The use of big data has matured to the point where analytical services are being offered that allow predictive maintenance. Mario Pierobon reports
Data analytics is helping to reveal insights which empower a wide range of possibilities for reducing uncertainty in air operations.
With real-time data and new analytics tools and applications, the practice of predictive failure analysis (PFA) is making it possible to move from looking primarily for emerging fleet issues to preventing or reducing schedule disruptions from unscheduled maintenance events per aircraft in real-time.
The use of data analytic methods, such as statistical analysis, in helping to predict and avoid unscheduled maintenance events, or to improve the performance and design of aircraft, is not a new practice in the industry. What is relatively new – and developing critical mass – is the level of sophistication of the analytics tools available, which now enable work on a near to or actual real-time basis. John Maggiore, Managing Director of Maintenance and Leasing Solutions at Boeing Digital Aviation and Analytics, says: “Boeing has a long history of using analytics to design and build more reliable, efficient aircraft, and in providing services to help improve the operation and maintenance of commercial and non-commercial aviation. In the mid-1990s, Boeing, in collaboration with its 777 customers, began sharing data via the In-Service Data Program. Sharing operational data allowed for additional analysis that helped further boost the dispatch reliability of this successful family of airplanes. This programme has since expanded to cover most Boeing models. We recently cored up these capabilities into our Boeing AnalytX suite of products.”
He continues, “One of our key analytics tools in commercial aviation, Airplane Health Management (AHM), applies descriptive and prescriptive analytics to real-time airplane data, providing maintenance data and decision support to airlines to increase operational efficiency. This enables proactive maintenance management and maintenance scheduling to avoid schedule disruptions. Proactive management is like having an over the horizon radar, it provides more time to assess, plan, and manage, rather than react to a situation. AHM uses analytics to evaluate two million conditions each day to determine when alerts should be generated across 4,300 airplanes. AHM is our key predictive fault analysis tool.”
Boeing’s AHM is a web-based decision-support tool designed to help airlines make more efficient real-time fix or fly decisions, and provides a range of predictive analytical tools to help airlines identify and act upon maintenance conditions before they turn into faults which can cause delays and cancellations. AHM is offered as a subscription service to operators. “AHM leverages existing systems on the airplanes and as such, in-production airplanes require no additional equipage use it. AHM is one of our Boeing AnalytX-powered products. Boeing AnalytX, which brings together various products powered by data analytics and the 800 data scientists who work on them, is accelerating innovation and our predictive analytics capabilities,” says Maggiore.
Boeing introduced AHM in 2003. At the time, the tool represented a first generation of health monitoring solutions focused on real-time identification and resolution of maintenance issues with aircraft in-flight. Over the years, the adaptation of AHM has changed customer operations to the point that using health monitoring to address real-time maintenance issues is standard practice today. “In fact, it is hard to find a 777 or 787 operator that does not use AHM. To develop reliable, actionable predictive alerts, Boeing identifies airplane system data that can be used to infer system or component degradation. Our engineers use this to develop a predictive alert, test and tune it until it is ready for integration into the AHM application for all to use. We have been doing this for over a decade, and now have thousands of diagnostic and prognostic alerts in our library, spanning all our production airplanes and including some legacy airplanes,” Maggiore says. “AHM is powered by Boeing AnalytX, which is the Boeing global programme that fuses data analytics and aerospace expertise together to create new insights, opportunities, and products that further advance the aviation industry.”
PFA has been around for a long time using what was traditionally known as the physics-driven modelling technique. “As aircraft systems and components generate and capture more data, connectivity becomes faster and more affordable. Moreover, as computer storage and memory becomes cheaper, we can complement the physics-driven models with data-driven models, to not only improve our failure prediction accuracies, but also to move into predicting sub-components failures,” says Fong Li Wee, Director of Information Technology, Connected Aircraft at Honeywell Aerospace.
PFAs are combinations of weak signals from aircraft systems that, once fed into models, can indicate if and when a system will fail. “Current Airbus predictive solutions in Skywise provide high dependability on the prediction enough in advance, providing the airline with the ability to take timely and informed decisions. Also, when connected with other systems in the airline, they provide actionable items so that benefits can be scaled and made systematic after every flight. Legacy systems were looking at a reduced number of parameters and therefore had limited coverage across the aircraft systems. Also, they were unable to systematically provide guidance without generating significant amounts of No Fault Found,” says Jaime Baringo, Head of Digital Business Development at Airbus.
The company recently launched Skywise, which collects data from across the systems in Airbus A320 Family aircraft and, using the Rockwell Collins Flight Operations and Maintenance Exchanger (FOMAX) program, transmits the data to Airbus.
Baringo continues, “The power of the analytics contained in FOMAX- Skywise allows airlines to identify predictive models which, combined with the OEM’s expertise, can quickly make new models available and hence grow prediction coverage exponentially.”
With new generation aircraft such as the A380, the 787 and the A350, more data is available to predict failures on components before they fail. “Now, with the quantity of data available, the automatic transfer of data, the decreasing cost to store data and to perform the corresponding analytics, AFI KLM E&M has been able to develop its own solution, PROGNOS, which predicts failures which are not seen by legacy solutions,” says James Kornberg, Director Innovations at Air France Industries KLM Engineering & Maintenance (AFI KLM E&M). “Being both an airline and MRO performing strong IT operational research with data scientists and engineers who know the aircraft and engine systems, AFI KLM E&M has been able to develop PROGNOS, which also relies on all maintenance data available such as line maintenance complaints, hangar findings, fault found confirmation, shop reports, etc.”
The legacy approach to predictive failure analysis has relied heavily on known failure conditions, whereas advances in data management and analytical methods are revolutionising how predictive analytics are applied. “Modern data recording and processing systems onboard the airplanes, combined with ground-based analytics platforms using technology such as machine learning and natural language processing, have resulted in sophisticated tools. These tools are used by airline maintenance teams and Boeing/supplier support teams during day of operation, and are used to resolve in-service issues and inform future aircraft design. Advances in onboard and offboard data processing and analysis can drastically reduce the flowtime for alert development. But a reliable infrastructure and process for predictive alert maturation is vital as airlines are using these alerts to make real-time decisions on whether to spend resources to change out a component – which can sometimes cost several hundred thousand dollars, prior to failure,” says Maggiore.
The primary data sources used for predictive maintenance alerting include aircraft fault data – flight crew messages and maintenance messages, parametric data – including sensor data, maintenance action and execution data, such as pilot write-ups, maintenance write-ups, component removals, and component shop findings. All aircraft systems have some level of fault reporting and data collection to enable predictive analytics.
“Over the years, improvements to aircraft hardware and software for onboard diagnostics and data recording systems have enabled improved health management and predictive analytics. Health monitoring is a key technology focus for all new aircraft design. Newer generation airplanes come fully equipped for health monitoring. For example, with the 737 MAX, AHM is available on delivery to the customer, with health management functions built into the onboard network. Looking toward the future, health monitoring and predictive analytics are fully integrated into the design process for the 777X,” says Maggiore.
Baringo notes; “In Airbus we have also seen the value of combining this data with operations information so that we can improve and create new models that work faster and are more effective; this is now possible with Skywise. With the expertise from our design office we are able to propose predictive models that bring significant efficiencies with the aircraft as it is. On top of that, we are introducing FOMAX, that natively connects to our Skywise platform, providing 20 times more data from sensors and systems that are not accessible otherwise. FOMAX is a minimal addition to the aircraft as it integrates seamlessly with existing avionics.”
“We use fault and sensor data generated from aircraft systems, flight operations and maintenance data provided by our aircraft operators, as well as repair and overhaul data collected in our shops for PFA. There are a lot of data that existed today which are not mined for meaningful insights. Our GoDirect Connected Maintenance service offering leverages existing data, sensors, and hardware to help operators reduce delays and cancellations due to unscheduled maintenance events,” Wee says.
“On the A380, for example, smart access recorder data is analysed on the systems which created more delays and cancellations, such as the fuel pump or the nose gear position sensor. Now, with the predictive PROGNOS solution in place, the fuel pump system is no longer on the top five delays and cancellations. Hardware and software upgrades are installed on older generation aircraft for the automatic transmission of data,” Kornberg explains.
A solid momentum
Even if it is only the beginning for the establishment of sophisticated PFA initiatives, airlines understand more and more the value of data available on their aircraft. “We are developing our predictive maintenance solutions on the aircraft and the engines of our customers,” says Kornberg.
The practical adaptation of predictive analytics, however, varies greatly among operators. Advances in data analysis tools and reduction in the cost of data collection, storage and management over the past decade have lowered the investment barriers to entry. “The evolution of companies towards a data-driven culture is also increasing the uptake of predictive and proactive analysis to support aircraft maintenance. Those who embrace the concept can also generate eye-popping results that directly impact the airline’s financial performance. Customers looking for new ways to save money and improve efficiency can rely on products powered by Boeing AnalytX to help them reach their goals,” Maggiore asserts. “An airline knows its own operation, but often lacks airplane design expertise and access to the global operating fleet data. Boeing aggregates maintenance actions data from the global fleet and provides that intelligence to our customers, making Boeing solidly positioned to provide reliable, predictive insights. This allows an airline to troubleshoot better. Airlines also use this data to benchmark their own operations versus the larger fleet to expose opportunities for greater maintenance efficiencies.”
“It has really taken off over the past year and we are adding new operators every month, now that they have heard from the proven field experience from the pioneering airlines. The users do not have to manipulate the data as it automatically flows to Skywise where it is ingested and pre-processed. They simply log into Skywise and get alerts directly on predicted failures for their fleet,” says Baringo. “The experience from the field is amazing, with direct short term results and proven savings as a result of the avoidance of operational interrupts, reduction of spares and overall decrease of the maintenance cost. Skywise also allows the airlines to benefit from the collaborative intelligence and experience across the various operators in the platform, while preserving their competitiveness.”
“We are seeing a lot of interest in our GoDirect Connected Maintenance offering due to the successful trials where we can quantify the operational savings for our customers and we currently have paying customers for this offering. Our offering sends a notification to maintenance teams to prescribe actions to prevent the unexpected product failures and explains the reasons behind it. Our customers love that our offering is prescriptive without having to do any troubleshooting. This offering also provides tools where they can dive deep into the data trends,” Wee explains. “Initially, our customers were sceptical of the recommendations and they did not act on them promptly because we were essentially telling the technicians who had been doing their work for many years to trust the suggestions generated by our data models. It only took one or two ignored recommendations to prove the value of our offering as those instances resulted in delays and cancellations due to unexpected product failures.”
Aggregating data across multiple operators helps to spot problems that are unique to one operator, and from this helps that operator improve their practices. “It also helps by generating larger data volumes that can train the predictive models and deploy solutions faster. Aircraft are engineered with high levels of reliability so the larger the sample size, the better it is for PFA, since specific part failures occur not as frequently as one might imagine,” says Wee.