University of Southampton
Predicting potential failure in real-time through monitoring and detection of anomalous behaviour using hardware performance counters
Abstract
dc:description.abstractSafety-critical embedded systems can be found in many application areas such as automotive control systems, medical devices, and nuclear systems. Failure in these systems can have catastrophic results and devastating effects on human lives and the surrounding environment. Variations in temperature and voltage, single event effects and component degradation are just some contributors that cause faults in these systems. Existing research into techniques that deal with errors due to the presence of faults has mostly focused on replication of hardware components, information redundancy or inclusion of additional components to perform self-testing. However, these techniques either have high overheads or are resource-intensive. This thesis presents a detection method that can predict potential failure in real-time by detecting a change in system behaviour using hardware performance counters that are readily available in a processor. The early detection and prediction algorithm consists of two main stages - one-step ahead prediction and anomaly classification. Evaluation on the early detection and prediction algorithm were performed on benchmarks that are perturbed by single bit flip faults.<br/>The analysis on the early detection algorithm shows that it achieves 99.7% accuracy and earliest detection time was recorded at 325µs, which is less than a typical time to failure about 4,000µs. The proof of concept results show that the detector manages to detect when the system had started to behave anomalously and is able to stop execution before the system encounters a critical failure. Analyses on the performance and size of the detector show that the detector can be realised with minimal computational time and resources.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2019
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Woo, Lai Leng
- Advisor dc:contributor.advisor
-
- Zwolinski, Mark
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng