Physical AI for Industrial SCADA and ERP Integration

Physical AI is AI that works from measurements of the real world, such as LiDAR scans, camera images, locations and sensor readings, and turns them into decisions about physical operations. In an industrial site it sits between the sensors and the business systems: it measures, predicts and flags, while SCADA keeps control of the process and the ERP keeps the business record.

SCADA runs the process, MES runs production execution and ERP runs the business. SCADA collects real-time data from PLCs and field devices and lets operators supervise equipment. An MES tracks work in progress, material movement and quality on the shop floor. The ERP records inventory, orders, maintenance and costs. In ISA-95 terms they sit at levels 2, 3 and 4. Integration moves the right measured values between them, for example a measured volume that updates both a SCADA tag and an ERP inventory record, so every system works from the same number.

Yes. In a recent consumer packaged goods installation, CRATUS integrated its two-LiDAR-per-bunker volume measurement system directly with the customer's SCADA, and the system integrator queries the edge CPU from SCADA and stores timestamped volume readings in their own database.

CRATUS designs connections to SAP S/4HANA, Oracle Fusion Cloud ERP and Oracle NetSuite, Microsoft Dynamics 365, Infor CloudSuite and Epicor Kinetic, using each vendor's published APIs and integration services. Other ERPs with a documented API can be connected the same way. The paths that apply to your project are confirmed in the site assessment.

CRATUS designs connections to Ignition, AVEVA System Platform and InTouch, Siemens SIMATIC WinCC, Rockwell FactoryTalk, GE Vernova Proficy iFIX, ICONICS GENESIS64, Schneider Electric EcoStruxure Geo SCADA Expert, Emerson Movicon.NExT and COPA-DATA zenon, mainly through OPC UA, MQTT and Modbus. The paths that apply to your project are confirmed in the site assessment.

The usual path has four steps. First, the LiDAR system measures the pile or bunker and calculates volume on an edge computer at your site. Second, the Physical AI layer converts volume to estimated weight with a density model for the material, time-stamps the result and checks it against limits you define. Third, a connector posts the value to the ERP through its published API, for example as an inventory adjustment or a stock level update, following your ERP's approval rules. Fourth, the same value goes to SCADA or a historian so operators see what the ERP sees. The exact objects and approval steps depend on your ERP and your inventory process, and are set in the site assessment.

OPC UA is a vendor-neutral protocol with a built-in information model and security. Most SCADA platforms and historians can read it directly, so it is the usual choice for connecting to SCADA. MQTT is a lightweight publish-subscribe protocol that suits sending data from many edge devices to a broker, which is why it is common in multi-site and unified namespace designs. Sparkplug B adds a standard topic and payload format on top of MQTT. Many projects use both: OPC UA toward the SCADA at one site and MQTT between sites. We choose the interface your destination system already expects.

ISA-95, published internationally as IEC 62264, is a standard that describes how business systems such as an ERP exchange information with manufacturing operations and control systems. It defines levels, from sensors and controllers at the bottom to the ERP at the top, and the objects that move between them, such as materials, equipment and work orders. For integration it gives every team the same vocabulary: a stockpile measurement becomes a material lot at a location, not just a number in a tag. CRATUS uses an ISA-95 style model to map sensors to sites, areas, assets and materials. ISA-95 is a reference model, not a certification.

No. Most projects start from a single data source, such as one bunker, one conveyor or one danger zone, connected to one destination system. The same data model then extends to a site-wide digital twin built on 2D and 3D facility maps in ASSET-Rx Workflow Studio.

CRATUS uses the interfaces the destination system already expects: OPC UA, MQTT and Sparkplug B, Modbus, REST APIs and SQL for SCADA and historians, and vendor APIs such as OData and REST for ERPs. On the device side, ASSET-Rx exposes REST, MODBUS, J1939, CAN and Wireless CAN.

It is designed not to. CRATUS designs simulations and test benches first, then deploys in phases while the operation continues. New systems are tested alongside the existing data sources, and each source is switched over only after the results match.

No. The AI in this layer monitors, predicts and recommends. Process control stays in your PLCs and SCADA, safety functions stay in your rated safety systems, and business transactions are posted for approval under your ERP's normal rules.

On site by default. CRATUS volume measurement and ASSET-Rx EDGE vision run on edge computers at your facility, and cloud services are optional. Your SCADA and ERP receive the results through your own network rules.

It depends on scope, which is set during the site assessment. A CRATUS site assessment takes two to four weeks, and a single-use-case pilot typically runs four to eight weeks from assessment to live data.

Cost follows scope, not the number of sensors alone. The main drivers are how many data sources and sites are included, which protocols and systems must be connected, how much data modeling and mapping to ERP objects is needed, whether your ERP approval rules need workflow changes, the network security zoning and review your IT team requires, and how much simulation and test-bench work is needed before cutover. A single-use-case pilot keeps all of these small, which is why many projects start there. The scope is set during the site assessment.

To scope a project we need the name and version of your SCADA platform and ERP, the data you want to move and the system it should arrive in, a list of the sensors or measurement systems involved, the ERP object types that should be updated (for example inventory or work orders), who approves postings, and the network rules for connecting an edge computer. A short call with your operations, automation and IT contacts is usually enough to start the site assessment.