Predictive maintenance in industry requires the secure and continuous transfer of large datasets from isolated operational technology (OT) networks to machine learning (ML) analytics platforms. The architect's primary challenge is to build a standardized pipeline without replacing existing SCADA systems, leveraging the OPC UA standard and a clear division of computing workloads.
Why traditional SCADA systems fall short for ML
SCADA systems and programmable logic controllers (PLCs) are optimized for real-time, low-latency process control. In contrast, machine learning requires large volumes of structured time-series data and high sampling rates. Querying industrial databases directly for analytics can disrupt process control stability. Furthermore, SCADA data is often siloed due to the use of proprietary protocols.
Standardization and distributed computing
To address compatibility issues, the OPC UA standard is used, enabling the transmission of not only raw values but also rich metadata. However, sending all telemetry to the cloud is inefficient. A hybrid approach is optimal:
- At the Edge: Gateways collect high-frequency data, filter noise, and calculate aggregated metrics locally, reducing network load.
- In the Cloud: Only filtered metrics are sent to a centralized repository for training heavy ML models and long-term maintenance planning.
OT security and practical implementation
Securing industrial environments is based on the ISA/IEC 62443 standards. Secure telemetry export is established through a demilitarized zone (DMZ) using a "push" principle, where connections are initiated exclusively from within the secure OT segment outward. This prevents direct access from the IT network to industrial controllers.
In practice, implementing such architectures requires specialized expertise. For instance, Softengi provides ML model development in compliance with the ISO/IEC 42001:2023 standard. To integrate analytical insights with enterprise operational processes (such as automatically generating maintenance work orders in ERP), the UnityBase low-code platform is used, ensuring reliable integration through a unified metadata model and built-in access control mechanisms.
Impact on the industry
Establishing secure, standardized pipelines allows industrial enterprises to unlock predictive maintenance without replacing legacy SCADA infrastructure. This transition minimizes unexpected downtime and reduces operational costs, but requires strict adherence to security standards to prevent exposing critical OT networks to external cyber threats.
Recommendations
- Standardize data: Use the OPC UA standard to transmit both raw telemetry and rich metadata.
- Implement hybrid computing: Filter and aggregate data at the Edge to reduce network load, sending only essential metrics to the Cloud.
- Secure the OT segment: Set up a DMZ and use a "push" principle where connections originate only from within the secure OT network.
- Ensure compliant integration: Develop ML models under ISO/IEC 42001:2023 standards and use platforms like UnityBase to safely connect analytical insights with ERP systems.
While no ML model guarantees absolute accuracy, a standardized pipeline based on OPC UA, edge computing, and security standards minimizes unexpected downtime and reduces operational costs.
Prepared by a Software Ukraine member. Original publication.