Factory Scheduling Demo

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.