Increasingly, manufacturers are exploring the benefits of analytics to support data-driven manufacturing. The focus is on collecting and connecting data at each step of the manufacturing journey to sustain growth, reduce costs and achieve operational excellence—all while taking the appropriate actions in moving toward a green and sustainable future.

This article summarizes a panel discussion on how artificial intelligence (AI) and machine learning can transform manufacturing. It was held during the Power of Unified Manufacturing event, where CGI’s Marcel Mourits was joined by Alexander Daehne, Manager, EMEA Manufacturing, SAS and Nathan Eskue, Professor of Artificial Intelligence in Manufacturing, Delft University, the Netherlands.

Managing expectations

Marcel notes that there is a misconception that AI can solve every challenge—from improving the quality of production to even predicting when personnel will fall ill. While AI may not have all the answers, it offers significant business benefits. However, its success is predicated on building the right environment, states Nathan.

“An organization needs to be prepared…. And in large part, that amounts to: do you have the right data, and do you have enough control of the process to be able to make use of that data?” states Nathan.

Organizations that have oversight of their data, control of their processes, and the skill sets to use data will be able to benefit from AI, he says. As a result, AI is best applied in those areas where an organization has already collected or can collect large amounts of data. For instance, he says that using AI for object detection and classification—either using photographs or other types of data for quality control and sorting—can offer immediate benefits. Another widely-heralded benefit is in predictive behavior.

Alexander notes that large transformational AI stories within manufacturing are still rare. A more routine application of AI is anomaly detection. For instance, machine learning can help identify the root cause of equipment failure and predict it. An additional approach Marcel suggests is hybrid AI, where human decision-making is enriched by augmented reality (AR). Here, AI doesn’t make the decisions but provides humans with insightful information to make better-informed decisions. At the same time, Marcel advises analyzing the cost-benefit ratio. “AI for AI’s sake is not always the correct approach.”

Addressing the challenges of AI adoption

Data is core to the success of any AI initiative. “A model can only be as good as the training data,” states Alexander. However, even with the right data in place, manufacturers need to address several other challenges that are related to people and processes:

  • Change management: This entails ensuring organizational readiness to embrace new ways of working, access to the right skill sets and training, and inclusivity to ensure all voices are heard, which in turn, will lead to greater trust and uptake of the solution.
  • Knowledge of operating procedures: This calls for a deep understanding of what will support and benefit the worker, what information they need, and when.

Becoming people and culture-led for successful outcomes

Alexander believes that organizations that scale their AI initiatives have a “culture of working together and collaborating with the end users.” This way of working ensures end users of the solution are in the loop and involved in the design. The benefit is that change management teams don’t have to “force acceptance.” It also helps to prevent an “us vs. them” dynamic between IT and the shop floor.

Nathan agrees. He says setting up an AI project requires actually going onto the shop floor, walking through the process, and truly understanding how people work and when they make decisions. This insight can inform AI initiatives through real-world practice.

He says this process can also transform suppositions of what can and can’t be done to advance true innovation. It will help identify “real wins” and “tangible goals” to help personnel perform their jobs better and give them the information needed to succeed. Often, there are simple solutions that come down to how and when data and information are processed across the organization. Taking a collaborative process for solution design, Nathan explains, could even result in discovering that a simple excel spreadsheet is the answer.

Becoming people or culture-led starts with understanding what is needed and, equally importantly, when.

“There is a real misconception that everything needs to be real-time. I prefer talking about the “right time,” because if you overflow people with information, you will not gain anything,” says Alexander.

Eliminating the fear associated with AI

There is a real concern that people will lose jobs to AI. Alexander says that addressing and alleviating this fear can create a work culture that is more receptive to AI. One way to achieve this is by conducting workshops with core users to explore and discuss use cases from other companies. This creates an environment where people can open up and express any concerns. It will allow for a better understanding of priorities and the feasibility of different approaches based on expected outcomes and key KPIs. Once one or two small use cases have been implemented, they will generate confidence and curiosity to explore other possibilities and applications.

As Nathan shares, the fear is lessened when a company says: “Hey, let's become a culture that cares about data, that cares about using data to make the right decisions.” I think if you take that approach, and if there’s some education in terms of what AI is and isn't, what it can and can't do. I think that helps in terms of the fear of being replaced.”

Openness is key. Manufacturers must be transparent about their motivation, which includes freeing up the workforce so that people can build their skill sets and focus on more high-value work that will benefit the company.

The future of AI in manufacturing

Currently, AI solutions in manufacturing are bespoke. In 5-10 years, could AI be “baked” into integral manufacturing systems such as enterprise resource planning (ERP) and manufacturing execution systems (MES) to automatically make predictions and identify new ways of achieving efficiencies? The panelists agreed that manufacturing operations are far too complex, with too many unique set-ups and variables for a “generalized” form of AI. However, in very standard manufacturing processes—ERP, for example—it may be possible to apply some off-the-shelf AI models and fine-tune them to the organization.

Nathan cites genetic algorithms among future possibilities, which is the ability for code to evolve independently based on a set of criteria. He says it could have significant implications for optimizing manufacturing processes or designing particular products. He adds that it would result in artificial creativity, i.e., removing human assumptions to conceptualize unique solutions in short time frames.

Alexander sees sustainability as a critical area where analytics and machine learning offer exciting opportunities. In the past, most engineering designs were arrived at by trial and error, which took time and money. Today, with new regulations looming, manufacturers don’t have the luxury of time to update whole product catalogs to be more sustainable. By grouping and analyzing products together and applying the learnings of one to another, manufacturers can significantly reduce time and cost, while preparing for upcoming changes.

Nathan summarizes that the real endeavor must be to understand how AI, people and industry can come together to solve challenges—sustainability being one of the most pressing.