Q&A
Condition Based Airline Fleet Maintenance explained in 7 questions

Current aircraft maintenance is carried out following two strategies:

  • Reactive, also known as ‘fix when it fails’. The disadvantage of this maintenance strategy is that it leads to unexpected maintenance that could cause delays for the passengers.
  • Preventive, also known as pre-scheduled or fixed-interval-maintenance. According to the maintenance planning document from the aircraft manufacturer parts are replaced after a defined number of flight hours, flight cycles or calendar days, whichever comes first. The disadvantage of this time-based maintenance strategy is that parts are replaced while still in good health. This leads to waste of time and resources, thus ‘over-maintenance’.

In this video, we explain the available aircraft maintenance strategies.

With Condition Based Maintenance the health of a component or structure is monitored and are only repaired when damaged or replaced when close to failure. The number of sensors present in a modern aircraft, the accessibility and fast communication of the vast data obtained from these sensors, and the increasing capability of data analytics create the ideal context for Condition Based Maintenance in the aviation industry as you can read in this article.

In ReMAP data collected from sensors from dissimilar aircraft systems and structures will be analysed through machine learning diagnostic and prognostic algorithms to create real-time adaptive maintenance plans. A fleet-level approach will be followed since the potential of CBM can only be achieved by monitoring the health of all aircraft in a fleet, managing maintenance resources and operational needs. The result is an Integrated Fleet Health Management (IFHM) solution that replaces fixed-interval inspections with adaptive condition-based interventions.

In this video, we explain the unique selling points of ReMAP.

According to ACARE, it is expected that by 2035 the Condition Based Maintenance philosophy will be accepted as a standard approach to monitor aircraft health and to plan aircraft maintenance. By 2050, all new aircraft will be designed for Condition Based Maintenance. CBM will result in a significant 40% reduction in Maintenance Repair & Overhaul (MRO) process time and costs, increase in aircraft availability, and maximization of asset utilization. As Ludovic Simon (Thales), Member of the ReMAP Advisory Board, states in his blog: “Now we must prove that condition-based maintenance is mature enough to implement into aviation

In this ReMAP welcome video, stakeholders explain the importance of CBM in aviation and the purpose of ReMAP.

Listen also to this interesting Podcast from Project Leader Bruno Santos with more details on ReMAP.

A generalized application of CBM is still far from feasible. Despite the existence of thousands of sensors in modern aircraft, the attempts, so far, have focused on system health diagnostics or prognostics health management for specific systems, mainly engines and auxiliary power units. Moreover, there is a lack of knowledge on how to incorporate these diagnostics and prognostics for efficient maintenance management. To fully exploit CBM benefits a systematic end-to-end approach is necessary: from sensor data to fleet maintenance. ReMAP does that. ReMAP research is split into the next technical working packages:

  1. Development of an open IT ecosystem of cloud services that will enable information load, data management, processing, visualization and sharing. Learn about its features in this video.
  2.  The procurement, development and integration of the most promising sensor technologies for damage monitoring in aeronautical composite structures (SHM), such as Piezo sensors and Lamb Wave Detection Systems LWDS. Watch our lamb waves detection system video at: https://h2020-remap.eu/lamb-waves-detection-system-lwds-cedrat-technologies/
  3. Development of Structural Health Management (SHM) Diagnostics and Remaining Useful Life Prognostics. Discover the ReMAP SHM-approach in this video.
  4. Development of the core analytics technology chain for system and component level diagnostics, prognostics and health management (PHM).
  5. Development of the Maintenance Decision Support Tool that delivers the adaptive maintenance plans
  6. Assessment of the safety of the condition-based maintenance technologies in ReMAP. Discover more here.
  7. Demonstration test in a relevant environment (KLM and KLM City Hopper). Discover more here.
Find all technical details and the research progress in ReMAP’s bi-annual newsletter. You can download them here: https://h2020-remap.eu/news/ ReMAP scientific results can be downloaded here: https://h2020-remap.eu/documents/

When a new technology, such as CBM using artificial intelligence, is being introduced, one has to first identify the effects of this technology on the safety of people, goods and environment. This is called ‘hazard identification’. “According to the regulatory framework defined by EASA and FAA, once you are aware of potential hazards, the next step consists in finding solutions to deal with these hazards and assessing the efficiency of the proposed solutions”, says ReMAP-partner Pierre Bieber (ONERA) who shares his investigation in his blog.

All defined hazards are translated into models for aircraft maintenance, with and without CBM incorporated by PhD Juseong Lee, his supervisor Mihaela Mitici (TU Delft), and Floris Freeman (KLM). From that point, the impact on aircraft safety when CBM is incorporated is assessed.


These field tests will focus on the KLM fleet of Boeing 787 airplanes (twenty per 2020) and KLM City Hopper Embraer 175 (ten per 2020).

Data sharing is one of the key aspects to incorporate CBM. With CBM data acquired from sensors is used to calculate the remaining useful life of components, through a model-based approach. To create such a model huge amounts of data have to be gathered. To increase the accuracy of the model and to create a company independent IFHM-solution, multiple airlines need to participate and share data. This can be challenging due to data privacy and data ownership issues.

In ReMAP, strategies are explored to train and share CBM models without the need to share the data. Federated Learning is a new technique that has the potential to solve the data sharing problem among companies and stakeholders while offering data-private model learning. Advanced settings increase the convergence speed hence models can be generated faster by reducing the amount of data needed to be sent to the server. 

Watch the ReMAP video in which we explain the architecture of the IT-platform that offers the Integrated Fleet Health Management-solution using Federated Learning.

Considering an airline’s profitability goal, the objective of the maintenance organization is to contribute to maximizing the airline fleet’s earning potential: maximizing the fleet availability and dispatch reliability (the proportion of planned flights that can be realised), at minimum cost. CBM has the potential to contribute to this but estimating that actual potential is challenging. “Nobody in the industry is able yet to perform a robust economic analysis, simply because there is no sufficient and extensive CBM experience and data to support such analysis”, states Roberto Hirschman (Embraer), Member of the ReMAP Advisory Board in his blog.“ That’s why ReMAP is an excellent initiative and opportunity to exercise CBM addressing all aspect of its implementation and adoption.” But estimating the ultimate benefit that CBM technologies will provide in an airline environment can be challenging, as Floris Freeman of KLM states. An assessment is needed on what is realistic in terms of technology performance, and what benefits are to be expected (download ReMAP-deliverable)


ReMAP will test its IFHM solution in an unprecedented 6-month operational demonstration, involving more than 12 systems in two different aircraft fleet of KLM and KLM City Hopper (twenty Boeing 787 airplanes and ten Embraer 175). Research will be done on the expected impact from CBM on the short- and long-term on fleet availability, maintenance cost and system complexity and weight benefits. Complementary the potential of CBM to deliver operator benefits will be examined.

For structures, health prognostics algorithms will be demonstrated in a laboratory setting. Representative structural composite subcomponents are subjected to a spectrum of fatigue loading. Part of the demonstration will estimate the structural weight benefits out of the SHM technology adoption in primary structures.

The ambition for both demonstrations is to prove that, under the current regulatory framework, ReMAP’s IFHM solution will lead to:

  • a 4.5% reduction of direct maintenance costs (110 thousand euros less per year for a long-haul aircraft);
  • a 3 to 7% weight reduction of structural composite components; and
  • a 10% missed failures reduction, reducing the need for systems redundancy and reducing complexity.

Current aircraft maintenance is carried out following two strategies:

  • Reactive, also known as ‘fix when it fails’. The disadvantage of this maintenance strategy is that it leads to unexpected maintenance that could cause delays for the passengers.
  • Preventive, also known as pre-scheduled or fixed-interval-maintenance. According to the maintenance planning document from the aircraft manufacturer parts are replaced after a defined number of flight hours, flight cycles or calendar days, whichever comes first. The disadvantage of this time-based maintenance strategy is that parts are replaced while still in good health. This leads to waste of time and resources, thus ‘over-maintenance’.

In this video, we explain the available aircraft maintenance strategies.

With Condition Based Maintenance the health of a component or structure is monitored and are only repaired when damaged or replaced when close to failure. The number of sensors present in a modern aircraft, the accessibility and fast communication of the vast data obtained from these sensors, and the increasing capability of data analytics create the ideal context for Condition Based Maintenance in the aviation industry as you can read in this article.

In ReMAP data collected from sensors from dissimilar aircraft systems and structures will be analysed through machine learning diagnostic and prognostic algorithms to create real-time adaptive maintenance plans. A fleet-level approach will be followed since the potential of CBM can only be achieved by monitoring the health of all aircraft in a fleet, managing maintenance resources and operational needs. The result is an Integrated Fleet Health Management (IFHM) solution that replaces fixed-interval inspections with adaptive condition-based interventions.
In this video, we explain the unique selling points of ReMAP.

According to ACARE, it is expected that by 2035 the Condition Based Maintenance philosophy will be accepted as a standard approach to monitor aircraft health and to plan aircraft maintenance. By 2050, all new aircraft will be designed for Condition Based Maintenance. CBM will result in a significant 40% reduction in Maintenance Repair & Overhaul (MRO) process time and costs, increase in aircraft availability, and maximization of asset utilization. As Ludovic Simon (Thales), Member of the ReMAP Advisory Board, states in his blog: “Now we must prove that condition-based maintenance is mature enough to implement into aviation
In this ReMAP welcome video, stakeholders explain the importance of CBM in aviation and the purpose of ReMAP.

Listen also to this interesting Podcast from Project Leader Bruno Santos with more details on ReMAP.

A generalized application of CBM is still far from feasible. Despite the existence of thousands of sensors in modern aircraft, the attempts, so far, have focused on system health diagnostics or prognostics health management for specific systems, mainly engines and auxiliary power units. Moreover, there is a lack of knowledge on how to incorporate these diagnostics and prognostics for efficient maintenance management. To fully exploit CBM benefits a systematic end-to-end approach is necessary: from sensor data to fleet maintenance. ReMAP does that. ReMAP research is split into the next technical working packages:
  1. Development of an open IT ecosystem of cloud services that will enable information load, data management, processing, visualization and sharing. Learn about its features in this video.
  2.  The procurement, development and integration of the most promising sensor technologies for damage monitoring in aeronautical composite structures (SHM), such as Piezo sensors and Lamb Wave Detection Systems LWDS. Watch our lamb waves detection system video at: https://h2020-remap.eu/lamb-waves-detection-system-lwds-cedrat-technologies/
  3. Development of Structural Health Management (SHM) Diagnostics and Remaining Useful Life Prognostics. Discover the ReMAP SHM-approach in this video.
  4. Development of the core analytics technology chain for system and component level diagnostics, prognostics and health management (PHM).
  5. Development of the Maintenance Decision Support Tool that delivers the adaptive maintenance plans
  6. Assessment of the safety of the condition-based maintenance technologies in ReMAP. Discover more here.
  7. Demonstration test in a relevant environment (KLM and KLM City Hopper). Discover more here.
Find all technical details and the research progress in ReMAP’s bi-annual newsletter. You can download them here: https://h2020-remap.eu/news/ ReMAP scientific results can be downloaded here: https://h2020-remap.eu/documents/

When a new technology, such as CBM using artificial intelligence, is being introduced, one has to first identify the effects of this technology on the safety of people, goods and environment. This is called ‘hazard identification’. “According to the regulatory framework defined by EASA and FAA, once you are aware of potential hazards, the next step consists in finding solutions to deal with these hazards and assessing the efficiency of the proposed solutions”, says ReMAP-partner Pierre Bieber (ONERA) who shares his investigation in his blog.

All defined hazards are translated into models for aircraft maintenance, with and without CBM incorporated by PhD Juseong Lee, his supervisor Mihaela Mitici (TU Delft), and Floris Freeman (KLM). From that point, the impact on aircraft safety when CBM is incorporated is assessed.

These field tests will focus on the KLM fleet of Boeing 787 airplanes (twenty per 2020) and KLM City Hopper Embraer 175 (ten per 2020).

Data sharing is one of the key aspects to incorporate CBM. With CBM data acquired from sensors is used to calculate the remaining useful life of components, through a model-based approach. To create such a model huge amounts of data have to be gathered. To increase the accuracy of the model and to create a company independent IFHM-solution, multiple airlines need to participate and share data. This can be challenging due to data privacy and data ownership issues.

In ReMAP, strategies are explored to train and share CBM models without the need to share the data. Federated Learning is a new technique that has the potential to solve the data sharing problem among companies and stakeholders while offering data-private model learning. Advanced settings increase the convergence speed hence models can be generated faster by reducing the amount of data needed to be sent to the server. 

Watch the ReMAP video in which we explain the architecture of the IT-platform that offers the Integrated Fleet Health Management-solution using Federated Learning.

Considering an airline’s profitability goal, the objective of the maintenance organization is to contribute to maximizing the airline fleet’s earning potential: maximizing the fleet availability and dispatch reliability (the proportion of planned flights that can be realised), at minimum cost. CBM has the potential to contribute to this but estimating that actual potential is challenging. “Nobody in the industry is able yet to perform a robust economic analysis, simply because there is no sufficient and extensive CBM experience and data to support such analysis”, states Roberto Hirschman (Embraer), Member of the ReMAP Advisory Board in his blog.“ That’s why ReMAP is an excellent initiative and opportunity to exercise CBM addressing all aspect of its implementation and adoption.” But estimating the ultimate benefit that CBM technologies will provide in an airline environment can be challenging, as Floris Freeman of KLM states. An assessment is needed on what is realistic in terms of technology performance, and what benefits are to be expected (download ReMAP-deliverable)


ReMAP will test its IFHM solution in an unprecedented 6-month operational demonstration, involving more than 12 systems in two different aircraft fleet of KLM and KLM City Hopper (twenty Boeing 787 airplanes and ten Embraer 175). Research will be done on the expected impact from CBM on the short- and long-term on fleet availability, maintenance cost and system complexity and weight benefits. Complementary the potential of CBM to deliver operator benefits will be examined.

For structures, health prognostics algorithms will be demonstrated in a laboratory setting. Representative structural composite subcomponents are subjected to a spectrum of fatigue loading. Part of the demonstration will estimate the structural weight benefits out of the SHM technology adoption in primary structures.

The ambition for both demonstrations is to prove that, under the current regulatory framework, ReMAP’s IFHM solution will lead to:

    • a 4.5% reduction of direct maintenance costs (110 thousand euros less per year for a long-haul aircraft);
    • a 3 to 7% weight reduction of structural composite components; and
    • a 10% missed failures reduction, reducing the need for systems redundancy and reducing complexity.
The following questions were made by the participants of our first webinar.

You can watch (or rewatch) the video here. We will continue answering questions, so, if you have any query, please use the messaging tool at the bottom of the page. 

With CBM, faults won’t occur less frequently (we are not adjusting the inherent reliability of the fleet). However, these faults won’t lead to a disruptive failure as often. The idea to keep the algorithms’ valid’ over time is to critically assess the condition once the component has been brought to the repair shop and see if there was indeed a fault.

Federated Learning is a technology that allows training models without the need to send private data. The model’s training is done locally by the clients that hold the data, and only the updates of the model are sent to a server. The server gathers all the updates of the multiple clients and aggregates them to generate a new model. For example, different airline companies can create better health prediction models for their systems without relinquishing valuable private data.

The model is trained at the locations of the airline(s). Of course, the airline is free to share a dataset, but that isn’t needed in this ‘federated analytics’ framework. For practical purposes, the model builder may have a small subset of the data to speed up development. But actual training on the full dataset will be done iteratively in the Nodes.

These tasks would be pushed to the maintenance system of the airline. However, in this proof-of-concept version of the platform, that has not yet been implemented. This means that the tasks (possibly after an optimal schedule was computed) need to be copied to the maintenance system manually.

The platform keeps a record of all the installed components in the aircraft, including their installation dates. The platform also holds average utilisations for each aircraft type. So yes, RUL estimators can make use of age (hours/cycles/days) in addition to sensor data

At this point, it does not have an API with ECTM systems. In fact, the scope of this project is focused on systems and structures (not engines). Connecting ECTMS systems could be possible though, by pushing diagnostic output from ECTM to the platform. There, it can be picked up and used in one of the scheduling models that is installed in the node.

The platform accepts sensor data from any source, either from a customised retrofitted sensor or from the already available sensors from the aircraft. In our demonstration, we rely mostly on the latter, which is 1Hz CPL data.

For the 6-months demonstration, we cannot and will not deviate from the approved maintenance program. Hence the preventive tasks originating from this demonstration are performed ‘on top of’ the maintenance program. In the future, on-condition tasks may substitute time-based tasks from the maintenance program.

The airline provides a continuous export of all its open tasks, including those from the maintenance program. Task attributes include (amongst others) tools & material required, zones, access panels, ETOPS restrictions, labour requirements, etc. The scheduling models use these data to come up with an optimal solution.

In ReMAP, we have developed models for about 12 systems from 4 aircraft types. In the live demonstration, we demonstrate 7 systems from 2 fleets. The data is collected from sensors present in aircraft from these fleets.

Ultimately, an efficient algorithm should either i) improve aircraft availability, ii) improve dispatch reliability, or iii) reduce maintenance costs. On a lower level, we use prognostic performance indicators like ROC curves in combination with the anticipated operational benefit from avoiding an unscheduled failure.

Yes. The airline may also have developer rights.

There are similarities, yes. However, in ReMAP, the airline keeps full ownership and control over its data. Learning is facilitated by federated analytics. ReMAP also integrated prognostic output into scheduling optimisation. ReMAP is also open-source and independent from OEMs. The ecosystem can be populated with models or algorithms developed by third parties.

In ReMAP, an extensive cost-benefit analysis is done in three steps. Firstly, current maintenance procedures are analysed in a benchmarking exercise. This will stipulate where CBM will have most value. Secondly, feasible prognostic and scheduling performance indicators will be found during a 6-months, real-life operational demonstration. Thirdly, future (cost-) benefits will be estimated by running various scenarios in an MRO/Airline simulator. These results will be made available in May 2022.

There are two OEMs involved in this project (Embraer and Collins). Whether they will get access to the model is up to the owner of the model. The OEMs are welcome to build and test their models on airline data.

Currently, we are making tests with flight data provided by KLM, which amounts to around 1 GB per flight. Internally, the system is based on a MongoDB database for storing the data, and during the demonstration phase, we intend to increase the amount of data collected/stored to test the possible limitations and bottlenecks of the current implementation.

Yes, ReMAP comes with its own data model. From here, aircraft taxonomy is translated to the ReMAP data model, to ensure compatibility amongst airlines.

Prognostic models have inherent uncertainty, which can be reduced by training on more run-to-failure data. Economic break-even points can be calculated in multiple ways, some of which are published by ReMAP research (example: https://doi.org/10.36001/phme.2021.v6i1.2865)

This is currently outside of the ReMAP scope. But we foresee that pre-computation of diagnosis can be part of CBM solutions. For this reason, we have developed edge solutions in which a diagnostic algorithm is run with the limited data available in flight. This has been developed and tested in a virtual environment.

This is a topic of ongoing work in our Safety-Risk work package. During the demonstration, we focus on economic tasks only. Nevertheless, since we are only addressing economic-related tasks (not safety-critical), we are not deviating from the approved maintenance program, and the solutions run outside the aircraft. These solutions can, in principle, run without being certified.

Yes, the platform tracks the expected availability date of required material (including spare components).

In principle, not. The data ingestion process has been implemented with a queueing system. Thus, data sources are ingested one after the other.

ReMap IT platform has been implemented using several open-source software components. However, it is currently under discussion how it will be released (open-source or not, and under which conditions). This will be addressed in the final exploitation plan.

We are working with and discussing our results with partners involved in the MSG. We hope that our effort can contribute to the discussion of best practices and requirements for the implementation of CBM in aviation.

With CBM, faults won’t occur less frequently (we are not adjusting the inherent reliability of the fleet). However, these faults won’t lead to a disruptive failure as often. The idea to keep the algorithms’ valid’ over time is to critically assess the condition once the component has been brought to the repair shop and see if there was indeed a fault.

Federated Learning is a technology that allows training models without the need to send private data. The model’s training is done locally by the clients that hold the data, and only the updates of the model are sent to a server. The server gathers all the updates of the multiple clients and aggregates them to generate a new model. For example, different airline companies can create better health prediction models for their systems without relinquishing valuable private data.

The model is trained at the locations of the airline(s). Of course, the airline is free to share a dataset, but that isn’t needed in this ‘federated analytics’ framework. For practical purposes, the model builder may have a small subset of the data to speed up development. But actual training on the full dataset will be done iteratively in the Nodes.

These tasks would be pushed to the maintenance system of the airline. However, in this proof-of-concept version of the platform, that has not yet been implemented. This means that the tasks (possibly after an optimal schedule was computed) need to be copied to the maintenance system manually.

The platform keeps a record of all the installed components in the aircraft, including their installation dates. The platform also holds average utilisations for each aircraft type. So yes, RUL estimators can make use of age (hours/cycles/days) in addition to sensor data

At this point, it does not have an API with ECTM systems. In fact, the scope of this project is focused on systems and structures (not engines). Connecting ECTMS systems could be possible though, by pushing diagnostic output from ECTM to the platform. There, it can be picked up and used in one of the scheduling models that is installed in the node.

The platform accepts sensor data from any source, either from a customised retrofitted sensor or from the already available sensors from the aircraft. In our demonstration, we rely mostly on the latter, which is 1Hz CPL data.

For the 6-months demonstration, we cannot and will not deviate from the approved maintenance program. Hence the preventive tasks originating from this demonstration are performed ‘on top of’ the maintenance program. In the future, on-condition tasks may substitute time-based tasks from the maintenance program.

The airline provides a continuous export of all its open tasks, including those from the maintenance program. Task attributes include (amongst others) tools & material required, zones, access panels, ETOPS restrictions, labour requirements, etc. The scheduling models use these data to come up with an optimal solution.

In ReMAP, we have developed models for about 12 systems from 4 aircraft types. In the live demonstration, we demonstrate 7 systems from 2 fleets. The data is collected from sensors present in aircraft from these fleets.

Ultimately, an efficient algorithm should either i) improve aircraft availability, ii) improve dispatch reliability, or iii) reduce maintenance costs. On a lower level, we use prognostic performance indicators like ROC curves in combination with the anticipated operational benefit from avoiding an unscheduled failure.

Yes. The airline may also have developer rights.

There are similarities, yes. However, in ReMAP, the airline keeps full ownership and control over its data. Learning is facilitated by federated analytics. ReMAP also integrated prognostic output into scheduling optimisation. ReMAP is also open-source and independent from OEMs. The ecosystem can be populated with models or algorithms developed by third parties.

In ReMAP, an extensive cost-benefit analysis is done in three steps. Firstly, current maintenance procedures are analysed in a benchmarking exercise. This will stipulate where CBM will have most value. Secondly, feasible prognostic and scheduling performance indicators will be found during a 6-months, real-life operational demonstration. Thirdly, future (cost-) benefits will be estimated by running various scenarios in an MRO/Airline simulator. These results will be made available in May 2022.

There are two OEMs involved in this project (Embraer and Collins). Whether they will get access to the model is up to the owner of the model. The OEMs are welcome to build and test their models on airline data.

Currently, we are making tests with flight data provided by KLM, which amounts to around 1 GB per flight. Internally, the system is based on a MongoDB database for storing the data, and during the demonstration phase, we intend to increase the amount of data collected/stored to test the possible limitations and bottlenecks of the current implementation.

Yes, ReMAP comes with its own data model. From here, aircraft taxonomy is translated to the ReMAP data model, to ensure compatibility amongst airlines.

Prognostic models have inherent uncertainty, which can be reduced by training on more run-to-failure data. Economic break-even points can be calculated in multiple ways, some of which are published by ReMAP research (example: https://doi.org/10.36001/phme.2021.v6i1.2865)

This is currently outside of the ReMAP scope. But we foresee that pre-computation of diagnosis can be part of CBM solutions. For this reason, we have developed edge solutions in which a diagnostic algorithm is run with the limited data available in flight. This has been developed and tested in a virtual environment.

This is a topic of ongoing work in our Safety-Risk work package. During the demonstration, we focus on economic tasks only. Nevertheless, since we are only addressing economic-related tasks (not safety-critical), we are not deviating from the approved maintenance program, and the solutions run outside the aircraft. These solutions can, in principle, run without being certified.

Yes, the platform tracks the expected availability date of required material (including spare components).

In principle, not. The data ingestion process has been implemented with a queueing system. Thus, data sources are ingested one after the other.

ReMap IT platform has been implemented using several open-source software components. However, it is currently under discussion how it will be released (open-source or not, and under which conditions). This will be addressed in the final exploitation plan.

We are working with and discussing our results with partners involved in the MSG. We hope that our effort can contribute to the discussion of best practices and requirements for the implementation of CBM in aviation.

This website uses cookies to ensure you get the best experience on our website.