First Advisor

Thomas Schumacher

Term of Graduation

Fall 2022

Date of Publication

8-30-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.) in Civil & Environmental Engineering

Department

Civil and Environmental Engineering

Language

English

Subjects

Reinforced concrete -- Imaging, Ultrasonic imaging, Nondestructive testing, Ground penetrating radar, Machine learning

DOI

10.15760/etd.8100

Physical Description

1 online resource(xv, 135 pages)

Abstract

Concrete structures may be exposed to a variety of loads and environments during their service life. Non-destructive testing (NDT) techniques can be helpful in evaluating the condition of a structure. Imaging provides a visual representation of the interior of concrete and its condition non-destructively. Ground penetrating radar (GPR) and ultrasonic echo array (UEA) using electromagnetic and stress waves, respectively, provide the data that can be used to reconstruct an image. In this PhD dissertation, image reconstruction and fusion algorithms, simulation, and a deep learning model were investigated with the goal to lay the foundation for enhanced imaging applications for concrete.

First, a multimodal 2D imaging pipeline is introduced that can process and fuse GPR and UEA data to enhance imaging of concrete. An algorithm, named extended total focusing method (XTFM) was developed that can reconstruct images from the raw signals collected with GPR and UEA instruments. This algorithm combines synthetic aperture focusing technique (SAFT) and total focusing method (TFM) concepts and can reconstruct images from multi-channel arrays with overlapping measurements. In addition, an image fusion algorithm is introduced that fuses GPR and UEA images using a multi-level wavelet decomposition and a NDT knowledge-based fusion rule. A novel local evaluation metric was developed to evaluate the output images of the algorithm. The results from three concrete specimens built in laboratory are reported and it is shown qualitatively, quantitatively, locally and globally that the reconstructed images represent an enhanced precise image of the interior of the concrete.

Second, the imaging pipeline was used to track damage progression in two full-scale reinforced concrete bridge column-footing specimens with different lap-splice detailing undergoing reverse-cyclic loading in the laboratory. A quantitative analysis revealed that changes in the images can be tracked as early as the columns undergo some initial damage. In addition, it is shown that changes along the height of column vary, i.e., the lower sections of the column exhibit more damage. This observation is in agreement with the internal force demand distribution of the column. A so-called backwall analysis suggests that the difference in the performance of the two tested columns can be captured using imaging.

Finally, GPR simulations and deep learning pipeline was developed that can be used to generate large datasets with different setups to employ deep learning to assist with imaging. A simulated dataset with 3000 data examples of B-scans was generated. A deep neural network model is introduced that can accurately predict two key required parameters for image reconstruction: the dielectric constant of the concrete and the time offset parameter of the GPR measurement. Tuning these two parameters is a cumbersome process usually done manually. Precise prediction of these parameters results in focused images where reflectors such as rebars in concrete, have correct shape and location. It is shown that the developed model can accurately predict these two parameters with an R2 > 0.999. The model was tested on data from the three experimental specimens and resulted in accurate images. The generalizability of the method is also discussed. Gradient visualization is used to highlight which part of an image is utilized most in predictions. It was found that the neural network pays the most attention to the angle of reflections to predict the dielectric constant, and the surface wave portion of GPR for the time offset parameter.

Rights

© 2022 Sina Mehdinia

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

Persistent Identifier

https://archives.pdx.edu/ds/psu/39159

Share

COinS