Sundown for the HH-60H Seahawk, responding to the Brumadinho dam disaster, Atlantic Airways, Mercy One in Iowa, and more!
We all wish we could foresee misfortune on the road ahead of us, allowing us the opportunity to avoid it. In many aspects of life, this ability remains firmly in the world of science fiction, but for those who work in helicopter maintenance, it’s fast becoming a reality. This is world of predictive maintenance, in which usage and condition monitoring data is used to predict potential issues before they happen. The rapid evolution of predictive maintenance is not only enabling measurable improvements in its effectiveness, but is democratizing its use among the global fleet, with even some smaller helicopters now able to benefit from it.
Most of the advancements stem from ever-increasing computing power, but a new, more science-based approach may allow manufacturers and operators to have an even earlier and deeper understanding of potential problems. Airbus Helicopters, Sikorsky and Sentient Science are among those investing to enhance the capabilities of prognostic tools.
Matt Tarascio, director for data analytics, prognostic health management and artificial intelligence (AI) for rotary and mission systems at Lockheed Martin (Sikorsky’s parent company), told Vertical that the manufacturer is using the extensive data collected from health and usage monitoring systems (HUMS) on the S-92 to predict negative events. “We have over 15 years of data on a 300-strong [S-92] fleet,” he said. The company uses machine learning-based tools to predict failures using data analysis, and the result is helping customers operate their fleet “safely and more affordably,” said Tarascio.
Such tools have been developed over many years, but that progression accelerated four or five years ago, he said. It “picked up in pace with big advancements in machine learning, when neural networks became popular again, when Google beat the world’s best Go player,” Tarascio said.
This has translated into an estimated five to 10 percent better fleet availability over the last 10 years. “We are now talking about extracting the last bit of availability; you will never get to 100 percent because there will always be something you cannot predict, but we are trying to get as close as possible,” said Tarascio.
Image recognition is one of the areas in which machine-learning tools have been leveraged for sustainment. It starts with photos of failed components of the same type. “Imagine numerous images of the same component in the fleet,” said Tarascio. “We can apply machine learning to looking at pictures, and detect a failure before it happens. The downside is you need a lot of data to train your algorithm well.”
Airbus Helicopters has also exploited recent progress in data analytics. “Last year, we launched Flyscan, a new data monitoring tool on dynamic systems,” Matthieu Louvot, Airbus Helicopters’ executive VP for customer support and services, told Vertical. “Before, such data analysis was used in a reactive mode; after a given threshold seen on HUMS data, you could not fly any longer and had to perform maintenance,” he said. “Now, in a proactive mode, weak signal analysis can indicate that a threshold will be crossed soon.” The warning comes several dozen hours early. “Therefore, the operator can plan a maintenance operation and thus avoid unscheduled works or even a mission failure,” he said.
As of late April, Airbus had found six customers for Flyscan, for a combined 26 rotorcraft. Those customers using HUMS have historically been offshore operators. “But Flyscan is suitable for every type of operation, such as EMS [emergency medical services],” Louvot emphasized.
It is available for all of Airbus’s twins, as the smaller H135 now has HUMS as an option — since the Helionix suite became the H135’s avionics standard last year. Louvot’s target for an offshore operator is one last-minute aircraft-on-ground situation avoided per year. Using Flyscan may also translate into a seven percent cost saving in unscheduled maintenance, he added.
In Airbus’s Flyscan, HUMS data is uploaded after each flight, instead of after a threshold is crossed.
Sikorsky has gone one step further. Last year, in collaboration with Outerlink Global Solutions, a Metro Aviation company, it launched in-flight, real-time HUMS data transmission to an operations control center via satellite. “We now have the ability to predict events by analyzing historical data in real-time,” Metro Aviation president Mike Stanberry said at the system’s launch. “That innovation helps operators more efficiently run their fleet and could very well be lifesaving.”
Engineers at launch customer PHI can view, assess, and track aircraft health data, and provide additional information to aircraft crew and ground support teams. PHI can thus improve operational and maintenance decisions, according to Sikorsky.
“We leverage maintenance, weather and other state data; this is an extra layer of safety,” Tarascio said. “A customer, such as PHI, can access this information throughout a mission and assess component health indicators.”
During flight, HUMS data is transmitted to PHI’s fleet management center. This includes flight manual exceedances, mechanical diagnostic health indicators, maintenance data collection alerts and avionics bus parameters. After the flight, the data is uploaded to the maintainer or operations center for further analysis and trending.
“HUMS data is owned by the customer,” Tarascio said. “The customer can choose not to share it, but there is an advantage — such as prognostics — for them to allow Sikorsky to apply advanced analytic models to their data.”
However, Louvot said he believed in-flight data transmission had limited relevance to rotary-wing operations. “In commercial [fixed-wing] aircraft, that could be interesting, but helicopters fly shorter legs, their utilization rate is lower than that of an A320, and satellite communications are expensive,” he said.
Increasing the amount of data available is key, according to Simon Gharibian, director of fleet management at Sikorsky. “The more data is available on the aircraft, the more Sikorsky is able to focus on parts that drive down availability,” he said.
Airframers monitor the removal rates of different parts that contribute to grounded aircraft. They track when the parts need to be replaced, how old they were, the reason they needed to be replaced, and how long they took to be replaced. Analytics tools are set to target parts that need to be replaced most often, or took the most time.
“In one case, we noticed that operators needed to replace their landing gear more often than we expected, so we took a look at the data and found that the issue was actually related to leakage and corrosion on the landing gear piston,” Gharibian said. “At this point, our engineering and manufacturing teams came in and developed on-aircraft repairs to address the corrosion and resulting leakage, thus reducing the number of times the landing gear needed to be replaced.”
Broadening the scope
Helicopter manufacturers are now working to extend the reach of predictive maintenance. What about adding further sensors? “The addition of sensors to various critical components of the aircraft has proven to be a crucial part in understanding the aircraft’s performance and potential maintenance needs,” a spokesperson with maintenance giant StandardAero said. Such components are relatively cheap.
Sikorsky’s Tarascio and Airbus’s Louvot agree, however, that installation can be expensive and impractical for existing aircraft.
A more relevant approach is to infer information from existing sensors, Tarascio said. It is the equivalent of creating virtual sensors around the aircraft. “We look at fully instrumented flight-test aircraft and compare [them] with production aircraft,” he said. On the latter, Sikorsky engineers try to predict loads and compare the predictions to measurements on instrumented aircraft. Once the load prediction is validated, component usage estimates can be improved.
Another way to expand the use of predictive maintenance is to look at those aircraft that were built without a HUMS. That was the purpose of the letter of intent Airbus and Safran Electronic and Defense signed earlier this year for the distribution of helicopter data monitoring systems.
Safran Electronic and Defense had developed Helicom, a system for the collection of usage and health data, certified through a supplemental type certificate. It is now part of Airbus Helicopters’ HCare Connected Services offering. “Safran had already sold the system to a great number of operators,” Louvot said. Helicom is believed to have become more attractive thanks to the services Airbus can provide.
The pair hope civil aviation authorities will approve the new methods and tools and eventually update regulations. “To extend a maintenance interval, you have to trust sensors and analyses, which involves work on aircraft certification,” Louvot said. Airbus has experimented with complementing scheduled checks with vibration monitoring. The company is aiming to receive certification of longer intervals in the coming years.
According to Sikorsky, the next step to implementing more advanced data analytics is regulation. Company executives hope the Federal Aviation Administration will be persuaded by the level of trust the airframer puts in its recommendations for repairs and replacements. Ultimately, new rules could lead to a 10- to 15-percent reduction in maintenance costs for commercial helicopter operators, Sikorsky estimates.
Another level where data analytics can be used is in part ordering. Sikorsky provides its customers with an Amazon-type environment to order parts: “Someone who ordered that component also ordered this one.” Why should the customer trust the recommendation? “The old style was very transactional, the spare parts department was incentivized to sell parts and not customer fleet availability,” Tarascio said. “But this is changing thanks to pay-by-the-hour fleet management models.”
Meanwhile, Sentient Science is claiming to have found a more science-based approach to predictive maintenance. “When you look at a system relying on sensors, it looks at deviation from the norm,” Jason Rios, vice president of Aerospace at Sentient Science, told Vertical. “It can figure that something is beginning to go wrong. The failure mode — such as a pit — has started to occur. So, your ability to react is already reduced.”
Sentient looks at the ability of materials to withstand operating conditions. Its software predicts “when these pits, cracks and other defects may start to occur before failure.” It takes into account the materials themselves, as well as heat treatment, surface finish and lubricant. It can tell whether a critical bearing will begin developing spall.
“It is a more science-based answer to questions, rather than interpolation of previous data,” said Rios. It can find the root cause of an issue. It thus “offers our customers the possibility to reduce stresses and extend life.”
Sentient’s software has been adopted by 10 percent of the world’s wind turbines in four years, said Rios. The first application in helicopters has been in the military, thanks to U.S. government-funded research and development. The 10- to 15-percent reduction seen in the operating costs of wind turbines is hoped to be surpassed in civil helicopters. “There are better records, a lot of data is available,” said Rios.
He deems Sentient’s work complementary to existing systems. “Our software programs would be continuously fed with HUMS data to keep refining our models,” he said.
Predictive maintenance is already proving itself in the field, reducing downtime and likely preventing more serious mechanical issues. But with an increasing amount of data at their fingertips, and evermore intelligent systems and software to analyze that data, a growing number of manufacturers, maintainers, and third party providers are making those predictions smarter, more accurate, and more accessible than ever before.