Air Cargo Ground Operations Optimization
Jenny Tonka - Univ. of Liege
Flight delays are one of the most important KPIs for air transport systems. Knowing that suboptimal ground operations have a highly negative impact on them, we seek to optimize the coordination between the different ground service equipment (GSE). As each ground operation requires a fleet of vehicles to serve different clients within a defined time interval, we decided to tackle the problem as a set of VRPs with time windows. Each VRP refers to a given ground operation and is characterized by heterogeneous fleets, multi-vehicle services, capacitated vehicles, multi-location services, forbidden and compulsory pairings, multi-operation vehicles, split deliveries, and dynamic data. Additionally, some vehicles are bound together through temporal, movement, or load synchronization constraints. We propose a heuristic approach centered on aircraft that uses recursive procedures to optimize the journey of the GSE, considering all these problem features.
Maintenance Cost Optimization for Engines, Airframe and Landing Gears
Keith Dugas & Rahim Akhavan – Air Canada
Our Maintenance Cost Optimizer is a sophisticated optimization platform that utilizes three advanced models to minimize the total cost of ownership. With a strategic planning horizon of up to 30 years into the future, this system considers not only our maintenance program defined by our world-class engineers but also incorporates our lease terms and conditions. This allows for effective optimization of both factors simultaneously, resulting in significant cost savings. The platform also enables digital twinning or cloning of our aircraft, simulating various scenarios using what-if parameters for our engine, airframe, and landing gear checks as well as RPA. By calculating costs at shop visits and end-of-lease adjustments, we can better predict and manage expenses. Overall, the Cost empowers our team to make data-driven decisions, optimize costs, and improve operational efficiency, ultimately resulting in better performance and profitability.
Application of Quantum Computing in The Field of Airline Revenue Management
Alexander Papen, Thomas Fiig, Antoine Boulanger, Mogens Dalgaard, Janus Halleløv Wesenberg,& Rune Thinggaard Hansen - Amadeus
Application of quantum computing in Revenue Management is evaluated, both considering the quantum devices that exist today as well as assessing the potential in the future with an increasing number of qubits and reduced error rates. We develop and implement a quantum algorithm for solving the single flight leg optimization problem by mapping the Bellman equation to an Ising model (spin model used to describe magnetism). This allows execution on existing hardware from D-Wave Systems. The results are limited to toy scenarios, but still demonstrate the state-of-the-art of quantum computing. Further, we develop a quantum algorithm for solving a Choice-Based Deterministic Linear Program (CDLP) for network optimization with customer choice. CDLP, although expected to be superior to existing approaches, is far too complex for existing classical computers. Utilizing a quantum computer, we obtain a quadratic speed up, which allow for more realistic problem sizes.
Onboard Wi-Fi Predictive Maintenance: Identifying Aircraft Wi-Fi Antenna Failure Using Big Data
Ehsan Rahimi, Shuang Ling, Keith Shackelford - United Airlines
This abstract explores the integration of onboard Wi-Fi in modern aviation, focusing on predictive maintenance by swiftly identifying aircraft component failures, specifically the Wi-Fi antenna, to enhance Wi-Fi performance. By utilizing Big Data encompassing historical Wi-Fi "heartbeats" (HBs) from United’s fleet over two years, we introduce the Normalized Wi-Fi Health Score (NWiHS), a robust indicator of aircraft-level Wi-Fi performance, which quantifies the percent of missing Wi-Fi HBs normalized to exclude the effect of Wi-Fi provider performance and global coverage. Using NWiHS, statistically driven thresholds, and noise reduction techniques, our alerting algorithm successfully captured Wi-Fi antenna failures in United's Airbus 320 fleet with a 76% success rate and an average 7-day lead time. This method shows promise in identifying and addressing Wi-Fi antenna failures, contributing to improved onboard Wi-Fi performance and passenger experience.
Combined Language Models with Optimization for Conversational Trip Planning
Mourad Boudia, Alexis Ravanel, Mohamed Habi, Massimiliano Pronesti, & Nicolas Bondoux - Amadeus
Planning trips requires travelers to interact with different sites to get relevant information before deciding to book their trips. To improve this experience, we propose a new chatbot channels, powered by state-of-the-art large language models (LLM) that today can provide much more information than before. However, these models are not designed to provide optimal travel plans and their suggestions are usually suboptimal in terms of costs or time.
In this work, we propose a novel approach to recommend travel plans by integrating language models and optimization techniques. Through interactive conversations, travelers express their interest enabling our engine to not only recommend possible destinations but also to optimize the order of cities to be visited. In the technical side, we build an orchestrator powered by LLMs combined with other fine-tuned LLMs that translate the conversation into optimization models that help to recommend an optimized travel plan.
Distributionally Robust Airport Ground Holding Problem Under Wasserstein Ambiguity Sets
Haochen Wu – University of Michigan
The airport ground holding problem seeks to minimize flight delay costs due to airport capacity reductions. However, the critical input of future airport capacities is often difficult to predict, presenting a challenging yet realistic setting. Even when airport capacity predictions provide a distribution of possible capacity scenarios, such distributions may themselves be uncertain. To address the problem of designing airport ground holding policies under distributional uncertainty, we formulate and solve the airport ground holding problem using distributionally robust optimization (DRO). We address the uncertainty in the airport capacity distribution by defining ambiguity sets based on the Wasserstein distance metric. We note that DRO can be a valuable tool for decision-makers seeking to design airport ground holding policies, particularly when the available data regarding future airport capacities are highly uncertain.
Conjointly Optimizing Multiple Airlines Operations – a Case Study on the Lufthansa Group
Marie Carré - Swiss
ATC strikes, or airport weather issues affect all airlines operating in the disrupted regions. Mandatory capacity reduction at an airport is a well-known optimization case for the OR community. As multiple airlines of a same group are facing the same mandatory capacity reduction, we assume that a joint optimization of their resources could lead to find better solutions for the airlines group and their passengers. This triggers sensitive questions towards the constraints and variables simplifications needed to meet the computational time requirements for a group optimization. To solve this problem, we decomposed the optimization core into two sub-problems: schedule recovery and passengers’ recovery, offering a semi-integrated optimization for the group of airlines. In this talk, we will focus on the optimization method, the simplifications chosen and will illustrate its benefits on data from recent strikes.
Combined Revenue Management and Fleet Assignment optimization
Fedor Nikitin & Bertalan Juhasz - Finnair
When revenue is optimized in revenue management, the capacities for the legs are fixed, which means that fleet is assigned beforehand. On the other hand, for fleet assignment optimization, revenue estimates are required to optimize the difference of the revenue and the costs on the network. In other words, optimizing capacity requires estimates of the optimized revenue, which require capacities which are already optimized. So what should be optimized first, revenue or capacities? To solve this chicken and egg problem, a model which combines revenue optimization and fleet assignment is considered.
Using An Autoencoder to Develop Health Indicators and RUL Prognostics for Aircraft Systems with Few Failure Instances and Varying Flight Conditions
Ingeborg de Pater, Mihaela Mitici – TU Delft
Most Remaining Useful Life (RUL) prognostics are obtained using supervised learning models trained with many labelled data samples (i.e., the true RUL is known). In aviation, however, aircraft systems are often preventively replaced before failure. There are thus very few labelled data samples available. In this presentation, we therefore introduce a Long Short-Term Memory (LSTM) autoencoder with attention to develop health indicators for an aircraft system. This autoencoder is trained with unlabelled data samples. The varying flight conditions of aircraft are also integrated in the autoencoder. These health indicators are further used to predict the RUL of the aircraft system using a similarity-based matching approach. We illustrate our approach for turbofan engines in the N-CMAPSS dataset, which contains only 6 failure instances. A 19% reduction in the RMSE is obtained using our approach in comparison to existing supervised learning models.
Dynamic Ancillary Pricing: Leveraging Contextual Bandits with Limited Historical Data
Bogusz Stefańczyk & Özcan Gündeş – Flyr Labs
Ancillary products are one of the key revenue drivers in airlines efforts to unbundle fares. However, most airlines do not leverage dynamic pricing to optimize ancillary revenue. We propose a reinforcement learning approach to dynamically price a-la carte advanced seat reservation using a contextual multi-armed bandit (MAB) adapted to the lack of variation in historical pricing. The proposed solution addresses both the cold-start problem due to limited historical data and the need to continuously optimize prices by combining exploration with exploitation. The solution efficacy is explored via simulation across multiple MAB architectures and demand scenarios.
Learning From the Crowd: Using Generative AI to Predict Willingness to Pay of Sparse Passenger Streams.
Giacomo Bonciolini, Guillem Sala Fernandez; & Florian Martin – Swiss
Conventional methods to estimate willingness to pay need extensive transaction history on O&D level. That task is challenging for passenger streams with insufficient history (new routes), low price variation or comparably sparse observations (connecting traffic, higher compartments). The presented generative AI model sidesteps these issues, only requiring two sets of readily available inputs: the O&D context (socio-economic indicators, geography), and a proxy of customers perspective given by ChatGPT or by the analyst via free text. Learning on all the O&Ds the link between their existing pricing policies and those inputs, the model crafts a set of novel alternative policies for any other O&D, given its own inputs and multiple realizations of a random seed. From those policies, the optimum can be achieved via online Reinforcement Learning, rewarding the best performing ones. That avoids the cost and complexity of experimenting directly with variation of prices.
Application of Large Language Models for Reliability Analysis
Soheil Parsa, Alexander Bellemare-Davis, Ying Xiong, Zu Sheng Zong – Boeing
Aircraft reliability analytics can provide insight on the best time to replace or repair a component to minimize flight disruption. Such analyses use different available data sources to monitor overall reliability, but maintenance records constitute the most valuable data if and where it can be mined successfully. Data collected during aircraft maintenance -- and the text narratives provided by maintainers themselves -- can reveal insightful patterns on component lifetime and performance monitoring, which are important parts of reliability analytics. However, analysis of unstructured text produced by operators and mechanics is not trivial. This presentation aims to introduce the application of Large Language Models for building CCLA (component, condition, location, and action) classifiers which produce higher accuracy on understanding what aircraft components have been causing certain maintenance activities.
Application of Alternative Economic Theories in the Airline Industry
John Lancaster & Troy Selland – Analytic Pricing
Airline commercial decision-making is rooted in individual decision-making microeconomics. It is well understood that the assumption of agent rationality and use of zero-sum games restricts the solution space. This implies the optimal solutions of RM and pricing systems may be local optima rather than global optima. Modern innovations using behavioral economics and AI retain the economic assumption that the relevant market dynamic is one of individual agents. In B2B marketplaces this assumption represents a major. Clearly, this does not reflect the reality of air cargo spot and tender markets. It makes sense to consider remodeling airline market dynamics using alternative economic frameworks which lead to more profitable and stable cash flows by all players. This talk presents two such models using 1) social welfare theory and 2) ecological economic theory. They lead to different distribution and pricing models and create more efficient and stable marketplaces.
Integrated Predictive Maintenance Tool for Business Aircrafts Through Continuous Monitoring in Practice
Ashwin Rai, Eden Macdonald, Kamrul Hassan, Qing Liu, Juraj Hresko, Jorge Mendoza, Guillaume Rabusseau, Soumaya Yacout, & Yossiri Adulyasak – Ivado Labs
This presentation introduces an advanced method to improve business aircraft maintenance. It combines real-time monitoring, diagnostics, and predictive techniques. The method employs anomaly detection to monitor conditions, spotting deviations from normal operations. Diagnostics link anomalies to components through domain mapping trained with expert data. Anomalies transform into continuous component degradation profiles as time-series data, essential for predictions. Prognostics leverage evolving degradation data, utilizing dynamic distributional parameters to predict component failure probabilities. By integrating these stages, aircraft operators can shift from time-based to efficient, cost-effective maintenance. This presentation highlights seamless technique integration and its potential for aircraft maintenance.
Using Machine Learning to Predict the On-Time Performance of a Flight Schedule
Pascale Batchoun, Quan Anh Bach, Lamine Baghriche, Nitesh Baswal, Pascale Batchoun, Vadlamudi Bhargav, Philippe Branchini, Zohreh Hajabedi, Marc Levangie, Trang Minh Nguyen, Narges Sereshti, Navjot Singh, Alexandre Vincart-Emard, & Tristan Waldie – Air Canada
An optimally designed schedule crafted by Network Planners to maximize network profitability might still overlook operational variability, potentially leading to delays, associated costs, and customer dissatisfaction. The first step toward establishing a resilient flight schedule is to accurately predict On-Time-Performance (OTP) and to evaluate the impact that both scheduling and operational changes could have on OTP. Our team at Air Canada has developed and implemented a machine learning-driven methodology that predicts system-level OTP key performance indicators (KPIs) while providing low level estimates for the block and turn durations. Augmented with a simulation engine, the “OTP Schedule Optimizer” evaluates various scenarios to simulate disruptive and cascading delays, and their impact on our aircraft performance, passenger connections and crew flow.
Ground Time Awareness in Fleet Assignment Models
Nitin Srinath, Ahmed Marzouk, Seza Orcun, & Raymond Lee - United Airlines
In airline Fleet Assignment Models (FAM), the most common objectives are to improve utilization, decrease the number of aircraft used or maximize profit, while ensuring the demand is met, all the flights are covered, etc. While these are key considerations, it is also important to consider how much aircraft ground time there is between two flights. Since these models typically work towards increasing utilization, solutions typically end up with minimum ground time, which has significant schedule reliability risks. An approach to incorporate ground time as a distribution (ie a few flights with minimum ground time and remainder with above minimum) is presented here, with FAM encouraged to force this distribution by adding MILP constraints with incentives or penalties.
Introducing ATL@GT and AirSim, a Competitive RM Simulation Tool
Laurie Garrow, Alan Walker & Jeffrey Newman – Georgia Institute of Technology
In January of 2023, a new center called ATL@GT was established at Georgia Tech under the direction of Laurie Garrow. The mission of ATL@GT is to lead research and education activities related to airline revenue management. The center is currently focused on conducting research related to offer management, ancillary pricing, and continuous pricing based on recent advancements in new distribution capability, cloud computing, and machine learning. In this presentation, we will give an overview of AirSim, a competitive RM simulator that we are developing to test different RM strategies. The competitive RM simulator is being designed with a highly flexible architecture that will enable students, researchers, and industry partners to write customized code to test out their own ideas while interfacing with the core simulator.
Revenue optimization for airline branded fares with machine learning
Adam Bockelie, Aldair Alvarez, Teodora Dan, Tianjiao Liu, Carl Perreault-Lafleur, Alan Regis, Yury Sambale, Sajad Aliakbari Sani, Cindy Yao, Emma Frejinger, Andrea Lodi, Guillaume Rabusseau - Air Canada and IVADO Labs
Airlines commonly offer their products in bundles, named branded fares, containing incremental ancillaries. The markup of each branded fare (relative to the other ones) should take into consideration the different needs and willingness to pay of the airlines’ customers due to the significant revenue impact that this can have. However, despite the rich body of literature on revenue management in the airline industry, we have observed a gap in branded fare pricing. In addition, commercial tools are not available to aid decision-makers in this context. In this talk we introduce machine learning and optimization models to capture willingness to pay and maximize the expected revenue from branded fares. The models developed are deployed in a decision support tool for pricing managers.
Willingness-to-Pay Estimation and Pricing Optimization for Airline Seat Assignment
Yury Sambale, Sajad Aliakbari Sani, Adam Bockelie, Tianjiao Liu, Alan Regis, Cindy Yao, Aldair Alvarez, Teodora Dan, Carl Perreault-Lafleur, Emma Frejinger, Andrea Lodi, & Guillaume Rabusseau – Air Canada and IVADO Labs
Within the booking flow, customers are presented with the option to purchase a preassigned seat. This option, presented as a la carte product, allows customers to select a seat on each segment of their travel journey. Multiple factors like the origin-destination and the seat product impact a customer’s willingness-to-pay (WTP). Decision makers had to rely on a method based on trial and error to set prices for preselected seats to attempt to maximize revenue while taking into consideration these dimensions. With very few seat assignment pricing optimization models presented in the literature that can be applied in a practical setting, we present a model designed at a major carrier that leverages machine learning to estimate the WTP and a price optimization model to generate optimal pricing points for each seat product. We also shed light on a field experiment methodology implemented to improve data quality and enable a data-driven approach.
Improved Decision Making for Disruptive Weather Events using Ensemble Forecasts
John Williams, The Weather Company, an IBM Business; Mike Robinson, The MITRE Corporation; Tim Niznik, American Airlines
Decisions made in the absence of probability-based analytics will not just usually be wrong, they will consistently be sub-optimal, especially when those decisions involve weather. An increasingly important need for both airline operation centers (AOC’s) and air traffic management (ATM) is for improved response to weather events, specifically in terms of quantifying the level of risk around potential courses of action. What is needed is a practical, data-driven process that explicitly considers weather uncertainty in balance with the costs and risks for decision options. Such a capability would yield more consistent and efficient ATM initiatives and AOC responses that reduce avoidable delay, flight cancellations, costs, and passenger disruption. This talk brings together the perspectives of The MITRE Corporation, The Weather Company, and American Airlines who will present a framework to “bridge the gap” from deterministic to probabilistic decision-making using ensemble weather forecasts and multiple “trajectories” of possible weather constraint scenarios.
Exploration vs Exploitation: The Case for using Aircraft as Weather Sensors
Lavanya Marla – University of Illinois at Urbana-Champaign, John Paul Clarke – University of Texas at Austin, Ankur Mani - University of Minnesota
Airlines and the FAA do not have access to accurate and timely weather information for every section of the U.S. airspace. Thus, aircraft must frequently navigate using limited information or utilize `nominal' paths that are likely to be sub-optimal with respect to fuel burn and/or travel time. This problem can be addressed by deviating aircraft from their nominal routes to serve as weather sensors, i.e., using them to collect real-time weather information as they travel through the airspace. We show that spatial-temporal correlations in weather can be leveraged to devise efficient algorithms that explore less traversed sections of the airspace, update information and reduce costs for the system. We also propose a systematic framework that translates desired outcomes (whether they be individual, group, or system-wide) into mechanisms that incentivize participation and enable collaboration between aircraft or airlines.
Wind Optimal Routing with Airspace Constraint Avoidance using CavanReports
Greg Feldman & Tim Myers – Cavan Solutions
Flight path planning and performance analysis considers factors including winds aloft, available routes (e.g., FAA playbook, FAA Coded Departure Routes), and airspace constraints (e.g., turbulence, convection). Access to estimated and actual wind mileage and the wind optimal route is not readilyavailable for real-time decisions and post operations analysis. We developed a wind optimal routing analysis capability using Djikstra’s shortest path algorithm (e.g., branching network graph between each city pair), groundspeed and estimated time enroute (ETE) for each arc segment based on wind, flight plan data, FAA Coded Departure Route (CDR) data, National Oceanic and Atmospheric Administration (NOAA) wind data, and airspace constraint data (e.g., SIGMETs). Key metrics are provided for the route analysis including wind miles, extra wind miles, ETE, extra ETE minutes, optimality compared to the Least Cost route, ground miles, and headwind miles.
Reserve Price Optimization for Airlines’ Ancillary Products
Maelle Zimmermann, Claudio Sole, Pierre-Luc Bacon, Maxime Cohen, Emma Frejinger, Andrea Lodi - Ivado Labs
In this talk, we consider the problem of pricing seat upgrades into premium cabins within a bidding system, where passengers can place a bid amount of their choice (above a certain reserve price) and be ticketed close to departure if their offer is the highest. We propose a stochastic optimization approach to select the reserve prices that maximize revenues. While the optimization problem can simply be solved by explicit enumeration, the key challenge is to learn the distribution of the passengers' willingness to pay for upgrades. The solution leverages a non-parametric empirical distribution estimator for censored data from survival analysis techniques.
Predictive AI/ML Analysis of Maintenance Operations Effectiveness
Vitali Volovoi, Dimitry Gorinevsky - Mitek Analytics
Predictive analyses of Maintenance Operations data can improve process effectiveness. This paper will describe AI/ML tools that extend Reliability Centered Maintenance methods and identify poorly performing parts and aircraft. Given reliability distribution, e.g., Weibull, we find outliers in recurrent failures of aircraft rotable parts by extending Statistics Process Control (SPC) tools to reliability problems. The SPC outliers are bad parts and aircraft that fail abnormally often. Addressing the bad actors improves fleet performance. Second, removing the outliers enables Robust Statistics learning of the reliability distributions going beyond Weibull. The third analysis predicts poor repair of rotable parts in Depot. The SPC outliers across the fleet over time are used to predict Failure-Not-Fixed process in the Maintenance Operations. These analyses are a part of developed AI/ML toolset successfully used at US Air Force and airlines in the last few years.
Meeting AA Scheduling Goals: How We Deliver Tools that Improve Profitability and Reliability
Ronald Chu and Hossein Dashti - American Airlines
It is well known that the objective of a profitable schedule and an operationally reliable one often works against each other, and it has been a long challenge for the OR practitioner to design tools to find a balance for both. In this talk, we discuss some recent successful enhancements to our scheduling tools that improve both objectives. We explain how we implement the Demand Driven Dispatch model without compromising operation integrity while capturing additional bookings and revenue opportunities. We will also discuss how we insert an Equipment-with-Crew process to bring crew pairing and aircraft routing together to enhance operation reliability while creating significant cost saving. While we are still far from a fully integrated model of fleet assignment, crew pairing, and aircraft routing, it is a small practical step towards the laurels everyone is working for.
Amazon Air's Aircraft Recovery Model
Chyi-Fu Hong, Tim Jacobs, Stefan Karisch - Amazon
Amazon air network is a key network for delivering millions of packages on a daily basis. To deal with uncontrollable operational disruption events (e.g., weather, aircraft maintenance, delayed inbound volume) and meet customer promise, Amazon has to develop novel disruption management capabilities for recovering multi-modal air shipments. This presentation will provide an overview of the Aircraft Recovery Model (ARM) which is a new mathematical optimization approach to support Amazon Air’s ability to efficiently recover from network disruptions by trading off between cost and packages saved. The ARM model generates multiple new aircraft routing plans to recover from flight schedule disruptions by re-routing and swapping aircraft while maximizing the number of packages arriving at their final destinations on time. ARM is central to Amazon Air’s novel disruption management system used to manage the network worldwide and generates schedule recovery plans.
Crew Scheduling & Fatigue Risk Management
Hector Barrocal - Copa Airlines
How can we, as Crew Management, contribute to mitigate the Fatigue Risk in our operations?