Summary

If you use GE iFIX, Tatsoft FrameworX, Ignition, OSIsoft PI, Canary, or other SCADA/historian systems — you already have the data you need for AI.

In this webinar, INS3 and Sorba.ai showed how to turn that data into predictive insights and measurable ROI — without a costly rip-and-replace project.

Missed it or want to revisit key points? Below you’ll find highlights and detailed answers to the top audience questions.

Key Takeaways

  • No Rip-and-Replace Needed: Use the systems you already have (SCADA, historians, ERP, MES, CMMS) as data sources.

  • Flexible Deployment: Sorba runs fully on-premises (below your firewall), in the cloud, or hybrid.

  • Broad Connectivity: 60+ built-in drivers plus OPC UA, MQTT, SQL, PI, Canary, Rockwell, Siemens, Modbus, and more.

  • Hardware-Friendly: Standard CPUs are enough; GPUs optional for high-speed compute.

  • Rapid AI/ML Adoption: No-code model building, auto-ETL, auto-retraining.

  • Real Use Cases: Predictive maintenance, digital twins, quality control, schedule optimization, advanced process control, and vision/optical ML.

  • Proven Expertise: INS3 brings 35+ years of solving manufacturing challenges, now helping clients scale AI with Sorba.

Download the Presentation

Q&A

Q: Can the solution be deployed 100% on-prem below the firewall?
Yes. Over 80% of Sorba customers run fully on-premises in air-gapped environments, below the OT firewall. Every feature — including upcoming LLM capabilities — is available on-prem.

Q: If we want to use the cloud is there a preferred cloud provider?
No. Sorba is cloud-agnostic. It has been deployed on AWS, Azure, and Oracle, and is available in the Azure Marketplace. It can run on any enterprise cloud you prefer.

Q: Does this solution require specific hardware to work?
No. Sorba is hardware-agnostic. Minimum requirements: 16 GB RAM, 8 cores, and 250 GB storage. It runs on standard CPUs; GPUs are optional and simply accelerate workloads. Supported deployments include VM, Docker, Ubuntu, Red Hat, and Windows.

Q: What systems can you connect to?
Sorba supports 60+ built-in connectors, including OPC UA, MQTT, Canary, OSIsoft PI, SQL databases, Rockwell, Siemens, Modbus TCP, Ignition, and more. Custom drivers have also been built for industry-specific use cases (e.g., trains, mining trucks).

Q: Is there a download available?
Yes. The Sorba Portal offers a download with a two-hour runtime trial (restart to continue).

Q: Is there a learning center?
Yes. Sorba University provides lessons and certifications. It is updated regularly with each new release.

Q: What is the database in Sorba which takes in all the data and what schema is created there including contextualization?
Sorba uses multiple databases, including its own built-in historian and InfluxDB for real-time data. The schema and contextualization depend on workload, but Sorba can also integrate with your existing databases via APIs.

Q: Different data come at different frequency — when you add all of them in one group how is the missing data filled for slower data signals?
Sorba includes built-in tools for data alignment and cleansing. For example, you can configure “fill with previous value,” discard nulls, or use other resampling methods to create consistent datasets across mixed tag frequencies.

Q: How fast can you ingest images — how many frames per second?
Sorba ingests as fast as the camera can provide. Standard CPUs handle most use cases; GPUs are recommended for high-frame-rate or millisecond-level image ingestion.

Q: In a Total Cost of Plant Replacement what is the ratio of such system including the physical and non-physical assets to operate if it is considered during the design stage of the manufacturing plant?
This depends heavily on plant design. Sorba is not a replacement but an augmentation layer on top of existing PLC/SCADA infrastructure. The key is ensuring systems are digitally ready (reliable data, connectivity, resolution) during plant design, so that Sorba can deliver maximum ROI.

Q: How much accuracy is lost in deploying the model to the edge?
None. Models deployed at the edge run with the same accuracy as in the cloud. You can also configure different data granularities (e.g., millisecond data at the edge vs. 1-second data upstream) to optimize bandwidth without losing accuracy.