Industrial IoT (IIoT)
Edge Computing in Indiana Manufacturing: What Belongs at the Machine and What Belongs in the Cloud
Published by IOT Indiana Research & Editorial Team

- Edge Computing in Plain Terms
- Five Tests for Deciding Where to Process
- The Four Tiers Where Processing Can Live
- Where Edge Computing Earns Its Keep in a Factory
- Indiana Realities That Shape the Edge Decision
- The Hidden Work: Managing Edge Devices
- Security at the Edge
- What It Costs
- An Illustrative Scenario
- How to Start Without Over-Building
- Common Mistakes
- Where Edge Computing in Indiana Manufacturing Is Heading
Edge computing sounds like a technology, but for a plant manager it is really a placement decision: where should a piece of data be processed, at the machine, somewhere in the building, or in a remote data center? Get the placement right and a camera catches a defective part before it leaves the station, a slow internet link stops mattering, and cloud bills stay reasonable. Get it wrong and you either pay to ship terabytes of useless readings offsite or wait two seconds for an answer that a press needed in twenty milliseconds. This guide gives Indiana manufacturers a simple way to make that call, describes the tiers where processing can live, and covers the management and security work that vendors tend to skip in the sales pitch.
Edge Computing in Plain Terms
Edge computing means processing data close to where it is created rather than sending everything to a distant server first. In a factory, the edge might be a small computer mounted in a control cabinet, an industrial PC on a production line, or a server room on the plant floor. The cloud is still useful. It is simply not the right place for every job. Plants that do well treat edge and cloud as partners, each doing the work it is best at.
Five Tests for Deciding Where to Process
For any task you are considering, run it through five questions. If the answer to one of them is a strong yes, that task probably belongs at the edge.
| Test | Question to Ask | Points Toward the Edge When |
|---|---|---|
| Latency | How quickly must the answer arrive? | A response is needed in milliseconds or a few seconds, such as rejecting a part on a moving line |
| Bandwidth | How much raw data is produced? | Signals like vibration waveforms or camera images are too large to ship continuously |
| Reliability | What happens if the internet drops? | Production or safety decisions cannot wait on an outside connection |
| Privacy and IP | How sensitive is the data? | Process parameters, recipes, or customer-controlled data should not leave the building |
| Cost | What does moving and storing the data cost? | Transmitting and storing raw readings would cost more than processing locally |
Tasks that score low on all five, such as monthly reporting across several plants or training a model on years of history, are natural cloud jobs. There is no need to force them onto local hardware.
The Four Tiers Where Processing Can Live
| Tier | Typical Hardware | Good For |
|---|---|---|
| Device edge | Smart sensor, camera, or drive with built-in processing | Simple filtering, thresholds, and event detection at the source |
| Gateway edge | Industrial gateway in a control cabinet | Protocol translation, buffering, aggregation, basic analytics |
| Plant edge | Industrial PC or small server on site | Vision inspection, historian, dashboards, models that serve the whole plant |
| Cloud or corporate data center | Remote servers | Cross-plant analytics, long-term storage, heavy model training |
Most useful designs use more than one tier. A camera flags a defect at the device edge, the plant edge logs the image and the cause, and the cloud accumulates months of results to spot which shift or supplier lot produces the most rejects.
Where Edge Computing Earns Its Keep in a Factory
Vision inspection. Cameras generate a great deal of data and often need a decision within moments. Running the inspection model on a nearby computer lets the system reject a bad part without waiting for a round trip to the internet, and it avoids uploading thousands of images that show nothing wrong.
Fast alerts and closed-loop responses. When a signal crosses a limit, such as a temperature spike in a heat-treat oven, a local rule can raise an alarm or adjust a setting immediately. The cloud can still receive the event for analysis, but the response should not depend on it. Any action that affects safety or product quality should be designed so a network failure cannot leave the process unprotected.
Data reduction. A vibration sensor sampling thousands of times per second produces far more data than anyone needs to store. Processing at the edge can compute a few summary numbers per interval, such as overall level and dominant frequency, and send those. Raw waveforms can be saved only when something unusual occurs.
Operating through outages. A plant that loses its internet connection should still keep collecting data and running local dashboards. Edge devices with local storage keep working and forward data when the connection returns, so history has no holes.
Protocol handling and standardization. Edge devices are where data from different machines gets translated and labeled consistently. This makes later analysis far easier, whether it happens locally or in the cloud.
Indiana Realities That Shape the Edge Decision
- Uneven connectivity. Many Indiana plants and processing sites are outside dense metro areas, where internet service can be slower or less dependable. Edge processing makes a system usable on modest connections, especially when paired with cellular backup.
- Mixed equipment ages. Older machines speak older protocols. Local edge devices are the natural place to translate them into a modern format without touching the machine itself.
- Customer and IP sensitivity. Suppliers to larger manufacturers may be cautious, or contractually restricted, about sending process data offsite. Processing locally and sharing only summaries or quality results can address those concerns.
- Lean IT staffing. Small and mid-size plants rarely have large IT teams, so edge equipment must be simple to deploy and manage remotely. Complexity that a big enterprise can absorb can sink a small shop.
The Hidden Work: Managing Edge Devices
Every edge device is a small computer that must be configured, updated, monitored, and eventually replaced. One is easy. Twenty across a plant become a real job. Before you deploy, settle these questions:
- Configuration. Can you create a template and roll it out to many devices, or must each be set up by hand?
- Updates. How are operating system, application, and security updates delivered? Can you test them on one unit first and roll back if something breaks?
- Monitoring. Will you know when a device goes offline, runs out of disk space, or overheats? Silent failures are the most dangerous kind.
- Spares. Keep at least one configured spare for critical devices, with a documented process to swap it in.
- Software packaging. Many modern edge platforms run applications in containers, which makes deployments repeatable and updates safer. Ask vendors how their software is packaged and how you would move it to different hardware.
- Ownership. Decide who is responsible: IT, maintenance, controls engineering, or an outside partner. Unowned devices decay.
Security at the Edge
Placing computers on the factory floor expands the number of places something can go wrong, so treat them as part of your security program from the start. Put edge devices on a network segment separate from office systems. Change default passwords and use unique credentials. Prefer connections that the device initiates outward, so you do not have to open inbound holes in the firewall. Keep software patched, and encrypt data in transit. Physical security matters as well: a device in an unlocked cabinet on a busy floor is accessible to anyone. Larger customers increasingly ask suppliers to describe these practices, so documenting them is good business as well as good hygiene.
What It Costs
Edge hardware ranges from inexpensive gateways to industrial computers that cost considerably more, and the price of the box is often the smallest part of the total. Count software licenses or subscriptions, installation and integration labor, network equipment, security measures, ongoing management, and the eventual replacement of hardware. Compare that to the cost it avoids: cloud transfer and storage fees, downtime from network dependence, and rework from slow defect detection. For many plants, the payback comes from a single well-chosen use case, such as vision inspection that cuts escapes to customers or local alerting that prevents a costly process excursion.
Indiana's Manufacturing Readiness Grants have listed advanced communications, industrial IoT, sensor technology, and big data and analytics among eligible investments when funded, though availability has depended on state budget cycles, so check the Conexus Indiana page for current status. Purdue's Manufacturing Extension Partnership offers consulting and training, including digital transformation topics, that can help a plant define its architecture before buying hardware.
An Illustrative Scenario
Imagine a mid-size plastics molder in a smaller Indiana city with a modest internet connection. It wants to inspect molded parts with cameras and monitor mold temperature. Sending continuous camera images to the cloud would overwhelm its connection and cost more than the project saves. Instead, the plant installs an industrial PC beside the line running the inspection model. Good parts pass without any image leaving the building. Rejected parts trigger an air-jet ejector within the cycle, and a picture of each reject plus the machine settings at the time is saved locally and sent to the cloud in small batches overnight. Mold temperature readings are averaged at a gateway and forwarded once per minute, with an instant local alarm if temperature leaves its band. When the internet goes down one afternoon, inspection and alarms keep working, and the buffered data uploads later. The plant did not adopt edge computing for its own sake. It solved a specific problem where the edge was the only sensible answer.
How to Start Without Over-Building
- Pick one use case that fails at least one of the five tests if done in the cloud.
- Run a small pilot with one edge device and one production line.
- Test failure behavior on purpose: unplug the internet, cut power, and see what happens.
- Document the setup, credentials handling, and recovery steps.
- Scale only after the first device has run reliably for a few months.
Common Mistakes
Treating edge and cloud as rivals. The best designs use both, with each doing what it does best.
Ignoring management. A device nobody updates or monitors becomes a liability.
Putting critical decisions behind the network. Safety and quality responses should survive an outage.
Buying too much hardware early. Start with what the first use case needs, not what the vendor says you will need eventually.
Forgetting the people. Someone must own the device, understand its behavior, and be able to fix it at two in the morning.
Where Edge Computing in Indiana Manufacturing Is Heading
Edge hardware is becoming more powerful and less expensive, which brings capabilities such as on-site vision inspection and local machine learning within reach of smaller plants. Management tools are improving, making fleets of edge devices easier to deploy and update. At the same time, customers and insurers are asking more questions about data handling, which favors designs that keep sensitive information on site. The plants that benefit will be those that decide placement deliberately, plan for management from day one, and design so that a lost internet connection is an inconvenience rather than a shutdown.
Frequently asked questions
Related articles
Industrial IoT (IIoT)
Industrial IoT (IIoT)
Predictive Maintenance Becomes More Important Across Indiana Factories
Industrial IoT (IIoT)
Cummins and the Connected Factory: What Industrial IoT Means for Indiana Manufacturing
Industrial IoT (IIoT)
