Sponsor presentation: Boeing

Technical presentation: Listening to Crew Feedback: The Artificial Intelligence Approach

by Alexander Motzek

Daily thousands of feedback comments are written by cabin and gate crew members. Those comments contain extremely valuable information, but often remain unheard. If analyzed, they reveal pain-stacking problems in handling, operations, catering, technical flaws, or general unhappiness which otherwise goes unnoticed. Unfortunately, the sheer number of comments to be read prevents any quick and holistic analyses. Often, only ad-hoc, single-case occurrences are followed without any clear situational awareness. If feedback is not responded to, initial reporters become frustrated, as, e.g., the same reported seat malfunction may still not be fixed when boarding the same aircraft, a week later; no feedback loop appears to be present. Fortunately, huge advancements in artificial intelligence in recent years have made it possible to turn thousands of written feedbacks into easily digestible insights. We have built an automated classification model that quickly highlights insights about emerging problem clusters and presents them for visual drill-down into clusters. On top of that, we solve the feedback-loop problem by transforming written action statements from mechanic inspections into a direct visualization of a cabin state for all crew members. In this talk, we present our insights, discuss approaches, and show future applications of artificial intelligence in cabin crew management.

Technical presentation: Practical Applications of Machine Learning on Roster Data: Predicting Absences and Measuring Satisfaction

By Sara Kamel, Karim Maarouf

Crew roster data in its raw form is quite challenging to make use of. However, transforming this data into an easily accessible format and combining that with the application of machine learning techniques unlocks its huge potential.

In this talk, we explore the application of machine learning algorithms on crew roster data. We will share some of the challenges and insights from our work on the OPSD project at Swiss. We will demonstrate how integrating crew roster data into a cloud analytical platform provides opportunities enabled by machine learning to improve operations. We focus on two use cases, namely predicting crew absences and measuring crew satisfaction. However, that is certainly not the end of the story. Once data is integrated and organized, this enables rapid ideation and prototyping of other use cases. The benefits are limited only by the creativity in applying some of these advanced techniques. We also discuss some considerations and lessons learned from this journey.

Sponsor presentation: TA Connections

Coffee break

Sponsor presentation: VeeOne 

Technical presentation: Innovation in crew rostering: standby’s, fairshare distribution and pre-assignments

by Steven Rushworth

We have developed a new crew rostering system that is now in production at several airlines. The pairing and rostering, typically separate steps in crew scheduling, are now calculated simultaneously in a single run. Our system generates personalized pairings for each crew member that take into account their preferences and preassigned activities. This leads to roster with a lower cost and a higher blockhour/duty hour ratio.

Based on these foundations we were able to continue to innovate and push the boundaries even further. Some of the recent innovations are:

• The standby planning is now integrated in the calculation. If a crew member executes a standby, there is no snow-ball effect on the remainder of his/her schedule. A single plan that includes all duties and standby duties is calculated in one shot.
• As a preprocessing step we can identify the critical pre-assignments, such as a training course, a meeting or SIM duty, that can cause challenges in the scheduling problem. By rapidly identifying these pre-assignments, we can suggest alternatives that will lead to a more efficient roster.
• Having personalized pairings allows a better fair share distribution of the work. For example, the number of nights away from base or block hours / week can be better distributed amongst the available crew members. Similarly, we can also take better into account the crew duty history to distribute the work more evenly over longer periods of time

During this talk we will discuss the details of this approach and discuss the benefits it brings for the operations and crew.

Technical presentation

by Franziska Burmester

Lunch

Technical presentation: Integrated planning

A Robust Pairing Model for Airline Crew Scheduling

By Vikrant S.Vaze

Delays and disruptions in airline operations annually result in billions of dollars of additional costs to airlines, passengers, and the economy. Airlines strive to mitigate these costs by creating schedules that are less likely to get disrupted or schedules that are easier to repair when there are disruptions. In this paper, we present a robust optimization model for the crew pairing problem, which generates crew schedules that are less likely to get disrupted. Our model allows adding robustness without requiring detailed knowledge of the underlying delay distributions. Moreover, our model allows us to capture in detail the delay propagation through crew connections and the complex cost structure of the pay-and-credit crew salary scheme, thus enabling us to find a good trade-off between the deterministic component of the planned costs on the one hand and the expected delay and disruption costs on the other hand. Our robust crew pairing model is based on a deterministic crew pairing model formulated as a mixed-integer linear program. The robust version that we propose retains the linearity of the constraints and objective function and thus can be handled by commercial solvers, which facilitates its implementation in practice. We propose and implement a new solution algorithm for solving our model to optimality. Several optimal solutions with varying robustness levels are compared for the network of a moderate-size airline in the United States. We test the model’s solutions in a simulation environment using real-world delay data. Our simulation results show that the robust crew pairing solutions lead to lower delays and fewer instances of operational infeasibilities, thus requiring fewer recovery actions to address them. We find that, with the inclusion of robustness, it is possible to generate crew pairing solutions that significantly reduce the delay and disruption costs with only a small increase in planned costs.

Presentation TBD

Coffee break

Presentation TBD

Awards for "best presentation" and "best innovation"

Conference recap and farewall