Building blocks for Autonomous Systems in manufacturing applications: Perception, Actuation, and Machine Teaching
Autonomous systems hold great promise in manufacturing and heavy industry. As technology has advanced, sensing technologies and automation systems have transformed what was once a highly manual sector into an area of advanced technology. Still, many tasks remain stubbornly labor-intensive in their execution due to their complexity or limitations in sensing and control. Autonomous systems hold the promise of expanding the role of technology in manufacturing by incorporating advanced sensing technologies with deep reinforcement learning. Autonomous systems have been used successfully by leaders in manufacturing such as PepsiCo, SCG, and National Oilwell Varco to help drive improvements in efficiency and profitability.
In this post, I will cover three key building blocks to make autonomous systems a reality in your manufacturing plant: perception, actuation, and machine teaching.
The key building blocks for Autonomous Systems: Perception + Actuation + Machine Teaching
Perception is the first building block in creating an autonomous system. When considering perception, it is important to take an inventory of the existing instrumentation on your manufacturing line.
The manufacturing line represents the environment of the autonomous system’s AI agent (aka Microsoft Project Bonsai “brain”). Powered by deep reinforcement learning trained neural network(s), the AI agent needs to measure the state of the environment through a series of sensors that measure variables such as temperature, pressure, and speed. It is important to identify any gaps in instrumentation that would prevent the agent from taking the desired action.
Think of any steps in the process where quality control personnel conduct manual inspections or take measurements. If these steps are candidates for autonomous systems, consider installing automated measurement and data collection systems to fill any gaps in perception.
Often there are measurements that human operators intuitively integrate into inspections that are not captured through instrumentation such as product dimensions and color. Fortunately, solutions such as intelligent IoT sensors or AI models running at the edge can help. For instance, PepsiCo built a product characteristics measurement system to monitor Cheetos quality for their autonomous systems project.
Actuation is the second building block. Actuation is at the heart of the actions that the agent will take in a production environment. For an autonomous agent to control a process the necessary actuators need to be in place to take the desired actions.
Think about the actions that an operator takes to operate your manufacturing line today. In some cases, they may program an action into a PLC while in other cases they may need to turn a valve or physically move materials. Actions that already have supporting automation (i.e., PLC) represent a shorter path to creating an autonomous system. To make manual actions part of an autonomous system, consider first expanding your actuators and PLC infrastructure to incorporate these actions.
Machine Teaching represents how to put the pieces together. Machine Teaching is based on strategies that subject matter experts use to translate information from perception systems into actions. This information is often codified in standard operating procedures or guides.
Consider the tips and tricks that subject matter experts use to tune and optimize the manufacturing line. Examples could include increasing the temperature in a chemical process to increase the speed of reaction or slowing down a feeder to avoid a blockage. To effectively apply autonomous systems to a manufacturing line the machine teaching strategy needs to be supported by perception and actuation systems.
The future of Autonomous Systems in manufacturing
Not all aspects of a manufacturing line are suitable for autonomous systems. However, with the advancement of technology such as Deep Reinforcement Learning, the boundary of what manufacturing steps can be made autonomous is changing.
In this post, I’ve covered some of the key building blocks to implement autonomous systems. Stay tuned to go deeper into the challenges and opportunities of leveraging autonomous systems to further empower your business.
To learn more:
- Machine Teaching group: https://www.microsoft.com/en-us/research/group/machine-teaching-group/
- Project Bonsai for autonomous systems: https://www.microsoft.com/en-us/ai/autonomous-systems-project-bonsai?activetab=pivot%3aprimaryr7
- Key elements to design DRL extruder
- AS 101 video: Machine Teaching
- Read more: Autonomous systems