AI, Cloud and 5G infrastructure driving advances—along with the embedding of subject matter experts’ knowledge
Until 2017, Schneider Electric faced a factory bottleneck at its breaker box plant in Lexington, Kentucky. When the automation cell that welded the boxes went down, all production could be forced to stop. One of a number of problems could cause a machine in the cell to malfunction. Diagnosing the problem required an experienced employee to dig into the HMI (human-machine interface) to get the fault code. Downtime ranged from 20 minutes to half a day.
“It was a complicated weld,” said Luke Durcan, EcoStruxure director at Schneider Electric. “When the machine went down, we were in big trouble. Once it went down, our ability to diagnose and fix the problem was complicated. We relied on a number of senior, experienced folks. If the machine went down when they were away, we had a problem. The downtime on that cell was killing us.”
OEE (overall equipment efficiency) was taking a big hit.
Then Schneider Electric introduced its Augmented Operator Advisor (AoA) tool at the plant.
With this augmented reality-based process-optimization solution now in place, when the welding machine goes down, the machine automatically communicates with a PLC (programmable logic controller) to diagnose the problem, he said. A worker—not necessarily a senior employee—points a tablet at the malfunctioning machine. The tablet connects directly to the PLC and receives the fault code along with the recommended fix for the problem.
The PLC also communicates with other PLCs upstream and downstream to pinpoint the failure among a combination of different fault codes for a more precise diagnosis, Durcan said.
“Unless you were an experienced service tech, you wouldn’t have been able to quickly put all the possibilities together,” he said. “It would haven taken a day to put it all together and figure it out.”
The factory cut mean time to repair by 20 percent. The software’s return on investment was less than one year, Durcan said.
“I looked at mean down time over a number of years,” he added. “We were slowly bringing it down, five points a year, three points a year. We introduced this and we saw this huge shot in the arm.”
Understand value of facility-wide OEE
Now, Schneider Electric has improved data integration that allows the facility to look collectively across the entire organization, he said.
“The ability to do OEE across the facility is absolutely key,” Durcan said. “Where can we get predictive? Where can we add value? Where can we leverage the ability for subject matter experts to embed their expertise into the process? The ability to improve a process is not dependent so much on which compute and Cloud services you can throw at it, but how much knowledge you can embed from SMEs (subject matter experts).”
Schneider Electric now has “better insight of downstream and upstream failures that, if eliminated, could drive better OEE,” he said. “The system drives synchronous OEE across the facility. People have been focused on individual OEE and individual cells. There could be 20 cells in that process. Looking at average cells across the site, that could be pretty good. Each individual cell can be ticking along nicely. But if cell eight goes down, cells nine and 10 go down. In the past, we got fragmented systems that didn’t allow us to get overall output.”
In the early stages of process optimization, the low-hanging fruit included leveraging learned optimal processes, efficiently managing data and change orders, efficiently reusing data and simulation of production, said Jay Gorajia, director of global services for manufacturing engineering, Siemens Digital Industries Software.
“Having a way to efficiently manage this data and related change orders to ensure prfaroduction is done on the right parts, with the right components and right processes, can only be done by stitching them together in a product lifecycle management system that understands the data and its contexts,” he said. “This could save companies from vast amounts of wasted material and time.”
Gorajia said tech driving these advances includes:
- artificial intelligence (AI) driving production data back for informed and “learned” best production practices;
- Cloud and 5G infrastructure to ensure efficient handling of data, and
- 3D printing to iterate and optimize tooling and fixture creation at an accelerated rate with turnaround in minutes instead of days.
Recognize it’s a ‘moving target’
Process optimization is a “continuously moving target,” not a once-and-done situation, said Mike Lackey, global VP of solution management for digital manufacturing at SAP.
Key steps, he said, include:
- bringing intelligence and automation to the entire manufacturing process;
- integrating that process with logistics, finance, sales and service, and
- connecting each link in the supply chain to the intelligent enterprise.
Leveraging the Cloud environment to lower total cost of ownership has moved process optimization further down the road, Lackey said. “Machines also have become smarter, with the ability to provide more data to predict maintenance, performance/quality trends, and possible downtime.”
‘Unlock’ knowledge of workforce first
Part of the success of the initiative at the breaker box plant has been buy-in from subject matter experts working in the factory, Durcan said.
“There’s a lot of cool technology sitting idle in facilities because people don’t adopt it. It’s not suited to their environment,” he said. “People have tried to solve process problems with technology but not leveraging the capability that’s been in the heads of their workforce. You need to unlock that before you think about AI. It’s great to sell technology but unless you can integrate technology into the people and process aspects, it’s a waste.”
At Schneider Electric, most of the implementation has been done by SMEs in the facility, Durcan said. “They’re the ones who understand the process.”
Generally speaking, a tool is designed 60 percent standardized out of the box with the ability to be customized 40 percent at the factory, he said.
“Bright people are bright people,” he added. “You need to allow people to take some risks. We didn’t put handcuffs on people. We just said, ‘Solve it.’ If you give them the tools and allow them to use them, they can be successful.”
Knowledgeable workers start by cutting and pasting work instructions and so-called cheat sheets into the software, Durcan said. Over time, senior workers continually enrich and improve the software.
“We have ‘Tony,’ the 30-year-guy who knows all this and constructed all his work flows, encapsulated all his knowledge in his tools,” he said. “Over months, we enriched, augmented and developed that tool. What was really cool was to see these experienced men and women look at the work flow and the problems and say, ‘It might be this. It might be that.’ They ran scenarios for a number of different problems. The more you see that problem, the more confident you are with the diagnosis.”
Reuse of production processes, knowing what worked well and what didn’t by leveraging feedback from production, provided a more efficient way to accelerate process optimization, Gorajia said.
Finally, that feedback can be put into simulation tools (discrete-event simulation for logistics, or kinematic for assets) to ensure the right KPIs, such as yield and throughput, are met. Creating the “as manufactured” data connected to the “as designed” data can dramatically support not only process optimization but can also drive improvements at design, he said.
Standardize integration of old, new tech
One challenge that remains is integrating new technology with what’s already on the factory floor.
“Most technologies came to be open and there usually is a way to connect them but it’s expensive,” Durcan said. “You end up with a Frankenstein of solutions, non-standard, non-repeatable. You’re always going to have a diverse base of vendors. You need to come up with a standard way of integration.”
In general, the best approach is to identify a problematic machine common across a facility—and then scale the software across the enterprise, he said.
After demonstrating how valuable its AoA (augmented operator advisor) was at its own breaker box plant, another use case was a Schneider Electric customer: a large cosmetics company, Durcan said.
The first thing senior workers did was to cut and paste their existing work instructions into the system.
“Implementation costs were very low,” Durcan said. “It took only a couple of hours.”
Consider more reliance on software
Schneider Electric also used its AoA augmented reality software to provide guidance from the experienced team in its Lexington plant to a newer facility in Monterrey, Mexico. Training remains important, but increasingly Schneider is relying on subject matter experts sharing knowledge via this new software.
“A big conveyor runs through the plant,” Durcan said. “If the product is not affixed properly on the line, it falls off. If it falls off the wrong way, it can start to take other products off the line.”
He recounted an employee at the Lexington plant sending a video about an event in Lexington to Monterrey, noting that “a couple of weeks later, the same event happened in Monterrey and the operator knew right away what had happened. That allowed the manager in Monterrey to make an intervention.”
“In the last couple of years, we’ve gone from training being a formal exercise to having more of a knowledge-sharing activity, having processes and systems to support that encapsulated tribal knowledge within the organization,” Durcan said. “We have a number of different solutions allowing people to cooperate and collaborate around knowledge rather than training.”
Even when slowdowns can’t be predicted, software can alert managers, who can deal with a work stoppage.
On one big conveyor assembly at the Schneider Electric facility in Lexington, the supervisor used to watch the conveyor to make sure there was enough product coming down the line, Durcan said. If that supervisor forgot to look or went on break, he’d miss the slowdown or stoppage.
“If you have a starvation event, then you have a group of people in a cell in a facility standing there doing nothing for five minutes, maybe an hour, before the product arrives and they can continue to produce. Once you lost it, that’s capacity you don’t get back,” he said.
Now, with its Avena Insight OT-data-integration technology in place, the manufacturing plant can first work to make sure the bottleneck cells are continually working, Durcan said.
“Our system sends alerts whenit sees an OEE blockage event in a particular cell and then calculates the overall impact on OEE for the facility. They can make operational decisions to put that cell team on break, for example,” he said.
The system also raises operational awareness throughout the facility to make investment decisions.
“If we’re continually seeing a starvation event at this cell, we can put in more machines. The goal was to drive OEE; we don’t want to build a new facility yet.”
Get ready for ‘a virtual knowledge base’
In recent years, manufacturers have achieved important advances in Cloud and data management, characterization of production systems with real production-data feedback, as well as smarter simulation of production systems, assets and logistics for material flow, Gorajia said.
“Over the last three years, data capacity limitations have almost disappeared,” he said. “If done well, vast amounts of data can be stored, organized and effectively shared with groups within a plant, country, or around the world, creating a virtual knowledge base.
“It is no surprise that there are pockets of deep domain expertise within any company. Soldering experts, metallurgy experts, component handling experts, board construction experts, metal part manufacturing and related mounting technology experts are everywhere—but not necessarily in the same place.
“Companies can effectively leverage Cloud and AI to provide this knowledge base and access to experts from around the world,” Gorajia said.