Deviation Metric
Statistical verification protocols calculate the square root of mean squared discrepancies between observed laboratory values and predictive numerical models. Applied across quality assurance and on-line scanner platforms, root mean square error quantifies the absolute dispersion of model estimates relative to verified physical paper measurements. The metric operates exclusively on continuous quantitative variables like basis weight, moisture content or caliper, ceasing to provide meaningful insight when evaluating categorical print defects or qualitative coating faults.
Computational Structure
Calculation begins by subtracting each experimentally observed paper property value from the corresponding value predicted by an on-line model or calibration equation. Technicians square every individual residual difference to convert negative deviations into positive values. Summing these squared residuals and dividing the total by sample count yields the sample variance of the prediction errors.
The algorithm concludes by taking the positive square root of that quotient, returning the error magnitude back to the original unit of physical measurement. Because the formulation squares each residual before averaging, root mean square error penalizes severe outlying measurement deviations far more aggressively than mean absolute error metrics. A single extreme off-weight reading on a scanner reel generates a substantial increase in the resulting error score.
Automated calibration routines across near-infrared moisture sensors rely on this mathematical sensitivity to detect sensor drift across changing ambient mill temperatures.
Calibration Accuracy
Quality control systems compare predicted automated scanner profiles against offline gravimetric and micrometer readings across multiple reels. Lower root mean square error figures confirm that automated headbox actuators and moisture spray booms track actual web variations with high fidelity. When predictive algorithms return elevated error values on high-speed newsprint or folding carton lines, automated stock feed valves over-adjust, inducing severe cross-direction basis weight variations.
Sensor recalibration brings these predictive models back inside tight mill operating tolerances, commonly below 0.2 percent moisture content or 0.5 grams per square metre basis weight variance. Precise root mean square error monitoring ensures that automated process models correctly reflect real physical properties during commercial paper manufacturing.