A modular validation engine that ingests live tag streams from your DCS, SCADA, edge devices, and historians — scores every value against statistical, physical, and contextual rules, and writes the resulting integrity verdict back into the historian itself. Downstream AI, optimisation, and operations applications then read every value with its quality flag already attached.
Each tier is independently scalable, independently deployable, and independently auditable.
A walk-through of what happens to a single sensor value as it moves from a transmitter on the unit through VALID8A and out to the consuming AI model or optimiser.
Certified protocol adapters pull live values from DCS controllers, SCADA front-ends, OPC-UA devices, and read historian backfill in parallel. The ingestion tier is stateless and horizontally scalable.
Each value is checked against multiple statistical envelopes — range, rate-of-change, CUSUM / EWMA drift, cross-tag correlation, and frozen-value detection. Failures are aggregated into a single health score (0–100) per instrument so engineers see one number, not twelve alarms.
Trusted rules handle the obvious: drop frozen values, suppress spikes outside instrument range, gap-fill short comms drop-outs. ML models then handle the subtle cases — analyser drift inside the calibrated range, persistent low-magnitude bias, multi-tag inconsistencies. Every transformation is reversible.
Each tag is enriched with its equipment class, process unit, operating mode, and alarm state, so downstream models can distinguish "the column is in a startup transient" from "the temperature transmitter is broken." Context is sourced from your existing tag dictionary and AMS hierarchy.
The verdict (VALIDATED · QUESTIONABLE · BAD) and its confidence score are written back into the historian alongside the value — using each historian’s native quality-flag mechanism (PI digital states, IP.21 status, etc.). Every downstream consumer reads the value with the quality verdict already attached.
Every transformation is logged with before / after value, rule that fired, model confidence, and analyst override (if any). Certified-clean streams are published to AI/MLOps pipelines, optimisers, MES, LIMS, and reliability platforms via REST, Kafka, or direct historian write — whichever your stack expects.
VALID8A inherits each sensor’s priority from the equipment and unit operation it supports — so safety-critical thermocouples are not treated the same way as an advisory wall-temperature indicator. Validation rigour, escalation speed, and audit depth all scale with priority.
VALID8A reads from and writes back into the data systems your engineers, operators, and analysts already use. No rip-and-replace; no parallel database for downstream apps to ignore.
Four reference deployments cover the patterns we’ve seen across the energy and chemical sectors. All four support the full validation, write-back, and audit pipeline.
Containerised stack inside the plant DMZ. Reads from historians and DCS over the L3 network; writes verdicts back through the same boundary.
Lightweight engine on a hardened industrial gateway, sized for a single unit or wellhead cluster. Ideal for remote sites with intermittent network back-haul.
Customer-managed AWS / Azure / GCP tenancy. Best fit for enterprises consolidating multiple sites into a single validation layer feeding shared AI pipelines.
Fully offline deployment for high-security environments. Audit-trail export via approved one-way data diodes; no inbound network dependencies whatsoever.
A 30-minute architecture review with our process automation engineers. Bring your historian, DCS, and AI/MLOps platforms — we’ll walk through exactly how VALID8A would sit in the middle.
Contact Us