Reuse a suitable host
Customer industrial computer
Qualify an existing Linux industrial PC, site server or managed virtual machine for the selected interfaces, workload, restart and recovery requirements.
Be Water Edge · Technical overview
Edge · Site connections and local runtime
Industrial foundation · New Be Water release in developmentCore team members have delivered iDOS data platforms and specialist applications in industrial projects. Be Water is developing that technology, those models and that delivery experience into products that are easier to deploy, upgrade and maintain.
Edge brings equipment data into usable form and runs approved calculations and operating actions close to the equipment. Deploy the software on a qualified industrial computer, or select an optional hardware configuration.
Software first, hardware to suit the site
Choose the deployment around the control network, installation conditions and work to be performed. An existing suitable computer can host Edge; a new box is one option.
Add a local node
Optional Edge hardware
Select a compact acquisition node or a larger computation node, with the network ports, isolated serial interfaces, storage and environmental range the installation requires.
Size AI separately
Shared local AI compute
Where local language or vision models are useful, several Edge nodes can use a shared compute host in an approved network zone. Each acquisition node does not need its own accelerator.
Where Edge sits in the system
Equipment & existing control
PLC · SCADA · meters · gateways
Source measurements and equipment status; retain local control and protection.
Site runtime
Edge
Connect, normalize, buffer and calculate locally; check authorized action requests at the site.
Industrial context & calculation
Core or existing platform
Organize objects, relationships, history and professional calculations.
People & ongoing tasks
Work or existing applications
Use evidence and calculations to coordinate work, approvals and follow-up.
Data can flow to existing applications as well as Be Water products. Where suitable SCADA or gateway connections already exist, reuse them. Equipment actions return through an approved local execution path; Edge does not replace PLC scans, emergency stops, protection relays or manufacturer safety functions.
Protocols, interfaces and access
| Interface | Intended use | Access and qualification |
|---|---|---|
| OPC UA client | Read measurements, source timestamps and quality codes from existing automation systems | Reading is the baseline. Any write operation is separately qualified and authorized. |
| Modbus TCP / RTU | Acquire registers from meters, PLCs and equipment over Ethernet or isolated serial links | Confirm register maps, data types, byte order and polling budget for the actual device. |
| MQTT | Exchange telemetry and events with gateways and applications | Confirm topic permissions, payload schema and reconnect behavior. Message delivery alone does not verify a business action. |
| HTTPS APIs & files | Exchange data with existing services and import batches | Agree identity, schema, data quality and source ownership. Read and write permissions remain separate. |
| SunSpec Modbus | Use defined information models for photovoltaic and storage equipment | Qualify the device model, firmware and supported SunSpec models. Standard data access does not imply control permission. |
| S7 · EtherNet/IP · BACnet/IP · IEC 60870-5-104 | Extend to site-specific automation and building or energy systems | Connector-based adaptation, subject to the selected protocol stack, license and target device qualification. |
| Northbound APIs & events | Provide normalized samples, object snapshots, model results and events | Batch API and MQTT interfaces are planned; an OPC UA server is an optional integration. |
| Serial & physical I/O | Connect serial instruments through the host or suitable adapters | Isolated RS-485 is the reference serial interface. RS-232, digital or analog I/O require suitable qualified hardware. |
A protocol name is a starting point for integration. The delivery scope identifies the actual vendor, model, firmware, data items, read/write operations and tested conditions. Physical writes use an explicit allowlist and a separate authorization path.
Reference hardware profiles
| Parameter | E-S · acquisition node | E-M · computation node | E-AI · shared AI host |
|---|---|---|---|
| Processor | 4-core, 64-bit ARM or x86 | 8-core x86-64 preferred | 8+ CPU cores and a supported accelerator |
| System memory | 8 GB | 16–32 GB | 32–64 GB; accelerator memory sized to the model |
| Storage | 256 GB industrial SSD | 1 TB industrial SSD | 1–2 TB, adjusted for models and event data |
| Ethernet | 2 × 1 GbE | At least 2 × 1 GbE; 4 ports recommended | At least 2 × 1 GbE; faster storage links as needed |
| Serial interfaces | 2 isolated RS-485 ports | 2–4 isolated serial ports, as needed | Usually uses connected Edge nodes |
| Cooling | Fanless preferred | Fanless preferred, or maintainable fan cooling | Maintainable active cooling |
| Power planning | 24 VDC; up to 30 W design allowance | 24 VDC; up to 65 W design allowance | Up to 150 W host allowance; supply depends on configuration |
| Operating temperature | −20 to 60 °C target | −20 to 55 °C target | 0 to 45 °C target |
These are design references for the planned product, not certified specifications or available hardware SKUs. The selected hardware and complete installation require qualification. CPU and memory figures do not establish data throughput or model capacity; those depend on the actual workload. Power allowances exclude external equipment.
Local work through a network interruption
The runtime design separates data collection, deterministic computation and action checks from optional AI services. The agreed deployment defines what continues locally and what waits for a connection or a person.
Capture with context
Retain source and receiving times, quality, original units and mapping versions. Missing, frozen or stale values remain visible instead of appearing as fresh measurements.
Buffer and calculate locally
Keep collecting and running approved local calculations within the node’s storage and resource limits. Size the buffer from update rate, record size, retention and outage duration; waveform and video storage need a separate allowance.
Check every operating action
Apply approved local constraints, input freshness, object permissions and valid operating authority. Manual site control takes priority. A language model cannot relax the local operating limits.
Reconnect and recover deliberately
Check identity, permissions, versions and time before restoring online work. Backfill history with its original timestamps and quality; discard expired physical commands. Record controller acknowledgment separately from an observed equipment response.
From site information to a deployment
Bring the site conditions
- Network zones and permitted data paths
- Device models, firmware, point lists and existing gateways
- Update rates, local calculations and any AI workload
- Required offline work, retention and recovery conditions
- Power, cabinet space, temperature and installation constraints
- Read/write scope, local responsibilities and operating limits
Define the engineering scope
- Node placement and qualified host or hardware profile
- Connector and device compatibility matrix
- Storage, computation and communication budgets
- Data mapping, action boundaries and network diagrams
- Commissioning checks, backup and recovery arrangements
- Maintenance ownership and agreed acceptance conditions
Separate network or operating zones can require separate nodes even when their data volumes are small. Closely connected equipment may share a node when isolation and workload conditions allow it.