Imperial College London Featured PhD Programmes
Xi’an Jiaotong-Liverpool University Featured PhD Programmes
University of Reading Featured PhD Programmes

Fully Bayesian Reinforcement Learning for Control of Continuous Industrial Processes (EPSRC CDT in Distributed Algorithms)


   EPSRC CDT in Distributed Algorithms

This project is no longer listed on FindAPhD.com and may not be available.

Click here to search FindAPhD.com for PhD studentship opportunities
  Dr A Listsa, Dr Bei Peng  No more applications being accepted  Competition Funded PhD Project (Students Worldwide)

About the Project

This PhD project is part of the CDT in Distributed Algorithms: The What, How and where of Next-Generation Data Science.

The University of Liverpool’s Centre for Doctoral Training in Distributed Algorithms (CDT) is working in partnership with the STFC Hartree Centre and 20+ external partners from the manufacturing, defence and security sectors to provide a 4-year innovative PhD training programme that will equip up to 60 students with: the essential skills needed to become future leaders in distributed algorithms; the technical and professional networks needed to launch a career in next generation data science and future computing; and the confidence to make a positive difference in society, the economy and beyond.

A new opportunity has arisen with the CDT looking at Reinforcement Learning, primarily looking at developing a control strategy which requires numerical Bayesian inference algorithms, Markov Chain Monte Carlo and High-Performance Computing. We’re looking for a student who thrives on challenging tasks and can work on developing relevant models from a variety of sources. 

This exciting and innovative PhD, in partnership with NSG, relates to settings where a continuous manufacturing process is monitored and so controlled with a focus on both guaranteeing the quality of the product and minimising the costs of doing so, e.g. by minimising the amount of excess material used to guarantee that certain specifications of the product (e.g. thickness or defect rate) are met. The focus is on manufacture and treatment of glass. In such settings there is often a significant latency (i.e. minutes) between the control input changing and the response being observable. It is challenging to apply feedback control in these contexts, so existing Engineering solutions often make use of physical models for the process and employ predictive model-based control. While this does make it possible to produce desired variations in the product, the approach relies on the physical models for the process and the models for the sensors to be known. These models are well understood in general, but there are aspects where it is not possible to build accurate models that, for example, can infer how the fine detail of the thickness profile is impacted by variation in the power applied to heating elements at some historic time. Furthermore, the real-world changes over time (e.g. because valves become worn or because scheduled maintenance has not occurred recently) and while it is possible to develop work-arounds to adapt to these changes, these work-arounds can fail. Such failures can result in sudden and significant degradation in the quality of product. 

The fundamental challenge is then to develop a control strategy that fully capitalises on: offline historic data; parameterised models that capture the extensive but incomplete understanding of the processes and sensors’ performance; offline simulated experience derived from those models; online data from sensors. Developing such a control strategy will require numerical Bayesian inference algorithms (e.g. Markov Chain Monte Carlo) to make inferences about the models in a way that exploits the historic data and domain experts’ existing understanding. Borrowing from recent successful applications of Reinforcement Learning (RL) in other domains, RL will then be used to learn how best to apply the control given the inferred model. Such RL is computationally intensive and will therefore require use of High-Performance Computing resources. 

We welcome UK nationals and international applicants from countries where NSG has a significant manufacturing presence – individuals from the following countries are eligible to apply: UK, EU, USA, Mexico, Brazil, Argentina, Chile, Malaysia, Japan & Vietnam.

Visit the CDT website for application instructions, FAQs, interview timelines and tips.

You must enter the following information to ensure your application is received and processed:

  • Admission Term: 2021-22
  • Application Type: Research Degree (MPhil/PhD/MD) – Full time
  • Programme of Study: Electrical Engineering and Electronics – Doctor in Philosophy (PhD)

The remainder of the guidance is found in the CDT application instructions on our website.


Funding Notes

We welcome UK nationals and international applicants from countries where NSG has a significant manufacturing presence – individuals from the following countries are eligible to apply: UK, EU, USA, Mexico, Brazil, Argentina, Chile, Malaysia, Japan & Vietnam.

References

Students are based at the University of Liverpool and part of the CDT and Signal Processing research community in the department of EEE. Every PhD is part of a larger research group which is an incredibly social and creative group working together solving tough research problems. Students have 2 academic supervisors and an industrial partner who provides co-supervision, placements and the opportunity to work on real world challenges. In addition, students attend technical and professional training to gain unparalleled expertise to make a difference now and in the future.
This studentship is due to commence 1 October 2021 (Covid-19 Working Practices available).