Why is your flight late? Mining airline data to assess root-causes and impact of delay propagation

Goran Stojkovic
Boeing Global Services

 

Air transportation systems are exposed to daily disruptions, which have significant impact on operations. Airlines operate tight schedules to maximize resource utilization. However, the lack of sufficient buffers often result in propagating delays. In this paper, we propose a framework for automatic detection of root-causes of delays and their propagation effects. We test our framework on empirical data of several airlines. Presentation on one case study where airline is prone to delay propagation through passenger connections. Additionally, majority of those delays are related to airport capacity, resource allocation, and passengers, and mainly originate from the hub. Obtained results and future research focus areas are  discussed.

Kick starting your analytics program, key tools and data required for maintenance optimization

Helmuth Naumer

IBM

 

Predictive maintenance, knowing parts will fail before they fail, is the Holy Grail for airline technical operations.  Many airlines struggle with getting cognitive solutions up and running. Cognitive tools have matured, and offer a great foundation for analyzing task cards, log books, and other maintenance data. By providing an analysis of historical data to line maintenance or maintenance control, technical operations can resolve faults faster, which will reduce their maintenance costs, aircraft downtime, and AOG’s.  Predictive maintenance combines this analysis with reliability and operational data/analysis/modeling.  Those airlines that take advantage of these new technologies will greatly outperform their competitors and show much higher operational margins.  This presentation will summarize the steps technical operations can take to initiate the journey to predictive maintenance while gaining many benefits along the way.

 

Digital Operations Transformation

Fernando Bosch

BCG

 

The BCG - KLM partnership has advanced since last year with optimization and AI tools now integrating advanced machine learning to have decision support tools interacting with each other across silos (OCC, Crew, Ground Services) and across time horizons (long/ medium term planing and day of execution. We would like to discuss how the models interact with each other and how that enables changes in the way of working, required skill sets, planning processes, etc.

Schedule planning with hub infrastructure constraints

Alexandre Pizzut

Air France KLM

Network planning is not only a difficult commercial offer problem. All kinds of operational constraints should be accounted for too. They can range from gate allocation to limiting activity on hubs to keep costs low. Air France has prototyped the CASTOR optimization tool to find the best trade-off between operations and revenue.

Machine Learning Application for Automatic Analysis of Aircraft Operations Interruption Reports

Wlamir Olivares Loesch Vianna, Juliano Elias Cardoso Cruz

Empresa Brasileira de Aeronáutica S.A (EMBRAER)

 

Events such as aircraft failures or operation interruptions are recorded to provide reliability data to authorities, manage operations performance and provide directives for airframers and equipment manufacturers to engage in engineering campaigns for product improvements. In general, these events are reported as free text and must be interpreted and structured for further statistical analyses and relevant information creation. This process is repetitive and time consuming. This work proposes a novel method to automate this process by the usage of text mining and supervised machine learning techniques for several indications, such as: fault class; contingency action performed; and technical accountability reporting the potential responsible for the interruption (e.g. weather). A numerical implementation using historical field data is used to demonstrate the performance and potential benefits and drawbacks of the proposed method.

OnD based Scheduling - Stop scheduling flights, start scheduling networks

Felix Windisch, Zuehal Aydin

Lufthansa Systems

This presentation aims to provide different concepts of how to provide OnD PAX forecasts to schedulers within a scheduling system for enhanced decision support. We outline general prerequisites of OnD decision support features and differentiate possible concepts by their requirements for new business logic (connection building) and historic data (with/without MIDT data). For each concept we discuss strengths and weaknesses, opportunities and threats (associated challenges).

Key objective of presenting at AGIFORS SSP is to initiate a scientific discussion about the presented concepts.

Short-Term Forecasting of Global Passenger Traffic Given Schedule Changes

Ankur Mawandia, Prashanth Sriram

Amadeus

The prediction of changes in route-level traffic in the near future, is necessary for airlines to take the right decisions to minimize losses or maximize capture of revenue opportunity. While the estimation of historical passenger traffic on all the global O&Ds by airline and routing is a problem with relatively mature solutions in the industry, the same cannot be said of forecast passenger traffic.

The key challenges include: responding to changes in the schedule; prediction at a sufficiently granular level allowing for interpretation through the O&Ds flowing over the route; and automating the entire process.

We present a forecasting approach that meets these challenges, through the following key components:

1. A connection building algorithm;

2. A quality of service index (QSI) model;

3. A capacity constrained global error minimization approach to traffic estimation;

4. A time series-based forecasting approach using the QSI score as an exogenous regressor in addition to the passenger time series;

5. A gravity model to arrive at a rough estimate of passenger numbers for hitherto unseen O&Ds and itineraries.

Forecasting passenger aircraft demand over a 20 year horizon

Romain Doucet, Rohan Nanda

Airbus

Airbus annually produces a 20 year passenger aircraft demand forecast based on market assessments, to provide recommendations for product policy. In this presentation, we give an overview of our methods and show how we abstract away complexity at the most granular level to ensure robustness of the forecast. This includes the choice of region aggregation for estimating traffic growth based on economic factors, network development to simulate fragmentation, and extrapolating trends in aircraft utilization - without attempting to replicate the complexity of daily airline operations & constraints. We then highlight the ways in which optimization models are used to effectively allocate both existing & forecasted aircraft to their appropriate routes/airlines to match the predicted traffic demand. Finally, we give a brief introduction to the system architecture & developments underway to enable scenario analyses and add additional robustness into the process.

Choice-Based Airline Schedule Design and Fleet Assignment

Vikrant Vaze

Dartmouth College

Chiwei Yan, Cynthia Barnhart

MIT

We study an integrated airline schedule design and fleet assignment model by simultaneously selecting optional flights and assigning aircraft types to scheduled flights. This is a crucial tactical decision, which greatly impacts airline profitability. In particular, we study a model that incorporates passenger choice behavior in selecting from the available fare products and the revenue management policies. This integration is achieved by some recent work on compact sales-based deterministic linear programming formulations of network revenue management under discrete choice models. However, the model suffers from formidable computational challenges, which prevent it from being adopted in the airline industry despite its potential for considerable profit improvement. To tackle this challenge, we develop a reformulation, decomposition and approximation scheme to maintain the tractability of such model. Computational experiments confirm that, for a full-size instance from a United States airline, the developed approach produces significant profit improvements over state-of-the-research baseline approaches under the same computational budget.

Robust Flight Operation Planning Using Delay Simulation

Mohammad Kouhi, German Gutierrez, Francesc Pons Llopis, Icamaan Bothelo, Rita Barata

Vueling Airlines

In order to assess the robustness of the flight schedule, a Monte-Carlo simulation model has been developed where the propagation of the delay through the network is mathematically modeled. The flight process including Turn Around Time (TAT) as well as Block Time (BT) is simulated along the flight rotations in order to estimate arrival and departure delays. TAT and BT are considered as the random variables extracted from the historical data of the previous years for each departure airport/arrival airport/departure time window/aircraft type configuration. The comparison of the simulation results with the actual value demonstrate the capability of the developed simulator.

Efficient Network Planning using Heuristics and Machine Learning

Deepak Agrawal, Jataveda Dasgupta, Hari Sreekumaran

American Airlines

Network creation and demand forecasting are the key steps to any airline’s schedule planning. Although this problem has been studied for more than three decades, the existing methodologies are either inaccurate or time expensive. In this talk, we will present a network constructor algorithm which can efficiently produce worldwide network while minimizing both Flown not Built (FnB) and Built not Flown (BnF) itineraries. We will also present the drawbacks of traditional QSI models for passenger demand forecasting and propose an ensemble based machine learning model for more accurate prediction.

Predicting Schedule Reliability

Jeff Warren, Benjadol Chakravut, Ramakrishna Thiruveedhi, Sergey Shebalov

Sabre Airline Solutions

Airlines traditionally use profitability forecasting methodologies as a main guiding principle during schedule development. However, schedule reliability is an equally important objective and hence airlines need to be able to predict it ahead of operations. We examine the strengths, weaknesses, and potential of Sabre's simulation tool for evaluating the reliability of a flight schedule. We consider its accuracy as a forecasting tool and its ability to inform business decisions. We also explain on specific examples how this capability can be used for making recommendations on optimal schedule design.

How much to Invest in Block Times?

Interactive Session led by Cristian Ricardo Carrizo Vergara - Latam Airlines