In the previous article, we introduced the Batchgram concept—an approach that applies batch process thinking to grade changes in continuous paper manufacturing. By treating each transition as a repeatable, time-bound sequence, Batchgram helps operators visualize where they are within that window and how closely they’re following successful historical paths. The result? Faster grade changes, less waste, and more consistency across shifts. But once a Batchgram is in place and the “green zones” are visible, an important question follows: can those operational windows be made even tighter?
That’s exactly what we explore in this second article. Because while visualizing process variation is a powerful first step, the real gains happen when that variation can be minimized in real time. Here’s where soft sensors come into play. Soft sensors—or virtual sensors—use real-time data to estimate hard-to-measure variables. Combined with Batchgram, they enable adaptive, tighter operational windows that bring production closer to ideal performance. In this post, we explore how soft sensors enhance Batchgram’s power and what mills can gain by moving from monitoring to true optimization.
Soft sensors, also known as virtual sensors, are models that estimate process variables that are difficult or impossible to measure directly. Unlike physical sensors, which rely on instruments like temperature or pressure transmitters, soft sensors derive values from existing real-time data using machine learning or other computational methods. This enables access to crucial process insights that would otherwise remain hidden or delayed.
In pulp and paper manufacturing, soft sensors are commonly used to monitor quality-related parameters such as bonding strength, brightness, or chemical concentrations. Many of these variables can't be measured continuously online, either because suitable instruments don't exist, the cost is too high, or lab analysis takes too long. Soft sensors fill this gap by providing reliable, real-time estimates that support both quality control and process efficiency.
These models make it possible to “see” into the process with greater resolution. Instead of relying on slow lab results or making decisions with incomplete data, operators gain continuous visibility into critical quality indicators. This real-time insight helps reduce raw material overuse, avoid off-spec production, and make more confident control decisions.
If you're interested in learning more about the fundamentals of soft sensors and how they are developed, we’ve covered the topic in detail in earlier article What are soft sensors?
Soft sensors open new possibilities for defining and optimizing operational windows more precisely. In grade changes, for example, historical data helps identify what constitutes a successful transition versus a suboptimal one. These become the basis for defining the "green zones” (periods of optimal performance) and the "red zones" where process deviations lead to inefficiencies. Traditionally, these zones are defined with expert input and fixed thresholds. However, when soft sensors are integrated, they provide real-time insight into critical variables like product quality, allowing the boundaries of these windows to be adapted dynamically.
This dynamic calibration means that green zones can safely be made narrower. Because soft sensors estimate key quality metrics in real time, it's possible to know more precisely when a grade change is complete, even if there's a lag in physical measurements. Operators can then respond more confidently and precisely, bringing the process to its target state more quickly and with less raw material waste. In essence, soft sensors reduce the guesswork by providing trusted, real-time information—tightening the window without increasing the risk.
Illustration: Advantages of Soft Sensors
Naturally, tighter targets come with the need for caution. Overreacting to real-time data or relying on unvalidated models can lead to unwanted disruptions. For example, aggressive setpoint changes might destabilize the process, or optimizing for a single quality parameter might inadvertently compromise another. That’s why it’s essential to validate soft sensors thoroughly and implement changes with a clear understanding of interdependencies between different quality metrics.
Despite these considerations, the combination of soft sensors and Batchgram offers a powerful way to move from static best practices toward more adaptive guidance. Whether the data is used to inform real-time decisions or to refine shift instructions and standard operating procedures, the impact is measurable. Operators gain better visibility into what matters most, and organizations can transition from merely monitoring to actively optimizing their grade changes.
While this exact combination of soft sensors and Batchgram is still under exploration, we’ve already seen clear benefits from each component individually. Real-time soft sensor data can help customers avoid overuse of chemicals and fiber during transitions by reducing the need to "overcompensate" for quality. Similarly, Batchgram analysis has highlighted inefficiencies and enabled better benchmarking between successful and failed transitions. Combining the two is a natural next step—and one that’s already showing promise in early discussions with our papermill partners.
If the combination of Batchgram and soft sensors has caught your attention, the good news is this: getting started is easier than you might think. It all begins with a clear understanding of what you're trying to achieve. Whether the goal is to shorten grade change times, reduce raw material usage, or increase process stability, a well-defined objective provides a strong foundation. With a clear target in mind, a technology partner can help translate ambition into action—defining the right approach and identifying the most impactful opportunities.
One of the most common misconceptions is that existing process data isn’t good enough for this kind of work. In reality, paper machines are typically well-instrumented from wet end to winder, and the available data, especially when combined with lab measurement, is often more than sufficient for training soft sensors. Batchgram, for its part, can analyze historical production data to uncover patterns in past grade changes, highlighting where performance has been strong and where there is room to improve.
Perhaps the best part is that none of this requires disrupting operations. Soft sensors and Batchgram models can be developed and tested entirely offline using historical data. This allows for a lightweight, low-risk starting point, and provides concrete evidence of value before any changes are implemented. Once the results are validated, mills can decide how to move forward; whether by updating operating instructions, refining grade change recipes, or enabling real-time decision support.
When integrated into daily operations, these tools offer more than just insight, they enhance confidence and control. Real-time soft sensor estimates replace guesswork with data-backed clarity, allowing operators to make faster and more precise decisions. This not only supports better outcomes but also frees up cognitive bandwidth for higher-value activities like problem-solving, optimization, and continuous improvement.
And the value doesn’t end when the grade change is complete. Soft sensors continue to provide live feedback, helping ensure the process remains in control and within target parameters throughout the entire run. In other words, smarter changeovers are just the beginning. If you’re curious about what this approach could deliver in your own environment, let’s talk—reach out to book a short walkthrough.