A nervous system, taken literally
In physiology, afferent describes signals that travel from sensory receptors toward the central nervous system. The word names the direction of travel: inward, from the body's edge to the place where sensation becomes awareness.
That is the architecture of the platform. NODE devices are the receptors. Networks carry the signals. AFFERENT Cloud is where signals from many receptors are combined and interpreted. The analogy is useful because it is structurally exact, and because it explains a design decision that might otherwise look arbitrary: a great deal of interpretation happens at the edge, before anything reaches the center, just as a reflex does not wait for the brain.
The six stages
1. Physical asset
Everything begins with a machine or a piece of infrastructure: a pump, a motor, a chiller, a transformer, a generator. The asset is the source of truth. It produces physical signals continuously as a by-product of doing its work.
2. NODE
NODE is the device mounted on or beside the asset. It is designed to capture several kinds of signal at once, because the combination carries far more information than any single measurement.
3. Edge AI
Analysis is designed to begin on the device. Edge intelligence establishes what normal looks like for this particular asset in this particular place, and identifies departures from it. This keeps response fast, keeps data volumes manageable and keeps the device useful when connectivity is poor.
4. AFFERENT Cloud
AFFERENT Cloud is designed to aggregate what devices report. It is where a single finding becomes part of a larger picture: the health of a facility, the history of an asset, the comparison between one site and another.
5. Machine intelligence
Across many assets and long periods, patterns of degradation become learnable. Machine Health Intelligence is our name for models that are intended to recognize those patterns and to estimate what they mean: what is likely failing, how soon and with what confidence.
6. Action
The final stage is a human decision. The platform is designed to end in something specific: an alert, a work order, an entry in a maintenance plan, a message to another system through an API. Understanding that does not reach a person who can act on it has not finished its journey.
Why build hardware and software together
It would be simpler to build only software and rely on sensors made by others. We think the harder path is the right one, for three reasons.
Data quality is decided at the sensor. A model can only be as good as the signal it receives. Controlling how a signal is captured, sampled and synchronized across modalities is the foundation of everything built on top of it.
Edge intelligence depends on the device. Running useful analysis on a small, power-constrained device in a hot, electrically noisy environment is an engineering problem that spans the processor, the firmware and the model. It cannot be solved from one side.
Consistency compounds. When every device captures signals the same way, what is learned from one asset can be compared with what is learned from another. That consistency is what allows knowledge to accumulate across deployments.
How a deployment is intended to unfold
The architecture is easier to judge when it is followed through an installation. The sequence below describes the intended experience.
- Select the assets. An operator chooses the equipment to cover, usually beginning with machines that have a history of trouble or a high cost of failure.
- Mount the devices. NODE devices are attached to those assets. The design goal is installation on running equipment, without a shutdown.
- Learn the baseline. Each device observes its asset through a period of normal operation and learns what healthy behavior looks like across its usual range of conditions.
- Watch for change. From then on the device compares what it observes with that baseline, on the device, continuously.
- Report findings. When a meaningful change appears, the device reports it to AFFERENT Cloud with the supporting evidence.
- Set it in context. The cloud is designed to compare the finding with the asset's own history and with the behavior of similar assets.
- Act. The operator receives a prioritized, explained recommendation, ideally inside the maintenance system already in use.
- Close the loop. What the technician finds on inspection is recorded. That outcome is the most valuable data in the system, because it is how models learn whether they were right.
The last step deserves emphasis. A prediction that is never checked against what actually happened teaches nothing. Designing for that feedback from the beginning is what allows the system to improve.
What exists today
AFFERENT is a new company. The platform described here is the system we are building, and each product page states its current status plainly. We would rather describe an ambitious design accurately than describe an early product generously.
