Factory Scheduling Demo
avtor: Wood
Sample bonsai project to optimize scheduling operations in a factory.
Factory Scheduling Optimization
Operations scheduling optimization is an important logistics problem that touches many industries. In this sample, you will teach a Bonsai brain and harness the power of Deep Reinforcement Learning and Machine Teaching to optimize scheduling inside a paint manufacturing facility.
This sample is offered by Wood and offers the following:
· A Discrete Event Simulation environment created out of Simply
· The Bonsai brain training inkling file
The problem statement for the sample is as follows:
1. There is an aggregate order list of products with volumes and due dates from customers
2. The manufacturing process consists of 4 steps: Mixing, Dispersion, Thin Down and Filling. None of these process steps can be skipped
3. Each process stage has 4 different units with different capacities and processing speeds. There is also a list of compatible products for each process unit. Not all products are compatible with all machines
4. When switching from 1 product to another, there is a fixed changeover time of 30 minutes
5. The aim is to optimize the scheduling by reducing either the total makespan (time to manufacture all batches), total number of changeovers or total delay (You can choose which of these to optimize alone or in parallel while training your Bonsai brain)
The training of the Bonsai brain has been done with “action masking”, a new Bonsai feature because the number of legal actions at each iteration varies in this problem. The simulator and Brain have been architected in a way that you designate a process train for each batch of product you want to manufacture (as opposed to choosing jobs for each unit).
This sample is very handy for testing scheduling optimization use-cases with Bonsai. Such an approach can be fitted to many other manufacturing operations and is not limited to paint manufacturing. Please feel free to reach out to Wood for any questions regarding this sample.