My research sits at the meeting point of signal processing, machine learning and engineering.
The same core question runs through all of it: given a stream of noisy measurements from something valuable and hard to inspect, how healthy is it — and what is about to go wrong?
Increasingly, the difficulty is one of scale. Much of my work has been about making that question answerable on data too large to handle naively — a thread running from my doctoral work on fast novelty detection through to engine health monitoring today.
Research interests
- Health & condition monitoringDetecting, diagnosing and anticipating faults in high-value systems
- Signal processingExtracting meaningful features from noisy, high-rate sensor data
- Machine learning & data miningAnomaly detection, fault classification, data-driven modelling
- Embedded systems & architecturesFrom on-engine devices through to cloud infrastructure
- Mathematical modellingPhysics-informed and data-driven models of system behaviour
Current focus
As lead of the Systems and Architectures work theme within the Rolls-Royce UTC, I work on health monitoring systems for the next generation of large civil gas turbine engines. This spans the whole pipeline: embedded and wireless sensing hardware that survives a demanding operating environment, the software and system architectures connecting on-board devices to cloud infrastructure, and the analytics and machine learning that turn the resulting data into actionable health assessments.
Selected projects
Embedded wireless sensor systems and health-management infrastructure for large civil gas turbine engines.
Project lead / developer for the NJR research-data platform and a personalised surgical-risk calculator.
A ray tracer running on SpiNNaker neuromorphic hardware, with tools to distribute scenery across cores.
Ultrasonic ground-condition algorithms for assessing buried pipes from the inside.
Automated novelty detection on wind-turbine sensor data — my first research post.
Publications
For the complete and current list, see my Google Scholar profile. My work spans aerospace, healthcare and infrastructure.
Engine health monitoring & prognostics aerospace
- Industrial Internet of Things for Health Monitoring of a Gas Turbine Engine: Wireless Sensor Networks for Aerospace Propulsion. Mills A., Hills A., Kadirkamanathan V. & Wall D. 26th International Symposium on Air Breathing Engines (ISABE 2024), 3, 1819–1831.
- Scalable Data-driven Availability Management. Mills A. R., Frison M., Jacobs W., Mansor M. M. & Hills A. F. STO Meeting Proceedings STO-MP-IST-160, NATO S&T Organization. PDF
- A likelihood ratio-based approach to bleed valve event detection in gas turbine engines. Li P., Jayasuriya J., Hills A., Mills A. R. & Kadirkamanathan V. UKACC 12th International Conference on Control (CONTROL), 408–413.
Patents aerospace
Medical data modelling & decision support healthcare
- JointCalc: a web-based personalised patient decision support tool for joint replacement. Zotov E., Hills A. F., de Mello F. L., Aram P., Sayers A., Blom A. W., McCloskey E. V., Wilkinson J. M. & Kadirkamanathan V. International Journal of Medical Informatics, 142, 104217.
- Estimating an Individual's Probability of Revision Surgery After Knee Replacement. Aram P., Trela-Larsen L., Sayers A., Hills A. F., Blom A. W., McCloskey E. V., Kadirkamanathan V. & Wilkinson J. M. American Journal of Epidemiology, kwy121.
- An evaluation of survival modelling approaches for personalised risk prediction after hip replacement for osteoarthritis. Aram P., Trela-Larsen L., Sayers A., Hills A. F., et al. Osteoarthritis and Cartilage, 26, S281.
Infrastructure & condition monitoring civil
- Assessing Ground Support of Plastic Pipes Using Ultrasound. Zhu J., Boxall J. B., Hills A. F., Dwyer-Joyce R. S., Anderson S. R. & Collins R. P. Infrastructures, 6(2), 30.
- A novel baseline model-based technique for condition monitoring of wind turbine components. Hills A. F., Lang Z. Q., Soua S., Gan T. H., Perera A., van Lieshout P. & Grainger B. Insight — Non-Destructive Testing and Condition Monitoring, 53(8), 434–438.
Thesis doctoral
- Cover Tree-based Fast Novelty Detection and Classification for Applications with Big Data. Hills A. F. PhD thesis, Department of Automatic Control and Systems Engineering, University of Sheffield. Supervised by Prof. Visakan Kadirkamanathan. Used the cover tree data structure to make novelty detection and classification tractable on large datasets: paired with random projections for fast dimensionality reduction, with a Bhattacharyya-distance feature-selection method for interpretable feature ranking, and embedded within Gaussian process regression to approximate the kernel matrix. Demonstrated on chemoinformatics, gas turbine engine and wind turbine data, and on single-cell breast cancer imaging.