Manufacturers are recording human workers and using AI to understand their strengths and weaknesses. It’s a win-win—but privacy has to be respected.
Over 72 percent of factory tasks are still completed by human workers, according to a 2018 report by Kearney, a global consulting firm, and Drishti, a Mountain View-based AI company. Robots have not taken over all aspects of our factories just yet.
In many ways, humans still have the upper hand. People’s bodies are more flexible than those of robots. Our hands are more dexterous. We learn new directions more quickly. We also avoid risks more easily, and can come up with new solutions without additional training. For example, if a workstation were lacking certain parts, a person would know to get more. A robot might not be able to do so.
Even though robots have not surpassed people in agility and intelligence, AI can help people do their jobs better. AI can analyze human performance and reveal important metrics like average assembly time, as well as highlight potential improvements or common errors. Further, AI can assist in minimizing the risk of injury and accidents. Ultimately, a well-integrated deployment of AI offers the potential to keep people happier—on and off the line.
How AI is advising manual assembly
Using AI to help human workers is what Drishti set out to do with its workstation monitoring technology, which starts with a camera that monitors an assembly station. The camera captures exactly how a worker assembles a widget, recording the steps they take, the order in which they take them, and how long each step takes.
The video feed is annotated using Drishti’s proprietary AI, which determines whether a worker took the correct steps and did so in the correct order. It can identify what the worker could improve and where they create or experience bottlenecks. It can also identify creative workers who find ways to assemble a widget more quickly or effectively, so that others can learn too.
A side benefit of Drishti’s tool is that it can document how machines on the assembly line perform. “AI and video do a better job of explaining just how a machine or tool malfunctioned. They also show what human workers might have done to trigger the line going down,” Prasad Akella, Drishti CEO, told engineering.com.
Such information could be worth millions, because the cost of restarting a line is extremely high. In addition, combining AI and video can help engineers cut the number of hours they spend conducting manual time and motion studies. The Kearney and Drishti report found engineers typically spend 37 percent of their time on these studies. Often the studies yield inaccurate data, which reduces their utility. In addition, human workers can experience stress or alter their behavior if engineers stand near them to observe.
Freeing up time for engineers could allow them to perform more plan-do-check-act (PDCA) cycles and shorten the length of time for each cycle. This is because an engineer would not need to put one PDCA cycle on hold to gather data for another.
Ensuring that AI helps rather than harms
Recording so much information may cause workers to be uneasy, worried that the video invades their privacy. They could also be concerned that management would use the data to penalize individuals and teams that perform poorly.
Helping workers feel comfortable with being recorded and having notes generated on their performance takes effort. Drishti says it works in partnership with management and individuals on the assembly line to ensure both perspectives are heard.
One of the keys was to record video primarily from an aerial perspective. Drishti also focused the camera on the work area rather than the person. It trained the camera on the tasks that a person performed with their hands. All of these actions gave a worker privacy. Drishti doesn’t record audio as workers had shared concerns that their conversations with colleagues would be picked up.
Drishti explained to individuals that the videos and annotations would be used to correct and reward workers. Its engineers talked with line associates to inform them about the technology and get their insights on how best to benefit from its insights.
“In parallel, we explained to AFL-CIO [American Federation of Labor and Congress of Industrial Organizations, a national trade union] leaders that AI and video can help their membership by benefiting from their insights and making them globally competitive while keeping them safe. We let the unions know that we need their help in this effort,” says Akella.
Techniques to ensure that videos are being used in a positive way include recording workers before and after a training, before and after a work stoppage, and before and after broken equipment has been fixed.
For the five years that Drishti has offered its services, it sought out customers that recognized the benefits of cross-pollinating learnings. This helped Drishti disseminate generic yet validated AI models for each vertical. That way, an entire vertical like electronic manufacturing could benefit.
“We could do that without revealing proprietary information like the setup of the workstations or process details, which our clients tended to treat as proprietary,” says Akella.
Improving the assembly line
The combination of AI and video also helps quality managers dealing with defective products and warranty issues identify which units need to be recalled. The Kearney and Drishti report found root cause investigations take up 39 percent of staff time, over 62 hours a month.
Companies that use AI to assist human workers typically want to make the technology extremely simple for clients to use but offer them complex insights. The goal is to deliver easily consumed reports as needed, be that on a daily, weekly, monthly, quarterly or yearly basis. The reports can explain how a manufacturer can fix systemic problems.
One of the other benefits of recording workers is that companies can take excerpts from exceptional or poorly performing workers and use them as examples in training videos.
“Watching yourself on video helps you do a better job. Also, most people are visual learners. Watching another person do something well gives others a chance to see how to do it correctly,” says Akella.
In the future, fine-tuning tools involving AI and video could allow workers in new sectors guarantees that processes are performed more accurately.
“On the assembly line, people can usually solve problems quickly. However, culturally, they do not often turn in improvement ideas in a company’s suggestion box. Not surprisingly, AI and video are an engineer’s partner. They can collect and evaluate raw data and identify ways to help everyone,” says Akella.
