Engine Prophets

Predictive Engine Maintenance is Turning Data Into Better Maintenance Decisions

Global GDP growth, rising urban populations and increasing middle classes are driving the upwards trajectory of demand for air travel. Over the next two decades passenger numbers are predicted to grow by up to 4% annually resulting in a doubling of global air traffic, ultimately reaching about 10 billion passengers per year.

Satisfying this appetite for wanderlust will require the global aircraft fleet to grow by around 42,000 new deliveries, with approximately 81% being new-generation single aisle aircraft.

But there’s chokepoints in the supply chain that are affecting airline operations. At the time of writing, the aircraft order backlog currently stands at over 18,000 and the average fleet age has reached a record 15.2 years, as airlines retain aging aircraft as a hedge against delivery uncertainty. The result is a shortage of 5,000 more fuel efficient replacement aircraft that the industry had counted on.

Willie Walsh, the then outgoing Director General of IATA told attendees to its 82nd AGM in Brazil in June this year, that this ‘ghost fleet’ means missed efficiency gains, higher lease rates and increased maintenance costs.

“Why? Because the aerospace supply chain continues its failure to deliver aircraft and engines as promised,” Walsh pointedly remarked.

Single-aisle hold up
IATA’s June 2026 study, Single Aisle Aircraft Engines MRO: Strategic Levers to Address Supply Chain Challenges, conducted with Emerton, found that the bottleneck is particularly acute for the latest engine models equipping the new generation of single aisle aircraft: A320neo (powered by CFM’s LEAP-1A or Pratt & Whitney’s Geared Turbo Fan GTF engines), 737 MAX (LEAP-1B), and A220 (GTF engines).

The study further highlights that the single aisle aircraft market is in a transition phase between legacy aircraft (A320 powered by CFM56-5 or Pratt & Whitney’s V2500 engines and 737 NGs powered by CFM56-7 engines) and new-generation aircraft (A320neo powered by LEAP-1A or GTF, 737 MAX powered by LEAP-1B and Comac C919 powered by LEAP-1C).

At the end of 2024 over 23,000 CFM56 (introduced in the early 1980s) and 5,000 V2500 (introduced in the late 80s) were still in service, while around 7,000 LEAP1A (EIS of 2016) EIS) and 5,000 GTF engines (EIS early 2016) had already been delivered by this date.

However, a ramp-up in new aircraft delivery targets, combined with a series of engines design maturity and production quality issues have significantly reduced the engines Time On Wing (TOW). Indeed, the annual number of shop visits required through 2040 will increase further, reaching a peak of over 5,000 shop visits for the LEAP and 2,000 for the GTF respectively, according to the IATA/Emerton study.

While engine OEMs are expanding their overall shop-visit capacity, another lever to enhance engine service is the progress being made in maintenance strategies.

The promise of predictive
As Vikram Gupta, Founder and CEO of India-based Awiros explains, preventive maintenance follows predefined schedules, while condition-based maintenance relies on operational parameters and sensor readings to determine when maintenance is required. For him, predictive maintenance represents the next evolution.

“It combines historical inspection data, engineering knowledge, operational context and artificial intelligence to anticipate degradation trends before they become operational concerns,” he tells Aerospace Innovations.

Aero NextGen’s Founder and CEO, Monica Badra expands. “In practice, it means pulling real-time telemetry from sources like ACARS, FADEC (full authority digital engine control), EGT (Exhaust Gas Temperature) margins, oil debris monitoring, and vibration signatures, then running that data through algorithms trained on failure patterns to estimate remaining useful life (RUL) of critical components.”

She says that the distinction from other strategies matters. “Preventive maintenance operates on fixed TBO (Time between overhauls) intervals — effective but wasteful, because you’re replacing parts based on calendar time, not condition. CBM (condition-based maintenance) is reactive to current state: you act when readings fall outside tolerance. Predictive maintenance goes a step further, it identifies degradation trajectories before limits are breached, giving you a planning window rather than an emergency response.”

And she says the operational advantages are real. “Reduced unscheduled removals, better parts inventory positioning, fewer AOG events, and more efficient shop scheduling. For engine MROs specifically, the ability to sequence work against predicted shop visits rather than reactive inductions is a significant throughput and cost lever. The transition from time-based to RUL-based (Remaining Useful Life) thinking is the core shift,” Badra states.

From Awiros’ perspective, the missing layer in today’s maintenance ecosystem is inspection intelligence. “Every manufacturing image, borescope inspection, assembly photograph, configuration verification record and overhaul image contains valuable engineering knowledge that often remains locked inside disconnected systems,” says Gupta. “Vision AI transforms these visual records into structured, searchable intelligence that can be compared across years of engine operation.

Rather than replacing engineers, Gupta says Vision AI augments them by retrieving similar historical cases, highlighting subtle progression patterns, and providing explainable recommendations supported by historical evidence. “Over time, every inspection contributes to a continuously growing digital memory of each engine, enabling more informed maintenance planning, greater consistency, reduced unplanned events, and improved asset availability while ensuring certified engineers retain responsibility for all airworthiness decisions,” he comments.

A coming of age
According to Badra, predictive engine maintenance has moved from a research-grade capability to an operationally deployable one, though maturity varies sharply by organisation and asset class.

“Early implementations, largely in the 2010s, were concentrated at OEMs and Tier-1 carriers with the data infrastructure to support them. The tooling was expensive, proprietary, and difficult to integrate outside of new-generation platforms. The broader MRO community, particularly independent shops and mid-sized operators, had limited access.

“The shift over the last five years has been meaningful. Cloud-based analytics platforms have dramatically lowered the cost of entry. HUMS adoption has expanded beyond helicopters into fixed-wing applications. The volume of labelled historical failure data available for model training has grown substantially, improving prediction accuracy. ACARS data is now more widely parsed and acted upon, not just logged. And vendor ecosystems have matured enough that purpose-built solutions exist for specific engine types, CFM, V2500, RR Trent families, rather than generic condition-monitoring tools requiring heavy customisation.

“The gap that still exists is between pilots who can collect the data and organisations who can turn it into decisions. Data capture has scaled faster than analytical capability within MRO operations. That’s the frontier: operationalising the insight, not just generating it.”

The Delta TechOps Predictive Technology and Innovation team focuses on on-wing predictive/health monitoring to identify engine degradation that could lead to in-flight effects or operational disruptions.

A Delta TechOps spokesperson explains that Delta TOC has been utilising tools across its fleet of engines for over 20 years – “always enhancing our analytics and data sources to provide more insight into the performance of our engines.”

Agile learning
The spokesperson continues that predictive maintenance can provide meaningful value when it is continuously measured, validated and improved. “This is why Delta TechOps evaluates predictive models using performance metrics that measure their precision and effectiveness, and we regularly assess whether the analytics are accurately identifying emerging issues,” they state.

“It’s important to note,” they emphasise, “that the industry does not treat algorithms as static tools. We can refine and retrain models based on new operational data and real-world outcomes. Predictive maintenance is an iterative process, where every success and every missed prediction can help improve future performance. Our objective is a disciplined process that continuously increases the reliability and value of the insights being provided.

“As predictive maintenance capabilities continue to evolve, a common engine health monitoring platform, or a true “single pane of glass,” will be critical to providing a comprehensive view of engine performance across Delta’s large and diverse fleet. Combined with prioritised access to our operational data and a deeper understanding of how our engines operate, this foundation will help improve reliability, accelerate insights and support more informed maintenance decisions.”

As an example, its APEX MRO program, which is utilised by Delta Airlines, collects and analyses real-time data throughout an engine’s lifecycle. This capability enables Delta TechOps to optimise material planning, shop visits, and production control, reducing repair turnaround times, improving operational efficiency, and generating substantial cost savings while keeping more aircraft flying.

At MTU Maintenance, Dr Christian Keller from the Engine Trend Monitoring Program says predictive maintenance is a central part to the company’s strategy to further evolve engine services from reactive or interval-based maintenance toward proactive, data-driven decision-making. Its predictive maintenance capability is primarily a digital and engineering-intelligence capability. “That means, we include operational engine and aircraft data, full-flight data, engine models, neural-networks, digital twins and AI-assisted fleet simulation,” he states.

As Dr Keller explains, MTU has been working in this field for many years. Our proprietary Engine Trend Monitoring was introduced in 2006 and has evolved from “just” monitoring into a key pillar of our digital services that includes diagnosis, prognosis and real-time insight, as well as feeding vital information into our fleet management software.”

myETM is MTU’s engine trend monitoring environment. It tracks parameters such as exhaust gas temperature, fuel flow, shaft speeds, oil parameters and bleed settings. If the system detects a trend deviation, MTU experts assess the data and provide qualified recommendations. These can range from a simple watch item to on-wing action, urgent repair or earlier preparation of a shop visit. “This data can also feed into our wider fleet-management tools, where it supports optimised removal planning, material forecasting and scenario analysis,” he says.

As a lever amongst a range of strategies to help avoid delays and optimise maintenance costs across an entire fleet, predictive engine maintenance is not without
its challenges.

Predictive pickle
Whilst these challenges are well understood; the difficulty is in solving them at scale inside real MRO operations, claims Badra.

“Data quality and completeness remain the most persistent barrier. Predictive models are only as good as the data they consume. Missing ACARS reports, inconsistent borescope documentation, or gaps in oil debris monitoring logs introduce noise that degrades prediction reliability. Many shops are working with partially digitised records that weren’t captured with analytics in mind,” she says.

Dr Keller agrees, adding that predictive maintenance also has to recognise the limits of prediction. “Even with strong data, the final workscope can only be fully determined once the physical engine is inspected,” he states.

For Gupta, much of the enormous volumes of visual inspection generated by the aerospace industry remains fragmented across OEMs, MROs, suppliers and airlines. He credits the difference in image quality, inspection methods, data formats and documentation practices as making large scale analysis challenging.

Legacy system integration is closely related. Most MRO environments run a mix of ERP, MES, and engineering systems that don’t communicate natively says Badra. Getting sensor data, shop findings, part traceability, and maintenance history into a unified data layer is therefore a significant integration project before any predictive algorithm can run.

An enthusiastic supporter Gupta believes Vision AI addresses these challenges by standardising visual interpretation, creating digital traceability and converting unstructured imagery into engineering knowledge. “Combined with domain expertise and robust governance, this creates the foundation for future predictive maintenance without compromising regulatory compliance or engineering oversight,” he says.

Regulatory acceptance is a real constraint. Airworthiness authorities require documented evidence that a predictive-driven maintenance decision meets or exceeds the standard established in the MRB. Getting that approval pathway defined adds time and cost to deployment says Badra.

“Model interpretability matters for maintenance engineers. A black-box prediction that says “remove in 200 cycles” gets challenged unless engineers understand the driving indicators. Explainability isn’t just a technical feature, it’s a change management requirement,” Badra says.

“Organisations we work with through Aero NextGen’s Solution Finder — a free, vendor-neutral matching tool that connects MROs with vetted predictive maintenance technology providers — consistently cite integration complexity and data readiness as their top blockers, ahead of cost.”

Multi-layer approach
The technology stack for predictive engine maintenance has several interconnected layers.

Awiros believes Vision AI will become a foundational layer of future aircraft maintenance. “Our visual inspection capabilities transform images into structured engineering intelligence through anomaly detection, configuration verification, inspection documentation, historical comparison, and digital traceability,” asserts Gupta.

At the data acquisition layer, Badra says the foundation is IoT-enabled sensor networks feeding continuous telemetry — EGT trends, vibration signatures, oil debris particle counts, and FADEC parameter outputs. ACARS remains the primary air-to-ground data pipeline for in-service monitoring, supplemented increasingly by onboard HUMS where installed.

At the analytics layer, machine learning models — specifically anomaly detection, regression models for RUL estimation, and classification models for failure mode identification — process the telemetry against trained baselines. Digital twin frameworks are emerging as a structuring approach: building a virtual representation of the engine that updates against real operational data and surfaces deviations before they manifest as physical symptoms.

Delta TechOps uses engine parameter data collected in-flight, physics-based performance analysis to identify deviations from expected engine behaviour, and machine learning models that evaluate engines against fleet performance. It also uses domain adapted large language models to draw information from unstructured sources such as maintenance records and shop findings. These capabilities are being consolidated onto a common engine health monitoring platform.

AI-assisted borescope analysis deserves specific mention. Computer vision tools trained on thousands of labelled inspection images are now commercially available and reduce inter-inspector variability significantly, an important data quality input to predictive workflows.

“Our experience working with leading global aerospace organisations demonstrates that consistent, AI-assisted visual inspections can improve quality, reduce variability and establish the digital foundation required for predictive maintenance,” remarks Gupta. “As the ecosystem matures, this visual intelligence will integrate with operational, sensor, and maintenance data to create a holistic view of engine health.”

Several threads are converging in a way that will materially change how engine maintenance is planned and executed over the next decade.

Looking ahead, Dr Keller sees the industry moving toward larger data sets, full-flight data analytics, predictive digital twins, AI-assisted inspection and more prescriptive maintenance planning. “We do not expect a single moment when this future suddenly arrives. It will be a gradual progression from better monitoring to better prediction, then to more integrated and eventually more automated decision support,” he opines.

For Delta TechOps, condition-based maintenance is another key area of focus, where engine data is used to determine the optimal time to perform maintenance rather than relying solely on fixed intervals. This approach has the potential to support maintenance decisions based on actual engine condition, improving efficiency while maintaining safety and reliability.

“As these capabilities evolve, we are focused on expanding our analytics from snapshot-based engine data to full-flight data, providing richer insight into engine performance across all phases of flight. Achieving this vision will require technology platforms capable of processing significantly larger and more complex data sets,” a spokesperson says.

This evolution underscores the importance of data quality, consistency, and reliability. Predictive maintenance programs are only as effective as the data that supports them, making a strong data foundation critical to delivering accurate, actionable insights.

Embedded diagnostics and self-reporting engines are the most significant structural shift. As Badra notes, new-generation platforms, and retrofitted sensor packages on legacy platforms, are moving toward engines that continuously report their own health state, narrowing the gap between what the engine knows and what the MRO sees. “The vision of a truly self-inspecting engine, where internal sensing replaces scheduled borescope access for routine assessment, is technically plausible; the certification path is what determines the timeline,” she says.

Badra also highlights autonomous maintenance decision support —AI systems that generate a recommended maintenance action, not just a flag — which is already deployed in limited forms. “The near-term evolution is less about full autonomy and more about integrating predictive outputs directly into work order generation and parts procurement, eliminating the manual translation step between data and action,” she remarks.

A final observation from Badra, is the growth of fleet-level predictive models that aggregate data across operators to train better predictions, enabled by OEM data-sharing frameworks and third-party analytics providers. The network effect here is powerful, each additional engine in the dataset improves prediction accuracy for all.

By 2035, every aircraft engine could have a continuously evolving visual digital twin that learns from manufacturing through retirement. That’s the belief of Gupta, who says engineers will work alongside AI copilots capable of retrieving historical evidence, comparing inspection findings across fleets, and identifying degradation trends long before they become operational issues.

“Awiros envisions a future where every inspection strengthens the intelligence of the next one,” he says. “Vision AI will not replace human expertise; it will amplify it, enabling safer operations, more efficient maintenance planning, improved sustainability through optimised component life and a continuously learning aerospace ecosystem.”

When does this fully arrive? For Badra, the technology is ahead of the operational and regulatory infrastructure needed to exploit it. “For leading organisations, meaningful predictive-driven maintenance is operational now. For the broader MRO market, the 2028–2032 window is realistic for widespread adoption, driven by platform refresh cycles, regulatory clarity on data-driven airworthiness, and a generation of digital-native maintenance engineers entering the workforce.”

By Alex Preston