Reniway Edge Architecture

From fragmented machine data to scalable industrial intelligence.

In short

The bottleneck in most factories is not analytics — it is reliable, contextualized industrial data. Reniway Edge collects from machines of any age, standardizes them into one namespace, analyzes the stream as it arrives, and serves the result to dashboards and your other systems. It runs on one appliance on your own network: your data stays on site, it works with the internet unplugged, and a fleet of devices is managed and patched from one place.

The problem this solves

Machines, PLCs, suppliers, data formats and customer environments rarely agree with each other. That fragmentation is what makes digital initiatives slow, custom, expensive and hard to scale — long before anyone gets to the analytics.

Reniway standardizes and contextualizes machine data before it reaches OEE, MES, BI, AI or enterprise systems — turning it into a reusable data foundation rather than another point solution. In one OEM programme, that replaced custom engineering with reusable product infrastructure and brought time to market down from a year to a month.

Connect what you already have

Most plants are a mix: a machine bought last year that speaks OPC UA, and one from 1998 with a serial port and an opinion. Both hold data worth having, and replacing the old one to get at it is rarely the business case.

Family	Examples
Industrial standards	OPC UA, MQTT, Modbus, Ethernet/IP, MT-Connect
Controller-specific	Beckhoff TwinCAT ADS, SINUMERIK, FANUC FOCAS, HEIDENHAIN DNC
Retrofit sensors	Vibration, sound, temperature, energy — for machines that expose nothing
Systems and databases	SQL, InfluxDB, Redis, MES and maintenance systems

Connectivity is configuration, not a development project. See Connectors.

Thirty machines of the same type?

Configure one, export it as a template — a YAML file holding the connectors, data mappers, transformations and OPC UA nodes — and upload it to recreate the whole set in one go.

Standardize into one namespace

One controller calls it Channel1.PartsProduced, the next st_prod.iCount, the old one register 40001. Anything built directly on those names breaks when the machine is replaced.

Data mappers resolve that into one model that is the same across every machine — a unified namespace. Vendor tags are renamed, units and datatypes normalized, and machine states and alarm codes mapped onto a common set of labels.

Analyze at the edge

Not every question should wait for a round trip to a data centre. Reniway runs the analytics where the data is produced.

Stage	What happens
Stream processing	Filtering, transformations and aggregations on live values as they arrive.
Feature engineering	Windowing, statistics and normalization — turning raw samples into something a model can use.
AI and anomaly detection	Isolation Forest, Extended Isolation Forest and custom models flagging behaviour that departs from normal.
Edge inference	Predictions calculated on the device, in real time.

The results are first-class values: a health score or a detected anomaly can be stored, charted, raised as an alarm, pushed as a notification, or published back out over OPC UA or MQTT like any other value. KPIs — availability, utilization, yield, OEE — are calculated against the standardized model, so they are defined once and mean the same thing on every line.

Calculating here keeps latency low enough to act on, keeps working when the network does not, and ships conclusions upstream instead of raw sample streams.

What this looks like in practice

CNC machining. Anomaly detection across energy consumption, axis and spindle load, operational data and program information produces a daily report of parts that deviated — so tooling, fixtures or material can be corrected early, cutting scrap risk and shortening the hunt for process issues.

Pump monitoring. Low-cost vibration, sound, temperature and magnetic-field sensors on ship-to-tank transfer pumps, with runtime data and anomaly alarms fed straight into the customer's cloud maintenance management system.

Store and visualize

Data lands in PostgreSQL with TimescaleDB on the appliance: time-series in hypertables partitioned by time, master data in ordinary tables beside it. It is plain PostgreSQL — your BI tool, a Python script or a warehouse ingestion job can read it directly, with no proprietary format and no export step.

Grafana ships with the stack at https://[HOST]/dashboards and queries that database in SQL, so there is no second copy to fall behind. Dashboards are delivered as versioned configuration, not hand-built on the box, so a set can be rolled out to another line and a change can be rolled back. Labels can be driven from translation tables for multi-language operator screens, the appliance can drive a plant-floor display directly, and the included report service renders PDFs from the same data. See Included components.

Retention is your decision

How long history is kept is a per-project setting agreed with you, not a platform default.

Deployment and fleet management

Reniway Edge is deployed as a self-contained physical appliance or virtual machine on the plant network. The application services run as containers on a hardened Linux operating system, while database, configuration and application data are kept separately so they survive application updates.

One appliance is a project; twenty is an estate. Reniway Edge treats a device as three independently versioned parts, so an OS security patch does not disturb a dashboard change:

Reniway Cloud is the optional management layer: it answers which release is available to which device and from when, which is what makes staged rollouts and pilot groups possible. It collects no machine data and cannot act on a device by itself — Cloud controls what is available; the operator controls when. Devices without internet are commissioned from a USB stick and updated through their own local interface; the cloud path is automation on top of that, not a dependency.

Security

Layer	What protects it
Device	Hardened Linux, read-only root — the running system cannot be changed in place.
Updates	Signed and verified on the device. Unsigned is refused.
Protocols	HTTPS for the interface, OPC UA certificate trust, MQTT over TLS with authentication.
People	Named accounts with per-area View / Manage permissions.
Data	Stays on your network. It leaves only over integrations you configure.

Where the machine is the limit

A 1998 controller cannot be encrypted, and no platform changes that. Reniway terminates those protocols at the edge and re-publishes over authenticated ones — so the unencrypted hop is a short cable inside your plant, not a path across your network.

Planning a rollout?

Contact sales@reniver.eu to walk through your target environment, network topology, or a specific integration requirement.