Self-Regulating Process Control Under Sensor Degradation
The control loop that automatically widens its safety margin when instruments degrade — and tightens again when they recover.
- ±0.3%
- SolvSRK max setpoint deviation
- ±4.7%
- Standard PID max overshoot
- 0
- SolvSRK safety trips (100 cycles)
- 2
- Standard PID safety trips (100 cycles)
The scenario
Set the picture
A continuous chemical reactor holds temperature within ±0.5°C of a 185°C setpoint. The control loop reads a bank of four thermocouples, averages them, and drives a steam valve. Over a 6-month campaign, thermocouples drift, foul, and occasionally fail — but the PID gains were tuned on day one and never change.
The same pattern shows up across every process plant: extrusion lines, pharmaceutical reactors, food-processing pasteurizers, paper-mill dryers, semiconductor CVD chambers. Fixed-gain loops running on degrading sensors.
Cost today
Fixed-gain PID under sensor degradation either overshoots into a safety trip (lost batch, 4–8 hours of recovery) or limps at conservative setpoints (2–5% throughput penalty per shift). Both cost money.
Adaptive-gain schemes exist but require a model of the degradation mode, re-tuning on a schedule, and engineering hours per loop per quarter. Plants with 2,000+ loops cannot keep up.
What changes with SolvSRK
SolvSRK tracks state-space uncertainty inside the integration step. When thermocouple readings diverge (one drifting, three stable), the uncertainty envelope widens — and the controller automatically backs off gain, widening the proportional band. When the faulty sensor is replaced, uncertainty tightens and the loop returns to tight control.
Across 5 degradation profiles (drift, noise spike, bias shift, intermittent dropout, correlated double fault), SolvSRK held ±0.3% of setpoint with zero trips. The standard PID exceeded ±4.7% overshoot and tripped twice in 100 challenge cycles.
No re-tuning. No degradation model. No per-loop engineering. The safety margin is live, not designed-in.
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.
- ±0.3% setpoint accuracy under all 5 sensor degradation profiles (standard PID: ±4.7%).
- Zero safety trips across 100 challenge cycles (standard PID: 2 trips).
- No per-loop tuning or degradation-mode model required.
- Uncertainty envelope visible to operator — confidence interval on every reading, not just the point estimate.
- Compatible with existing 4–20 mA / HART / Profibus instrument buses.
The demo
What was tested. How. What the simulation printed.
Simulated continuous reactor with 4 thermocouples, a steam valve, and a 185°C setpoint. 100 challenge cycles across 5 degradation profiles: slow drift (0.1°C/hour), noise spike (±2°C burst), bias shift (+1.5°C step), intermittent dropout (5-second gaps), and correlated double fault (2 of 4 sensors drifting in the same direction).
Two controllers run in parallel on the same plant model: standard industrial PID (Ziegler-Nichols tuned) and SolvSRK self-regulating control. Measured: max overshoot, trip count, time to recover after sensor replacement.
Captured benchmark output
The numbers the simulation actually printed.
| Profile | SolvSRK max dev | PID max dev | SolvSRK trips | PID trips |
|---|---|---|---|---|
| Slow drift | 0.21% | 2.1% | 0 | 0 |
| Noise spike | 0.30% | 4.7% | 0 | 1 |
| Bias shift | 0.18% | 3.2% | 0 | 0 |
| Intermittent dropout | 0.27% | 3.8% | 0 | 1 |
| Correlated double | 0.24% | 4.1% | 0 | 0 |
Standard PID: Ziegler-Nichols tuned on clean instruments. SolvSRK: single configuration, no per-profile tuning.
Composes with
Where this POC sits in the benchmark suite
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.
- SolvSRK self-regulating control validation suite
- SolvSRK Real-Time UQ — calibrated uncertainty envelopes
- Robotics vertical — validated claims for self-regulating control
- SolvSRK dynamics validation — stability under contested noise (91%+ survival)
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.