Embedded predictive maintenance · Réunion Island
We build connected devices that analyse your equipment where it stands. The intelligence sits inside the module, not in a remote server : an anomaly is detected and reported on site, even when the link is down. Without replacing your equipment, without a carrier subscription, without your data ever leaving your infrastructure.
Bearing degradation stages Detected at stage one
The context
Industrial SMEs, network operators, local authorities : unplanned downtime costs far more than the part that broke. Continuous monitoring is the only answer, but market solutions generally assume you start from scratch. That is neither affordable nor desirable.
The solution
They deploy together or separately, depending on what your installed base requires.
Anticipation
Analysing vibration, ultrasound and power consumption reveals drifts before they become failures. The models run inside the module, on the remote site : detection depends neither on a permanent link nor on a server. Server side, machine learning consolidates the history and refines the prediction.
Prediction on site, even with the link down
Metering
Meters, dials, needle gauges : embedded AI reads the analogue values of your equipment through image analysis. Nothing to remove, no interruption to operations, an entry cost far below a fleet replacement.
No equipment replacement
Communication
Our LoRaMesh stack : every module relays its neighbours and recomputes its route if one of them goes down. It covers sites scattered over kilometres just as well as a factory floor, where thick walls and electromagnetic noise defeat other links. No subscription, no carrier coverage, no single gateway. Wi-Fi or a wired link takes over where the infrastructure allows, with PLC interfacing over Modbus.
Scattered outdoors or noisy indoors
Route : N01 · N04 · N06 Link established
The mesh in action
When a relay goes down, the data does not stop. It changes route.
Every module knows its neighbours and recomputes its route continuously. No central gateway can bring the network down : there isn't one.
Local processing
When a rolling element passes over a spall, it produces a brief impact that makes the structure ring in the ultrasonic band, around 35 to 45 kHz. That ringing says nothing about the fault. What identifies it is the rate at which it repeats — about a hundred times per second, set by the bearing geometry and the shaft speed. Extracting that rate from the carrier is envelope demodulation : it brings out the harmonic family of the fault, weeks before the overall level moves.
Ultrasound 35–45 kHz · demodulated on site BPFO signature · outer race
This computation cannot be moved off site. It is not an architecture choice, it is a physical constraint.
Day to day
The modules do not merely report readings : they decide on the spot, alert immediately, and feed a platform you host yourself.
What sets us apart
The industrial IoT market sells standardised products, designed for well-covered sites and recent equipment. Our design starts from the opposite assumption.
The computation happens in the field, inside the module. Immediate detection and alerting, with no latency and no dependence on a permanent link to a server. A site cut off from the world keeps monitoring itself.
A self-healing network with no single point of failure. A node going down does not cut the flow of data : it simply changes the path.
We instrument what is already there. No removal, no production stop, no proprietary sensors imposed as replacements.
A proprietary communication stack, independent of carrier networks, with end-to-end encryption. The platform is hosted on your premises. Your infrastructure data passes through no third-party platform.
The company
Insularity, relief, scattered assets, tropical climate : within a few kilometres the island concentrates everything that defeats an industrial solution. We validate on the spot, in real conditions.
A solution that holds here carries over to comparable territories — the Indian Ocean region, the French overseas departments, isolated sites in mainland France.
Founder & president
Trained in electronics and then in industrial computing, he spent several years running infrastructure and information-system projects, up to management roles, before returning to his original discipline.
Robotic 974, the association he founded around ten years ago, put him back in direct contact with the needs of local industry. Edge Processing grew out of that meeting between electronics, software and concrete problems in the field.
Contact
The number of sites, the distances, the nature of the equipment and what you are trying to monitor or anticipate. We will tell you plainly what is feasible — and what is not.
Edge