Research Topic: Using Deep Learning and Computer Vision Technologies to Detect and Analyse Defects on the Wind Turbine Blade Surfaces
Studentship: Railston & Co Ltd + Loughborough University
PhD Thesis
1st Year Contribution:
1. Image Enhanced Mask R-CNN: a proposed pipeline to detect the WTB defects and classify the defect types
2. Proposed new evaluation metrics for defect detection
3. Compared defect detection performance between our proposed pipeline and current state-of-the-art algorithms (YOLOv3, YOLOv4 and Mask R-CNN)
Published on MDPI J. Imaging
2nd Year Contribution:
1. Defect Characteristics: a new set of features for describing the WTB defects in common vision-based knowledge
2. AI Reasoning: a proposed Decision Tree-based framework to reason and explain the outputs of the AI models (generic to any AI models) by using Defect Characteristics
Published on SSCI2023
3. Forest Monkey Toolkit: a Python package to provide functionalities, including Defect Characteristics extraction, AI reasoning capability, AI model improvement suggestions, etc. Also, the research successfully applied the toolkit to other datasets with different types of defects
Published on SSCI2023
3rd Year Contribution:
1. Proposed an image retrieval framework, utilising the Defect Characteristics as features, to retrieve defects/irregular patterns from images by computing their similarities.
2. Developed a web-based toolkit for industry uses based on the Flask framework
3. Integrated all PhD research outcomes into the toolkit
4. Designed an AI model training and testing management system to maintain the operation efficiency
Published on MDPI J. Imaging
