Comparing and Implementing Job Shop Scheduling Techniques with AI-powered Simulation

Comparing and Implementing Job Shop Scheduling Techniques with AI-powered Simulation

Problem

The Model Group develops, produces, and supplies intelligent, innovative, and high-quality display solutions made of solid and corrugated cardboard. It operates in 15 locations in seven countries with a total of 4,500 employees.

In their factory in Weinfelden, Switzerland, they produce about 200 jobs a day on 19 different machines. All their daily scheduling was being done by four production planners, who tried to work out the best schedule for each day. These planners relied on their experience to make decisions but did not have a way to validate and compare their designed schedules to understand which one was better.

Initial job shop scheduling technique illustrated showing the steps

Initial job shop scheduling technique

The planners wanted to improve their job shop scheduling to make it better and more efficient. Schedule creation was a complicated process because of the dependencies between the elements of the production line system.

Solution

ProSim is a company that works mostly with manufacturing and logistics companies to help them make better business decisions using data and simulation. They are also working on AI-powered simulation, as they see a lot of potential in it.

To solve the job shop scheduling problem presented by The Model Group, ProSim developed an AnyLogic simulation model incorporating different job shop scheduling techniques to provide feedback for the planners’ schedules.

The planners continued to manually create the schedules and then input them into the model, comparing and validating them every day. The schedules were therefore being improved every day.

Job shop scheduling steps using the simulation model illustrated

Using one of the job shop scheduling techniques with the simulation model

Following on from this, ProSim decided to implement an AI agent with the simulation model in order to create the schedules without any input from the planners.

Job shop scheduling steps using AI-powered simulation illustrated

Using one of the job shop scheduling techniques – AI-powered simulation

This worked well and produced good results, but they decided to develop this further by using a genetic algorithm in the simulation model to schedule the jobs.

A genetic algorithm solves an optimization problem based on natural selection. There is a generation consisting of individuals with their own strengths and weaknesses. Because of these, some individuals survive, and some do not. Those that do survive try to produce new mutated individuals that are stronger and better than what came before.

After this, there is a whole new generation, and this process repeats and repeats until there are no longer any better individuals. In this case, we are talking about new schedules as the individuals.

Job shop scheduling steps using a genetic algorithm illustrated

Using one of the job shop scheduling techniques – a genetic algorithm

The first step in the job scheduling process is to randomly generate schedules in the model. Every time this process occurs, about 200 schedules are generated at the beginning. Then, using the genetic algorithm, the schedules are improved over each generation.

The model will usually run for about 100 generations, reducing the number of schedules each time. At the end, the best schedule will be available. The user can then look at the best solution in the simulation model to find out in more detail the production schedule, total setup time, total production time, and total delay time.


The model interface illustrated The production schedule from the model illustrated
The model interface and a production schedule including total setup time, total production time, and total delay time (click to enlarge)

Results

Of the job shop scheduling techniques implemented, the genetic algorithm has proven to produce better results than AI-powered simulation and manual scheduling. The job shop schedule improved by 18% based on KPIs that were defined by The Model Group. Currently, there are three production lines that have been modeled using a genetic algorithm in the simulation model, but more are planned to be done in the future. ProSim also continues to investigate both AI-powered simulation and genetic algorithms to further solve problems.

The case study was presented by Patrick Kehrli of ProSim, at the AnyLogic Conference 2022.

The slides are available as a PDF.



Похожие проекты

Другие истории успеха

Сборник историй успеха от пользователей AnyLogic

Скачать