In food and beverage, several products are made in batches — especially any product made through a fermentation or a drying process where parameters are difficult to control due to changing environmental conditions or fluctuations in the process itself.

The problem was that in the early days, batch samples were often taken to the lab after the batch was finished — too late to rescue an errant batch. Even when a grab sample was taken to the lab before the batch was finished, it was often too late to turn a batch around once it drifted outside of specified parameters.

Recognizing this issue, the FDA supported PAT (process analytical technology), which was designed to use real-time monitoring of process variables and automated data analysis in order let operators know if a process was going out of control so they could make the necessary changes to correct the process — either manually or automatically. This was all based on the “perfect batch.”

Today, applying artificial intelligence with machine learning (ML) and combined with online, real-time monitoring and control of key process variables can not only fix problems in process, but also predict where the batch is headed at any given moment and prevent future unwanted, divergent changes. AI also provides smarter recipe execution and adaptive setpoint control, precise ingredient management and loss reduction, predictive equipment performance (and maintenance), CIP and energy optimization, improved traceability and batch genealogy, food safety and microbial risk prediction and more. Maybe summed up, AI improves batch control primarily by being predictive, adaptive, consistent and efficient — able to turn on a dime when necessary.

Some batch control suppliers have integrated AI into batch control, and batch processing keeps getting better and better — certainly worth the investment if you’ve spent months or years trying to get consistent batches as the seasons, plant environments and ingredients change.

Already, AI tools can address the challenges of batch processing in the food and beverage industry by combining real-time data capture, intelligent document processing and advanced analytics, says Jon Knisley, head of process AI, ABBYY. “This enables manufacturers to achieve greater control, efficiency and predictability in their batch operations.”

Software tools can integrate seamlessly with existing process control systems to analyze real-time data from sensors and equipment. This allows manufacturers to predict batch outcomes and intervene proactively to prevent deviations from the “perfect batch.” For example, during fermentation or drying processes, AI can identify subtle changes in environmental conditions or ingredient quality that might lead to inconsistencies, Knisley says.

Fermentation: A Difficult-to-Control Operation without AI Supervision

AI is revolutionizing the food and beverage industry, turning batch processing into a precise, predictive and adaptive science. By leveraging real-time data and advanced analytics, AI ensures consistent quality, reduces waste, and sets a new standard for efficiency and innovation. Continuous monitoring powered by AI is the secret ingredient to smarter recipes, safer food and sustainable operations.

This is true especially of applying AI to difficult-to-control batch applications such as fermentation. Batch control has evolved from reactive to proactive, thanks to AI’s ability to see what’s coming and adjust in real time.

A good example is ABBYY’s work with a leading beverage manufacturer, which implemented ABBYY technology to improve the consistency of its fermentation process. By analyzing real-time data from sensors and automating the processing of COAs, the company reduced batch failures by 25% and achieved a 15% increase in production efficiency. The tool also provided actionable insights for optimizing energy usage, contributing to the company’s sustainability goals.

AI solutions enhance traceability by creating a digital thread that links raw materials, process parameters and finished products. This not only supports compliance with food safety regulations but also provides valuable insights for continuous improvement and root cause analysis in the event of quality issues.

— Jon Knisley, Head of Process AI at ABBYY

Batch, Machine Learning and AI

While AI has been successfully applied in enterprise applications — e.g., inventory control, logistics, product and production planning, maintenance, etc. — AI and ML are welcome technologies at the ISA-88/IEC 61512 batch control layers. In recent years, automation suppliers have added new sensor inputs to batch control systems to monitor critical process variables and observe how they affect one another throughout the process. This heralded the beginning of machine learning and was usually applied at the PLC/SCADA/industrial layer. Today AI/ML is an integral part of PLC and industrial PC control.

For example, Siemens provides AI-driven capabilities through its SIMATIC PCS 7 and PCS Neo batch control systems, enhanced by Siemens Industrial Edge, and SIPAT for advanced analytics, says Ed Montgomery, industry manager - RC-US DI S VSP CPG VS. These solutions combine real-time process data with machine learning models to improve batch consistency and efficiency, and are commonly used in:

• Fermentation processes in breweries and dairy plants
• Drying operations for powders and snacks
• High-value ingredient blending in food and beverage
• Batch processes with variable environmental conditions

Additionally, Siemens offers two SCADA-type ISA-88/95 process control and batch solutions: Siemens BRAUMAT — which is short for “BRewing in AUtoMATic” and is marketed to the brewing industry — and SiStar, which is short for “Siemens Standard Tools for Automation of Recipes controlled processes” and is geared towards the entire process control/batch industries. Siemens is invested in leveraging artificial intelligence and machine learning (AI/ML) to enhance industrial processes, moving towards the “Autonomous Plant” or “Digital Enterprise.” When it comes to batch control, the integration of AI isn’t typically a standalone AI-assisted batch control product as much as it is a suite of intelligent functionalities and software tools that augment the BRAUMAT/SiStar batch control system.

Enhancing batch control with AI/ML begins with models that analyze historical batch data (e.g., raw material properties, process parameters like temperature, pressure, mixing speed, reaction times) and correlate them with final product quality attributes, Montgomery adds. They can learn complex relationships that human operators might miss. The data for these models is acquired from the open interfaces for BRAUMAT/SiStar where every reading from the control system is available for analysis.

Example capabilities include:

• Predicting the quality of an ongoing batch before it’s finished, allowing for early intervention if deviations are detected
• Suggesting optimal setpoints for process parameters to achieve desired quality targets with fewer off-spec batches. This is particularly valuable in industries like brewing for consistent taste and quality, or pharmaceuticals for purity and yield.
• Identifying process conditions that produce higher yields, reducing waste and increasing profitability

Not only can AI monitor the batch process itself, but it can also monitor equipment reliability and maintenance. In these systems, AI algorithms continuously monitor sensor data from equipment (pumps, valves, agitators, heat exchangers) within the batch process. They establish “normal” operating patterns and flag deviations that could indicate impending equipment failure or process issues. This helps plants move from time-based to condition-based maintenance, reducing upkeep costs and extending asset life, Montgomery says.

While traditional batch control follows predefined recipes, AI can enable more dynamic and adaptive control strategies. AI models can learn how the process responds to changes and adjust control actions in real-time to maintain optimal conditions, even when faced with disturbances or variations in raw materials.

Go Takami, manager, Consulting Division, AI Consulting Department, Yokogawa Digital Corporation, describes a project for a craft brewing process. “We utilized our Factorial Kernel Dynamic Policy Programming (FKDPP) technology to develop a temperature setting plan for a production process. Jointly developed by Yokogawa and the Nara Institute of Science and Technology (NAIST), FKDPP is an autonomous control AI protocol that makes use of reinforcement learning technology. In a proof-of-concept experiment, we applied this approach to a craft beer fermentation process and successfully reduced the fermentation time by 28%.”

Takami adds, “When utilizing our FKDPP technology, we proceed through the following steps:

1. Build simulator
2. Use FKDPP to examine conditions
3. Validity checks by human experts
4. Application with actual equipment

“We believe this technology can be applied to planning temperature profiles for batch processes,” Takami adds, “and we expect it could be extended to the production of fermented foods such as sake and yogurt, as well as pharmaceuticals like antibiotics that use microorganisms, enzymes or yeast.”

What Processors Expect from AI-Assisted Batch

Food and beverage companies expect AI-assisted batch control to deliver consistent product quality, optimize key performance indicators (KPIs) like efficiency and waste reduction, enable real-time decision-making, ensure compliance with regulations and support sustainability goals, says Robert Purvy, DCS technical consultant manager, Siemens. They also seek scalable and flexible systems that adapt to changing demands and provide actionable data insights for continuous improvement.

“They also expect improved traceability, compliance with FDA PAT and tools that fit into existing automation systems and user workflows without major disruption to maximize productivity and employee retention,” Purvy adds.

“We believe there are the following expectations, says Yokogawa’s Takami:

• Improving production efficiency while maintaining quality
• Standardization and reproducibility to reduce variability in human operations
• Co-creation of new approaches by humans and artificial intelligence (AI)

“To meet these expectations, we aim to use AI to develop temperature plans and provide temperature plan optimization as one of our solutions,” Takami adds.

From Traditional Batch to AI/ML-Based Batch Control

Traditional batch control focused on recipe execution and maintaining setpoints, says Carlos Felipe, Siemens Digital Industries vertical architect, CPG. It worked well for stable processes, but it could not accurately predict deviations or adapt to changing conditions. AI-based batch control adds predictive and adaptive capabilities using these types of sensors:

• Inline quality sensors (NIR, pH, turbidity) for real-time quality monitoring
• Advanced flow and level sensors for precise ingredient dosing
• Condition monitoring sensors for equipment health (vibration, temperature)
• Environmental sensors for humidity and ambient conditions

Modern batch control software typically provides compliance and real-time quality analytics features. Meanwhile, processors are frequently installing edge controllers for local AI/ML processing and cloud applications for advanced analytics, predictive maintenance statistics and operational optimization insights.

In the brewing example described by Yokogawa’s Takami, by manually implementing the temperature setting schedule created by AI, brewers were able to shorten the fermentation process time from 336 hours to 240 hours, (28% reduction) while maintaining the aroma, taste and mouthfeel of the product produced by experienced brewers.

Additional sensors were installed on the existing system to measure specific gravity and temperature data, which were then used as input for developing the temperature plan, Takami says.

Remember: AI/ML is Only as Good as its Data

AI models thrive on data, but if that data is poor quality, incomplete, inconsistent or insufficient, the AI’s insights will be flawed, leading to incorrect recommendations or predictions, says Siemens’ Felipe. To avoid these sorts of issues, processors must establish clear standards for data collection, storage and maintenance from the outset, especially with operational personnel who are manually entering data into any destination.

Implementing automated checks and manual reviews to ensure data accuracy and consistency is desired so that the AI models can adapt to the best set of data. When implementing a new batch optimization strategy, it is best to phase the installation, avoiding a “big bang” approach. Start by integrating a small set of data points and a simple AI application, then gradually expand. By thoughtfully addressing these potential risks, companies can unlock the immense potential of AI-equipped batch control systems, providing more consistent quality, process optimization and a more efficient workforce.

While poor quality data is one issue, Yokogawa’s Takami points out that insufficient or biased data when creating the simulator for AI to develop temperature plans can lead to overfitting and difficulty in assessing the validity of the plans proposed by AI.

Methods to minimize or avoid these risks include collecting a diverse range of data during data aggregation and not accepting AI-generated results blindly, but instead having a human verify their validity, adds Takami.

Planning Expenditures

AI/ML-based systems may certainly appear to be worth the investment, but some preliminaries are needed before taking action. Processors should begin by auditing their current automation system state, says Siemens’ Montgomery. To derive value from AI-based control, data transparency must exist among all levels from the shop floor to enterprise management software. Once this data is in place, AI-assisted batch control systems become remarkably scalable, capable of handling small and hyper-complex batch processes alike.

On the hardware side, the main additional investment will typically be for the collection of data by sensors, but Takami believes that large-scale investments can be avoided by leveraging existing systems.

“Although we cannot provide specific ROI figures,” Takami says, “we believe that the shortening of production time will enable flexible capacity increases during peak periods, and that the provision of insights to plant personnel will help reduce research and development costs, contributing to increased ROI.”