Anomaly detection in industrial robotic arms using edge-based IoT systems

Kayan, Hakan 2025. Anomaly detection in industrial robotic arms using edge-based IoT systems. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (Hakan Kayan, PhD, Thesis) - Accepted Post-Print Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (4MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (205kB)

Abstract

Industrial robotic arms are valued for their precision and versatility in performing complex tasks. However, unexpected deviations, or anomalies, can cause significant damage. These anomalies may result from misconfiguration, degradation, or environmental disturbances. Detecting anomalies requires understanding operational context. In this thesis, “context awareness” refers to recognizing expected movement patterns of robotic arms within specific timeframes. Deviations from these patterns are contextual anomalies. The proposed framework uses embedded machine learning to process externally gathered sensor data for real-time anomaly detection. This thesis contributes to anomaly detection in industrial robotic arms by introducing a context-aware framework. This framework uses externally gathered IMU data, including accelerometer, gyroscope, and magnetometer readings. It detects movement-based anomalies in real time, such as joint velocity changes, collisions, platform impacts, and magnetic field disruptions. The framework uses external sensors to enhance detection reliability in connected environments. Optimized 1D-CNN and LSTM models were deployed on ultra-lowpower edge devices and compared with state-of-the-art methods for industrial applications. The anomaly detection system was tested in two use cases: a pick-and-place task and a multitasking scenario with screwdriving, painting, and pick-and-place. Results confirmed its ability to detect anomalies in complex operations. We addressed challenges in deploying machine learning on edge devices, such as limited RAM and computational overhead, through quantization techniques, RAM-optimized firmware, and an edge-to-cloud IoT architecture supporting over-the-air model updates, real-time data monitoring. The research provides datasets, tools, and methodologies to demonstrate the feasibility of edge-based solutions in dynamic industrial environments involving robotic arms. The datasets, experiments, and insights from this dissertation demonstrate the potential of embedded machine learning and externally gathered data for real-time anomaly detection. All work is replicable through a GitHub repository, supporting further research and application in diverse industrial contexts.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software
Funders:	Republic of Turkey Ministry of National Education Scholarhip
Date of First Compliant Deposit:	11 July 2025
Last Modified:	11 Jul 2025 15:07
URI:	https://orca.cardiff.ac.uk/id/eprint/179739

Actions (repository staff only)

Edit Item