At Kubrick, research forms part of how our data, AI, and engineering teams explore practical ways to solve complex operational problems. It helps us test new methods, build deeper technical understanding, and translate emerging approaches into capabilities that clients can use in real-world environments.
Our team’s research on evaluating airline operational resilience using discrete event simulation was accepted for presentation at the 7th International Conference on Data Analytics for Business and Industry on October 29-30th in Bahrain.
The challenge of optimized vs resilient schedules
Airlines invest heavily in optimization to create efficient crew rosters, but an optimized schedule is only part of the challenge. Once operations begin, crew sickness, unexpected delays and resource shortages can quickly undermine even the most carefully prepared plan.
Our research explores a different question: but how resilient it remains when disruptions occur.
Building an airline simulation
To address this, we developed a cloud-based simulation framework that combines synthetic flight schedule generation, crew rostering, disruption modeling and recovery logic to evaluate airline operations under uncertainty.
The framework creates a realistic operating environment with a focus on crew scheduling.
Flights, aircraft rotations and duties are generated and optimized into a feasible roster before being exposed to hundreds of random disruption scenarios involving crew absences and flight delays. Recovery actions, such as reserve crew deployment and crew reassignment, are then applied automatically while operational and regulatory constraints remain enforced.
Built on a scalable cloud architecture, the framework is also designed for integration with existing airline scheduling, crew management and operational systems.
Testing resilience before operations begin
The approach enables planners to perform what-if analysis before schedules are deployed. Rather than waiting for real-world disruption, airlines can test how scheduling decisions respond under varying conditions and quantify the consequences of different workforce strategies.
For example, planners can evaluate questions such as:
- What if crew sickness rates increase this month?
- What if flight delays become more frequent due to adverse weather?
- What if reserve crew levels are reduced to lower staffing costs?
Evaluating schedules before implementation helps increase confidence that staffing decisions are robust, resilient and supported by evidence rather than intuition alone.
Key findings
The results show that different disruption types affect operations in different ways. Crew absence was the primary driver of duty coverage and flight completion, while flight delays had the greatest impact on accumulated delay and recovery performance.
One particularly important finding was that adding reserve crew as additional capacity consistently improved completion rates, with benefits increasing as absence levels rose. However, reallocating active crew into reserve roles produced diminishing returns, demonstrating that resilience depends on balancing standby flexibility with day-to-day operational capacity.
Behind the scenes: what’s next
Research implementation
Based on this research, we are now extending the cloud-based simulation framework on Databricks. Existing flight, crew and roster data is read from Delta tables governed by Unity Catalog. In parallel, Python and Spark assemble duties and construct a feasible roster from operational data, while Lakeflow Jobs analyze historical operational performance to estimate crew absence rates, flight delay characteristics and other disruption inputs. The simulation model is then calibrated by tuning disruption and recovery parameters until simulated operational performance aligns with historical observations.
Once calibrated, a discrete event simulation executes hundreds of disruption scenarios involving crew absences and flight delays. For each scenario, recovery logic deploys reserve crew and reassigns duties while enforcing operational and regulatory constraints, including duty limits, rest requirements, connection rules and rolling-window restrictions.
The simulation evaluates how the roster performs under uncertainty rather than generating a replacement roster. Results are tracked in MLflow, and scenario outputs including duty coverage, flight completion, delay propagation, total delay and recovery performance are written to governed Delta tables for analysis.
The approach is designed to evaluate schedule resilience through what-if analysis before deployment and support integration with existing scheduling and crew management systems
Workflows
Scenario results are surfaced through a Databricks App with AI/BI dashboards that provide a resilience assessment of the roster. The assessment presents operational metrics, diagnostic outputs and decision-support insights across duty coverage, flight completion, delay propagation, total delay and recovery performance. The purpose is to evaluate how resilient a roster is under different disruption scenarios, not to create a new roster.
Planners can use the assessment to compare staffing strategies, identify operational vulnerabilities and understand the consequences of different workforce decisions before implementation.
The selected resilience assessment, associated staffing recommendation, scenario assumptions and roster version are recorded in Delta tables with full lineage and auditability.
Planning teams then use these insights to inform decisions within existing scheduling and crew management systems. Unity Catalog governance applies throughout the workflow, providing access control, lineage tracking, auditing and separation of service identities.
Because every assessment is linked to its underlying data, calibrated parameters, simulation assumptions and roster version, staffing decisions are supported by transparent simulation results and auditable assumptions.
From optimization to confidence
The research provides a foundation for integrating live scheduling, crew management and flight operations data to support a digital twin of airline workforce operations. This type of capability would allow organizations to continuously evaluate resilience, test recovery strategies and assess staffing decisions before implementation. Ultimately, the goal is not simply to build better schedules, but to build greater confidence that those schedules will continue to perform when disruption inevitably occurs.
