The molecular frontier: extending the boundaries of process design

Lead Research Organisation: Imperial College London
Department Name: Department of Chemical Engineering

Abstract

In designing new chemical manufacturing processes, the molecules or materials required (e.g., solvents or catalysts) are often chosen prior to optimising the topology or operating conditions of the product or process. This sequential decision-making process can lead to poor performance of the overall process because all these factors are intrinsically linked. For example, what is the best solvent for a reaction in a pharmaceutical manufacturing process depends on the temperature and pressure of the reactor, but also on what comes next in the process. If it is another reaction, it may be best to find a solvent which works reasonably well for both reactions, in order to avoid expensive additional processing steps such as swapping one solvent for another. By making decisions simultaneously, we can significantly improve the economics of a process and reduce its environmental impact through decreased material use and increased energy efficiency. Such an approach is referred to as integrated product and process design.

There are three elements needed for integrated product and process design: predictive models that can relate changes in the materials and in the process to performance; optimisation formulations that capture mathematically the trade-offs inherent in such complex systems; reliable algorithms that can solve the resulting design problems efficiently. In recent years, we have developed predictive models that have opened up new possibilities in design. The aim of this proposal is therefore to propose new formulations and algorithms for integrated product and process design and to apply them to a series of design problems.

The key challenge in problem formulation is to ensure that innovative (but unknown) solutions are embedded within the optimisation problem so that they can be uncovered. One way to do this is to allow the structure of the molecules or the materials to be part of the decision process by representing them through discrete decisions. Another complementary approach is to develop formulations that allow the identification of optimal mixtures. By mixing known molecules, one can tune the performance of the process. We will propose generic formulations for such problems. We will also tackle the simultaneous design of molecules/mixtures and processes.

In the optimisation algorithms used to solve these design problems, the main issue is to identify the very best (global) solution reliably and in a reasonable amount of time. This is difficult due to the nature of the integrated product and process design problem: it is nonlinear and combinatorial, which means that many local solutions may exist. We will develop robust algorithms for such problems, tackling the different types of mathematics that may be encountered, such as differential equations and/or discrete variables. These generic algorithms will be applicable to large classes of problems and will therefore be useful to solve other optimisation problems.

The findings of this research will be implemented and tested on a set of design case studies we have gathered in recent years through collaboration with industrial partners and other academic groups. Ongoing collaborations will ensure that our formulations and approaches are captured in software tools and suitable to tackle realistic design problems.

Planned Impact

Two main types of users will benefit from the results of this research.

First, the chemical and allied industries (pharmaceutical, oil & gas, ...) will benefit from the ability to design processes that are better - improved economics, decreased environmental impact and safety risks, increased sustainability. This is because they will be able to treat the design problem as a whole, rather than to break it down into smaller problems. By considering at the same time what are the best molecules or materials to use in a process, together with what the best process is, many new options open up. This project will give them tools to pose these important questions, and to answer them reliably.

Second, any business or public body that uses nonlinear optimisation can benefit from this research. Nonlinear optimisation covers a very large set of applications, from process design, to the identification of optimal delivery plans for a disease therapy, to the design of networks, to resource allocation problems. We will develop more robust techniques to solving optimisation problems, with the potential to identify much better solutions than those can be found today. It is not possible to assess a priori how much better the solution might be, but experience has shown us that it can be orders of magnitude better. The applications of optimisation are technological, economic and environmental, and as a result, the novel techniques we will develop in this project can resonate across all of these impact categories.

The impact of the project will be realised through a diverse set of routes in order to be sure to reach a wide audience. This will include training of skilled researchers, dissemination in international refereed journals and at international conferences, collaborations with industry through internships and case studies, workshops and short courses accessible to the academic and industrial communities, and the development of prototype software.

Publications

10 25 50
 
Description We have developed new theories, algorithms and tools to help engineers design new processes and products. These tools have been used to design experiments and reduce uncertainty and cost in process development. They have been used to identify better solvents for crystallisation and liquid-liquid separation, so that pharmaceutical and bio-processes can become more efficient. They have been used to develop new adhesives and drug delivery systems. Furthermore, we have made significant advances to the ability to solve nonconvex problems to global optimality, especially in the field of nonlinear optimisation, where we have developed novel bounding techniques and algorithms, and in the field of bilevel nonlinear and bilevel mixed integer nonlinear optimisation, where we have proposed novel algorithms and demonstrated new theoretical results.
Exploitation Route Others can use our software, models and methods to solve their own design problems, in chemical engineering, product and process design, but also in many other areas of engineering and economics that require optimisation.
Sectors Agriculture, Food and Drink,Chemicals,Energy,Environment,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology

URL http://www.imperial.ac.uk/people/c.adjiman
 
Description Research findings have been used to support process development across several industrial sectors. They have also led to the development of two major pieces of software, one which is available publicly and another which is being commercialised as software.
First Year Of Impact 2013
Sector Chemicals,Energy,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology
Impact Types Economic

 
Description Manufacturing Research Hubs
Amount £10,330,544 (GBP)
Funding ID EP/P006965/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 01/2017 
End 12/2023
 
Description Systems-based methods for pharmaceutical process development
Amount £5,300,000 (GBP)
Organisation Eli Lilly & Company Ltd 
Sector Private
Country United Kingdom
Start 10/2016 
End 12/2022
 
Title A Library Of Nonconvex Bilevel Test Problems With The Corresponding Ampl Input Files 
Description This library provides a description of nonconvex bilevel test problems in AMPL format, compatible with the Branch- And- Sandwich Bi Level solver - BASBL: http://basblsolver.github.io/home/ Kindly note, that this is a growing collection of nonconvex bilevel problems meant as a resource for researchers in the field, including problem statement, analysis, solution(s) and input file(s). For more details on these problems please visit wiki-page: https://github.com/basblsolver/test-problems/wiki We welcome contributions and corrections to this resource! 
Type Of Material Database/Collection of data 
Year Produced 2016 
Provided To Others? Yes  
 
Title A library of test problems for bilevel optimisation 
Description This resource is a collection of test problems, their solutions, and input files for over 90 bilevel optimisation problems. 
Type Of Material Database/Collection of data 
Year Produced 2017 
Provided To Others? Yes  
Impact None yet 
URL http://basblsolver.github.io/BASBLib/
 
Title Big-M mixture design problems 
Description Big-M formulations of mixture design problems (from a list) following formulation as generalized disjunctive programming problems. 
Type Of Material Computer model/algorithm 
Year Produced 2017 
Provided To Others? Yes  
Impact None yet 
URL https://zenodo.org/record/35234#.Wpl92ejFK70
 
Title Hull relaxation of mixture design 
Description A set of GAMS models that represent the hull relaxation of mixture design problems (select from a list) formulated as generalised disjunctive programming 
Type Of Material Computer model/algorithm 
Year Produced 2017 
Provided To Others? Yes  
Impact None yet 
URL https://zenodo.org/record/45297#.Wpl9EejFK70
 
Title Models for mixture design from atom groups 
Description This set of models (in GAMS) enables the design of mixtures of novel compounds to achieve better processing 
Type Of Material Computer model/algorithm 
Year Produced 2018 
Provided To Others? Yes  
Impact None yet. 
URL https://zenodo.org/record/1154203#.Wpl8bOjFK70
 
Description Argonne 
Organisation Argonne National Laboratory
Country United States 
Sector Public 
PI Contribution We are providing feedback on software developed at Argonne National Lab
Collaborator Contribution They are providing software for us to build on and develop new techniques Training of PhD student and postdoc Advice on optimisation techniques
Impact Chemical engineering and mathematics
Start Year 2012
 
Description Ashutosh Mahajan, IIT Mumbai, India 
Organisation Indian Institute of Technology Bombay
Country India 
Sector Academic/University 
PI Contribution Extended the capabilities of Minotaur to nonlinear and bilevel global optimisation
Collaborator Contribution Providing and supporting the Minotaur platform
Impact BASBL code and NLP code. This collaboration involves chemical engineers, computer scientists and operations researchers
Start Year 2015
 
Description CMU - Grossmann 
Organisation Carnegie Mellon University
Department Department of Chemical Engineering
Country United States 
Sector Academic/University 
PI Contribution Dissemination of research findings
Collaborator Contribution Advice on solution strategies Training of PhD student
Impact None yet
Start Year 2017
 
Description Eli Lilly - Leadership Fellowship 
Organisation Eli Lilly & Company Ltd
Country United Kingdom 
Sector Private 
PI Contribution Design of a method to support process development and reduce the cost of experiments
Collaborator Contribution Provision of case studies and understanding of industrial context Delivery of short course on multivariate statistical analysis (2015 and 2016)
Impact Implementation of experiment design approach in GAMS modelling environment
Start Year 2014
 
Description Syngenta 
Organisation Syngenta International AG
Country Switzerland 
Sector Private 
PI Contribution Knowledge dissemination Advice on optimisation problems Training on optimisation methods Provision of solutions to specific industrial problems
Collaborator Contribution Provision of case studies and insights on application areas of relevance. Testing of methods developed to case studies provided by the company.
Impact Creation of a tool based on multi-objective optimisation to analyse the risks of application strategies for different agrochemicals and to mitigate those risks. Training of an undergraduate student
Start Year 2012
 
Title BASBL 
Description Implementation of the Branch-and-Sandwich Algorithm for the solution of nonconvex bilevel optimisation problems. 
Type Of Technology Software 
Year Produced 2015 
Impact None yet, it has just been completed. 
URL https://github.com/basblsolver/home
 
Company Name Octeract 
Description This start-up company develops and markets massively parallel optimisation software. 
Year Established 2017 
Impact None yet
Website https://www.octeract.co.uk/
 
Description Mathematiques appliquees 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Schools
Results and Impact Presentation on the role of mathematics in engineering to an audience of 15 year old students.

After my talk, several students decided to pursue engineering studies
Year(s) Of Engagement Activity 2013,2014,2015,2016
 
Description PSE APMF 2014 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact interest in research and use of software tool deriving from research

setting up collaboration with new research partner
Year(s) Of Engagement Activity 2014
URL http://www.psenterprise.com/events/uk/2014/apmf/overview.html
 
Description The Branch-and-Sandwich Algorithm : A Deterministic Global Optimization Algorithm for Bi-Level Programming Problems 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Other academic audiences (collaborators, peers etc.)
Results and Impact Presentation to Optimization Research group at Argonne National Lab, as part of a collaborative visit.

Collaboration on software with Argonne National Lab
Year(s) Of Engagement Activity 2012