top of page

Two new ORbit articles

Great news! Two new articles were accepted for ORbit!


Back in January, we decided that articles are now distributed among our members as soon as they are accepted, and a physical publication will follow once there are enough articles. Therefore, all our members received the fresh articles in their mailbox. This time we are sharing two excellent articles.


Michael Lindahl wrote about the usage of Large Language Models combined with Operations Research. Michael has worked with applying OR and Analytics for 12 years across various industries, such as airports, container shipping, and offshore wind farms. He has a PhD from the Technical University of Denmark, and was the president of DORS from 2015-2018. He will be at the upcoming EURO Conference talking more about the topic (Tuesday Jul 2nd 8:30, B324, Room 40).


Ella Pirttikangas, who won the DORS Prize 2023, wrote about the Assignment Problem considering operational uncertainty in maintenance. Ella has a background in Process Engineering, specifically in the industrial maintenance and service sector. She is now working at DecisionBrain as a Business Analyst, creating decision support solutions and adding operational and business value.


Please find abstracts for both articles below! Full articles are available only for DORS members at this point, so if you are not yet a member please register with us using the link in our website.


Remember that submissions to ORbit are open all year round, so do reach out to us if you have something in mind that could be submitted: orbit@dorsnet.dk



Best regards, Joao Fonseca ORbit Editor The newest tool in the OR practitioner’s toolbox: Large Language Models - Michael Lindahl

Since the release of ChatGPT in November 2023, Large Language Models (LLMs) have seen unprecedented adoption by users. These models can work with unstructured, incomplete data and output in easily understandable text, but the downside is that they cannot reason or apply logic. 


This is very opposite to Mathematical Optimization Models, which require structured and complete data but, on the other hand, can provide optimal solutions to complex and large decision problems. A challenge when using these models to build real-world applications to help an organization make better decisions is often that getting structured and complete data can be very difficult. This leads to projects never making it past the proof-of-concept stage or a decision application is built, but doesn't get the expected adoption because it is cumbersome for the users to input the required data or interpret the output.


This article explores the potential role of LLMs in Operations Research and how they can lead to faster development cycles and more intuitive decision applications. It shows how LLMs fit into the different parts of such development processes and applications. This is illustrated with examples and a case study of the application, www.findgaven.ai, which combines LLMs and simple optimization.



Case Study: Assignment Problem under Operational Uncertainty in Maintenance

Decision-making in Maintenance Management contains different decision levels, where former levels give guidelines for the following decision levels. These former-level decisions are commonly made under uncertain outcomes at the following level. Therefore, there is a high motivation to integrate the uncertainty in the former decision-making levels.


This study focuses on the decision-making problem of an industrial maintenance provider, where the tactical decision of assigning equipment to technicians is made under the uncertainty of the operational circumstances. The uncertainties are unplanned maintenance activities that occur into the plan. These tasks additionally induce uncertainty in the maintenance tours, as this unplanned task disrupts the tour. The aim is to study how to account for this uncertainty when creating the assignments so that the negative consequences of such events are minimised.


The problem is formalised through a Stochastic Program, where two alternatives for the recourse are created. The formalised problem is solved using a metaheuristic, which is designed for solving large-scale problem instances of the case company. The algorithm uses a combination of Local Search and Large Neighbourhood Search to leverage principles of diversification and intensification. The neighbourhood functions use a graph formulation to quantify the qualitative empirical knowledge of the structure of optimal assignments. The uncertainty is modelled through simulation, where, depending on the recourse function, either just Monte-Carlo simulation is used or a combination of Monte-Carlo simulation and Simulation Optimisation.


The objective is to minimise the imbalance in workload and overtime. For the instances in the case study, the designed metaheuristic is able to improve the objective by an average of 90 %. The two recourse functions are studied, and it concluded that Simulation Optimisation, which modelled the recourse decision, creates more reliable results. Solving this model created more balanced workloads in terms of different workload types, had fewer negative effects on the utilisation of the technicians and generally created smaller service districts compared to the other recourse function.

10 views0 comments

Comments


bottom of page