Manufacturing is notoriously slow to adopt new technologies, and artificial intelligence is no exception. Given the lack of specialized AI talent in-house and the difficulty in leveraging complex models to optimize and automate routine tasks, deep learning models are becoming increasingly popular among all but the largest manufacturers. was out of reach for manufacturers of
The lack of universal industrial data is also a major obstacle slowing the adoption of AI in mainstream manufacturers. Manufacturing data is often localized or specific to a particular industry domain or company operation. As a result, there is not much relevant data available to build reliable AI models. Especially compared to those available in industries such as finance and retail, where large volumes of trading and stock market data can be easily transferred across sectors.
At the recent MIT Machine Intelligence for Manufacturing and Operations symposium, Kalyan Veeramachaneni, principal investigator at the MIT Schwartzman College of Computing, said that large language models can be understood and understood by many people. He said it was built on a readily available universal language that follows certain rules and sentence structures. “This is not the case in manufacturing operations, where data from turbines, cars and other signals is not universally available,” he said.
Bridging the AI Gap
Generative AI, data-centric AI, and synthetic data are making AI more accessible and better suited to solving challenges in manufacturing operations. Generative AI tools such as ChatGPT offer a more intuitive way to model complex datasets and images, making AI technology more accessible to a wider range of manufacturing use cases and user types. Similarly, data-centric AI and synthetic data, which focus on engineering the data needed to build AI systems, transforms highly specialized algorithmic models into building optimal data sets for training AI systems. shift focus to These approaches make AI accessible to factory workers and manufacturing engineers who understand everyday production requirements and process challenges, but are not necessarily familiar with the language of mathematics and complex modeling. increase.
Consider the example of a factory maintenance worker. He knows how the shop floor works, but he’s not very digital. Employees may struggle to utilize information from computer dashboards, let alone analyze results to take specific actions.
The scenario is very different with generative AI. “What if that worker could talk to the system?” [through generative AI] You get information instead of understanding charts and metrics,” said Manoj Kotillal, partner at Boston Consulting Group and head of digital technology for the company’s AI manufacturing platform. “Many of these AI and machine learning models can now be augmented with generative AI, resulting in faster deployment and easier change management.”
Generative AI and other advances will accelerate the use of AI in many manufacturing use cases, including:
continuous operationFor example, it enables plant floor personnel to quickly identify specific machines operating outside preferred boundaries. This allows real-time adjustments to prevent downtime and quality issues.
companion for maintenancehelps field staff with maintenance operations by digitizing paper manuals and using AI to provide step-by-step real-time instructions based on the problem at hand.
Defect detection and inspection. This means augmenting human inspectors, or possibly replacing them with AI-enabled visual inspections. This improves accuracy, reduces inspection time, reduces recalls and rework, and translates into significant cost savings.
Eliminate repetitive tasks A process for improving employee productivity.
Adopting AI in Manufacturing
Understanding the potential value of AI in manufacturing is another. The actual implementation is another story. Experts advise:
Focus on data. Compared to high-value AI initiatives in other industries, manufacturing use cases tend to be more individualized and less profitable, making them more difficult to fund and execute.
An alternative to custom-built AI solutions are data-centric vertical AI platforms that can facilitate specific use cases. For example, automated anomaly detection tools may replace or augment human workers responsible for quality control.
“Data-centric AI is a systematic effort to get the right data to train AI systems, rather than focusing on coding the right algorithms,” said Vice President of Products at Landing AI. said Kai Yang at the recent EmTech Digital conference hosted by MIT Technology. review. “This enables users with solid domain knowledge to prepare datasets suitable for training AI models without deep machine learning knowledge.”
Related article
Prepare for organizational change. Much has been said about AI taking over human jobs, but the pervasiveness of AI will require new roles and operating models. If companies rely on AI-generated insights, they will need a layer of humans to systematically manage data quality and automation results. “We will have to do a lot of organizational redesign,” Cotillard said.
Start by experimenting with ROI in mind. It’s still early days, but things are moving fast. Manufacturers should start applying generative AI or other technologies to targeted initiatives to ensure learning, skill development, and early success that can be used to build organizational momentum and buy-in. I have. “It’s about introducing knowledge into the organization about how to use and implement AI,” he said. Mentioned at the MIMO symposium.
At the same time, any approach to AI must be more than just a learning exercise. “At the end of the day, it’s about making money,” Cotillard says. “If we can show an ROI, we get the budget, and then we can start doing something more experimental and risky.”
Read next: Why the era of “data-centric artificial intelligence” has arrived
