by Bas Versluis, Commercial Director, KSE, Netherlands
For more than fifty years, batching in the animal nutrition industry has relied on a combination of mechanical precision and operator expertise. That model is under pressure. Ingredient lists keep growing, inclusions become smaller and more variable, and quality and traceability expectations are rising. At the same time, experienced operators are increasingly hard to find.
Artificial Intelligence (AI) offers a way to address these challenges directly in the process. AI can be used internally for tasks such as data analysis, support tools and software development, but the real impact for customers comes from AI-driven control algorithms embedded in batching equipment and factory control systems. These algorithms directly steer dosing and weighing performance in real time.
This article explains how AI is applied within the KSE’s machines and software, the transition from traditional self-learning control to fully self-steering AI and what this means for performance and operations in compound feed, aquafeed, petfood and premix plants.
Rule-based control to self-learning
Conventional batching control is based on fixed parameter settings: target speeds, tolerances, cut-off points and correction factors. These parameters are configured by process engineers and then adjusted over time by experienced operators. This approach works, but assumes relatively stable conditions and a high level of process know-how. In practice, feed plants face:
- Variations in ingredient behaviour (bulk density, flowability, moisture)
- Changing environmental conditions (temperature, humidity, dust)
- Mechanical changes over time (wear, friction, alignment)
- Frequent recipe changes and short batch runs
To deal with this variability, the company introduced self-learning algorithms in its machines years ago. These algorithms adjusted parameters based on historical performance data. For example, if a scale consistently overshot the target weight, the control system would gradually adjust cut-off timing or fine-dosing behaviour to reduce that overshoot.
This reduced manual tuning and improved performance under changing conditions. However, these systems still required expertise to configure and maintain. Operators or process engineers had to understand learning rates, boundaries and reset conditions. The system supported the operator, but did not replace continuous human optimisation.
Fully self-steering AI control
Since 2025, the company has introduced AI-based algorithms that are fully self-steering. Instead of asking operators how the system should learn, the AI determines this itself, within the boundaries defined by the plant. Key characteristics of this AI-based control are:
- No manual tuning parameters: The AI itself is not configurable by operators. There are no learning parameters to adjust or maintain.
- Autonomous adaptation: The system continuously observes performance—dosing times, achieved weights, overshoot, correction cycles and adjusts control behaviour without human intervention.
- Applicable to new and existing equipment: The AI layer can be integrated in new installations and retrofitted on existing dosing and weighing lines, provided the necessary signals are available.
The AI focuses on two core KPIs:
- Dosing time: Typical improvements range from 30–50 percent, with a minimum of around 20 percent, depending on process and ingredients.
- Accuracy: Accuracy improvements of about 25 percent are achieved through continuous optimisation of coarse and fine dosing, cut-off moments and stabilisation times. These gains are realised without long start-up periods. The system starts optimising from the moment it is activated and keeps adapting as conditions change.
High-level technical view
Without disclosing implementation details, the AI can be described as an adaptive, performance-focused control system with machine-learning elements. It uses real-time weight signals to track approach speed, overshoot and stability, captures ingredient flow behaviour during coarse and fine dosing and learns from historical batch data to refine control decisions.
The model adapts continuously, with each dosing cycle contributing new data and improvements applied immediately, avoiding any separate learning phase. It responds to both gradual changes, such as equipment wear or material drift, and sudden changes, such as new raw material lots, while operating strictly within the process limits defined by the plant’s higher-level control system.
Integration in the control architecture
The value of AI in batching depends heavily on how it is integrated into the plant’s control structure. In typical installations, it is integrated in the following ways:
- AI on a separate server next to the PLC: The AI runs on a dedicated server that operates alongside the machine PLC. This server is isolated from the outside world, ensuring the AI logic is shielded from external interference and unauthorised access.
- Optional integration in SCADA: The same functionality can be hosted within a SCADA environment, for closer interaction with supervisory control and existing visualisation.
- Process parameters from MES/SCADA: The AI does not decide targets or tolerances. Batch sizes, setpoints, permitted tolerances, cycle constraints and other process limits are supplied by MES/SCADA. The algorithm’s sole task is to achieve the best possible speed and accuracy within those limits.
- Separation from safety logic: Safety functions such as emergency stops, interlocks and safety PLC logic remain entirely separate. The AI does not replace or modify safety layers; it operates inside the envelope defined by the plant’s safety and quality design.
In other words, the plant continues to define ‘what’ must be achieved and ‘how far’ the process may go. The AI only optimises ‘how’ the dosing and weighing is executed within those boundaries.
Boundary conditions and safeguards
Because the AI operates as a performance-optimising layer, its behaviour is strictly constrained by limits defined higher in the control architecture. Targets, tolerances and cycle constraints are supplied by the MES/SCADA, and the AI optimises strictly within these boundaries.
If sensor data become unreliable, for example due to signal loss or inconsistent readings, the system is designed to remain stable by reverting to baseline PLC behaviour, while operators are alerted through existing alarm mechanisms. All AI-driven adjustments are applied within the standard control environment, ensuring that process values, setpoints and events are logged as usual, preserving full transparency and traceability without introducing separate or opaque logging systems.
Premix plant in Spain
A premix factory in Spain provides a practical illustration of how this approach works in a real plant. The facility handles a wide range of ingredients, including numerous micro-components, within strict batch-time constraints. AI-based control was implemented on five machines, delivering clear results without any mechanical modifications or hardware changes.
Average dosing times were reduced by approximately 35 percent, achieved entirely through AI-driven optimisation of dosing behaviour. At the same time, weighing accuracy improved by around 25 percent, leading to more consistent batches and reducing the risk of rework or formulation deviations.
With shorter dosing times and higher accuracy, the plant is now able to automatically dose more ingredients within the same batch window, reducing the need for manual additions, particularly for small inclusions and increasing the overall level of automation. This use case demonstrates how AI can increase both capacity and quality on existing assets, without major mechanical intervention.
Impact on operators & organisation
AI-driven control changes the role of operators without eliminating it. Previously, experienced operators or process engineers spent much of their time continuously tuning parameters, reacting to changes in ingredient behaviour and adjusting settings between shifts. With AI in place, this fine-tuning is largely automated, shifting the operator’s role from tuner to supervisor.
Operators focus instead on overseeing performance, monitoring key performance indicators and managing exceptions. Because the AI algorithm itself is not operator-configurable, user interfaces can be simplified to concentrate on monitoring dosing times, deviations and alarms, verifying that production targets from manufacturing execution systems and supervisory control and data acquisition systems are being met, and supporting troubleshooting and maintenance activities.
In a labour market where industry-specific expertise is increasingly scarce, this approach also enables faster onboarding, as new operators can be trained to monitor processes and respond to issues rather than developing detailed control strategies themselves.
Why AI matters now
The operating environment for feed, premix, pet food and aquafeed factories is becoming increasingly complex, driven by a growing number of ingredients and product variants, smaller and more demanding inclusions with tighter tolerances, pressure to reduce hand additions and manual handling, and higher expectations for consistency and predictable output. Traditional control strategies and manual tuning struggle to keep pace with this complexity, particularly when experienced personnel are scarce.
By embedding intelligence directly into dosing and weighing control, artificial intelligence enables plants to increase throughput on existing equipment, improve accuracy and stability across shifts and operating conditions, reduce dependency on individual operator expertise and raise the overall level of automation within existing batch times.
Looking ahead
The introduction of fully self-steering AI algorithms in dosing and weighing is a major step, but it is also a foundation for future developments. As sensor technology, connectivity and data availability continue to improve, AI can extend from individual machines to plant-wide optimisation, coordinating lines, balancing buffers, and eventually integrating energy and maintenance considerations into control decisions.
For now, the most immediate value is achieved where AI has direct impact on performance: real-time control of batching operations that define the capacity, efficiency and quality of feed production. By turning machines into genuinely self-steering systems rather than merely operator-assisted ones, AI helps factories operate at a consistently high level under increasingly demanding conditions.
