Mr Craig Pirie

Computing Science: Application Software Development
Bachelor's Degree

Status Complete

Part Time No

Years 2018 - 2020

Project Title Using Computer Vision to Identify Corrosion In Underwater Images For Inspection Engineering Applications

Project Description Inspection engineering is regarded as a highly important field in the Oil & Gas sector to analyse the health of their assets offshore. Corrosion, anatural phenomenon, is the degradation of metal over time due to a chemical reaction with its environment. Costing the global economy US $2.5trillion per annum, the destructive nature of the occurrence is clear.
Following the downturn experienced by the industry in recent times, the need to combat corrosion in an increasingly efficient was exaggerated as companies are forced to look for ways to reduce costs without compromising on important tasks. This thesis attempts to explore tackling the inspection problem by using computer vision techniques and deep learning. The investigation uncovered that there is potential in the application of computer vision and machine learning to identify corrosion in images. Mask RCNN, a de facto instance segmentation algorithm, was found to be most useful, with the paper reporting a mAP of 77.1%. Adopting this technique will reduce the time an engineer must spend assessing the presence and location of rust in an image. Thus, it will reduce the human workload and in turn reduce costs for the industry, without compromising the validity of the inspection process.

Awarding Institution Robert Gordon University

Director of Studies Carlos Moreno-Garcia

Thesis Image pre-processing and segmentation for real-time subsea corrosion inspection.

Status	Complete
Part Time	No
Years	2018 - 2020
Project Title	Using Computer Vision to Identify Corrosion In Underwater Images For Inspection Engineering Applications
Project Description	Inspection engineering is regarded as a highly important field in the Oil & Gas sector to analyse the health of their assets offshore. Corrosion, anatural phenomenon, is the degradation of metal over time due to a chemical reaction with its environment. Costing the global economy US $2.5trillion per annum, the destructive nature of the occurrence is clear. Following the downturn experienced by the industry in recent times, the need to combat corrosion in an increasingly efficient was exaggerated as companies are forced to look for ways to reduce costs without compromising on important tasks. This thesis attempts to explore tackling the inspection problem by using computer vision techniques and deep learning. The investigation uncovered that there is potential in the application of computer vision and machine learning to identify corrosion in images. Mask RCNN, a de facto instance segmentation algorithm, was found to be most useful, with the paper reporting a mAP of 77.1%. Adopting this technique will reduce the time an engineer must spend assessing the presence and location of rust in an image. Thus, it will reduce the human workload and in turn reduce costs for the industry, without compromising the validity of the inspection process.
Awarding Institution	Robert Gordon University
Director of Studies	Carlos Moreno-Garcia
Thesis	Image pre-processing and segmentation for real-time subsea corrosion inspection.

Data Science
Master of Science [MSc]

Status Complete

Part Time No

Years 2020 - 2021

Project Title Application of Generative Adversarial Networks in Concept Car Design

Project Description Due to the economic fallout of the COVID-19 pandemic, businesses including car manufacturers will benefit from smarter modes of working to reduce costs. This project
investigates the state-of-the-art in deep generative modelling — namely Generative
Adversarial Networks (GANs) — to explore the possibility of their application in concept car design. The research conducted shows that without suitable resources and a respectable budget, GANs can be tricky to train in the automotive domain. Out of the three models trained: a Deep Convolutional GAN (DCGAN), Wasserstein GAN with Gradient-Penalty (WGAN-GP), and Variational Autoencoder (VAE), none were able to produce realistic looking images of cars. In terms of Fr´echet Inception Distance (FID), the DCGAN was the best performing with a score of 367.43. However, via manual inspection, the only model that was able to produce any visible signs of learning was the WGAN-GP but results remained very poor and inconsistent. Further work with a larger budget is a necessity in order to more thoroughly explore alternative architectures with higher quality data and to afford comprehensive optimisation of networks.

Awarding Institution Robert Gordon University

Director of Studies Carlos Moreno-Garcia

Status	Complete
Part Time	No
Years	2020 - 2021
Project Title	Application of Generative Adversarial Networks in Concept Car Design
Project Description	Due to the economic fallout of the COVID-19 pandemic, businesses including car manufacturers will benefit from smarter modes of working to reduce costs. This project investigates the state-of-the-art in deep generative modelling — namely Generative Adversarial Networks (GANs) — to explore the possibility of their application in concept car design. The research conducted shows that without suitable resources and a respectable budget, GANs can be tricky to train in the automotive domain. Out of the three models trained: a Deep Convolutional GAN (DCGAN), Wasserstein GAN with Gradient-Penalty (WGAN-GP), and Variational Autoencoder (VAE), none were able to produce realistic looking images of cars. In terms of Fr´echet Inception Distance (FID), the DCGAN was the best performing with a score of 367.43. However, via manual inspection, the only model that was able to produce any visible signs of learning was the WGAN-GP but results remained very poor and inconsistent. Further work with a larger budget is a necessity in order to more thoroughly explore alternative architectures with higher quality data and to afford comprehensive optimisation of networks.
Awarding Institution	Robert Gordon University
Director of Studies	Carlos Moreno-Garcia

Explainable AI
Doctor of Philosophy [PhD]

Status Current

Part Time No

Years 2022 - 2028

Project Title Towards Solving The Disagreement Problem In Explainable AI

Awarding Institution Robert Gordon University

Director of Studies Nirmalie Wiratunga

Second Supervisor Carlos Moreno-Garcia

Status	Current
Part Time	No
Years	2022 - 2028
Project Title	Towards Solving The Disagreement Problem In Explainable AI
Awarding Institution	Robert Gordon University
Director of Studies	Nirmalie Wiratunga
Second Supervisor	Carlos Moreno-Garcia

Mr Craig Pirie's Qualifications (3)