How Artificial Intelligence and Sensor-Based Measurements Are Revolutionising Quality Control and Production Efficiency in the Paper Industry

by Luca Orizio

The paper industry—although one of the oldest branches of manufacturing—is undergoing a period of profound technological innovation. The convergence of process digitalisation, advanced sensing, and artificial intelligence (AI) is radically transforming the way production is managed, with a direct impact on both product quality and operational efficiency.

The Critical Role of Grammage in Papermaking

One of the most delicate and decisive aspects of papermaking is the control of grammage—the basis weight of paper expressed in grams per square metre (g / m²). This parameter has a direct effect on the functionality of the finished product, whether printing paper, packaging, or tissue. Traditionally, grammage verification has relied on manual sampling followed by laboratory analysis. Today, thanks to the integration of sensors and machine-learning algorithms, this value can be predicted and controlled in real time with a level of accuracy and speed unimaginable just a few years ago.

In a sector that runs at break-neck speed and faces ever-thinner margins, grammage stability is synonymous with efficiency and competitiveness. Any deviation from the optimal value can trigger a cascade of problems—from non-compliance with customer specifications to material and energy waste—with inevitable environmental repercussions.

Take a packaging line, for example. Excessive grammage leads to over-consumption of pulp and chemical additives, as well as heavier transport weights. Conversely, insufficient grammage may not provide the required mechanical strength, causing the customer to reject the batch. Even tiny variations can affect cost, logistics, and sustainability.

For these reasons, the paper supply chain has devoted the past decades to enhancing quality control, aiming at standardisation that minimises variability and maximises performance.

From Manual Sampling to Smart Sensing

The traditional method for determining grammage involved collecting paper samples at set intervals and analysing them with precision scales and laboratory instruments to calculate basis weight from sheet mass and area. Accurate, yes—yet disconnected from the real-time production flow.

High-resolution industrial sensors now make it possible to monitor variables such as paper thickness, residual moisture, surface density, and even beta- or gamma-ray absorption, all indirectly correlated with grammage. Mounted along the production line, these devices gather continuous data, providing a dynamic picture of the entire process.

But gathering data is not enough. To make sense of such information streams, something more is required: algorithms capable of learning, adapting, and predicting. This is precisely where machine learning (ML) comes into play.

Machine Learning at the Service of Paper Mills

Machine learning—a subset of AI—rests on the idea that a system can learn from observed data, recognise hidden patterns, and generate forecasts. In papermaking, that means building a model able to estimate grammage directly from sensor readings, without waiting for lab results.

The workflow unfolds over several stages:

Data acquisition – Capture large volumes of data: thickness, temperature, humidity, stock pressure, web speed, chemical composition, and more. Align these with actual grammage values obtained from reference measurements.

Pre-processing – Clean and prepare the data: remove anomalies, harmonise units, and synchronise datasets.

Feature engineering – Identify the most influential variables.

Model building – Choose and train a predictive model (linear regression, neural network, random forest, etc.), then validate it.

Deployment – Implement the model in the production chain to deliver real-time estimates and instantly adjust processing parameters.

Algorithms in Action: Selecting the Right Model

Choosing the appropriate algorithm is never trivial. It depends on the process type, data volume, operational variability, and in-house expertise.

Algorithm Best suited for Notes

- Multiple linear regression Simple contexts with few, well-defined relationships Easy to interpret

- Decision trees / Random Forests Scenarios with many interacting variables, even non-linear Robust and versatile

- Artificial neural networks Complex pattern recognition Require large datasets and significant computing power

- Support Vector Machines (SVM) Noisy environments or highly correlated variables Effective at handling outliers

Whatever the choice, continuous monitoring, updating, and retuning are essential—especially when materials, equipment, or product requirements change.

Why Machine Learning Is a Game-Changer

Introducing ML into a paper mill is more than a technical upgrade; it is a paradigm shift. Key benefits include:

- Continuous monitoring – Grammage is estimated instant by instant, shortening reaction times.

- Scrap reduction – Corrections are applied immediately, before the product becomes unusable.

- Material & energy savings – Precise dosing of raw materials directly boosts sustainability.

- Traceability & certified quality – Every decision is documented, every datum preserved, simplifying audits and regulatory compliance.

- Customer loyalty – Consistent product quality enhances brand reputation and cuts complaint rates.

Challenges Beyond Technology

Naturally, adopting ML is not without hurdles. The main difficulties concern:

- Data reliability – Poorly calibrated sensors or corrupted data can undermine the entire process.

- Specialist skills – Professionals must combine papermaking expertise with data-science abilities.

- IT/OT integration – Communication between corporate IT systems and shop-floor machinery must be stable, secure, and flexible.

- Model maintenance – Algorithms need periodic retraining to reflect evolving conditions.

- Corporate culture – Personnel must be engaged in the transition, overcoming scepticism toward automated decision-making.

Real-World Cases: Experiences and Outcomes

Numerous mills across Europe and beyond are already applying AI to grammage control with notable success. In Scandinavia, one of the largest coated-paper producers implemented a neural-network predictor that cut grammage fluctuations by 40 % and saved several thousand euros per month in energy costs.

In Italy, packaging-focused mills use regression and ML models to tailor grammage to each packaging grade and brand requirement. Meanwhile, some industrial-solution providers now offer turnkey packages that bundle sensors, cloud dashboards, intuitive GUIs, and predictive-maintenance tools—paving the way to digitalisation even for SMEs.

Looking Ahead: The Smart Paper Factory

What today appears as an advanced application will soon become the norm. The next step is embedding ML within a full smart-factory logic: interconnected plants, real-time data processing at the edge, self-adapting algorithms, and systems that learn from their own mistakes.

The ambition is not merely to boost productivity but to transform the very way paper is conceived, produced, and traced. Grammage prediction becomes just one of many tiles in an intelligent, sustainable, highly competitive production mosaic.

Conclusion

Machine learning applied to grammage prediction is no passing trend; it is a strategic lever for transforming the paper sector. Through the synergy of sensor data and predictive models, mills achieve fine-grained process control, cutting scrap, costs, and environmental impact.

Despite its challenges, this technology heralds the dawn of a new era—one of autonomous manufacturing in which human expertise and AI power collaborate to produce more, with less. Those who seize this opportunity will be well-prepared to meet the challenges of Industry 5.0.

© All rights reserved