Data-driven optimisation of production planning: A study on mathematical production planning at Oatly
(2026) MIOM05 20261Department of Industrial and Mechanical Sciences
Production Management
- Abstract
- Background: The global plant-based food shift puts high pressure on manufacturing capacity at facilities like Oatly's Landskrona site. This environment faces high combinatorial complexity due to growing product variety and sequence-dependent cleaning requirements. Relying on manual, experience-based planning to navigate these rules limits throughput and introduces organisational risk.
Purpose: To develop a mathematical optimisation model that minimises production makespan for Oatly's high-volume filling lines, providing a decision-support tool to assist planners rather than fully replacing human expertise.
Research Questions: (1) How can Oatly's specific production constraints, including sequence-dependent cleaning rules and routing... (More) - Background: The global plant-based food shift puts high pressure on manufacturing capacity at facilities like Oatly's Landskrona site. This environment faces high combinatorial complexity due to growing product variety and sequence-dependent cleaning requirements. Relying on manual, experience-based planning to navigate these rules limits throughput and introduces organisational risk.
Purpose: To develop a mathematical optimisation model that minimises production makespan for Oatly's high-volume filling lines, providing a decision-support tool to assist planners rather than fully replacing human expertise.
Research Questions: (1) How can Oatly's specific production constraints, including sequence-dependent cleaning rules and routing dependencies, be translated into a model to create a technically feasible production schedule? (2) To what extent can the total production time for a given order list be minimised by utilising an optimisation model compared to the current experience-based manual planning? (3) Is it possible to reduce the number of CIP cycles with a MILP model? (4) How may the potential transition to an automated optimisation model affect the stability of the planning process, specifically regarding the reliance on individual human expertise?
Methodology: The study utilises an Operations Research framework to develop a Mixed-Integer Linear Programming (MILP) model in Python. Primary data was gathered through semi-structured interviews, while quantitative validation was performed by retrospectively benchmarking the model's schedules against 39 weeks of historical manual production plans from 2025.
Findings: The MILP model handled sequence-dependent setup constraints and generated faster schedules in 26 of the 39 evaluated weeks under a 2000-second baseline, resulting in a 2.9% reduction in total makespan. The 13 weeks where manual planning resulted in shorter makespans were characterised by high combinatorial complexity. However, extended computational testing (25,000 seconds) on some of the most complex weeks demonstrated that the model has potential to outperform the manual planner when given adequate processing time. The study concludes that a hybrid approach delivers the optimal practical outcome. Using the mathematical model to generate a baseline schedule allows the human planner to focus on managing stochastic disruptions, upstream dependencies, and practical exceptions. This combination of mathematical optimisation and human expertise ensures high capacity output while maintaining system stability. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9233208
- author
- Lewerin, Mattias LU and Olefjord Widding, Emil LU
- supervisor
- organization
- course
- MIOM05 20261
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Mixed-Integer Linear Programming, Production Planning, Semicontinuous Manufacturing, Sequence-Dependent Setup Times, Oatly, Combinatorial Optimisation, MILP
- other publication id
- 26/5349
- language
- English
- id
- 9233208
- date added to LUP
- 2026-06-10 09:59:17
- date last changed
- 2026-06-10 09:59:17
@misc{9233208,
abstract = {{Background: The global plant-based food shift puts high pressure on manufacturing capacity at facilities like Oatly's Landskrona site. This environment faces high combinatorial complexity due to growing product variety and sequence-dependent cleaning requirements. Relying on manual, experience-based planning to navigate these rules limits throughput and introduces organisational risk.
Purpose: To develop a mathematical optimisation model that minimises production makespan for Oatly's high-volume filling lines, providing a decision-support tool to assist planners rather than fully replacing human expertise.
Research Questions: (1) How can Oatly's specific production constraints, including sequence-dependent cleaning rules and routing dependencies, be translated into a model to create a technically feasible production schedule? (2) To what extent can the total production time for a given order list be minimised by utilising an optimisation model compared to the current experience-based manual planning? (3) Is it possible to reduce the number of CIP cycles with a MILP model? (4) How may the potential transition to an automated optimisation model affect the stability of the planning process, specifically regarding the reliance on individual human expertise?
Methodology: The study utilises an Operations Research framework to develop a Mixed-Integer Linear Programming (MILP) model in Python. Primary data was gathered through semi-structured interviews, while quantitative validation was performed by retrospectively benchmarking the model's schedules against 39 weeks of historical manual production plans from 2025.
Findings: The MILP model handled sequence-dependent setup constraints and generated faster schedules in 26 of the 39 evaluated weeks under a 2000-second baseline, resulting in a 2.9% reduction in total makespan. The 13 weeks where manual planning resulted in shorter makespans were characterised by high combinatorial complexity. However, extended computational testing (25,000 seconds) on some of the most complex weeks demonstrated that the model has potential to outperform the manual planner when given adequate processing time. The study concludes that a hybrid approach delivers the optimal practical outcome. Using the mathematical model to generate a baseline schedule allows the human planner to focus on managing stochastic disruptions, upstream dependencies, and practical exceptions. This combination of mathematical optimisation and human expertise ensures high capacity output while maintaining system stability.}},
author = {{Lewerin, Mattias and Olefjord Widding, Emil}},
language = {{eng}},
note = {{Student Paper}},
title = {{Data-driven optimisation of production planning: A study on mathematical production planning at Oatly}},
year = {{2026}},
}