Implementing artificial intelligence (AI) often involves a leap of faith, especially at this juncture. The ever-cautious baking industry is poised to take its first steps, according to a pulse survey of 80 commercial bakeries and industry suppliers conducted by the American Bakers Association (ABA) and NDP Analytics in March.

The key takeaway from the research indicated that nearly 70% of survey respondents have adopted or plan to adopt AI in some capacity, including pilot programs, during the next 12 months.

Many bakers, however, still have doubts that their companies are ready for it, while others haven’t quite figured out how it can be adapted to their operations to drive efficiency.

That said, perhaps one of the greatest adoptions of AI in the manufacturing sector can be found in robotics, where the machine learning and improved hardware are making these pick-and-place and automated processing and packaging systems more versatile and easier to run.

“Machine learning and AI have profoundly impacted the performance of robotics in the baking and snack industry,” said Saman Farid, chief executive officer of Formic. “These technologies enable robots to learn from their operations, improving efficiency and precision over time.”

At the same time, a greater number of bakeries are exploring adding more affordable robots and cobots to alleviate myriad workplace issues, such as repetitive product handling, manual scoring or labor-intensive packaging.

“The baking industry is experiencing a democratization of robotics, with a growing number of bakers and manufacturers investing in robotic solutions,” observed Felix Pang, robotic solutions specialist, ABI Ltd. “Robots offer consistent performance and easy monitoring, capable of operating for extended periods, as needed. With a persistent shortage of skilled labor and the associated high costs, bakers are increasingly drawn toward adopting robotics to address these challenges.”

AI algorithms can analyze data from robotic operations to optimize movement paths and automatically determine gripping positions from camera information in much the same way a person would.

“This leads to enhanced performance in handling delicate products and precise packaging tasks, minimizing the need for large, clunky and custom hardware, and improving throughput,” Farid explained. “It also makes systems more adaptive to variances in the production process that are a part of every manufacturer’s daily life. Many robotics systems now can include self-learning capabilities, allowing them to adapt to new tasks with minimal human intervention, thus streamlining the setup process for new products or operations.”

Many bakeries are examining ways to lower costs by eliminating lifting, repetitive tasks and other non-value-added work while alleviating operators’ fatigue that can diminish product quality.

“Leveraging machine learning and AI technology, robotic systems can now undertake these tasks with precision,” Pang said. “Training a machine learning model involves exposing it to an extensive array of images depicting both compliant and defective products, enabling it to autonomously process data. The integration of AI-powered systems into bakeries not only provides real-time insights into production performance but also empowers swift action at the earliest stages of operation.”

Schubert is combining AI and its vision systems to make its high-speed robots and cobots more flexible. Giorgio Calorio, sales account manager at Schubert North America, said the tog.519 robotic model, which provides up to 90 cycles a minute, can be moved using a simple pump truck. The system can self-adapt to new products and different shapes of packaging using AI-deep learning.

Overall, the emergence of machine learning and AI now offers greater accuracy and simplicity with robotic product handling and packaging operations, said Dan Alper, product marketing manager, AMF Workhorse, a brand of AMF Bakery Systems.

The company’s Automated Motion Optimizer (AMO) software eliminates the need for a teach pendant — or control box for programming the motions of a robot — for simple operation through recipe and pattern management. A PLC controls all operational and pattern-forming functions. 

“Machine learning algorithms allow robots to adapt to different shapes, sizes and types of food items,” he said. “When the robots are able to efficiently adjust to more delicate products, there is typically less damage to the product. AMF’s robotic product loading solutions integrate soft-touch, end-of-arm tools with auto-adjustment features for product collation, pass-through hand packaging or automated mix-pick/rejection.”

Alper added AI enables robots to quickly switch between tasks and products, adapting to changes in production schedules without extensive reconfiguration. Downstream on the production line, AMF’s robotic product and pan handling can be customized to handle a number of configurations for maximum flexibility.

To improve product handling, machine learning algorithms enable robots to optimize their movements and tasks based on real-time data, said Jamie Bobyk, marketing manager, Apex Motion Control. AI is also being used to enhance predictive maintenance.

“These algorithms can analyze product size, shape and weight to ensure precise handling and packaging,” he explained. “The result is a decrease in errors which in turn reduces product waste and improves overall quality control.”

Some tasks, he added, are simpler to implement than others. In fact, once a robot or cobot learns a certain function, it’s easier to build upon an existing program to install other automated systems to handle additional workloads.

“When talking about strictly repeatable tasks, such as stacking boxes on pallets or moving trays from conveyors to racks or racks to conveyors, much of the standard programming that goes into machine learning can be used in many of the standard applications,” Bobyk said. “The more that robotics are implemented into production facilities, everyday repeatable tasks will become part of the programming norm.”

Machine learning, along with similar tools, now offers shorter programming times for cobots and robots. Used properly, they can simplify what were once complex, often expensive and even frustrating processes a few years ago.

“The Schubert Cobot tog.519 is able to self-learn product characteristics, packaging dimensions and shapes, and thanks to deep learning and high-speed processing data in combination with vision systems, robots are even able to pick overlapped products,” Calorio said. 

Joe Crompton, vice president of robot technology, BluePrint Automation, noted AI and deep learning allow for better inspection and grading of products.

“Now systems are capable of learning the subtle differences between an acceptable and unacceptable number of defects,” he said. “This means less false rejects as well as improving the overall quality of the product being produced.”

Alper pointed out AMF UX, the company’s global standard for AMF’s HMI systems, provides operator-friendly, visual recipe-management options.

“Operators can quickly select pre-programmed pattern configurations for efficient product changeover without tool change, ensuring precise package or pan placement and maximum reliability,” he said.

He added that advanced sensing and vision systems enable robots to detect variations in product size, shape and orientation, and then easily adjust their actions accordingly.

“Vision systems, particularly those powered by AI, allow robots to perform quality checks and recognize different items, enhancing their flexibility across various production lines,” he said. “AMF’s AMO software uses preconfigured patterns and recipe selection to simplify the operator’s process for changeovers.”

This article is an excerpt from the June 2024 issue of Baking & Snack. To read the entire feature on Robotics Packaging, click here.