Addressing Complexity and Intelligence in Systems Dependability Evaluation

Aslansefat, Koorosh

Abstract

Engineering and computing systems are increasingly complex, intelligent, and open adaptive. When it comes to the dependability evaluation of such systems, there are certain challenges posed by the characteristics of “complexity” and “intelligence”. The first aspect of complexity is the dependability modelling of large systems with many interconnected components and dynamic behaviours such as Priority, Sequencing and Repairs. To address this, the thesis proposes a novel hierarchical solution to dynamic fault tree analysis using Semi-Markov Processes. A second aspect of complexity is the environmental conditions that may impact dependability and their modelling. For instance, weather and logistics can influence maintenance actions and hence dependability of an offshore wind farm. The thesis proposes a semi-Markov-based maintenance model called “Butterfly Maintenance Model (BMM)” to model this complexity and accommodate it in dependability evaluation. A third aspect of complexity is the open nature of system of systems like swarms of drones which makes complete design-time dependability analysis infeasible. To address this aspect, the thesis proposes a dynamic dependability evaluation method using Fault Trees and Markov-Models at runtime.
The challenge of “intelligence” arises because Machine Learning (ML) components do not exhibit programmed behaviour; their behaviour is learned from data. However, in traditional dependability analysis, systems are assumed to be programmed or designed. When a system has learned from data, then a distributional shift of operational data from training data may cause ML to behave incorrectly, e.g., misclassify objects. To address this, a new approach called SafeML is developed that uses statistical distance measures for monitoring the performance of ML against such distributional shifts. The thesis develops the proposed models, and evaluates them on case studies, highlighting improvements to the state-of-the-art, limitations and future work.

Citation

Aslansefat, K. (2023). Addressing Complexity and Intelligence in Systems Dependability Evaluation. (Thesis). University of Hull. Retrieved from https://hull-repository.worktribe.com/output/4500562

• Thesis Type: Thesis
• Deposit Date: Jan 5, 2024
• Publicly Available Date: Jan 18, 2024
• Keywords: Computer science
• Public URL: https://hull-repository.worktribe.com/output/4500562
• Additional Information: School of Computer Science; University of Hull
• Award Date: Dec 6, 2023