Five integrated layers — from inline hardware to physics-informed models — each designed for continuous industrial operation from day one.
The quantropIQ system is not a sensor with a dashboard. It is a full-stack measurement and inference architecture — hardware, compute, representation, physics-informed modelling, and closed-loop applications — designed so that each layer creates the conditions for the next.
The sensor mounts via standard industrial flange — no flow interruption, no sampling, no modification of the process vessel. A measurement grid spans the full cross-section, resolving spatial distribution of phases, concentrations, and flow structures at millisecond resolution.
Conductivity and permittivity are captured simultaneously across the full cross-section. The output is a space-time field where gas holdup, mixing quality, and phase instabilities are directly observable — not inferred from surrogate signals.
Embedded GPU compute per sensor handles real-time signal processing at the acquisition rate. Multiple sensors are aggregated and synchronized with auxiliary signals — DCS, flow meters, temperature — in a unified data stream. Deterministic low-latency processing runs fully on-premise, with no dependency on cloud connectivity for operation.
Each measurement frame is represented as a physical fingerprint — a structured spatial signature capturing gradients, flow patterns, and dynamic changes. Similar states produce similar fingerprints; deviations are measurable distances from a known baseline. This representation enables process states to be compared, classified, and tracked over time in a physically meaningful way.
Physics-Informed Neural Networks (PINNs) combine measurement data with physical conservation equations — mass, momentum, energy — enabling predictions, what-if analysis, and causal interpretation that purely data-driven models cannot provide. Process knowledge is encoded in the model structure, not only learned from historical data. This makes models that generalize reliably across operating conditions and remain interpretable to process engineers.
The physical understanding built in L1–L4 enables a progression of operational capabilities: stability assurance and early warning; energy optimization; autonomous closed-loop control; scale-up acceleration with quantified process fingerprints; and cross-plant learning as deployments accumulate. Each application builds on the same physical foundation — no separate system required.
Physical AI requires dense, spatially resolved, high-speed data. Most conventional measurement technologies cannot deliver this combination — which means the models trained on them cannot learn physical behaviour, only correlations.
The technology stack described above is not process-specific — it applies wherever multiphase flow has an economic consequence. What changes between applications is the physical phenomenon being measured and the operational decision being informed.
See how the same sensing and inference infrastructure creates distinct value in different industrial contexts.
Sensor acquires — full cross-sectional measurement at 1,000+ frames per second, continuously, inline.
Compute processes — on-premise GPU transforms raw measurements into spatial state representations in real time.
Model interprets — PINNs translate spatial fingerprints into physically meaningful process states, predictions, and anomalies.
Operations act — process engineers receive actionable insight; closed-loop systems receive control signals directly.
A pilot deployment is configured for your specific process geometry, fluid system, and operating conditions — and delivers working results your team can evaluate.