Link To Latest Report : Coming Soon.
Background:
In traditional physical testing of structures, post-processing of the data, which often includes dozens or hundreds of point measurements, is an unglamorous, but often significant, component of the work. It requires extensive pre-test planning, and inevitably at least some post-test troubleshooting, to ensure that the data set is properly synchronized, scaled, and analyzed to determine quantities of interest. The same challenges are often present in health monitoring. Ultimately, traditional measurement provides only snapshots of what researchers typically desire, which is accurate information about the original and deformed configuration of structures.
LiDAR is the most promising tool to modernize physical testing and health monitoring. It has matured rapidly to deliver repeatable and rich data sets in a broad range of applications such as planetary science, forest management, snow hydrology, and slope stability. However, tradeoffs remain between accuracy, point-cloud density, scanning speed, memory requirements, and post-processing workload.
Objectives :
The activities will focus on producing a software tool that is capable of: identifying structural elements, cleaning data (i.e., removing all data aside from the structural elements), quantifying the structural configuration and deformation (differences in configuration over time), and interpretating deformations and outputting quantities of interest. It will seek to use open-source Python-language tools to create an interactive GUI that automates the process. The target is that the user will be able to focus on engineering considerations instead of data cleaning tasks.
Scope :
Task 1: The first task will be to automate the identification of structural elements from point cloud data. The work will focus on wide-flange and rectangular sections. These geometries produce large relatively flat surfaces that may be identified in an automated manner. An example, taken from recent work on roll stability of precast girders by the PIs, is shown in Fig. 1. The points belonging to the girder in the raw point cloud in part (b) are relatively easy to discern for an analyst, however, manual isolation of the girder data is time consuming. Both open-source ML tools (multiple Python libraries are available), and direct analysis with in-house algorithms will be used to automate this process. Training and evaluation data will come from the vast library of results already published by the NHERI RAPID center hosted in the UW CEE department. The PIs also have immediate access to the RAPID center to produce new data if necessary.
Task 2: The second task will consist of developing an automated data cleaning process to isolate points belonging to individual girders and columns, and components of interest within the elements (flanges and webs of wide-flange sections, and faces of rectangular columns). Element and subcomponents will be placed in separate layers for subsequent analysis. For example, Fig. 1 (c) shows the isolated web of the girder. Some iteration between Tasks 1 (identification) and 2 (isolation) will likely be necessary during the research.
Task 3: The third task will involve fitting surfaces to the structural element webs and flanges (for wide-flange sections), and the outer surface (for rectangular sections). As shown in Fig. 1(d), the resulting point clouds are typically smooth, and well-suited to surface fitting methods. They are also well-suited to interpolation or extrapolation to extend the information to obscured regions. For example, if a portion of a web is obscured at one end of a member, but the web depth is observed elsewhere, extrapolation methods will be considered.
Task 4: The final task will be to determine lines representing the structural element midlines, and perform analysis to obtain deformation, slope, curvature, and twist. This task seeks to simplify the surfaces created in task 3 into beam-theory quantities such deformation, slope, curvature, and twist, which are usually described in terms one-dimensional theory. For example, in Fig. 1(e), which highlights the weak-axis deformation of the web (i.e., relative to a plane in its local axes) the deflection at the top and bottom of the web are typically not sufficiently different, and can be represented by a line model.
Research Team :
Principal Investigator : Richard Wiebe, Ph.D.
Co- Principal Investigator : John Stanton, Ph.D.