Thesis
Autonomous inspection of composite materials with multiple intelligent robots through thermal imaging and machine learning
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2026
- Thesis identifier
- T18062
- Person Identifier (Local)
- 202052405
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Composite materials have become key to the development of modern lightweight manufacturing, with applications in aerospace, automotive, marine, and defence industries. Their superior strength-to-weight ratio, durability, and corrosion resistance make them indispensable, yet their production remains dominated by manual processes. Manual hand lay-up techniques are time-consuming, hazardous to workers, and prone to defects that compromise the mechanical and cosmetic performance of final components. Automated Fibre Placement (AFP) machines offer an alternative by automating the layup of pre-impregnated carbon fibre tows. However, AFP remains susceptible to defects such as gaps, overlaps, wrinkles, twists, and misalignments, which can significantly reduce structural integrity. Furthermore, most existing defect detection approaches rely on post-process inspection, meaning defects are only discovered once a part has already been completed, resulting in wasted materials, higher costs, and missed opportunities for real-time repair. These challenges establish the imperative for a robust, flexible, and cost-effective in-process defect detection system that ensures higher quality, reduces waste, and increases industrial use of AFP. To this end, this thesis investigates the integration of thermal imaging and machine learning to achieve automated, real-time AFP defect detection. A review of existing defect detection techniques was first undertaken, identifying current methods such as visual inspection, profilometry, and ultrasonic testing, alongside their limitations in detecting in-situ and hidden defects. For this work, a dual-robot experimental platform was developed in which a KUKA KR120 industrial AFP robot carries out layup operations, while a UR10e collaborative robot, equipped with an AMETEK LAND ARC thermal camera, follows in real time to capture thermal imagery of the process. The resulting thermal data was collated, cleansed, segmented, and augmented into a dedicated dataset, which was then used to train and evaluate a series of machine learning and deep learning models. The experimental evaluation considered four machine learning models: KNearest Neighbours, Support Vector Machines, Decision Trees, and Naïve Bayes as well as two deep learning models, ResNet101 and ResNet152. Comparative analysis across metrics such as accuracy, precision, recall, F1-score, and Receiver Operating Characteristic Area Under the Curve (ROC AUC) demonstrated that ResNet101 offered the most balanced and effective performance, achieving an accuracy of 78.57% and a ROC AUC score of 0.82. This outcome validates the feasibility of employing deep learning based thermographic analysis for real-time AFP defect detection. In addition to model evaluation, this thesis addresses the issue of system flexibility by designing and building a Smart Sensor Box (SSB), a modular and low-cost sensing platform with accelerated machine learning that allows fast integration of different sensors into production work cells. The SSB lowers both the technological and financial barriers for smaller manufacturers. Furthermore, multi-robot coordination was explored using RoboDK simulations, demonstrating the potential for end-to-end collaborative systems that integrate AFP robots and inspection cobots into a single defect detection framework. The contributions of this thesis are: a comprehensive literature review identifying gaps in real-time AFP defect detection research; the creation and use of a custom new thermal imaging dataset tailored to AFP defect detection; the design, development, and validation of a low-cost Smart Sensor Box for flexible sensing integration; and the implementation of a multi robot system that combines thermal imaging and machine learning for in-process defect detection. These contributions of knowledge in automated composite manufacturing provide both theoretical insights and practical tools that enhance quality, reduce waste, and enable scalable adoption of AFP defect detection technologies across the lightweight manufacturing industry.
- Advisor / supervisor
- Yang, Erfud
- Resource Type
- DOI
关系
项目
| 缩略图 | 标题 | 上传日期 | 公开度 | 行动 |
|---|---|---|---|---|
|
|
PDF of Thesis T18062 | 2026-08-12 | 公开 | 下载 |