From your data to your maintenance action. Built for industrial reliability.
TruPrognostics AI combines physics-based models, machine learning, and contextual AI to diagnose specific fault modes, forecast remaining useful life, and recommend the next inspection or maintenance action grounded in relevant operating manuals.
Every layer is documented. Every diagnosis cites its evidence. Every recommendation cites its source. Reliability and process engineers know what they are getting before deployment begins, not after.
How the models work together.
TruPrognostics AI combines physics-based models, machine learning, and contextual AI into a single end-to-end pipeline. The architecture below shows the path from data already in your historian, through the modeling and diagnostic layers, to a sourced maintenance action that reaches your reliability team.
From a list of fault candidates to a coherent diagnosis.
A ranked list of fault confidences is a starting point, not an answer. On any one machine, the diagnostic library may flag several related candidates with overlapping evidence. An interstage seal leak and a loss of efficiency on the same stage may be one issue expressed in two signals. A bearing wear indication and a misalignment pattern may share a root cause, or may not.
TruPrognostics AI's agentic reasoning layer takes the diagnostic library output, the prognostic projections, and the operating context, and synthesizes them into causal chains. It produces multiple ranked root cause hypotheses — each with supporting and contradicting evidence — plus a summary recommendation tied to the most likely cause. Decision authority stays with the engineer; the system narrows the problem space and shows its work.
Process data and waveform data, in the same diagnostic workflow.
Most monitoring tools cover process data or vibration data, rarely both. TruPrognostics AI runs both analytical domains in a single platform, with one integration and one diagnostic workflow.
Built from the data you already collect.
Polytropic efficiency, pressure ratios, temperature relationships, flow consistency, hydraulic performance. Detects and diagnoses process-related faults like valve degradation, intercooler fouling, leakage, and packing wear.
Spectral signatures that map to specific failure modes.
Vibration waveforms, high-frequency in-cylinder pressure, motor electrical signatures. Resolves precise mechanical and bearing faults: bearing wear, journal bearing instability, individual valve identification, crosshead and rod wear.
Three tiers, defined by data.
All TruPrognostics AI models are pre-built to support all three tiers. A site starting at Base can unlock Plus or Premium by adding sensors. No new software, no new integration, no redeployment.
| Tier | Data Required | What It Enables |
|---|---|---|
| Base | PI / time-series process data (pressures, temperatures, flows, speed) already collected by your SCADA / historian system. | Process-domain diagnostics and prognostics. Detects efficiency loss, compression changes, and some bearing conditions. Works with data you already have. |
| Plus | Base data + 1 high-frequency vibration sensor (raw waveform data) on select machines. | Precise mechanical and bearing diagnostics on critical machines. Valve leakage diagnosis on reciprocating compressors. Oil whip / whirl detection on journal bearings. |
| Premium | Multiple high-frequency sensors and modalities (e.g., vibration + high-frequency cylinder pressure + crosshead / piston rod sensors). | Maximum diagnostic precision: individual valve identification, crosshead wear, rider band wear, piston ring leakage, and more precise efficiency diagnostics. |
Capability scales with instrumentation. Day-one value comes from data already in the historian. Investment in additional sensors is targeted at the assets where the diagnostic uplift earns it.
AI that shows its work.
Every TruPrognostics AI model is documented by data tier and fault mode with explicit capability statements: which fault modes it detects, which it diagnoses, which it forecasts a remaining useful life for, and what data it needs to do each.
The same standard applies across all three layers. Physics models publish what they can detect from which inputs. Machine learning models publish their performance benchmarks and calibration approach. Contextual AI recommendations cite the OEM manual section, customer SOP, or historical work order they draw from.
Validated, not asserted.
Diagnostic true positive rate routinely above 90% with low false positive rates. Performance varies by machine type and data tier and is documented per model version. The technical guides include the full fault coverage matrices.
See it on your own assets.
The fastest way to evaluate TruPrognostics AI is on your own historian data. We can scope a proof of value in a 30-minute conversation.
