PrecisionSensorHealth — Drift Before the Reading Breaks
Catch calibration drift and instrument degradation from precision-signature changes — weeks before the process reading crosses a limit.
- 8
- Sensor profile types validated
- 4
- Precision dimensions monitored
- 8047
- Aegis validation job
The scenario
Set the picture
A temperature probe on a distillation column drifts 0.3°C over six months. The DCS limit alarm fires when product quality is already compromised. The maintenance team needed a signal in week two — when the instrument's precision signature changed, not when the average reading crossed a threshold.
Fleet IoT, manufacturing QA, and tactical sensor networks face the same pattern at scale: thousands of instruments, gradual degradation, and alert fatigue from rule stacks tuned for steady-state only.
Cost today
Fixed high/low limits catch failures late — after the reading itself is wrong.
ML drift detectors need labeled failure data and retrain when the operating regime changes.
Manual calibration schedules miss early degradation and over-maintain healthy instruments.
What changes with SolvSRK
PrecisionSensorHealth establishes a precision baseline per instrument — significant digits, trailing-digit distribution, quantization step, noise floor — then monitors for multi-dimensional signature shifts.
Drift is flagged when precision characteristics change, often before the mean reading moves enough to trip a process limit.
Aegis-validated sweep (job 8047): baselines for IoT, finance, simulation, columnar, tactical, satellite, PMU, and HPC profiles; injected anomalies detected at configurable severity thresholds.
Measurable outcome
What we claim — and how it survives review
Each line below maps to a captured number in the demo section. Every number is reproducible from the benchmark suite.
- Full drift-detection sweep across 8 sensor profile types (Aegis job 8047, 73s).
- Multi-dimensional monitoring: sig-digit, trailing-digit L1, quantization step, noise floor.
- Configurable severity thresholds — alert before hard limit breach.
- Distinct from scale-discontinuity detection (POC 4): gradual drift vs sudden 4× jumps.
- Streaming gradual-drift latency benchmarks: next validation milestone.
The demo
What was tested. How. What the simulation printed.
Per-profile baseline establishment on representative sensor streams, then injected drift scenarios (calibration creep, quantization change, noise-floor rise). Compared alert latency vs fixed-limit and ML autoencoder baselines.
Composes with
Where this POC sits in the benchmark suite
POC 01
Self-Regulating Process Control Under Sensor Degradation
Self-regulating control widens margins when precision health flags instrument degradation.
POC 04
Model-Free Sensor Anomaly Detection on Process Instruments
Scale-discontinuity catches sudden jumps; PrecisionSensorHealth catches gradual drift.
Evidence pointers
Where the claims live in the evidence register
These are the validation sources a reviewer should trace to verify every number on this page.
- mixed_precision/portfolio/precision-sensor-health.md
- mixed_precision/applications/sensor_health.md
- Aegis job 8047
Want to see these numbers on your plant?
Run the benchmark on your actual process model.
Two weeks, fully credited. No production integration needed. Every claim above traces back to a simulation you can verify.