Scribd
Upload
106 views20 pages

2021 - Application of A Reinforcement Learning AI Solution For Optimization of Screening Shuttles Control (Presentation)

The document discusses applying a reinforcement learning AI solution to optimize the control of screening shuttle conveyors. The objectives are to improve bin level control, increase plant utilization, and reduce operator interventions. The AI control strategy uses a cognitive architecture and neural network speed controller. It was trained over 38 months of simulated time and tested on the plant. Results so far show improved bin variability, reduced bin starvations, faster start-ups, and increased throughput. The conclusion is that reinforcement learning can be successfully applied to real-time process control.

Uploaded by

rodrigofmarques
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
106 views20 pages

2021 - Application of A Reinforcement Learning AI Solution For Optimization of Screening Shuttles Control (Presentation)

The document discusses applying a reinforcement learning AI solution to optimize the control of screening shuttle conveyors. The objectives are to improve bin level control, increase plant utilization, and reduce operator interventions. The AI control strategy uses a cognitive architecture and neural network speed controller. It was trained over 38 months of simulated time and tested on the plant. Results so far show improved bin variability, reduced bin starvations, faster start-ups, and increased throughput. The conclusion is that reinforcement learning can be successfully applied to real-time process control.

Uploaded by

rodrigofmarques
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

Application of a Reinforcement

Learning AI Solution for


Optimisation of Screening
Shuttles Control
Bruno Versiani
Principal Control Systems Engineer,
BHP
Introduction
• A process control application using Reinforcement Learning AI.
• The scope – Motion control of two shuttle conveyors.
• Objectives:
• Improve bin level control performance;
• Increase plant utilisation rates;
• Reduce the number of operators interventions.
Process and Conventional Controls
AI Control Strategy
• The state of the
plant is abstracted
into clusters.

• Clusters are
adjusted by their fill
level and status.
AI Control Strategy
• Two major components:

• Cognitive BDI
Architecture /
Behaviour Hierarchy

• Neural Network
Speed Controller.
AI Control Strategy

• Creation of a
gap in feed on
the belt that
allows the
shuttle to cross
over
unavailable
bins.
AI Control Strategy

• Similar concept
applied to the
‘ore-break’
however
without creating
a gap in feed
on the belt.
AI Control Strategy

• Shuttle departs
as fast as
possible from a
bin which
becomes
inhibited.
AI Control Strategy

• If surrounding
bins are
inhibited, the
agent will fill the
isolated bin.
AI Control Strategy

• Standard mode
of operation.

• The shuttle
moves over a
cluster of at
least two bins.
Reinforcement Learning
Machine
Learning

Supervised Unsupervised Reinforcement


Learning Learning Learning
Reinforcement Learning
• Hybrid approach
– NFQCA

• Alternation
between online
exploration and
batch learning
RL Controller Design and Development
• Dynamic simulation model of the plant
RL Controller Design and Development
• AI Training on
cloud computing
architecture.

• 38 months worth
of simulated time
logged.
RL Controller Design and Development
• Utilisation of a
‘Scenario Dispatch’
tool.

• 440 / 500 scenarios


tested on scalping
and products
shuttle respectively.
RL Controller Deployment
Plant HMI

• Integration with plant control systems.


SCADA
driver
RL AI Server comms

• AI commands validation. OPC


Comms PLC

Field
• Back-up conventional controls. I/O

Shuttle
Drive
AI Performance Results
• Long term results
still under validation.

• Overall bin
variability improved
from 11.1% to
10.6%.
AI Performance Results
• Bins starvations reduced by 17% in AI Mode.

• Start-ups duration decreased by 15%.

• Ore Break: average throughput increased by 40%.


Conclusions
• Reinforcement Learning can be applied to real time process control;

• The technique enables the implementation of complex functions;

• The results indicate improved performance achieved by the RL AI


deployed solution.

You might also like