Artificial Intelligence and the AI Smackdown: Case Studies in Implementation of AI in Advanced Manufacturing

Artificial intelligence (AI) can be traced to the father of computer science, Alan Turing, who first posed the question, “Can machines think?” in 1950 in his seminal paper, “Computing Machinery and Intelligence.” The continued impact of the pandemic, chronic labor shortages, and supply chain issues is changing the interest in AI, and spurring a digital transformation.

To look at the current state of AI and its potential, Jim Beretta of Customer Attraction recently hosted a webinar for the Association for Advancing Automation (A3), tapping into the expertise of three leaders in the field of AI.

“There is a prevailing misconception that AI is still a ‘new’ or emerging technology. One of our first tasks when meeting with a potential customer is to demonstrate how mature AI is. Many customers are wary of AI, and are not sure it can be trusted,” said Hugues Foltz, co-owner and executive vice president of Vooban.

North America is lagging behind the rest of the world in the adoption of AI, but chronic labor shortages that impede productivity have forced many companies into a position where they need to innovate.

“The problem then becomes where to start. Manufacturers like to invest in things they can see, like a robot or a machine,” Foltz said. “You cannot see AI — but without AI and the Internet of Things (IoT), the robot or machine will not work as it should. It requires a shift in thinking that software is neither cheap nor free, and worth the investment in something that cannot be seen.”

The working environment for food manufacturing and production can involve hours spent working in in extreme heat, temperatures of 34℉–38℉, and employee turnover is significant, leading to chronic labor shortages. When food production is automated, the status quo for ingredient delivery is some form of depositor, such as an auger, gravity dispenser, or pneumatic head that works best for high volume runs of the same ingredient as it moves along a conveyor. Humans can react to changes in orientation or missing containers, but mechanical dispensers cannot, leading to spills, wasted materials, or missed ingredients.

“Chef Robotics uses collaborative robots with AI and 3-D computer vision cameras to react to the physical space. For example, if the tray has shifted on the conveyor, AI will adjust the delivery, and if there is no tray or container, no ingredients are delivered. The advantage of AI is the central ’brain’ algorithm leverages the data from the operation to continuously improve over time,” said Rajat Bhageria, founder and CEO of Chef Robotics.  

Sina Afrooze, founder and CEO of Apera.ai added that an automation digital transformation is more than “just buying a robot and adding AI. AI is the enabler, but it’s more than just technology. It may require an evaluation of the entire manufacturing process to determine what can and cannot be solved with AI.”  

“Manufacturing is hardware-centric and AI disrupts that thinking. In food manufacturing, for example, adding a new recipe or ingredient to an existing line usually requires new tooling, a $20K investment, and downtime on the line while the new tooling is installed,” said Bhageria.

 In contrast, an AI-based collaborative robot (cobot) solution could add a new ingredient with a couple of days of training and a software upgrade, with no disruption to the manufacturing process.

 “We offer a different business model of robots as a service, via yearly subscription rather than outright purchase. We offer end-to-end hardware and software service, including weekly software updates. With a culture of continuous improvement, we increase throughput and reduce waste and spillage. Sometimes it’s with better software, sometimes better hardware, and sometimes with different cameras or sensors.”

Apera’s AI-enabled computer vision solutions can be retrofitted to existing robotic solutions, extending the life of existing equipment. Traditional vision systems can experience microstops that can result in significant losses. Sensors are sensitive to changes in light, vibration, dust, or other environmental changes, whereas Apera’s Ai-guided solution can work in any environment or lighting.

 “We installed a retrofit for an automotive OEM over a weekend for a laser-etching machine that had a conventional 3D vision system that was having issues with the lighting in the factory at certain times of the day that resulted in mispicks, dropped parts, and poor laser etching. Our solution eliminated the problem, and resulted in more accurate laser etching,” said Afrooze.

One of the advantages and challenges of AI is the amount of data it consumes and generates. The algorithms that power AI “learn” from data provided to it, but first the data must be obtained.

 “Local, state, and federal governments and departments, and governments of all levels across the globe provide access to free data on a wide variety of topics. For example, demand forecasting and sales projections are one of the first tasks we look at automating using AI. It’s an old concept that can be done in a new way, and we can access data to train the algorithm along the entire supply chain,” said Foltz.

While humans are very good at managing four to five variables when problem-solving, this ability diminishes with each additional variable. One of the big advantages of AI is the ability to process millions of bits of data quickly and accurately, and find solutions. Foltz described a solution that Vooban created that was able to schedule different-sized construction cranes in different geographic locations with employees who had the proper accreditations to ensure that all jobs were covered and all employees were working every day. Instead of spreadsheets and paper, the process was handled by AI and a web/mobile app that increased accuracy and profits by ensuring the right machinery in the right location was dispatched with the right operator.

Apera takes a different approach to data management, training AI using CAD data in simulation. It allows them to create a perfect digital twin for each customer’s requirements, and determine all parameters before any physical testing is completed onsite. “The challenge with real data is the minute variations that can occur naturally. It’s easier to train AI with SKUs and then train it to recognize the anomalies,” said Afrooze.                                                                              

Advances in generative AI (Gen AI), which is deep-learning AI that can generate text, images, and other content using the data that trained the algorithm, have the potential to “automate automation.” The engineering involved in custom automation of any form is expensive and time consuming. Gen AI has the potential to disrupt the process by generating everything from the PLC codes to specifications and risk assessments in a fraction of the time.

“AI can be a significant enabler to move from a problem-specific solution to an automation-assisted solution. In current medical diagnostics, the results are dependent on the experiences of the person conducting the diagnosis, and factors such as previous experience, time of day, level of fatigue and other human limitations that AI does not experience. Imagine the possibilities if gen AI can access all known diagnostic information. That being said, we will need to ensure there are adequate standards, checks and balances in place,” said Afrooze.

Bhageria added in the short-medium term he sees AI-enabled flexible automation cobots can fill positions that may not suit traditional robots, or are chronically under-staffed, allowing employees access to higher value positions, such as robot operation. “I foresee new industries will be created that don’t currently exist, in a similar way that social media access created new opportunities.”

“We are in the midst of a global shift from data-driven to AI-driven decisions. There is often ‘it can do that?’ when working with new customers, but as AI provides the proof of concept and shows results and solutions, the C-suite is more comfortable with making AI-based decisions across the business and supply chain. While some companies have been forced into a position to innovate, no one wants to be last to adopt solutions,” said Foltz.

In a whimsical nod to the summer of football (soccer) events, the webinar ended with the following question: If the full implementation and adoption of AI in manufacturing is measured like a 90-minute football (soccer) game, where are we in that game?

Foltz: North America has joined the game late, but we are well past the half. In North America, when we attend trade events, we might be the only vendors, while in Europe and Asia, AI is everywhere. It’s 3-0, and it’s halftime.

Bhageria: It feels like we are 5-10 minutes in. AI is not a new idea; Turing talked about it. There’s a funny idea in computer science that as soon as something starts to work, it’s not called AI anymore. OCR (occular character recognition) used to be considered AI, and now it’s not. Statistical method used to be considered AI. Now it’s not AI. So for that, we’re well into the match. However, for things like artificial general reasoning (AGR), we’re still early in a macro sense. The existing models are predictive; they aren’t really “thinking.” It’s an exciting time for AI with no clear path for the potential of AGR.

With the adoption of AI increasing in advanced manufacturing, innovation and digital transformation is likely to continue. With Gen AI and AGR still in developmental stages, the next few years in AI and advanced automation could be exciting and unpredictable. Hey Siri, make popcorn!