Skip to main content

IoT in manufacturing: Coordinating machines with edge computing

Coordinating machines to run autonomously using edge compute technology.
Robots assembling a car in a factory

Photo by Lenny Kuhne on Unsplash

One way that companies use edge computing is to monitor and control equipment in physical plants such as factories and warehouses. Computation on the edge is vital in tracking industrial IoT data to perform tasks like predictive maintenance, where an algorithm might monitor the temperature, vibrations, and output of a machine to estimate its remaining useful life.

Use cases like that tend to be centered on individual machines, much like a heart monitor is centered on a single person.

But a new group of use cases involves tracking and coordinating multiple pieces of equipment at once. Rather than being machine-centric, you could say they are system-centric or team-centric algorithms that collect data from several robots in order to make them aware of each other as they make coordinated choices about what to do next.

Machine coordination for real-time benefit

One application based on this machine team focus is a Level 2 control system, which deals with product localization, physical movements, and batch management. These systems, which people also refer to as SCADA systems and manufacturing supervisors, almost always rely on fixed, embedded, and suboptimal human rules to work.

Here, edge technology can make a difference quickly because it doesn't require any change management in how scheduling or planning are implemented, and it's immediately applicable in production with a "real-time decision" benefit.

An example might be a job shop with a group of CNC machines. The job shop has to process a stream of items every day, and that stream changes based on orders the shop receives from its clients.

Some CNC machines and their operators will be more suited to processing certain items than others, because of the machine's capabilities, the operator's skills, and the current workload of the devices.

"In our work supporting manufacturers to redesign their workflows and factory floors, we see significant gains to be made by applying reinforcement learning to control devices on the edge," said Luigi Manca, head of digital twin practice at Engineering Ingegniera Informatica, based in Italy.

Team-centric coordination amongst machines isn't exclusive to manufacturing. This same kind of synergy can be created in other industries. Take, for instance, the rate at which Chick-Fil-A's edge devices can fulfill customer orders to meet high demand. The operational efficiency achieved by the communication between these fryers, grills, and refrigeration systems—responsible for preparing one sandwich every 16 seconds—can only be attributed to the telemetric relationship between them. That telemetry is the basis of coordination, where individual pieces of equipment can learn emergent behavior.

Combining edge computing that monitors the progress of a set of machines in processing jobs with an algorithm that can see the varying stream of orders and route those orders to the correct machine-operator pairs is a good example of optimization that has to be done on the scale of a team.

Using edge and cloud for greater throughput

Frequently, edge computing is part of a hybrid system that shares information with a larger data processing cluster in the cloud. Some data is processed locally for decisions that require low latency, with models that fit in the memory of an edge device. Other data is sent to the cloud, where larger ETL jobs and machine learning models can run.

Other examples of edge coordination include:

  • Routing fleets of autonomous guided vehicles in factories
  • Orchestrating the movement of components on an assembly line
  • Choreographing the movement of cranes in a warehouse to solve the putaway problem
  • Directing drones that fly in a swarm
  • Moving trains through a shunting yard or haul trucks through a mining site

These coordination activities help surface emergent behavior. In the context of an optimization problem, if you are trying to maximize throughput in your factory, being able to leverage edge and cloud computing together to make decisions for a whole team of machines at once can lead to actions that look like sacrificing for the team. One robot, for example, may decide to get out of the way of another robot carrying an urgent payload. This kind of team coordination can lead to real gains that aren't possible if you only focus on individual pieces of equipment.

Scheduling and routing for productions and operations present many interesting problems where edge computing can be applied and combined with computer vision for intelligent sensing, as well as reinforcement learning for team coordination and system optimization.

Author’s photo

Chris Nicholson

Chris Nicholson is the founder of Pathmind, an AI startup that applies deep reinforcement learning to supply chain and industrial operations. Pathmind optimizes performance in warehouses and on factory floors, using cloud and edge compute. More about me