Development and Demonstration of an Effective Optimisation Approach for Large-scale Chemical Production Scheduling
Lead Research Organisation:
University of Manchester
Department Name: Chem Eng and Analytical Science
Abstract
The UK chemical industry plays a vital role in the UK economy with a total annual turnover of £50 billion. To remain competitive both regionally and globally, UK chemical companies have moved towards product customisation and diversification, which in turn have resulted in a large number of low-volume, high-value products. Furthermore, UK chemical manufacturers have started to employ flexible multiproduct/multipurpose facilities, which allow for higher utilisation of resources, lower inventory costs and better responsiveness to a fluctuating manufacturing environment. However, these advantages have not been fully achieved due to the use of poor heuristic rule-based production scheduling methods, which could cause the sector to lose potential annual profits estimated in the hundreds of millions of pounds.
Existing optimisation-based methods for large-scale real-world chemical production scheduling in the literature require significant computational cost while also struggling to provide optimal or near-optimal solutions, which restrict their capability to achieve the aforementioned advantages and industrial application. This research is to develop a novel and effective optimisation-based method to address these challenges. It will combine the advantages of the mathematical programming approach and a new machine learning technique, Gene Expression Programming (GEP), for systematic generation of robust and high-quality dispatching rules in an offline manner, which are expected to be applicable for a variety of scheduling problems. These high-quality dispatching rules will then be used to generate optimal or near-optimal schedules for scheduling in an online manner with improved profit and substantially reduced computational effort when compared to existing optimisation-based methods. The proposed solution approach will be tested in a practical context with the industrial collaborator Flexciton Limited and an improvement in profit of at least 5% and up to 20% will be demonstrated.
This research is significantly different from previous work in this area in that it will be based upon the combination of the new machine learning method and the mathematical programming approach. It will advance the state of the art in the use of optimisation methodologies in chemical production scheduling and lead to significant advances in solving a variety of large-scale production scheduling problems, opening new avenues of research in smart manufacturing. It will help strengthen the leading expertise of the PI in this field. It will also allow the UK to take a leading position in developing the cutting-edge optimisation-based solution approach to improve chemical manufacturing competitiveness and thus continue to remain the leading position in chemical industries.
Existing optimisation-based methods for large-scale real-world chemical production scheduling in the literature require significant computational cost while also struggling to provide optimal or near-optimal solutions, which restrict their capability to achieve the aforementioned advantages and industrial application. This research is to develop a novel and effective optimisation-based method to address these challenges. It will combine the advantages of the mathematical programming approach and a new machine learning technique, Gene Expression Programming (GEP), for systematic generation of robust and high-quality dispatching rules in an offline manner, which are expected to be applicable for a variety of scheduling problems. These high-quality dispatching rules will then be used to generate optimal or near-optimal schedules for scheduling in an online manner with improved profit and substantially reduced computational effort when compared to existing optimisation-based methods. The proposed solution approach will be tested in a practical context with the industrial collaborator Flexciton Limited and an improvement in profit of at least 5% and up to 20% will be demonstrated.
This research is significantly different from previous work in this area in that it will be based upon the combination of the new machine learning method and the mathematical programming approach. It will advance the state of the art in the use of optimisation methodologies in chemical production scheduling and lead to significant advances in solving a variety of large-scale production scheduling problems, opening new avenues of research in smart manufacturing. It will help strengthen the leading expertise of the PI in this field. It will also allow the UK to take a leading position in developing the cutting-edge optimisation-based solution approach to improve chemical manufacturing competitiveness and thus continue to remain the leading position in chemical industries.
Planned Impact
The potential impact of the research includes the impact on industry, the economy and society, knowledge, people and policy, all of which is summarised below.
1. Impact on industry
The UK chemical industry loses a significant profit (approximately hundreds of millions of pounds) every year largely due to the use of the 'trial and error' simulation scheduling methods or the poor heuristic rules based on the operator's experience, which are utilised in an attempt to find a feasible operating schedule. The effective optimisation-based solution approach developed in the research will identify an optimal or near-optimal solution, which is expected to lead to an improvement in profit of at least 5%, and up to 20%, with zero capital expense (an estimated annual increase in profit from £90 million to £450 million), higher resource utilisation, lower inventory costs and better responsiveness to a fluctuating manufacturing environment. This will help the chemical industry to improve its competitiveness and enhance its current leading position.
2. Impact on the economy and society
The expected improvement in profit through the proposed solution approach in the research will be at least 5%, and up to 20%, with zero capital expense (an estimated annual increase in profit from £90 million to £450 million for the UK chemical industry, which would add an estimated additional value of £800 million to the UK economy annually). This will in turn help create more employment opportunities in the sector, provide more training and improve welfare for society, contributing to a better quality of life in the UK.
3. Impact on knowledge
This research will advance the state of the art in the use of optimisation methodologies in process operations, opening a new avenue of research in process systems engineering. The results from this research will be published openly in high-impact journals and also disseminated at seminars and leading national and international conferences. The results of this research will in addition be used to develop educational materials for a graduate-level course that PI teaches.
4. Impact on people
This research will help develop a future generation of leading academics in the UK through training of the post-doctoral research associate (PDRA) in the field of process modelling, machine learning, and optimisation. The high-level skills and expertise gained from this research will enable the PDRA to make valuable contributions in both industry and academia in key areas that suffer from significant shortages of highly-skilled professionals. The expertise of the PI in this field will be strengthened through high-impact journal publications and presentations at leading national and international conferences. The developed mathematical models and solution approach will be of interest to academics and researchers, particularly those working in the field of process scheduling. Both the code and the library of problems developed during this research will be made publicly available so that other academics and researchers in this field can replicate (or improve upon) the results. This research will also offer the PI an opportunity to further develop his MSc and MEng research program, which allows MSc and MEng students to use some of the cyber-infrastructure tools developed in this effort to carry out computational studies and receive training for optimisation-based chemical production scheduling.
5. Impact on policy
The optimal strategies for scheduling of flexible chemical facilities in the chemical industry can be used for the creation of scheduling decision policy. They can also help better understand existing decision policies, challenges and the needs in developing new scheduling policies. These insights will be shared with relevant policymakers who will be engaged in different ways to ensure that any policy recommendations are relevant and effective.
1. Impact on industry
The UK chemical industry loses a significant profit (approximately hundreds of millions of pounds) every year largely due to the use of the 'trial and error' simulation scheduling methods or the poor heuristic rules based on the operator's experience, which are utilised in an attempt to find a feasible operating schedule. The effective optimisation-based solution approach developed in the research will identify an optimal or near-optimal solution, which is expected to lead to an improvement in profit of at least 5%, and up to 20%, with zero capital expense (an estimated annual increase in profit from £90 million to £450 million), higher resource utilisation, lower inventory costs and better responsiveness to a fluctuating manufacturing environment. This will help the chemical industry to improve its competitiveness and enhance its current leading position.
2. Impact on the economy and society
The expected improvement in profit through the proposed solution approach in the research will be at least 5%, and up to 20%, with zero capital expense (an estimated annual increase in profit from £90 million to £450 million for the UK chemical industry, which would add an estimated additional value of £800 million to the UK economy annually). This will in turn help create more employment opportunities in the sector, provide more training and improve welfare for society, contributing to a better quality of life in the UK.
3. Impact on knowledge
This research will advance the state of the art in the use of optimisation methodologies in process operations, opening a new avenue of research in process systems engineering. The results from this research will be published openly in high-impact journals and also disseminated at seminars and leading national and international conferences. The results of this research will in addition be used to develop educational materials for a graduate-level course that PI teaches.
4. Impact on people
This research will help develop a future generation of leading academics in the UK through training of the post-doctoral research associate (PDRA) in the field of process modelling, machine learning, and optimisation. The high-level skills and expertise gained from this research will enable the PDRA to make valuable contributions in both industry and academia in key areas that suffer from significant shortages of highly-skilled professionals. The expertise of the PI in this field will be strengthened through high-impact journal publications and presentations at leading national and international conferences. The developed mathematical models and solution approach will be of interest to academics and researchers, particularly those working in the field of process scheduling. Both the code and the library of problems developed during this research will be made publicly available so that other academics and researchers in this field can replicate (or improve upon) the results. This research will also offer the PI an opportunity to further develop his MSc and MEng research program, which allows MSc and MEng students to use some of the cyber-infrastructure tools developed in this effort to carry out computational studies and receive training for optimisation-based chemical production scheduling.
5. Impact on policy
The optimal strategies for scheduling of flexible chemical facilities in the chemical industry can be used for the creation of scheduling decision policy. They can also help better understand existing decision policies, challenges and the needs in developing new scheduling policies. These insights will be shared with relevant policymakers who will be engaged in different ways to ensure that any policy recommendations are relevant and effective.
People |
ORCID iD |
JIE LI (Principal Investigator) | http://orcid.org/0000-0001-5196-2136 |
Publications
Cheng X
(2023)
Multi-scale design of MOF-based membrane separation for CO2/CH4 mixture via integration of molecular simulation, machine learning and process modeling and simulation
in Journal of Membrane Science
Han D
(2021)
An Efficient Augmented Lagrange Multiplier Method for Steelmaking and Continuous Casting Production Scheduling
in Chemical Engineering Research and Design
Li D
(2022)
Novel Multiple Time-grid Continuous-time Mathematical Formulation for Short-term Scheduling of Multipurpose Batch Plants.
in Industrial & engineering chemistry research
Pan Q
(2023)
Automatic creation of molecular substructures for accurate estimation of pure component properties using connectivity matrices
in Chemical Engineering Science
Rakovitis N
(2022)
Novel approach to energy-efficient flexible job-shop scheduling problems
in Energy
Teymourifar A
(2022)
An Open-Source Simulation Model for Solving Scheduling Problems
in Universal Journal of Applied Mathematics
Description | (1) A novel multiple time-grid continuous-time mathematical formulation has been developed for short-term chemical production scheduling. It is found that the proposed models require a smaller number of event points in many cases to achieve optimality than existing unit-specific event-based models. It is interesting to find that no task is required to span over multiple event points to reach optimality for all addressed examples. The best variant developed is superior to existing unit-specific event-based models with the same or better objective values by a maximum improvement of 67%. The computational effort is significantly reduced by at least 1 order of magnitude in some cases. (2) A novel hybrid GEP-based solution approach has been developed for chemical scheduling problems. It is found that the proposed solution approach can generate high-quality dispatching rules, which can be used to achieve near-optimal solutions. The computational experiments have demonstrate that the hybrid algorithm can reach optimality for all considered moderate-complexity examples within seconds, and lower energy consumptions (with maximum of 37.2%) using significantly reduced computational time (with maximum of 94%) in majority (>72%) complicated instances relative to the existing approaches. |
Exploitation Route | The developed models and solution approach have been published in several journal papers. The results have been disseminated in several key PSE conferences and Process Integration Research Consortium of The University of Manchester. The dataset and open-source codes are available for other academic researchers or industrial practitioners to be used for solving their scheduling problems. |
Sectors | Chemicals,Digital/Communication/Information Technologies (including Software),Manufacturing, including Industrial Biotechology |
Description | UoM/CSC Joint Scholarship |
Amount | £157,600 (GBP) |
Funding ID | 202106440020 |
Organisation | Chinese Scholarship Council |
Sector | Charity/Non Profit |
Country | China |
Start | 09/2021 |
End | 08/2025 |
Title | Hybrid Knowledge-guided machine learning scheduling optimisation method |
Description | The research method is a novel hybrid solution approach for industrial scheduling problems, which integrates a machine learning technique (i.e. gene expression programming, GEP), variable neighbourhood search (VNS) and mathematical programming formulations. The GEP algorithm is used to determine operation sequences and machine assignments. The VNS is well designed to improve the searching capabilities. The mathematical programming formulations (i.e. event-based and sequence-based mixed integer linear programming models) are adapted to the integrations with GEP-VNS and elucidates the determinations on machine mode to further improve the solution quality. |
Type Of Material | Technology assay or reagent |
Year Produced | 2023 |
Provided To Others? | Yes |
Impact | The proposed research method will advance the state of the art in the use of optimisation methodologies in process operations, opening a new avenue of research in process systems engineering. The developed mathematical models and solution approach can be used by academics and researchers, particularly those working in the field of process scheduling, to solve their scheduling problems. The research method can also be used by the industrial practitioners to solve industrial scheduling problems. It is expected that the hybrid method can reach optimality for all considered moderate-complexity problems, and lower energy consumptions (with maximum of 37.2%) in majority (>72%) complicated instances. |
Title | An open-source simulation model for solving scheduling problems |
Description | This open source code is developed based on the GEP-based machine learning algorithm. It can be used to generate different dispatching rules, which can be used to solve industrial scheduling problems. |
Type Of Material | Computer model/algorithm |
Year Produced | 2022 |
Provided To Others? | Yes |
Impact | The developed simulation code can be used by chemical industries to improve their chemical production scheduling. It can also be used by researchers to solve their scheduling problems. |
URL | https://github.com/aydinteymurifar/Scheduling-Source-Codes |
Title | Generic mathematical model for chemical production scheduling |
Description | This model addressed all drawbacks of the existing continuous-time models for chemical production scheduling of multipurpose batch plants. It introduced a novel concept of direct and indirect material transfer and allowed intermediates to be partially transferred to storage and downstream processing units instead of fully transferred. This model will share with others after it is published in the near future. |
Type Of Material | Computer model/algorithm |
Year Produced | 2021 |
Provided To Others? | No |
Impact | The developed mathematical model will complement the chemical production scheduling theory and model library. it will also improve the efficiency of resource utilisation and increase the productivity output by 67%. The developed model will be published openly in high-impact journals and be used by academics and researchers working in the field of process scheduling. |
Title | Research dataset for job shop scheduling problem |
Description | This dataset consists 58 problem instances with a variety of jobs, operations, and machines, ranging from small-scale to industrial scale. |
Type Of Material | Database/Collection of data |
Year Produced | 2021 |
Provided To Others? | Yes |
Impact | This research dataset can be used by other researchers to support their research related to job shop scheduling problems. |
URL | https://doi.org/10.1016/j.energy.2021.121773 |
Description | Distributed Digital Twin Shop-floor Scheduling |
Organisation | Zhengzhou University of Aeronautics |
Country | China |
Sector | Academic/University |
PI Contribution | My expertise in machine-learning based scheduling optimisation methods, supervision, and training of the staff |
Collaborator Contribution | The main contributions made by the partners include their expertise in digit twin technology and intelligent algorithms, and industrial data for job-floor scheduling, |
Impact | This collaboration is multi-disciplinary. it involves Mechanical Engineering, Electronic Engineering, Operations Research and Chemical Engineering. Currently, a new framework for distributed digit-twin job-floor scheduling has been proposed. Corresponding resource optimisation mechanism has been designed for the next development of intelligent optimisation algorithms. |
Start Year | 2022 |
Description | Optimisation of Air compression system |
Organisation | Chinese Academy of Sciences |
Department | Institute of Process Engineering |
Country | China |
Sector | Academic/University |
PI Contribution | My expertise, intellectual inputs, supervision |
Collaborator Contribution | Data and staff time |
Impact | The current output is the development of an optimisation model for air compression system in Xinjiang Tianye Co. Ltd. The research has led to the funding from Xinjiang Tianye Co. Ltd. |
Start Year | 2021 |
Description | Production scheduling in semiconductor manufacturing |
Organisation | Flexciton |
Country | United Kingdom |
Sector | Private |
PI Contribution | My expertise, intellectual input including solution algorithm development, and results generated |
Collaborator Contribution | Data and staff time |
Impact | The current output is the development of genetic algorithm based solution approach for production scheduling in semiconductor manufacturing. |
Start Year | 2021 |
Title | scheduling source codes |
Description | This product is the MATLAB source code for scheduling problem, which involves in the simulation are mainly based on generating dispatching rules or using them to solve problems. It is capable of solving different benchmarks for industrial scheduling problems. It is also open to development and can be easily modified by users to solve other types of problems. |
Type Of Technology | Webtool/Application |
Year Produced | 2022 |
Open Source License? | Yes |
Impact | This product can be used by industrial practioners to solve their scheduling problems. It can also be used by other researchers in the field of scheduling to solve their scheduling problems and further development. It can also be used by academics to implement |
URL | https://github.com/aydinteymurifar/Scheduling-Source-Codes |