamic tests. In fact, the growing interest in the use of machine vision-based technologies in this field of investigation, as well as in others, is mainly due to rapid advances in instruments performances available at falling costs. On the one hand, some issues related to the optimization of data processing and filtering algorithms still make these methods more complicated and less attractive than the use of more simple consolidated measurement techniques. On the other hand, many advantages in the vision-based instrumentation setup and its higher potentialities are hardly deniable. In particular, the potentialities of integration of such instrumentation output with computer graphics applications to visualize real-time data and results make these methods extremely effective in terms of science communication, with tremendous impact for technology transfer and diffusion purposes.

Several vision-based techniques have been developed in recent years and are quickly consolidating in their application to structural engineering. Some of them are derived from traditional techniques, but their impact and potentialities improved dramatically by latest hardware technology advances and new data processing algorithms. This is the case of the revival of stereo-photogrammetry applications due to the innovative Structure-from-Motion (SfM) data processing method (Micheletti, 2015) and the diffusion of relatively low-cost Unmanned Aerial Vehicles (UAV) equipped with vision-based systems for structural inspection and surveys (Remondino, 2011; Morgenthal, 2014; Mongelli, 2017). This kind of methods is particularly useful to provide an objective 3D documentation of the crack pattern, which is otherwise detected and documented by subjective visual inspection of experts. Application to shaking table testing is limited to spot surveys in static conditions to document the evolution of the crack pattern after each shaking trial.

Another practical and effective methodology called Digital Image Correlation (DIC) is now widely accepted and its use is already consolidated in many fields of experimental mechanics for quantitative deformation measurements of planar object surfaces. By comparing the digital images of a test object surface acquired before and after deformation the DIC directly provides full-field displacements to sub-pixel accuracy and full-field strains (Pan, 2009). This technique requires the preparation of the specimen surface with well-contrasted dots, for example by painted brush or spray, which makes it suitable for the study of small-scale objects subjected to slow pseudo-static deformations in lab tests. It was later successfully utilized for tensile and bond tests on composite reinforcements (Tekieli, 2017). Similar methodologies based on taking periodic photographs of tags or targets positioned on both sides of a crack were developed to monitor cracks aperture in damaged structures (Mangini 2017).

The application of computer vision and image processing techniques to shaking table tests in dynamic conditions, i.e. to detect motion parameters during seismic trials, has been investigated and developed in the last 20 years to remedy some drawbacks of conventional instrumentation, such as encumbrance, limitations in the number of sensors, range restrictions, and risk of damage of the devices (Berladin, 2004; Lunghi, 2012; Doerr, 2005; De Canio, 2013; Hutchinson, 2004). Moreover, a typical limitation of displacement devices, such as the Linear Variable Differential Transformers (LVDTs) and laser sensors, is due to their capacity of providing accurate data in only one direction. Consequently, the need of locating three sensors for tri-axial acquisition at each measurement point causes more difficult and time-spending set-up, also taking into account the usual range and encumbrance of conventional sensors. Among the several systems and techniques of this kind, particularly successful experiences were recently conducted in shaking table testing with the use of passive 3D motion capture systems. This kind of systems is capable of acquiring the positions of more than a hundred passive markers reaching interesting accuracy levels (0.01-0.1 mm) with a constellation of a proper number of high-resolution cameras. A crucial role for such accuracy levels is played by appropriate sub-pixel location estimation algorithms (Fillard, 1992).