skip to primary navigationskip to content
 

Current research themes in the PSE group

PSE Group: examples of research areas

The PSE area can be separated into 3 major domains: Applications, Models and Algorithms. Following this structure, we can examine the research topics in the PSE group under 3 headings.

The sections below highlight different older and current topics, and are meant to be an example. In the PSE group we are open for people to choose original work in either established wider domains, such as the ones listed in this section, or to propose even their own ideas based on their interests.

A list of past publications for the research themes presented below can be found here.

 

1. Applications

1.1 Scheduling of maintenance of decaying performance processes and preventive maintenance

The PSE group at Cambridge has been leading research on Heat Exchanger Network (HEN) maintenance for more than 20 years now.  The original proposal was for long-term horizon planning of cleaning actions, so as to take into account long-term economic impact of local decisions for cleaning and maintenance.

Recent work published in the area has shown a new approach to solve these Mixed-Integer Nonlinear Optimisation models via a direct approach, based on the formulation of the underlying cleaning/maintenance problems as bang-bang optimal control problems (more details to be found in the modelling section below).

The maintenance problem is a general one in the process industries, as well as other key sectors, where the smooth operation and safety issues are guaranteed through systematic maintenance of production units which either decay in performance, or suffer from reduction in reliability.

 

1.2 Treatment of chronic illnesses

Chronic illnesses are an excellent case where PSE tools can be brought in to provide critical answers to both understanding of the illness mechanisms as well as for their treatment.

For example, earlier work in the PSE group at Cambridge has produced some very original work for the understanding of mechanisms underpinning the HIV infection, and showing through optimal control studies how interrupted treatment was the best policy emerging directly from the model built.

Future work is aimed to continue in this domain, considering other illnesses that are of critical importance to enhance our understanding, treatment policies, and at the end contribute to the improvement of quality of life.

 

1.3 Model Predictive Control of thermal runaway reaction systems

Process intensification is very often hindered by stability constraints of the underlying processes, such as the case of thermal runaway systems this work covers. Loss of control in such processes results in catastrophic events and loss of lives. Detecting the loss of stability of strongly nonlinear systems is a challenging field in literature.

Many processes are inherently unstable and yet produce high-added-value products making the necessity to operate more tightly very important to enhance productivity, improve energy usage and avoid wastage. These are two conflicting goals, which ongoing work in our group aims to cover with this first publication and others to follow.

It was demonstrated in a first-ever worldwide basis how new optimisation problems can be formulated for online control which incorporate totally novel stability criteria as part of the optimising control algorithms put forward. As such it is a contribution which has impact on many levels: industrial productivity, safety of process operations, and social impact in terms of efficiency and minimisation of waste of resources and energy.

 

1.4 Simulation and optimisation of biochemical reaction systems

Biotechnological processes, particularly for industrial scale production and the generation of renewable energy sources, are increasingly playing an important role for both present technologies as well as for future evolution in terms of where society is moving through high-level decisions as well as new policies being introduced.

The establishment of new energy sources is very crucial for the future of society and industry at the same time. The operation of continuous processes, through which energy sources and valuable by-products are to be produced, depends heavily on the ability to control them tightly online, ensuring both high levels of productivity and quality of products.

This is the theme of this work, which presents continuously ongoing research work in our research group in large scale optimisation and Model-Based Real-Time Control methodologies, such as advanced development of Model Predictive Control (MPC) for challenging and important Bioechemical Processes.

 

2. Models

2.1 Reformulation of multistage decaying performance models as bang-bang optimal control problems

The case of Heat Exchanger Network (HEN) scheduling for cleaning is a case where through a change in modelling approach it becomes possible to achieve integer solutions using standard nonlinear optimisation algorithms.

Indeed, this shows that a good understanding of the process, the solution procedure and the model can lead to very novel results – and ones that can have impact in making solution of challenging engineering problems essentially very simple and requiring only use of standard and established solvers.

Work in this area continues on different topics which bear similarity to the HEN problem in being bang-bang multistage optimal control problems.

 

3. Algorithms

3.1 Gradient flow algorithms for nonconvex nonlinear programming (optimisation) (NLP)

The motivation behind this work is to introduce advanced nonlinear optimisation methods capable of addressing problems of industrial importance. So far nonlinear optimisation finds limited use in chemical industry primarily because of not being "robust" enough. Indeed, despite worldwide efforts to make nonlinear optimisation algorithms reliable for large scale problems, such as flowsheeting and online control problems, the reality remains that these algorithms can suffer severely from the inability to converge when the models contain large numbers of nonlinear constraints.

Ongoing work in our group, including this which is among the very few published in the domain of gradient flow worldwide in decades, aims to address these issues. The importance of nonlinear optimisation in industry is that all processes involve strong nonlinearities (kinetics, thermodynamic properties) and at the same time their accurate modelling and solution is very important for their intensification and productivity enhancement.

 

3.2 Development of novel simulation and optimisation environments

The work in the various domains presented in this website shows the diversity of applications possible within the PSE area, particularly the way we organise our work and address it in creative ways.

Part of that creativity is facilitated by modern computing platforms, such as Python.  Python, although can be slow computationally compared to numerical applications done in FORTRAN or C++, it is nonetheless the ideal environment for prototyping and for mixing importantly symbolic and numerical computation together in a seamless fashion.

It is for this reason we are currently open to exploring projects that will produce generic simulation and optimisation tools in Python aimed at both specific type applications, as well as more general-purpose type simulation and optimisation of process systems models.

 

3.3 Large scale optimisation problems via decomposition approaches

The concept of plantwide or production-wide optimisation was introduced in Process Systems Engineering after the requirements for tight operational cost managements and to enhance productivity globally in industrial plants as well as in multisite operations. The work was produced in this spirit, in that such wide optimisation tasks result in huge-scale optimisation / decision making models that current solvers can have difficulty in converging – or even loading to begin with.

Earlier work presents advancements from ongoing research in our group in large-scale optimisation, and introduces totally automated ways of decomposing optimisation models into manageable size subproblems. As such it is probably the very first published work that shows the task to be possible to automate completely, with exceptionally good computational time performance -- and more importantly without requiring expert modellers to use it. It is both a totally original contribution to engineering practice and to optimisation theory alike.

 

3.4 Large scale solution of Mixed-Integer Nonlinear Programming problems (MINLP)

Many decision making optimisation problems in PSE models in Chemical Engineering involve discrete decisions, typically represented as binary variables (0,1), along with continuous design parameters.

Solution of these problems requires special procedures and algorithms, and its uniqueness can only be guaranteed for convex problems.  Application of existing algorithms to a nonconvex problem, will produce a solution that cannot be guaranteed to be the global one – hence these algorithms although based on strong theoretical principles for convex problems, they are no more than any other heuristic when it comes to nonconvex problems.

Worse than this even, for nonconvex MINLP problems it may be impossible to guarantee an easy discovery of an all-feasible point – let alone guarantee global optimality.  As such it is very important to continue research in this area either for special-purpose, application-specific solution procedures, or also for a general-purpose reliable and fast solver.

 

3.5 Solution of nonlinear optimisation problems to global optimality

Nonlinear optimisation plays a paramount role in PSE research activities and their applications to real-life problems in industry.

However, most problems of interest are nonconvex in nature, thus a local solver (gradient based or otherwise) will only be able to guarantee a local solution.  The global solution is often of paramount importance, as for example in discovering the 3-dimensional active form of a protein (enzyme) and many other instances where it is important to know the exact global optimum to an optimisation problem (e.g. the pooling-blending problem).

Such problems are the hard global optimisation problems requiring a guaranteed global solution, while for others we could simply accept a better solution than the one found with other heuristic means – the latter being an area where stochastic search (or “optimisation”) methods can be devised and utilised.

Work in this domain is open within our group for interested applicants to carry out novel research.