As more applications surface for AI, its potential within manufacturing has gained significant attention. However, with a big picture view, manufacturing is barely scratching the surface. And that situation is not because of a lack of interest. Instead, there is a lack of preparedness among most manufacturers, Ople.ai CEO Pedro Alves tells IndustryWeek.
“Manufacturers express a desire for big data analytics and predictive maintenance, but many are not actively collecting the required data,” he says. “They have sensors in place when something breaks, but this would only alert a tech to come in and fix the problem. This is reactive instead of a proactive dynamic using data from past occurrences to accurately predict future results.”
However, Alves suggests that manufacturing is close to a breakthrough for using AI to solve a multitude of problems. “More companies understand where the data resides and what they want to do with it. They just need the right technology tools to help them move past some hurdles and begin leveraging AI’s benefits,” he says. “And the potential benefits are there. McKinsey reported in 2019 that asset productivity increased by 20 percent with AI predictive tools and maintenance costs also fell by up to 10 percent.”
A lot of the problem scenarios in manufacturing involve predicting problems before they happen to keep the plant at full capacity. Predictive modeling and maintenance require an array of sensors and an understanding of past conditions and how that relates to future instances. A common shortcoming is the sensors were built for human monitoring, not for machines, explains Alves. “Better data is the most important next step for manufacturers to transition into AI and better sensors is one of the main ways to achieve that. Another common shortcoming in manufacturing is error detection, a critical yet difficult effort, especially when manufacturing parts where tiny imperfections matter,” he says.
For example, consider a company that builds convertible tops for cars. Every top is tested by humans and the facility also uses sensors to detect vibrations. “The human operator would look at the sensor data and try to determine the cause of unacceptable vibrations. When they could not spot the problems, that created the perfect opportunity for machine learning,” says Alves. “With a machine learning platform in place, the various sensor data points are pooled together which allows for further testing models. These models can spot vibration signals that are not visible to the naked eye.”
According to Alves, manufacturing forecasting efforts also need to improve with AI and machine learning. “Both pricing and demand forecasting are typically completed using simple statistical models,” he says. “With AI technology, firms can greatly improve the forecasting accuracy for greater efficiency and simpler scaling.”
Building on best practices
Manufacturing is ideally suited for computer vision applications such as inspecting for defects or optimizing products, explains Alves. “For a manufacturer producing material, for example, there are often manual tasks to cut the material, evaluate it, and see if it passes various grades. Manual testing is slow and expensive especially when the line is turned off,” he says. “With the right sensors and a robust AI platform, the data is collected and provides the company with real-time readings. The quality of the produced material is predicted ahead of time, which limits stoppages and inefficiencies. As manufacturers expand their AI capabilities, they will also uncover additional use cases including adding machine learning to smart robots to improve performance over time, integrating consumer data into product development choices, and uncovering bottlenecks in production lines.”
For manufacturers starting the AI journey, Alves suggests erring on the side of collecting more data. “Do not throw away data or miss an opportunity to put more sensors in place. The needs for AI continually evolve, and manufacturing firms that continually build their data sets are the ones best positioned to improve and grow at scale,” he says.
However, manufacturers diving into AI should carefully consider their needs before deciding to hire an entire data science team, explains Alves. “There must be good reasons for building an expensive team compared to using proven software tools. If a manufacturer needs to innovate in the field of AI to accomplish some new project never done before in manufacturing, then they might need a team,” he says. “But for the more common occurrences with predictive maintenance and monitoring, an adaptable and powerful AI tool from a software company will deliver quick wins.”
When there is an advancement, the market leaders will overcome the biggest hurdles. This dynamic is playing out in manufacturing, with the early adopters taking on the risks and narrowing down the use cases and best practices. The first groups to follow the market leaders are the ones that will turn the tide to the bulk of the manufacturing industry using and pulling considerable benefits from AI.