Mondragon Unibertsitatea – Warehouse-Optimised Work Order Generation
Mondragon Unibertsitatea
Sector: Logistics
Business Case
The challenge is to generate optimised manufacturing orders for the operators of a company that distributes industrial vehicle parts. Currently, these orders are generated almost manually and based on the experience of the person generating them. The use of optimisation techniques will allow this person or role to be freed from this task and create work orders that better fit the needs of the company and operators.
Objectives
Develop a system that proposes optimised work orders to the operators at the beginning of each shift and presents these orders in an intuitive way for the production planner role.
Use case
The optimisation problem was approached from two different points: by training reinforcement learning (RL) agents and using evolutionary multi-objective optimisation algorithms. Based on the results obtained, a decision was made to implement a GUI (desktop application) to generate work orders using an OMM algorithm (Decmo2). The optimisation algorithm is fed from the database already deployed in the company.
Infrastructure
On premise
Technology
Machine learning and deep learning
Data
Datos tabulares obtenidos de la base de datos de la empresa (privados)
Resources
1 project leader, 1 RL expert researcher, 2 MOO researchers, 3 people with expertise in the area. Iterative process to define the problem, objectives and constraints.
Difficulties and learning
Difficulty in defining clear objectives and restrictions. Difficulty in training RL agents and defining a rewarding system suitable for the use case. A solution based on the use of multi-objective optimisation algorithms is more maintainable and requires fewer resources than an RL-based system.
KPIs (business impact and metrics of the model)
Measures for optimisation: Hypervolume, IGD. Visualisations of Pareto fronts: scatter plot, chord plot, parallel coordinate plot. KPIs: number of wrongly packed parts, package filling times.
Funding
The project was financed by the Provincial Council of Gipuzkoa within the Etorkizuna Eraikiz strategy.
Collaborators, Partners
The project was developed with the Gipuzkoan company Industrias Onyarbi S.L.
