Statistical Process Control Using Weibull Residual Filtering for High Speed Converting Substrates
Weibull residual filtering isolates true substrate defects from machine noise in high-speed converting, reducing false alarms and protecting against supplier liability.

Tension
High-speed converting lines running flexible paper webs between 800 and 1,200 meters per minute expose structural non-linearities that break conventional statistical process monitoring. Paper fibers relax under strain. Standard rolls of virgin unbleached kraft liner, recycled testliner, and supercalendered release papers carry significant thickness variations and moisture gradients across the reel length.
Continuous sensor arrays measuring load cell strain or optical displacement produce high-frequency time series packed with non-stationary baselines. When converted under heavy mechanical draw, substrate web behavior deviates sharply from classic Gaussian assumptions.
Standard Shewhart charts fail here. Conventional control limits depend on independent, identically distributed normal observations. Tensile strength measured under ISO 1924-2 and burst resistance evaluated via ISO 2758 exhibit right-skewed failure distributions due to localized hydrogen bond defects and shive inclusions.
Machine-direction to cross-direction strength ratios ranging from 1.8 to 2.6 further complicate real-time measurement. Applying standard three-sigma boundaries to high-speed optical measurements sampled at 1,000 Hz generates thousands of false alarms per shift. Line operators routinely desensitize alarm thresholds or disconnect automated web-break interlocks to maintain production volume, leaving converting equipment exposed to catastrophic wrap-around failures.
A 120 gsm unbleached kraft roll operating at 85% relative humidity exhibits a 22% reduction in machine-direction tensile energy absorption compared to standard ISO 187 conditioning at 50% relative humidity.
Material non-uniformity manifests across specific physical degradation paths during converting operations. Tensile forces vary across reels. Understanding these defect origins allows process engineers to categorize signals before attempting statistical filtering.
- Cross-direction grammage streaks yield localized zones of low elastic modulus, causing micro-skewing across nip rollers during high-speed acceleration ramps.
- Pinhole web fractures initiate at localized shive inclusions, propagating micro-cracks under peak dynamic pull before optical edge sensors register displacement.
- Moisture transient spikes soften the hydrogen bonding network within recycled fibers, dropping structural yield stress below line setting limits.
- Edge slitting micro-burrs generate stress concentration points along outer reel boundaries, precipitating full web separation under minimal flexural torque.
| Substrate Grade | ISO Test Standard | Mean Tensile Value | Weibull Shape Beta | Primary Failure Mode |
|---|---|---|---|---|
| Unbleached Kraft (120 gsm) | ISO 1924-2 | 8.4 kN/m | 4.12 | Fiber matrix tension shear |
| Recycled Fluting (90 gsm) | ISO 1924-2 | 4.2 kN/m | 2.18 | Shive-induced brittle crack initiation |
| SC-A Release Paper (58 gsm) | ISO 1924-3 | 3.1 kN/m | 5.64 | Calender nip over-compression fracture |
| BOPP Laminated Paperboard (220 gsm) | ISO 2758 | 580 kPa | 3.85 | Inter-ply lamination shear separation |
Disabling automated web-break sensors to bypass false alarms leads to catastrophic wrap-around events on gravure print cylinders, incurring direct tooling replacement costs of 45,000 euros per incident.

Noise
Continuous strain readings on converting equipment combine deterministic machine frequencies with genuine material variations. Unwind stands with out-of-round reels generate sinusoidal force oscillations at 15 to 30 Hz. Mechanical vibration hides material flaws. Backing roll eccentricity, drive gear pitch errors, and suction box vacuum fluctuations imprint repeating signatures onto raw load cell streams.
A raw sensor reading equals the additive sum of the underlying trend, the periodic mechanical component, and the stochastic substrate residual.
Filtering requires isolating deterministic rotational cycles without distorting true substrate weakness signals. Applying Fast Fourier Transform algorithms identifies dominant machine frequencies tied directly to line velocity and roller circumferences. High-speed continuous converting lines deploy adaptive notch filters alongside autoregressive integrated moving average models to strip baseline trends.
Continuous sensors collect raw data. Extracting the stationary residual series isolates raw structural variability from mechanical line dynamics.
- Record continuous load cell signals across thirty seconds of steady-state line operational speed.
- Apply a fast Fourier transform algorithm to identify dominant mechanical frequencies corresponding to unwind reel diameters.
- Construct an adaptive notch filter centered at primary roller rotation frequencies to subtract periodic mechanical oscillations.
- Pass the filtered signal through a first-order autoregressive filter to eliminate non-stationary baseline drift driven by roll depletion.
- Extract the uncorrected residual sequence for statistical tail distribution parameter fitting.
Subtracting cyclic mechanical signals prior to statistical limit calculation prevents false out-of-control flags during unwind roll reel depletion.
Converting machine manufacturers often assert that periodic sensor oscillations stem entirely from raw material caliper variance rather than mechanical eccentricity in the unwinding shaft assembly.

Tail
Residual values extracted from paper substrate time-series data skew heavily toward extreme negative structural strength values. The residual distribution remains asymmetric. Standard normal distribution curves understate tail risk, undercalculating the probability of extreme low-strength events that cause instantaneous web breaks.
Parameterizing filtered residuals via a two-parameter Weibull probability density function captures the heavy-tailed behavior of paper substrates under dynamic load.
The Weibull shape parameter beta dictates the underlying failure regime governing the residual series. A shape parameter below one indicates early infant-mortality fractures driven by severe mill defects or razor slitting blade damage. A shape parameter near one points to purely random micro-tear events across the fiber matrix.
Shape parameters between two and six describe progressive structural fatigue under continuous web draw. Maximum Likelihood Estimation algorithms calculate shape and scale parameters in real time over sliding five-second observation windows.
Consider a 40-tonne conversion lot of 90 gsm recycled fluting running at 1,000 meters per minute. The total run length equals 444,000 meters of web. Continuous laser caliper and strain sensors sampling at 100 Hz generate 26,640,000 raw data points.
Under standard Gaussian SPC assumptions with zero mean and a standard deviation of 1.2 N/mm, standard three-sigma control limits set alarm thresholds at 3.6 N/mm. This traditional configuration flags 71,928 out-of-control data points, of which 98.4% represent false positives caused by benign surface roughness peaks. Under a two-parameter Weibull residual filter with shape parameter beta at 1.42 and scale parameter eta at 0.88 N/mm, setting the upper tail probability at alpha equal to 0.0027 yields exactly 142 actionable anomalies.
Line operators intercept 12 severe micro-cracks before full web separation occurs, saving 18 hours of unscheduled web-threading downtime.
| SPC Architecture | False Alarm Rate (per 100km) | Missed Micro-Tears | Downtime Hours | Scrap Rate (%) |
|---|---|---|---|---|
| Standard Shewhart X-Bar (Raw Data) | 16,200 | 41 | 24.5 | 4.8 |
| EWMA with Gaussian Limits | 1,480 | 19 | 11.2 | 2.1 |
| Weibull Residual Filtered SPC | 32 | 2 | 1.2 | 0.3 |
| Data derived from continuous laser caliper and load cell monitoring across 444,000 web meters under ISO 1924-2 strain rates. | ||||
Standard EN 13430 recyclability testing mandates structural material homogeneity that directly impacts the Weibull shape parameter derived from converted substrate tension logs.
Whether dynamic estimation of Weibull shape parameters across multi-ply laminate structures can isolate inter-ply adhesive delamination from face-sheet fiber failure remains an open analytical question.

Limits
Statistical process charts built on Weibull transformed residuals adjust probability boundaries to match skewed real-world failure thresholds. Dynamic sampling frequency prevents statistical degradation during speed transitions. Control boundaries map directly to inverse Weibull cumulative distribution functions.
Setting the upper and lower false alarm probabilities to match the traditional three-sigma alpha level of 0.0027 creates asymmetric upper and lower control limits around the median residual value.

What Dynamic Sampling Frequency Prevents Statistical Degradation?
Machine acceleration ramps change the strain rate applied to the paper matrix. ISO 1924-3 documents that higher strain rates increase brittle fracture susceptibility while reducing elongation at break. Control limit algorithms must adjust Weibull scale parameters dynamically as line speed scales from 200 m/min threading speed up to 1,200 m/min operational velocity.
False alarms desensitize line operators. Modulating sampling frequency proportionally with web line speed maintains constant spatial resolution across the substrate web.
- Reel changeover transitions require real-time re-estimation of scale parameter values during the first three hundred meters of unwinding.
- Speed acceleration ramps demand dynamic adjustments to probability control limits to account for elevated strain rates under ISO 1924-3 testing conditions.
- Relative humidity drops below forty percent trigger immediate narrowing of lower critical defect boundaries to mitigate brittle fiber fracturing.
- Blade wear tracking uses progressive downward drift in shape parameter estimates to schedule slitting knife replacements before web fraying occurs.
Setting control thresholds on raw web sensor data without subtracting mechanical frequencies guarantees operator desensitization within the first hour of production.
Establishing control limits without decoupling mechanical roll vibration from material yield variance guarantees false alarm saturation.

Compliance
Quality assurance files backed by filtered Weibull process logs protect converters during commercial dispute resolution. Mill certificates detail nominal properties. However, nominal batch averages stated on mill test certificates rarely prove compliance when high-speed converting runs suffer repeated web breaks.
Under Article 6 of the EU Packaging and Packaging Waste Regulation 2024/2528, packaging manufacturers must demonstrate strict process quality controls and defect minimization during conversion. FSC Chain of Custody standards under FSC-STD-40-004 require verifiable batch tracing through physical conversion steps.
Verification files containing high-frequency Weibull residual logs provide defensive proof of substrate condition. When a converter rejects a 50-tonne paperboard shipment for micro-tearing, mill technical service teams frequently attribute failures to improper tension settings on the converting line. Presenting time-series data demonstrating that machine frequencies were filtered out, leaving isolated Weibull tail anomalies, isolates material defect responsibility.
Wet end conditions dictate yield. Weibull filtering isolates true anomalies. The filtered residual log proves whether web failures resulted from poor mill fiber bonding or improper converting machine adjustments.
Incorporating ISO 186 paper sampling protocols alongside real-time Weibull residual filtering requirements into commercial purchase orders transfers direct financial liability for micro-tear downtime back to the substrate mill.

