A comprehensive technical analysis of the Aether fog computing architecture—from its Hub & Spoke topology and binary data protocols to its high-velocity ingestion engine and on-premise anomaly detection. Engineering industrial sovereignty from first principles.
In high-security industrial environments, the "Cloud-First" orthodoxy is not a preference—it is a liability. The centralized, remote nature of cloud computing introduces fatal vulnerabilities for facilities where operational uptime is measured in millions of dollars per hour.
Round-trip cloud processing takes seconds. Failure occurs in milliseconds.
Raw operational data is IP. Transmitting it outside firewalls is unacceptable.
Streaming raw audio from dozens of sensors saturates corporate networks.
Aether rejects cloud-centric models in favor of Fog Computing— a "Hub & Spoke" architecture that centralizes all heavy computation in an on-premise "Brain" while orchestrating a swarm of simple, dedicated sensors. Intelligence delivered where it's needed.
Decisions on the metal. Milliseconds from vibration to insight.
100% raw data stays on-site. Only alerts escape the perimeter.
Binary mathematics, not streaming audio. Negligible network impact.
A deliberate pivot from "Pure Edge" approaches where significant processing occurs on thermally-constrained edge devices. Aether centralizes all heavy computation in an on-premise Hub while the Nodes remain singular in purpose: capture and transmit.
Dedicated, ruggedized sensor nodes designed for single-purpose operation. Data capture and lightweight pre-processing. Custom firmware for maximum performance and reliability.
Listens for Hub's UDP broadcast (b"AETHER_HUB_BEACON") on port 5000 for zero-touch auto-pairing.
Captures machine audio. Performs Fast Fourier Transform (FFT) locally. Converts sound waves to frequency spectrum.
Converts 1,024 FFT floats into compact binary struct. Far more efficient than JSON.
Fire-and-forget UDP packet to Hub. Lightweight. High-volume capable.
On-premise "Heavy Edge" server. 100% of data processing, time-series storage, anomaly detection, and decision-making. The single source of truth within the facility.
Python asyncio UDP server. 1,000+ packets/second from the entire sensor swarm without drop.
TimescaleDB persistence. Batch inserts via asyncpg executemany. 100-packet buffer or 1-second timeout.
NumPy/SciPy analysis on CPU. Frequency peak detection (np.max). Threshold monitoring.
Grafana dashboard. Real-time machine health. No cloud dependency.
A containerized, multi-service software architecture orchestrated via Docker Compose, running headlessly on Ubuntu Server. Best-in-class open-source tools combined with custom high-performance code.
The Black Box. On-premise sovereign server with industrial-grade compute power. Purpose-built for high-velocity edge deployment.
Ruggedized acoustic sensor nodes with on-device pre-processing. Magnetically mounted. Industrial-hardened enclosures.
Private mesh network. Pre-configured and isolated. Zero-touch node discovery via UDP beacon.
Complete kit arrives in ruggedized weatherproof case. Plug in power. Connect to factory ethernet (outbound only). Sensor nodes auto-discover Hub via beacon. Zero manual configuration.
The network topology is the architectural answer to "Data Gravity." A two-tiered strategy creates a logical air gap isolating the sensor swarm from the corporate network.
The Hub broadcasts its own private WiFi 6 network. Nodes connect only to this network. They are completely invisible to—and inaccessible from—the factory's main LAN/WAN.
Hub's single physical connection to factory ethernet is used exclusively for outbound transmission of high-level "Alerts." No inbound connections required. Minimal, tightly controlled network exposure.
UDP beacon-based auto-pairing eliminates manual configuration.
UDP binary protocol chosen specifically for low overhead and high performance compared to text-based formats like JSON. Every byte transmitted is pure data.
Binary packet received by Python asyncio UDP server. Parsed instantly. Zero deserialization overhead.
Buffered in memory. Batch-written to TimescaleDB via asyncpg executemany. 100 packets or 1-second timeout triggers flush.
Each batch passed to NumPy engine. Spectrogram converted to high-performance array. np.max checks frequency peaks against operational thresholds.
The Aether platform is currently in active development. Join the waitlist for priority access, founding member pricing, and direct input on the product roadmap.