Novel AI-assisted computational solutions for GPR data interpretation and electromagnetic data fusion to detect buried utilities

Download Statistics

Downloads

Downloads per month over past year

Afrasiabi, Arasti (2023). Novel AI-assisted computational solutions for GPR data interpretation and electromagnetic data fusion to detect buried utilities. University of Birmingham. Ph.D.

Preview

Afrasiabi2023PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (12MB) | Preview

Abstract

This research presents a number of novel computational solutions using artificial intelligence (AI) to interpret ground penetrating radar (GPR) data as well as fusing GPR data with data from other sensing modalities, including electromagnetic conductivity (EMC) and electromagnetic locating (EML). The application of the proposed computational solution is predominantly for detecting and locating buried utilities (e.g. pipes and cables) and ground anomalies (e.g. ground disturbances) in the shallow subsurface environment although the work can be extended to detect other buried anomalies. Processing GPR data is usually a subjective and time-consuming practise which involves expert intervention. Thus, the quality of the interpretation of such data depends on user experience and knowledge. Whilst several numerical approaches are available in the literature for post-processing GPR data, they all suffer from various shortcomings including lack of accuracy and/or excessive computational time. The issue is similar (or often worse) for data fusion between GPR and other sensors e.g. EMC and EML. To tackle some of these issues, in this research, four new computational procedures were developed. Three of these computational procedures are based on Kalman Filtering (KF), a less-studied approach to process GPR radargrams despite its great potential in efficient data analysis, and genetic algorithm (GA) as a machine learning based global optimisation tool. The final computational procedure combines finite element modelling and genetic algorithm to infer fused EML-GPR data. For the first two numerical methods, new algorithms were developed to optimise KF parameters using GA to remove noises from GPR radargrams and detect targets. The proposed procedures were validated against data from field and their performance was assessed against additional unseen dataset different to that of the validation to identify their potential limitations. Furthermore, their performances were compared against existing GPR data processing methods and differences were highlighted. The other two computational packages focused on data fusion from GPR and EMC/EML. The first of these two, extended the above KF algorithm to fuse data from GPR and EML as well as GPR and EMC. The results showed that the proposed data fusion algorithm significantly enhanced the quality of locating conductors and conductive regions in the subsurface compared to the individual techniques which were either incapable of defining the material of the buried target or the geometry of conductive anomalies. Finally, a novel inversion algorithm was developed by integrating finite element modelling of a coupled magnetic field and GA for detecting and locating buried live cables using GPR and EML. It was demonstrated that the proposed inversion can successfully detect the location of the buried cables as well as their intensity.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):