Potential PhD Projects

I am happy to discuss projects in the general area of real-time systems. However I have specific interests in:

  • Mixed-Criticality Scheduling - Most of the work to date has considered situations where absolute guarantees about the WCET is known at the higher-criticality levels and that it is acceptable to completely lose lower-criticality services during a recovery period. I would be interested in supervising work that relaxes these assumptions. The challenge here would be to use real execution time profiles and overheads of systems to consider the real likelihood of missing deadlines at any particular time so that less jobs need to be dropped. This would require adaptive scheduling policies to be learned using state of the art machine learning and statistical analysis techniques.
  • Digital Twins - As systems become more complex, there is a growing acknowledgement that the best model of the system is the system itself. The Digital Twin is the concept of evolving a model of the system at run-time with the model then being used as part of making decisions - both at run-time and as part of maintenance. For example, models of execution time can be developed that are then used during allocation and scheduling. Key research questions here are: what abstraction and contents is right for the model; appropriate means of instrumenting the system so that the models can be refined; how to refine the models; and how to assure that the system is then dependable. This work can be focussed on scheduling and allocation, or Internet of Things.
  • Wireless Sensor Networks - Wireless sensor networks are used in a wide range of applications, many of which are dependable in nature. Most existing work is intended to increase the dependability but much less is focussed on how this is ensured. I would be interested in methods for ensuring the system is safe, providing evidence that the resulting system is safe, and how the information from the system can be used in the context of wider safety-critical systems, e.g. making decisions about care pathways for health-related systems. A number of projects could be proposed. For example,
    1. machine learning and statistical analysis could be used to determine activity patterns of people, e.g. whether someone has fallen, in a predictable and dependable way. The research challenges addressed could include: to determine the situations under which the classification may be incorrect, determining the confidence and uncertainties associated with a specific classification, and what acceptable secondary sensing could reduce the number of incorrect classifications and reduce the uncertainties. This requires appropriate means of validating the predictions.
    2. acceptable interventions could help improve people's Quality of Life. Based on appropriate sensed measures, possible interventions could be proposed however importantly these would need to be presented in an acceptable and meaningful way. Users of the technology may be put off if interventions are too frequent or presented in a confusing way. Adherence may be enhanced if appropriate explanations are given to the user. This requires a user centred approach to the design.
    3. look at compression and aggregation techniques that can be performed on lightweight sensor nodes. The challenge is to not remove important information as once the amount of data is reduced then it can never be recovered. This requires context-sensitive approaches to be produced.
  • Task Allocation - My previous work on task allocation has largely considered single criticality systems including how to produce maintainable systems. I would be interested in supervising work that extends this work to balance the flexibility that having mixed-criticality gives the designer whilst still ensuring the system meets its requirements and is still maintainable. The challenge here would be to allocate the software to the hardware such that the minimum loss of service is achieved. This requires accurate knowledge of execution times to be used and the temporal dependencies between tasks to be used as part of a search-based task allocation approach.