Skip to content

Measuring the right data, and context, to ditch costly unplanned maintenance

By Karen Haywood Queen Contributing Editor, SME Media

‘Historically, people … were drowning in data but starving for insight’

Until 2014, Delta Airlines often had little warning before an airplane airframe experienced a maintenance issue requiring additional ground time, resulting in delayed or canceled flights.

predictive_ Delta Lead 768x432.jpg
Delta now has predictive analytics/maintenance program for its engines and airframes. Above, maintenance technicians rig a door.

“What’s huge is unplanned maintenance where you have to fly parts and people to another location,” said Dan Lohmeyer, senior VP of product management at GE Digital.

“We were 100 percent reactive” when it came to airframes, said Shawn Gregg, general manager, predictive technology engineering at Delta. “We had to figure out what the problem was, what parts are needed to address the problem, and then acquire the parts.”

Then Jim Jackson, a Delta manager of technical services and engineering, lobbied for the company to start a predictive analytics/maintenance program for Delta airframes, the structure that typically includes the fuselage, undercarriage, empennage and wings and excludes the propulsion system. The carrier already had a predictive analytics program for aircraft engines.

“We started seeing some results almost immediately,” Gregg said. “Being able to predict a maintenance issue as opposed to waiting for one to happen is one of the principal keys to being able to provide a good customer experience.”

As part of a companywide effort to improve passenger satisfaction, Delta moved from No. 3, in 2014, to No. 1 in 2019, in The Wall Street Journal’s “middle seat scorecard” ranking passenger satisfaction. Other passenger-facing teams played bigger roles, he said, but, “predictive maintenance helps ensure the aircraft is available and ready to fly.”

Accurate predictions mean that Delta now can schedule a plane to be serviced when and where it makes the most operational sense.

“Our stations have unique capabilities, some more expansive than others,” Gregg said. “You want to do the best you can to perform the maintenance on that aircraft in a station that has that capability. Now, we get to decide when we work on the aircraft vs. the aircraft telling us when we work on the aircraft.”

Although the aircraft themselves typically have shown no signs of problems beforehand, the predictive analytics have proven accurate, he said.

“We find that when we issue a predictive maintenance log page, 95 percent of the time it results in removing a component that was about to fail,” Gregg said. “Predictive analytics is machine agnostic. Instead of having a plant full of manufacturing machines, we have aircraft. Our plant is the sky. Our machines move through the plant.”

Delta’s predictive analytics journey started with the 777 fleet because the company had more data available on that airplane. “The 777 at the time was the most data-rich aircraft we were flying and also had one of the most comprehensive predictive maintenance tools available,” he said.

Since then, Delta has broadened its scope to include other aircraft. One such case is the air conditioning in the 737, Gregg said.

“We were getting a lot of customer complaints about inconsistency in cabin temperature throughout the 737,” he said. “We had developed a very traditional maintenance program. For example, replace this part every 100 hours. We ended up with a lot of false positives.”

predictive_John Ferraioli 2.jpg
Companies are moving toward an outcome-based approach, Deloitte’s John Ferraioli said. “Now ... we’re able to have a tighter recipe for success vs. boiling the ocean.”

Gregg’s team designed a suite of sensors that measured temperature and pressure in the 737’s air conditioning system. Delta took advantage of longer planned maintenance downtime on each plane to install the system, he said.

“We are now able to predict failures one to two weeks before they would happen,” meeting the company’s predictive analytics goal, Gregg said.

‘Starving for insight’

Across manufacturing, data analysis has improved even as the volume of data has increased.

“Historically, people had the ability to get a lot of data and then did nothing with it,” said John Ferraioli, consulting managing director in Deloitte’s supply chain and network operations offering and the data leader in Deloitte’s digital supply networks practice. “They were drowning in data but starving for insight.”

Now, companies are moving toward an outcome-based approach, setting business goals and analyzing the data to help reach those goals.

“People are realizing they don’t have to measure, or even get, all the data,” he said. “They measure what matters. They measure the right data, the right context to get the outcomes they’re searching for. Now we’re able to tap into past learning and be more prescriptive: ‘We want to use this data from this date from this sensor capture point.’ We are able to have a tighter recipe for success vs. boiling the ocean.”

Awareness of data’s value growing

Predictive analytics is an emerging use case cited by 12 percent of companies in PTC’s 2019 “state of industrial IoT report,” said Jean-Philippe Provencher, VP for manufacturing strategy and solutions at PTC.

Companies that PTC surveyed said they were looking to add more predictive maintenance for improved functionality in the future, he added.

“Customers are more and more aware of the hidden value in their data and are looking to leverage it at scale,” Provencher said.

Analytics take time

GE Digital also worked with Intel to design a program to monitor the health of its fan filter units with predictive maintenance. In early pilots, Intel reduced unscheduled down time due to fan filter failure by 300 percent compared with manual inspection, the company said in 2018.

But many companies get stuck after running data analytics, Lohmeyer said.

“They learn that a certain variable is trending differently from before, which indicates something may be amiss,” he said. “But they don’t know what they should do: take the equipment out of service; schedule service in advance?”

The next key goal, Lohmeyer said, is: “what do you do with the results of those analytics?”

Achieving that key goal is a multi-year journey, he said. Some companies, such as Delta, are farther along the path. Companies first must create a digital twin blueprint to establish and track all the possible different failure modes.

“An asset could have 50 different failure modes—some severe, some less so,” Lohmeyer said. “Of those failure modes, you need to determine what are the consequences of failure and what are the mitigating activities you can take to prevent failure from happening.”

Building blocks explained

To build predictive maintenance models, Delta paired engineers with experienced mechanics committed to improving operations. That experience and institutional knowledge was critical as Delta built new maintenance alerts that weren’t covered in current manuals, Gregg said.

“My team has a responsibility to write a very specific prescription to address the issue, including parts, special equipment and time to get the aircraft to a Delta repair facility,” he said.

A digital twin is another important step in the process.

When developing a digital twin, many companies start with a generic digital blueprint, Lohmeyer said. GE has a catalog of over 300 digital twin blueprints for equipment, including pumps, aircraft engines, transformers and turbines.

The first phase of the process is setting up the generic digital twin, he said. Then experienced people within the company customize the digital twin to reflect how that company deploys equipment, consequences of particular failures, specific mitigation for failures, and other issues, he said.

Augmented reality also plays a role, Provencher said: “Although the digital thread primarily serves as the source of and contextualizing function for the generated data, the visualization component—namely augmented reality—catalyzes the synthesis of the data as it is put forward in a more interactive format.”

Cultural readiness examined

An aversion to culture change and a lack of interoperability were among the roadblocks to implementing the digital thread, companies told PTC. PTC has focused on helping its customers leverage their existing OT and IT systems, using an IIoT platform that wraps around these systems instead of replacing them, Provencher said.

In addition to the technical work, Delta focused on a cultural shift.

“We had a lot of work on the technical side but also a lot of work on the non-technical side,” Gregg said. “A lot of that is cultural based. We have some of the best mechanics in the world and you’re going to tell them they have to replace a part that isn’t broken. Why are they going to want to do that? That took a lot of work. The same manager who convinced us we needed to start doing predictive maintenance on aircraft (Jim Jackson) had ideas on how to start building trust. He understood what the hurdle was.”

To build trust and buy-in, Delta did more than merely issue replacement instructions to its mechanics in the field.

“My team stays in close contact with the component shops and our external repair vendors,” Gregg said. “Often times the part to be replaced has not failed yet. To determine if there’s a pending failure, we have to tear the part apart and inspect it in more detail than we would otherwise.”

After a team determines whether a failure was pending, the team informs the mechanics of the results.

“We send the mechanic the results of the tear down,” Gregg said. “We tell them, ‘We had you remove this part. Our shop took it apart and we found this issue. This is what we were seeing and why.’ The whole idea was to close the loop and make sure they understood they are as much a part of this change of predictive maintenance philosophy.”

Highly valuable knowledge will be walking out the door in next few years

Manufacturers that haven’t already started down the road to predictive analytics need to get moving quickly. That’s because a key element of the process is cataloguing the knowledge of an aging workforce, said Dan Lohmeyer, senior VP of product management at GE Digital.

Most companies don’t fully appreciate exactly what needs to be done and that the task will take years to complete, Lohmeyer said.

The median of manufacturing workers was 44 in 2018, according to the U.S. Bureau of Labor Statistics, up from 40.5 in 2000. About one quarter of the manufacturing workforce is 55 and over.

“A significant amount of the manufacturing workforce is retiring in five to 10 years,” Lohmeyer said. “A key concern is how do we get ahead of that? How do we codify their knowledge? You need to codify their knowledge of how equipment works, what causes failure modes, the consequences of failure modes, and what actions to take to remediate failure. If you don’t do that, these people and their knowledge are going to walk out the door and you’ll have issues 20 years from now.”

  • View All Articles
  • Connect With Us

Related Articles

Always Stay Informed

Receive the latest manufacturing news and technical information by subscribing to our monthly and quarterly magazines, weekly and monthly eNewsletters, and podcast channel.