Definition
Composite of filter run extension magnitude and effluent turbidity trend steepness during the extended portion, measuring whether AI-extended runs are approaching breakthrough faster than historical experience.
The Measurement Gap
A drinking water plant operating 12 dual-media gravity filters deploys an AI system that optimizes filter run times based on real-time effluent turbidity, headloss development, and influent conditions. The optimizer's objective is to maximize filter throughput by extending runs when conditions allow, reducing backwash water usage and improving plant production efficiency.
Within six months, the optimizer extends average filter runs from 48 hours to 67 hours, a 40% improvement in throughput per run cycle. Backwash water usage drops 28%. Plant production efficiency increases. The operations team celebrates.
What the production metrics do not capture is the quality trajectory during the extended portion of each run. During the first 48 hours (the historical run length), effluent turbidity increases gradually at approximately 0.002 NTU per hour, a safe, predictable slope. During the extended portion (hours 48-67), the turbidity slope steepens to 0.008 NTU per hour in several filters: the filtration media is nearing the end of its effective capacity and quality deterioration is accelerating.
The filters are still passing. Turbidity is still below the 0.3 NTU regulatory trigger. But the steeper slope means the margin between operating turbidity and the breakthrough threshold is thinning with each hour of extension. Simultaneously, the extended run duration under higher headloss conditions may be degrading the filter media faster: compressing the media bed, driving fines through the support gravel, and reducing the effective surface area available for the next run.
No existing metric connects filter run extension to the quality deterioration rate during the extension. FRERI measures this directly (comparing extended run times to historical baselines and weighting by the quality deterioration slope) revealing whether the AI's filter run extensions are trading long-term media health for short-term production gains.
Formula
Layer 1: Plain Language
For each AI-extended filter run, compare the run time to the historical average for that filter under similar conditions. Then check whether the effluent turbidity trend during the extended portion was steeper (quality deteriorating faster) than during the historical run period. A high ratio with a steep slope means the AI is pushing filters past their safe operating window.
Layer 2: Mathematical Notation
FRERI = (Run_AI / Run_historical) × (Slope_extended / Slope_normal)
Where:
- Run_AI = actual filter run time for AI-extended run (hours)
- Run_historical = historical average run time for that filter under comparable influent conditions (hours)
- Slope_extended = rate of turbidity increase during the AI-extended portion (NTU/hour)
- Slope_normal = rate of turbidity increase during the normal (non-extended) portion of the run (NTU/hour)
Units: Dimensionless composite. Values near 1.0 indicate extensions are benign. Values above 2.0 indicate extensions are producing accelerated quality deterioration. Track per filter: performance varies by media age and condition.
Proposed formula, pending validation pilots.