Here is a summary of the technical program as of 01 May 2018, subject to change and final approval.
Tuesday 22 May 2018
Vice President – Airport Operations
Reactive to Predictive – OCCs over the course of time
The Operations Control Center is the heart of each airline. Working 24/7, 365 days a year, the goal is to respond to irregularities of all kinds. Hundreds of decisions are taken every day. Aiming to optimize those, automated IT support becomes more and more relevant for OCCs. M2P conducted a worldwide market study to analyze the impact of different IT, organizational and process setups of OCCs, best practices and trends. One key insight was that there is nothing like the perfect OCC setup. A low level of IT support does not necessarily mean that the airline is not satisfied with the performance of their OCC and vice versa. An OCC needs to fulfill the specific requirements of its airline and these requirements vary depending on the airlines business model, size, etc. The presentation will contain the key results of the market study including a discussion of 7 identified core trends (Mission Management, Disruptions Recovery, Automated Decision Support, etc.)
A Supervised Machine Learning Approach for Solving the Aircraft Recovery Problem
Commercial optimization solvers have been widely used for solving airline operations recovery problems. However, obtaining high quality solutions within the limited time frames of online applications is very challenging. In this research, we develop a data-driven approach to rapidly and efficiently solve a broad class of aircraft recovery problems. This method can be applied to identify a near-optimal solution for a highly nonlinear, mixed-integer aircraft recovery optimization problem by combining the insights obtained from previously known good solutions generated by computational decision-support tools, human experts, or a combination thereof. The performance of our method was tested by applying it to a set of real-world networks for a moderately large US carrier. The results indicated that the obtained aircraft recovery solutions were, on average, within a small optimality gap and were calculated much faster than those using commercial optimization solvers.
A New Approach of Recovery for the Airline Industry
IBS Software Services
The ever changing factors that contribute towards effective management of disruptions and recovery is evidence enough that the solution lies in empowering and assisting the users with timely information and intelligent tools, rather than leaving it to a press-of-the-button universal solution. It must be instead an IT system designed to help the ops controllers in making decisions during severe disruptions, rather than trying to create a tool designed to replace their expertise. The ops controller will be guided through the entire situation and has full visibility on all relevant factors and disruption impacted aspects. The user applies the possible options and gets immediate feedback about the impact on connecting passenger, crew and aircraft rotations. Completing the recovery scenario with trade-offs and several options, the staff now has a real-time holistic view on the recovery solution.
Artificial Intelligence for Disruption Management
What are the best strategies for recovering from disruptions? How can Artificial Intelligence help? As most disruption situations are unique, it is impossible to detect best recovery strategies from data analysis of the past. With accurate simulations of delay causes, human interference measures and dependencies with other business processes, arbitrary sequences of recovery actions can be tested. At this point Artificial Intelligence comes into play. Artificial Intelligence will find the best possible strategies in such a simulation framework, which then can be applied to real-life problems. Applied to other problems Artificial Intelligence has often been able to identify strategies human brains never considered, as they were contradictory to their experiences or intuition.
Computer-Vision-Based Gate Activity Monitoring
Victor Cheng & Vicky Lu
Optimal Synthesis Inc.
This presentation will discuss research aimed at monitoring airport gate activities with the objective of improving aircraft turnaround. Activities related to turnaround of the aircraft represent a significant source of delay and variability and therefore impact the predictability of the National Airspace System (NAS). Extended from previous research in using computer-vision technology for surveillance of aircraft in the crowded ramp area of an airport, the Gate Activity Monitoring TOol Suite (GAMTOS) is developed to identify the various stages of turnaround, which in turn enables a gate turnaround prediction. This presentation will summarize our experience of using computer-vision technologies for detecting gate activities under different lighting conditions. Using video surveillance data obtained from existing cameras at Charlotte International Airport, we demonstrated that GAMTOS is able to detect various gate events/activities including aircraft arrival, jet bridge attachment to aircraft, jet bridge detachment from the aircraft, tug attachment to aircraft, cargo loading activity, pushback of aircraft, and start of aircraft taxi.
Schedule Recovery with Crew Considerations
Sureshan Karichery & Helder Inacio
Sabre Airline Solutions
We present advancements in solving the airline schedule recovery problem with successful adoption use-cases in the industry. The operational efficiency gained through a holistic, optimization-based schedule recovery solution that also minimizes the impacted crew, enables airlines to better serve their customers. A comparative analysis of the solution with crew considerations is presented. Further, solving this problem involves obtaining a feasible solution for the airline by minimizing several costs and within a short timeframe while also adhering to several operational constraints. Using a real-world dataset, we demonstrate the benefit of crew friendly schedule recovery solutions. The resulting application is able to solve disruption events optimizing costs and respecting operational constraints. The holistic approach can handle a broad range of disruption scenarios.
KLM & BCG Partnership for Operations Optimization
Fernando Bosch & Sander Stomph
Boston Consulting Group
KLM and The Boston Consulting Group have partnered to build on their “Profit Maximizing Operational Performance” or PMOP concept introduced in last year’s AGIFORS Operations conference and have jointly developed a series of disruption management and operations planning tools based on machine learning and decision optimization spanning OCC decision making, schedule optimization and crew disruption management. We will explore the journey we have taken together, the results and impact as well as the collaboration.
Wednesday 23 May 2018
Misconnect Passenger Analysis using a Visual Platform
Misconnecting passengers cause difficulties for our hub locations resulting in rebooking costs, protection issues, over sale situations, and reputation damage. With a constantly moving schedule, irregular operations, and multiple connection opportunities, it is difficult to consume the large amounts of data to identify the problem(s) in misconnections. As a result, we have developed an interactive and visual tool to help stations track progress, find most critical misconnecting city pairs, analyze past events, and recommend realistic minimum connection times based on history, location, time of day, and domestic/international types of connections. This tool not only focuses on the immediate problem, but also the larger solution of adequately using data visualization to help identify, explain, and solve complex problems.
Human Plus Artificial Intelligence for Robust Block Time Forecasting
Accurate block time forecasts are critical to building a reliable and profitable schedule. However, as these two metrics are opposing in nature, the planner has to balances the tradeoffs between reliability and profitability in the plan. In this talk, we present a different and novel approach to forecasting block times using Machine Learning. FICO uses historical on-time performance data and external data such as weather and airport operations to build out a feature set. Using this, we build, tune, and evaluate multiple Machine Learning models. We provide levers to the planner to then analyze the results and find the sweet spot that balances the tradeoffs between OTP and utilization before coming up with the final planned block time. As this is not a one-time exercise, we will present a decision support platform that allows the planner to work collaboratively with the ML algorithms and their human colleagues to rapidly iterate towards optimal planned schedules
Leveraging Data Science to Find Punctuality Quick Wins
Blaise-Raphael Brigaud & Solene Richard
Punctuality has always been a major issue for carriers. Multiple business objectives tend to create and propagate delays such as decreased aircraft turn-around times to improve aircraft usage, or peaked hub activities to improve the intrinsic connectivity of networks. Our study aimed at comprehensively using Air France historical data to make out best practices that could be shared broadly. Using state-of-the-art data science techniques, we ended up following two tracks that lead us to determine simple operational levers to enhance punctuality (or at least theoretically).
Resilient Airline Scheduling and Operations - Optimize On-Time Performance Based on Delay Risk Prediction
Judith Semar & Marius Radde
Lufthansa Systems & DLR
Flight schedules regularly get disrupted on the day of operations by events that were unforeseeable at the time when the schedule was built. While this can hardly be avoided, we aim to minimize delay risks on a standard day of operations. Delays caused by special events or irregularities cannot be taken into account by scheduling, they have to be addressed in a later process phase by ops control. Using operations data from different airlines over several years, we predict the lengths of ground and block times of typical operations days. Based on simulation results we assess the delay risk for each leg in a given rotation to support the scheduler in his daily work, thus enabling him to reduce the number of situations with high delay risks.
Improving On-Time Performance through Robust Routing
Nabin Kafle & Hakan Ergan
The talk will focus on explaining how the on-time performance and maintenance requirements can be improved with the help of robust aircraft routing. A mixed integer problem is formulated and solved which prioritizes the routes with higher buffer time in the ground turn around leg while adhering to maintenance requirements. Application of the model to the airline's schedules shows that at average 8% of the ground turn legs can be relaxed from having the minimum turn around time and can be provided an additional 10 to 20 minutes of buffer time. The maintenance schedule of the aircrafts also improves as the aircrafts touch the maintenance base more frequently with ground time long enough to carry out the periodic maintenances. All these benefits are realized without changing the schedule times therefore not requiring reduction to the aircraft utilization ratio.
Robust Airline Scheduling with Optimal Block and Ground Times
Airlines solve their schedule planning problem considering a little or no stochasticity and to maximize their profitability. Highly optimized schedules show an opposite trend in terms of operational metrics such as On Time Performance (OTP). We develop a simulation-based approach to measure the operability of a given schedule under various uncertainties. Schedule optimizers try to balance tradeoff between the time allocated on ground vs. flying time. Insufficient or excess ground time and block time can cause expensive delays or under-utilization of assets respectively. We aim to distribute right contingency at right places so that optimal OTP can be achieved at minimal cost. We apply an iterative framework using a mixed integer programming formulation and updating its parameters based on the simulation results. We test our approach using a major U.S. airline's daily schedule and obtain 1.81% increase in OTP without a significant decrease in profit.
Managing Airside Facilities and Operations: Adaptive Planning with Historical Perspective
The University of Missouri-St. Louis
With historical background on the forces which shaped the development of the St. Louis Lambert International Airport in the U.S., we discuss the challenge of right-sizing facilities and getting optimal use from airport assets as they approach their capacity limits. We then present a simulation model designed to foster collaboration among major airport stakeholders as they strive to balance airside and groundside activity, reduce delays and plan for future activity.
Working with Alert Management
Traditional systems used in operations control have been based around a Gantt view showing the tail assignments. The operator monitors the Gantt to understand where the most important disruptions are, and then starts addressing these disruptions. In this presentation, we will look at some concepts around alert management that shifts the focus from the Gantt view to web and mobile applications. We will also look at how analytics can be used to predict where and when certain disruptions will happen. Addressing alerts automatically is yet another area that will be covered in this presentation. And finally, we will look at how advanced business rules management can vitalize the Gantt for certain types of workflow.
Dynamic Airline Disruption Management under Airport Operating Uncertainty
The University of Illinois at Urbana-Champaign
Air traffic disruptions are amplified by the uncertainty and variability underlying airport operations. However, existing disruption recovery practices use static and deterministic decision-making schemes. We design proactive approaches to airline disruption recovery, by explicitly considering future operating uncertainty at hub airports. We develop, implement, and evaluate an original dynamic airline recovery paradigm that integrates a stochastic model of airport congestion into an optimization model of disruption recovery. Our novel stochastic optimization approach combines a queuing model of airport congestion, an engineering model of flight planning and a large-scale optimization model, formulated as a Stochastic MIP. Our approximation algorithmic method lowers expected operating costs by leveraging network-wide airport delay information. Results suggest that the integration of airport congestion scenarios can reduce expected recovery costs by 1% to 4%.
Thursday 24 May 2018
Leveraging Probabilistic Weather Forecasts and Machine Learning in Airport Capacity Prediction
The Weather Company – IBM
Although probabilistic weather forecast provides airline flight operations department with the ability to examine the confidence of a forecast, such forecast need to be translated to actionable system impact, such as the timing, duration and magnitude of airport capacity reduction and recovery scenarios, so that decision-makers can associate weather to actual National Airspace System impact. This presentation covers three major topics. First, we provide an overview to probabilistic weather forecast and its impact to airline operations. Second, we devise a novel approach using machine-learning methods to predict airport capacity at an airport. The approach focuses on predicting the timing of the capacity change since this objective is critical for airlines to proactively make decisions when facing irregular operations. Third, we discuss a method in consolidating different capacity reduction and recovery scenarios for human interpretations.
Improvement of On-Time Performance at Airline Hubs: A Case Study and Best Practices
Growth in demand for commercial air transport has only recently been outpaced by market supply. Predominantly Hub carriers are therefore increasingly pressured to focus on customer experience and service delivery. OTP is a major operational KPI and a key lever to attract and retain customers. Top ranking airlines perform close to a 90% level throughout the year, while most airlines set a minimum goal of 80%. Improving OTP has a qualitative and quantitative benefit. Aircraft delays are a lost monetary value which, for the average network carrier, is calculated around 50$ per minute. Influential factors on OTP are numerous, ranging from crew management, ground and terminal handling to network scheduling. M2P is experienced in identifying levers for OTP improvement by moderating a dialogue with all relevant stakeholders and selecting suitable IT solutions. We would like to share our success stories in the form of case studies and visualize our approach.
Prerequisites for Excellent Operational Control
Lufthansa Systems - Airline Consulting
The speed of decision making is the essential factor of influence to the quality of decisions executed by the OCC. Even if the operator is using expensive tools and masses of data the result of the problem solving and decision making process would be below expectations if processes and procedures are not lean and simplified beforehand. But simple processes need flexibility and flexibility needs outstanding competence and quality of the people and their tools. In this presentation the team of Airline Consulting of Lufthansa Systems would like to demonstrate the consequences of slow decision making in operational control and the methods improving the quality of the OCC result. The new ATM-concepts (i.e. SESAR, NextGen) will change the OCC decision making rules and standards already in the near future. The CDM-standards will increase the operational risk with increasing costs if the OCC’s have not prepared the organization for a fast and effective adjustment of data processing, competence development and the setup of tools.
Departure Readiness Prediction Using Machine Learning
Shana Parhizi & Tim Niznik
A frustrating but accepted reality at busy U.S. airports—the inevitability of a “conga line” of aircraft on the taxiway waiting for departure—could potentially fade into history before the turn of the decade thanks to a NASA -built scheduling system called ATD-2. This set of technologies and algorithms fuses data from the FAA, airports, airlines and others to meter the movement of aircraft from the gate to the runway and into the surrounding airspace. A key input into this system is an estimate of flight readiness provided by the aircraft operators. As the Lead Carrier for NASA’s ATD-2 project, American Airlines undertook an effort to improve the accuracy of its own departure readiness predictions. This talk will describe some of the initial approaches and analyses that led ultimately to the development of a machine learning model based on the Random Forest method. We will discuss the relevant features, the journey to deployment, the current performance, and the future directions of our model and this project.
Data Driven Insights Transforming Optimization of Network Operations
GE is actively working in airline operations and applying Data Science broadly from the planning process through the day of operations. We will highlight the use of Machine Learning and Simulation to advance the potential for Dynamic Optimization of operations fulfillment to meet the business goals of revenue & margin, passenger experience and operations performance.
Towards a Block Chain-enabled Aviation Ecosystem
Georgia Institute of Technology
Airline operations is replete with regulatory requirements with respect to the documentation of events, especially those related to maintenance, repair, and overhaul (MRO). Currently, however, much of that documentation is done manually. Thus, the information contained within the documents cannot be easily verified and validated, and the associated data is not readily available for analytics. In this presentation, we will explore the use of electric documentation and block chains as a basis for verification and validation of MRO events, and as the foundation of an analytics and optimization structure that could lead to greatly improved operations.
An Automated Disruption Management Tool
Embraer has been researching irregular operations management problems over the last 3 years and it has been developing an automated disruption management tool that aims to puts together aircraft routing recovery, crew pairing recovery and passenger re-accommodation in the same decision framework. This presentation will introduce the Embraer initiative and developments on the disruption management decision support system, briefly reviewing models and architecture adopted.
Operational Maintenance Planning
We present the achievements of a proof of concept developed for Qantas Airways as an extension of the already existing disruption recovery tool. A subset of Qantas fleet has been considered for planning the maintenance events over a period of 40 days following the day of operations. The decisions made must maximize the utilization of the tails but also respect the workforce available at the different maintenance bases. As schedule changes are proposed, the operations control centre needs approval from the maintenance operations centre to make sure maintenance deadlines are still met. In this capacity and time constrained context a fast and consistent resolution method is required. We show the mathematical model and the results carried out during the POC and discuss the integration with the current tool.
En Route Flight Time Prediction Under Convective Weather Events using Machine Learning
Iowa State University
Flight en route time is expected to be longer when traffic demand exceeds airspace capacities. When demand is high, flights can often be subjected to rerouting, Miles in Trail and vectoring, etc. During summer months, convective weather, which could severely affect the capacities of major air routes and airports, accounts for the most significant share of weather-relate delays. Flight en route time prediction could help airline dispatchers and traffic management coordinators make strategic plans, and alert travelers of potential delays. Our work aims to use machine learning techniques to model traffic volume and especially the effect of convective weather within 100 miles of great circle line between city pairs on the en route time. Several best off-the-shelf algorithms are tested and compared. These algorithms are trained and validated on four data sources: historical flight punctuality data, NEXRAD level 3 data, surface weather data and wind aloft data.