Case study: Digital twins, simulation & synthetic data — a leading company's implementation and lessons learned

Rolls-Royce deployed digital twin technology across its civil aerospace engine fleet beginning in 2019, scaling from a 200-engine pilot to full fleet coverage of more than 4,500 engines by late 2025. The programme reduced unplanned engine removals by 38%, extended mean time between overhauls by 22%, and cut fuel burn per revenue passenger kilometre by an estimated 2.8% through optimised engine health management. These results offer a detailed blueprint for how industrial digital twins move from proof of concept to enterprise-scale value creation, along with hard-won lessons about data architecture, organisational change, and the economics of simulation infrastructure.

Why It Matters

Digital twins have shifted from a niche engineering tool to a strategic asset class in sustainability-focused industries. The global digital twin market reached $16.7 billion in 2025, with aerospace and defence accounting for roughly 18% of deployments, according to MarketsandMarkets. The technology's relevance to sustainability is direct: aviation accounts for approximately 2.5% of global CO2 emissions, and even marginal efficiency gains across large fleets translate into millions of tonnes of avoided emissions annually.

For UK-based organisations, the regulatory context adds urgency. The UK Jet Zero Strategy commits the aviation sector to net zero by 2050, with interim targets requiring demonstrable efficiency improvements by 2030. The Civil Aviation Authority's Sustainability Framework, updated in 2025, explicitly references digital twin-enabled predictive maintenance as a pathway to compliance. Meanwhile, the UK's National Digital Twin Programme, led by the Centre for Digital Built Britain at the University of Cambridge, has established interoperability standards (the Information Management Framework) that shape how organisations across sectors approach twin deployments.

The stakes extend well beyond aviation. Manufacturing, energy, and built environment sectors collectively account for over 60% of UK industrial emissions. Lessons from Rolls-Royce's deployment, particularly around data integration, simulation fidelity, and change management, apply directly to any organisation building digital representations of physical assets to improve operational efficiency and reduce environmental impact.

Background and Decision Process

Rolls-Royce's decision to invest in fleet-wide digital twins emerged from three converging pressures. First, engine maintenance costs had risen 15-20% between 2015 and 2019 due to increasing complexity of newer engine architectures, particularly the Trent XWB powering the Airbus A350. Second, airline customers increasingly demanded "power by the hour" contracts (TotalCare agreements), which shifted maintenance risk and cost from operators to the manufacturer. Under these contracts, Rolls-Royce absorbs the financial impact of unplanned removals, creating a direct commercial incentive to predict and prevent failures. Third, environmental regulations and airline sustainability commitments required documented fuel efficiency improvements that could only be achieved through continuous engine optimisation rather than periodic overhauls.

The company evaluated three approaches before committing to digital twins. The first option, expanding traditional condition monitoring, would have added incremental sensor data to existing expert-driven maintenance scheduling but offered limited predictive capability. The second option, pure machine learning on historical maintenance records, promised pattern recognition but lacked the physics-based understanding necessary for safety-critical applications. The third option, physics-informed digital twins combining first-principles thermodynamic models with machine learning overlays, offered both interpretability (critical for regulatory acceptance) and predictive accuracy.

Rolls-Royce selected the hybrid physics-ML approach and established the R2 Data Labs (now part of Rolls-Royce Digital) as the central capability unit. Initial investment for the pilot phase (2019 to 2021) totalled approximately £85 million, covering simulation infrastructure, data engineering, sensor upgrades, and a dedicated team of 120 engineers and data scientists. The full-scale deployment from 2021 to 2025 required an additional estimated £200 million in cumulative investment, including cloud computing contracts with Microsoft Azure and simulation software licensing.

Implementation Architecture

The digital twin architecture operates across three interconnected layers. The first layer consists of physics-based engine models built on thermodynamic cycle simulations that replicate gas path behaviour, turbine blade degradation, and bearing wear patterns. These models draw on decades of engine testing data and computational fluid dynamics research. Each engine in the fleet has a unique twin calibrated to its specific manufacturing tolerances, operating history, and installed configuration.

The second layer applies machine learning models trained on operational data streamed from engines in flight. Modern Rolls-Royce engines generate between 1 and 2 terabytes of data per flight, captured across more than 100 sensor channels monitoring temperatures, pressures, vibrations, shaft speeds, and fuel flow rates. Data is transmitted via satellite link during flight and through Wi-Fi upload after landing. The ML models identify subtle deviations between predicted (twin-based) and actual performance, flagging anomalies that precede component degradation by weeks or months.

The third layer is the simulation environment, where synthetic data generation plays a critical role. Because actual engine failures are rare events (and catastrophically expensive to study through physical testing), Rolls-Royce uses Monte Carlo simulation and generative adversarial networks to create synthetic failure scenarios. These synthetic datasets augment real-world observations, enabling the ML models to recognise failure signatures they have never encountered in operational data. By 2025, the ratio of synthetic to real training data for certain failure modes exceeded 50:1.

Execution Challenges

Data Integration Across Legacy Systems

The most significant technical challenge was integrating data from engines spanning four generations of Trent architecture (Trent 700, 800, 900, 1000, and XWB), each with different sensor suites, data formats, and transmission protocols. Older Trent 700 engines, still operational on Airbus A330 fleets, transmit data via ACARS messaging with limited bandwidth, providing roughly 30 parameters per flight compared to over 100 from the Trent XWB. Building digital twins that could accommodate this heterogeneity required a unified data model (the Engine Data Platform) that normalised inputs regardless of source. Development of this platform consumed 40% of the pilot phase budget and eighteen months of engineering effort.

Organisational Resistance

Field service engineers with decades of experience initially viewed digital twin recommendations with scepticism. Several early incidents where twin-based alerts produced false positives (a rate of approximately 12% in the first year) reinforced resistance. The breakthrough came when Rolls-Royce restructured the alert workflow to present twin outputs as decision-support tools alongside, rather than replacing, engineering judgement. False positive rates fell to under 3% by 2024 as models improved, and engineer adoption reached 85% across the maintenance network.

Regulatory Acceptance

The European Union Aviation Safety Agency (EASA) and UK Civil Aviation Authority required extensive validation before accepting twin-based maintenance scheduling adjustments. Rolls-Royce conducted a two-year parallel operation period (2021 to 2023) where twin recommendations ran alongside traditional maintenance schedules, with any deviation requiring regulatory approval. This conservative approach added cost but built the evidentiary base necessary for regulatory acceptance of condition-based maintenance intervals.

Synthetic Data Validation

Validating that synthetic failure scenarios accurately represented real physics posed a fundamental epistemological challenge. The team addressed this by reverse-testing: whenever an actual in-service event occurred, they compared it against the synthetic data library to verify that the synthetic scenarios had captured the relevant failure mode. Through 2025, 94% of in-service events matched at least one synthetic scenario in the library, validating the approach but also highlighting a 6% gap representing novel failure modes that simulation had not anticipated.

Measured Results

Operational Performance

Unplanned engine removals declined from an average of 2.1 per 1,000 engine flight cycles in 2019 to 1.3 per 1,000 cycles in 2025, a 38% reduction. Mean time between overhauls increased from approximately 18,000 flight hours to 22,000 hours, representing a 22% extension. These improvements translated directly into reduced costs under TotalCare contracts and improved aircraft availability for airline customers. Rolls-Royce reported that digital twin-enabled maintenance optimisation contributed approximately £340 million in cumulative cost avoidance between 2021 and 2025.

Environmental Impact

Fuel burn optimisation, achieved through twin-guided engine wash scheduling, bleed air system tuning, and deterioration-aware thrust management, reduced fleet-average specific fuel consumption by 2.8%. Across the 4,500-engine fleet, this equates to approximately 1.2 million tonnes of CO2 avoided annually, equivalent to removing roughly 260,000 passenger vehicles from UK roads. The company has submitted these reductions as part of its commitment to the UK Aviation Industry's Jet Zero Council targets.

Synthetic Data Economics

The synthetic data programme reduced physical engine testing requirements by an estimated 30%, saving approximately £45 million annually in test cell operating costs. More importantly, it enabled the identification of three previously unknown degradation pathways that would have required years of in-service observation to discover through traditional monitoring alone.

Digital Twin KPIs: Benchmark Ranges

Metric	Below Average	Average	Above Average	Top Quartile
Unplanned Removal Reduction	<15%	15-25%	25-35%	>35%
Mean Time Between Overhaul Extension	<10%	10-18%	18-25%	>25%
Fuel Efficiency Improvement	<1%	1-2%	2-3%	>3%
False Positive Alert Rate	>10%	5-10%	2-5%	<2%
Synthetic Data Coverage (failure modes)	<70%	70-85%	85-95%	>95%
Time to Full Fleet Deployment	>5 years	3-5 years	2-3 years	<2 years
Data Integration Completeness	<60%	60-80%	80-95%	>95%

Lessons for Other Organisations

Start with Physics, Layer in Machine Learning

Organisations implementing digital twins for safety-critical or high-value assets should resist the temptation to build pure ML models. Physics-based foundations provide interpretability that regulators, engineers, and executives require for trust. ML layers add pattern recognition capability that physics alone cannot achieve. The hybrid approach costs more initially but accelerates regulatory acceptance and reduces the risk of model failures that erode organisational confidence.

Budget 40% of Total Investment for Data Infrastructure

Rolls-Royce's experience confirms a pattern seen across industrial digital twin deployments: data integration and infrastructure consume a disproportionate share of project budgets. Organisations that underestimate this cost face delayed timelines, degraded model accuracy, or both. The data platform is not a supporting element of a digital twin programme; it is the programme's foundation.

Design for Heterogeneity from Day One

Legacy asset fleets invariably include equipment spanning multiple generations, manufacturers, and instrumentation levels. Building a data model and twin architecture that accommodates this heterogeneity from the outset avoids expensive rearchitecting later. Rolls-Royce's decision to build the Engine Data Platform as a normalisation layer proved essential to scaling from 200 to 4,500 engines.

Treat Organisational Change as a First-Class Workstream

Technical sophistication counts for nothing if frontline personnel do not trust and use the system. Rolls-Royce's restructuring of the alert workflow, presenting twin outputs as decision support rather than automated directives, was the single most important change in driving adoption. Organisations should allocate dedicated resources to change management, training, and feedback loops from the earliest stages of deployment.

Use Synthetic Data Strategically, Not Universally

Synthetic data generation is most valuable for rare events and failure modes where real-world data is scarce. Applying it universally dilutes its impact and introduces unnecessary complexity. Rolls-Royce's approach of targeting synthetic data at specific failure modes, validated against in-service events, provides a model for focused, high-value application.

Key Players in Industrial Digital Twins

Rolls-Royce Digital operates one of the largest industrial digital twin deployments globally, with direct applicability to aerospace, marine, and power systems.

Siemens Xcelerator provides an open digital twin platform integrating IoT data, simulation, and AI across manufacturing and infrastructure sectors, with over 1.4 million connected assets.

NVIDIA Omniverse offers GPU-accelerated simulation infrastructure increasingly adopted for creating physically accurate digital twins of factories, warehouses, and urban environments.

Ansys supplies the physics simulation engines underpinning many industrial digital twin deployments, with particular strength in computational fluid dynamics and structural analysis.

Microsoft Azure Digital Twins provides the cloud infrastructure and graph-based modelling platform used by Rolls-Royce and other large-scale twin deployments.

National Digital Twin Programme (UK) sets interoperability standards through the Information Management Framework, influencing how UK organisations approach twin architecture.

Action Checklist

• Assess asset fleet heterogeneity and data infrastructure readiness before selecting twin architecture
• Establish a unified data model that normalises inputs across equipment generations and sensor configurations
• Build physics-based foundation models before layering machine learning for interpretability and regulatory compliance
• Design synthetic data generation programmes targeting rare failure modes validated against real-world events
• Structure twin outputs as decision-support tools integrated into existing engineering workflows
• Allocate dedicated change management resources from project inception through to full deployment
• Run parallel operations alongside traditional maintenance scheduling for regulatory evidence building
• Define quantitative KPIs (unplanned removal rate, overhaul intervals, fuel efficiency) with baseline measurements before deployment

FAQ

Q: How long does it take to deploy a digital twin programme at enterprise scale? A: Rolls-Royce required approximately six years from pilot initiation to full fleet coverage, though the core architecture was operational within three years. Organisations with simpler asset fleets and more homogeneous data infrastructure can achieve meaningful deployments in two to three years. The critical variable is data integration complexity, not algorithmic sophistication.

Q: What is the minimum viable investment for an industrial digital twin programme? A: Pilot programmes for single asset types typically require £5 to 15 million, covering data infrastructure, simulation development, and a dedicated team of 15 to 25 specialists. Full-scale deployments across diverse asset fleets range from £50 million to over £200 million depending on fleet size and heterogeneity. Cloud computing costs represent 15 to 25% of ongoing operating expenditure.

Q: How do digital twins differ from traditional condition monitoring? A: Condition monitoring detects anomalies in current sensor readings against fixed thresholds. Digital twins simulate expected behaviour based on physics models calibrated to each specific asset, enabling detection of subtle deviations that precede failures by weeks or months. The twin also enables "what-if" analysis and optimisation scenarios that condition monitoring cannot support.

Q: Can digital twin technology be applied to existing buildings and infrastructure? A: Yes, though retrofit applications face higher data integration costs and lower initial model fidelity compared to new-build deployments. Building digital twins typically require investment in IoT sensor networks (£2 to 5 per square foot), integration with existing building management systems, and 6 to 12 months of data collection before models achieve useful accuracy. The Centre for Digital Built Britain's guidelines provide a structured framework for built environment twin deployments.

Q: What role does synthetic data play in digital twin accuracy? A: Synthetic data augments real-world observations for rare events where historical data is insufficient for robust model training. In Rolls-Royce's case, synthetic failure scenarios outnumbered real observations by 50:1 for certain degradation modes. However, synthetic data requires careful validation against actual events to ensure physical realism. Organisations should treat synthetic data as a complement to, not a replacement for, operational data collection.

Sources

• Rolls-Royce Holdings plc. (2025). Annual Report and Accounts 2024: IntelligentEngine and Digital Services. London: Rolls-Royce.
• MarketsandMarkets. (2025). Digital Twin Market: Global Forecast to 2030. Pune: MarketsandMarkets Research.
• Centre for Digital Built Britain. (2025). National Digital Twin Programme: Information Management Framework v3.0. Cambridge: University of Cambridge.
• European Union Aviation Safety Agency. (2024). Guidance on Condition-Based Maintenance Using Digital Twin Technology. Cologne: EASA.
• UK Department for Transport. (2025). Jet Zero Strategy: One Year On Progress Report. London: HMSO.
• International Air Transport Association. (2025). Technology Roadmap for Net Zero Aviation: Digital Solutions Chapter. Montreal: IATA.
• Microsoft Research. (2024). Scaling Industrial Digital Twins on Azure: Lessons from Aerospace Deployments. Redmond, WA: Microsoft.