Predictive maintenance has progressed from industry buzzword into a goal for many operators. Today, several airlines and MROs are demonstrating how to use data to increase fleet reliability. But how are they able to fully benefit from the vast wealth of information available, and mine it effectively without incurring unmanageable costs?
Tata comes in many forms and from various sources in an airline – the vast amount available today created the term ‘big data’. Unless robust digital solutions are installed that can aggregate, distribute and analyse information, data is useless. Complex algorithms are required for this analysis, specifically machine-learning algorithms to handle aircraft and engine sensor information.
According to an Oliver Wyman MRO Survey, the global fleet of commercial aircraft could generate 98 million terabytes of data per year by 2026, due to big data. Aircraft data comes from sources including the flight data recorder (FDR), engine health monitoring (EHM) and airframe health monitoring (AHM); each receiving and transmitting thousands of parameters from in-built sensors, often down to component level. The amount of data has implications for transmission costs and for an airline’s connectivity and storage capabilities. That is, for the data to perform proactively, it needs to feed data regularly into maintenance (M&E) and operational systems to create a current picture. Having the infrastructure for this can feel cost-prohibitive for carriers.
Engine and airframe original equipment manufacturers (OEMs) were initially at the forefront of these digital solutions; as aircraft become more sophisticated, the intellectual property (IP) that governed them meant that OEMs were ideally positioned to generate software that could manage data effectively. However, airlines with multiple fleet types still sought solutions that could ingest different data standards and forms. To maximise the ability of big data in the industry, it can’t be kept in-house. “Today, OEMs, airlines and maintenance, repair & overhaul (MRO) operators are showing interest not just in gathering data, but sharing it for a number of different uses—predictive maintenance or health monitoring systems being key applications,” says James Elliott, Principal Business Architect, Aerospace & Defence at IFS.
Predictive maintenance is explored here. By utilising solutions that can interpret aircraft data, maintenance control centres can build a day-to-day picture of individual aircraft (and fleet-wide) performance. Overlaying this with historical information means one can forecast – using advanced analytics – when a part will fault. Moreover, this performance data will contribute to the historical data – meaning that predictive models generated become ever more accurate. By predicting fault behaviour, operators can schedule maintenance ahead of the fault being flagged in operation.
As Aerospace Technology Week approaches, ATR is reviewing the industry stance on predictive maintenance analytics – that is, how are airlines best utilising maintenance and operational data to maximise time-on-wing (TOW). “Ideally, predictive solutions shall reduce the overall cost of operation, reduce interruptions and increase the reliability of the fleet,” agrees Frank Martens, Head of Customer Development Digital Products at Lufthansa Technik (LHT). “There is no generic number available, but some predictive solutions reduce the number of unscheduled removals by 80%, and just one predictive solution can save an airline more than a million Euros per year, but this strongly depends on specific operational patterns.”
Before predictive maintenance can reach maximum potential in the industry there remain challenges pertaining to data ownership, connectivity and regulatory support.
Data Origins and Access
In addition to FDR, AHM and EHM data, predictive maintenance can utilise information from other sources to present a robust picture of aircraft and engine performance.
Honeywell’s digital platform – Honeywell Forge – supports its Connected Maintenance application. Connected Maintenance analyses aircraft data in order to generate trends, maintenance alerts and proximity warnings for failures and faults. Honeywell Forge then allows customers to assimilate and distribute data effectively, which are key for predictive maintenance. “There are a variety of data sources used for predictive maintenance, namely quick access recorder (QAR) Data (or a subset thereof), ACMS Fault Messages, ACMS Performance Reports, and Maintenance Tech Logs,” describes Josh Melin, product line director for Honeywell Forge Connected Maintenance at Honeywell. “The richest data set is direct sensor data from the 717 bus or 429 buses which can be pulled from the QAR, or tapped directly from the bus using wireless enablers. These can be installed on the aircraft.”
While wireless enablers can simplify data flow for airlines, Melin adds that data can be extracted in other ways for the operator, with no need for aircraft modification or retrofits. “We do find, however, that if data is not collected regularly, the value of predictive maintenance solutions is lower, because predictive maintenance relies on regular data feeds to predict failures,” continues Melin. “Furthermore, it is important that the airlines owns the data it generates, and can decide which elements to share and withhold. So Honeywell actually does not need a full set of QAR data to create a predictive solution, in fact, we only need a subset of data labels from the 717 bus which we can provide as a list to the airline.” Melin adds that Honeywell can offer wireless enablers to the airline which can tap the 717 bus and pull only the exact parameters needed to provide the service the airline requests.
If an airline has issues pertaining to cost, data ownership or distribution, Melin explains that Honeywell does not need to collect all aircraft data in order to provide a predictive solution. “Honeywell Forge has airlines providing everything from ACMS Performance Reports, QAR data, to Maintenance Tech Logs in order to formulate their solution. While all the data sets listed are ideal, it’s possible to get started with just a subset of data, such as ACMS data and then as ROI is established the data set can be expanded,” he adds. The solution started as a tool to analyze data coming from thousands of Honeywell APUs. In 30 years, just one model of Honeywell APU has amassed over 100 million hours of service data; an ideal starting point for predictive analytics involving the complex systems which make up the APU.
Saravanan Rajarajan (Saran), Associate Director for Aviation Practice at Ramco Systems explains that maintenance-related data on the Components / Aircraft recorded in their MRO platforms provide another data stream for predictive maintenance. “Non-routines, removals / NFF / minimum equipment list (MEL) occurrences and Operator Maintenance programs all enhance predictive data analytics,” he says. “Analysing both the operational data from the sensors and the MRO data is key for high accuracy.”
Due to the data now available from connected aircraft, Sander de Bree of Exsyn Aviation Solutions adds that operators can now go further than traditional maintenance and health data, to boost predictive and analytical capabilities. “Non-aircraft related data such as weather information and airport data are important data-sources to be used in predictive maintenance algorithms,” he says. “These can be used to detect the impact of operational conditions (such as dry or humid operations) on component health. Additionally maintenance data from MRO’s needs to be used to report back any failure data to an operator’s prediction models.”
Data platforms and advanced analytical capabilities aside, there is one digital tool that a growing number of operators use today: the electronic techlog (ETL). It was the implementation of this device for recording faults that gave rise to the potential for predictive maintenance to flourish. It is also the primary interface between operational and maintenance data; an area where data can become disconnected.
“Data for predictive maintenance is critical, as there are so many areas in which it can be exploited—if it can be collected,” explains Elliott. “Think about a paper technical logbook on the plane, which is only accessible by a single person at a time. Handwritten entries cannot be used in analytics, and cannot be mined for information.
“An electronic, connected logbook can be used by multiple people at the same time,” continues Elliott. “A mechanic can see what faults are on the aircraft, and arrange for proper parts and tools for arrival at the aircraft. And, of course, that digital data can be aggregated and mined. The Internet of Things (IoT) will also help, with sensors being used to measure and collect data.
Digital twins are one industry development linked inherently to predictive maintenance, and applications of the technology are becoming more prevalent. For example GE has helped develop a digital twin for an aircraft’s landing gear. “In this last scenario, sensors placed on typical landing gear failure points, such as hydraulic pressure and brake temperature, provide real-time data to help predict early malfunctions or diagnose the remaining lifecycle of the landing gear,” adds Elliott.
Preventative vs. Predictive Maintenance
There are two core approaches to data-based maintenance, each geared towards different connected capabilities of aircraft or component. For instance, an A320 Classic aircraft will not transmit the same level of operational data as the A320neo; therefore maintenance strategies are different.
Preventative maintenance relies more on ‘trend monitoring’; trying to prevent a fault from being flagged by a line maintenance team by removing a component in the next scheduled maintenance event. The onus is less on the data being transmitted ‘that minute’, or the condition of a specific serial number, but rather taking an intelligent look at historical patterns across a fleet with that part installed, and determining based on age and hours or cycles when that part should be removed for inspection. But is preventative maintenance less dynamic or effective than predictive maintenance? “Preventative maintenance is an age-based maintenance philosophy, not taking into account actual condition of systems & components,” explains Sander de Bree, founder of EXSYN Aviation Solutions. “Predictive maintenance aims to use the actual calculated condition of components (based on operational usage) to serve as triggers for maintenance requirements.”
“Effectiveness of the predictive maintenance (over preventative) lies in its ability to leverage the historical data alongside live operational data,” explains Saran. “This is purely aided by the latest developments on processing the high volume of dynamic data feeds and analysing with sophisticated statistical tools. Because preventive maintenance relies only on historical data it is less effective.” Moreover, the age-based approach often leads to parts being removed ahead of time; meaning ‘wasted’ time remaining on the part if not re-installed.
There are instances where preventative maintenance is more appropriate for operators. “It is a good option in the absence of insights into the actual condition of a component/system,” describes Melin. “But as those insights become available, moving from preventative to predictive can ensure that maintenance actions be prescribed to exactly what maintenance action is required to remedy the current issues and at the right time.”
Data Hurdles in Maintenance
One of the main hurdles preventing operators from investing fully in predictive maintenance initiatives is the data itself – the completeness of it, and the ability to synch data from different sources, departments and formats.
Melin of Honeywell states two primary hurdles that preventing airlines realising predictive maintenance potential. “Some Airlines have a wait-and-see approach to data sharing. This is understandable but unless it’s shared, it is difficult for a software provider to demonstrate potential,” he explains. “Moreover, airlines’ traditional decision-making processes are tough for the software, applications and platforms that can harness and interpret data.”
“Airlines should be in full control of their operational data and be able to share it with their partners like MROs for example,” elaborates Martens. “We doubt that the approach of certain OEMs to restrict operational data access and control will prevail, since all airlines have a strategic interest to control their data.”
“Feedback from component shops on the actual health of components once removed from the aircraft based on prediction models is another hurdle,” adds de Bree. “This information is not readily available to airlines either because they are in a parts pool programme, or have components contracts based on time on wing (power by the hour). For the latter, there is an economical incentive to classify parts removed based on predictions as no-fault-found (NFF). After all, the part did not fail on wing ‘yet’.
“Also, many airlines are looking into predictive maintenance; some with OEM’s, some independently. Currently it seems a race for the best possible algorithm and platform, meaning each initiative is siloed. To make predictive maintenance work we need OEMs & local CAA’s to approve changes to the MPD, airlines to make available operational data, MRO’s to make available maintenance records and solution providers to provide algorithms and calculations,” says de Bree.
Data Platforms & Infrastructure
One way to connect data from different applications and departments is via a data platform; a repository that can exchange information between applications and systems – for instance between an ETL and an operator’s M&E system. “The most data-driven often work with a provider that can cover their entire fleet,” says Melin, “which for many airlines consists of multiple aircraft types from multiple aircraft OEMs.”
“The responsibility for the maintenance of an airline is of the operator and its CAMO and not the OEM’s expertise,” explains Martens. “More airlines realise the potential of digital solutions and the requirement to adapt these solutions to the specific needs of their fleet and operations. Open digital platforms like AVIATAR enable operators to provide digital interfaces to MRO’s and other players in the market, who help them in maintaining their fleet.”
Elliott explains that airlines are starting to work on their own data platforms to get in on the benefits of sharing engineering data. These platforms were initially pioneered by airframe and engine original equipment manufacturer (OEMs) in order to support OEM-developed applications that are often chosen when operators order new aircraft types. Furthermore, OEM platforms benefit from having access to global customer data, thereby bolstering their analytical data provisions. “Airbus launched its cloud-based data platform, Skywise, in 2017 which collects data such as work orders, spares consumption and flight schedules from multiple sources across the industry for MRO operators to perform predictive and preventative maintenance. Early adopters included easyJet, Air Asia, Emirates and Delta Airlines, all of which are using the platform for predictive maintenance,” says Elliott.
Not all data is so readily available. “Sensor data from aircraft is still “locked-up” with the OEM’s as it mostly uses OEM IP in order to be decoded,” highlights de Bree. “You do see independent flight data acquisition avionics becoming available to work around this issue.”
According to Ramco, an M&E MRO system provides the foundational block to support predictive maintenance capabilities. “With the recent advancement on data processing power and ability to store TB of data , the key challenge is agility to connect to the external eco systems and leverage with inhouse data for prediction,” adds Saran. “API based protocol is essential for the organization to achieve software collaboration and encourage data sharing.”
“The number of airlines using the latest big data solutions is limited but growing fast,” adds Martens. “Many airlines are looking at the solutions, but the offerings of real predictive maintenance are limited. Many offerings just provide digital results without direct connection to maintenance actions. Connecting a data platform such as AVIATAR with different M&E System vendors like AMOS or TRAX and other airline IT providers such as Netline help to create the necessary solution.”
Much of the data required for predictive maintenance suggests a high level of data transmission; but to what extent does this need to be performed in-flight, which incurs a great cost? “Data synchronized in flight is mainly linked to EHM/AHM parameters or ACARS data and contain fault messages once a situation has already occurred,” explains de Bree. For instance, while LHT’s AVIATAR ingests data from multiple data sources in-flight and on the ground; the extent of this is defined by the operator. Engines and other components can send data via aircraft interfaces. “In many cases data such as fault messages is sent via ACARS in flight and Wifi/GSM on the ground, but this is up to the airline to define it, based on requirements,” says Martens. “For engineers it can be very helpful to receive these while the aircraft is inflight, since manpower, tools and spare parts can be ordered ahead of landing. This helps operators to save costs by avoiding AOGs (aircraft-on-ground).”
In general, airlines transmit the bulk of their data once on the ground, saving cost. “Honeywell Forge Connected Maintenance has been able to predict component failures days and weeks in advance,” says Melin. “The process of detecting an impending failure and alerting the relevant maintenance engineers can be automated. Typically, the process of then deciding when to complete that maintenance action and submitting the work order is still manual so that the airline can remain in control of that final decision.” Airlines can reduce operational disruptions with the current generation of systems, transmitting data on the ground. Honeywell believes that in future there will be a shift towards transmission of a subset of key data during flights, utlilizing existing satcom connectivity, in order predict a wider set of ATA chapters with high accuracy.
The ETL can provide the means to notify of faults inflight. “An effective ETL allows pilots to communicate with the whole team involved in flying an aircraft on the day of operations—spanning mechanics, maintenance control centres, engineers and more,” continues Elliott. “Once a pilot is flying, if they encounter any problems, they can log the fault in the electronic technical logbook app. On aircraft with in-flight internet connectivity the maintenance organization will receive a push notification in real time outlining the fault and start preparing work orders and parts, so they are ready to address it the moment the aircraft lands. From a more preventative perspective, on aircraft without in-flight connectivity, an electronic technical logbook can push updates to the maintenance department when the aircraft lands.”
IFS’s customer, China Airlines, has been utilising IFS Maintenix to optimize data sharing of real-time management of line and heavy maintenance events, as well as data capture at the point of maintenance across the airline and its subsidiaries. This included expanding third-party MRO services for the airline’s customers. “In addition to reducing operating costs by $3.5 million, IFS Maintenix has helped China Airlines significantly decrease its aircraft layover due to more efficient scheduled and unscheduled line maintenance,” adds Elliott. “This means that its aircraft spend more time in the air and less time in the hangar.”
“While real-time data transmission in-air is a benefit for EHM/AHM fault messages, for predictive maintenance trend calculations an offline datafeed is sufficient,” agrees de Bree. “In terms of wider infrastructure, server capacity is going to be critical to ensure timely processing of data and visualizing outcomes. As an airline you don’t want to wait 4 hours for a calculation to finish prior to giving indications on component condition.”
Unnecessary Part Removal
Removing parts if a fault arises is the traditional business model of the industry, and reactive rather than proactive.
An issue of predictive maintenance lies in the clinical and rigid nature of data if intelligent parameters aren’t built in; we run the risk of incorrect forecasts and erroneous ‘fault’ messages. For instance, if an operator forecasts that a component will fail within 200 hours, based on historical behaviour, it might schedule removal to prevent failure in operation. However, upon removal the part tests no fault found (NFF), costing unnecessary time and money for the operator.
How do we prevent parts being taken off for testing, only to be NFF? And is there risk of oversensitive data, causing unnecessary time off wing for testing? “No algorithm can be 100% reliable,” says de Bree. “The key is feeding back MRO shop data of actual components removed based on prediction models. This is the only real evidence if a failure of that component was imminent. Feeding back such data will make models more reliable.”
“Parts pre-emptively removed need to undergo longer troubleshooting time due to non -availability of fault code or maintenance findings,” says Saran of Ramco. “High sensitivity on the Part removals and longer turnaround time (TAT) will also lead to increased investment in float for airlines. The sensitivity can only be reduced over the time by a continuous closed loop data flow on maintenance findings on the removed part back into the prediction algorithms. It is also imperative that parts are sent to shops with the data leading the predicted fault which reduces the troubleshooting and turnaround time.”
“Ultimately, condition-based removal avoids costly AOGs, improves the fleet’s reliability and ensures high rates of passenger satisfaction,” counters Martens. “If MRO providers don’t know the predictive reason of the removal, it may lead to NFF, but the operator will save on operational cost. An AOG at the wrong location can cost more than €100,000.
“There are several examples where predictors are used successfully. The parameter of these parts are continuously tracked and analyzed, resulting in a trigger/information when the fill level/temperature/pressure parameters start to shift without causing a real aircraft failure. This helps us to change or service these parts preventively to avoid AOGs. Very often the work order can be transferred automatically into the maintenance information system,” adds Martens.
According to IFS, Rolls-Royce has disclosed high expectations for the accuracy of its own predictive analytics strategy. The OEM targets a 100 percent success rate in terms of ensuring they never miss something they are looking for, at the same time as zero false predictions including NFFs.
Predictive Maintenance vs. MSG-3
What effect might predictive maintenance have on scheduled maintenance? For instance, will airlines that maximise its potential still follow an MSG-lead maintenance programme? Or will we see an evolution away from this and scheduled shop visits? “Going forward condition-based maintenance will be used more often, but requires close collaboration between the authorities, operators, MROs and OEMs,” says Martens. “Predictive maintenance should result in less unscheduled, high priority repairs and eventually, we can make many checks obsolete because we calculate figures and probabilities per system which previously were checked manually.”
“Ultimately this might become a new maintenance standard, however no airline today is allowed by CAA regulation to deviate by themselves from the (MSG-based) approved maintenance program and OEM MPD,” explains de Bree. “As long as these are still leading, predictive maintenance initiatives can only impact on-condition components of an aircraft.”
“In the next few years, predictive maintenance can eliminate airline determined soft time maintenance intervals in order to optimize costs and efficiencies, however, it is less likely that predictive maintenance would be a substitute for hard time service requirements,” says Melin at Honeywell. “Ultimately, the change in maintenance practices must be spearheaded by airlines maintenance teams.”
Rolls-Royce is pioneering the concept of an adaptive and evolving maintenance programme, that can effectively go a step further than MSG-3 logic. In 2019 IFS partnered with the OEM to support its data exchange program with airline customers operating the Trent Engine family.
“The IFS Maintenix Aviation Analytics capability enables the automated provision of field data, which ensures that Rolls-Royce receives timely and accurate information on its Trent 1000, Trent XWB and Trent 7000 engines,” explains Elliott. “IFS Maintenix then acts as a gateway to automatically push maintenance program changes from Rolls-Royce back to the airline operators. As a result, life-limited engine part maintenance deadlines can be updated based on actual operating conditions and life consumed by each engine in use.”
Artificial Intelligence (AI)
AI is increasingly referred in conjunction with predictive maintenance. “The use of AI revolves around algorithms being used for predictive calculations to be become more reliable over time by themselves,” explains de Bree. “It allows systems to detect possible failures to monitor purely based on data supplied without any relation between parameters and component failure being known,” explains de Bree.
“AI is a valuable tool for analytics, along with machine learning and neural networks,” says Melin of Honeywell. “AI can be used to determine the state of a system (how it operates in given conditions) and then detect anomalies through time series data which can then be used to predict remaining life and recommend mitigation strategies. AI is different from the historically human-based maintenance systems in that it enables integration of contextual data as well as behavioural parameters of assets.”
In addition while AI can be used in predicting the item removal through predictive maintenance, it is also expected that it can offer additional services to increase the intelligence of the predictive maintenance solution, Saran of Ramco explains. “For instance, AI might also assist an M&E systems in suggesting part replacement options and other parts which might also be needed in replacement, therefore streamlining maintenance downtime. The confluence of Predictive maintenance, AI and Big data drives maximum benefit.”
For large-fleet commercial operators and aviation MRO providers alike, AI is now an essential tool. “Recent examples of airlines such as Delta, and MROs such as Lufthansa Industry Solutions working on adopting AI and machine learning (ML) into their aircraft maintenance strategies highlight the transition organizations are already making towards digital and predictive-focused maintenance strategies,” continues Elliott. “The reduced maintenance technician and engineering labour hours spent analysing data makes intelligent maintenance strategies particularly desirable.”
Want to learn more? Several presentations will take a deeper dive into predictive maintenance at Aerospace Tech Week. See page 43 for the full Show Guide.