Common Machine-Learning (ML) approaches for scene classification require a large amount of training data. However, for classification of depth sensor data, in contrast to image data, relatively few databases are publicly available and manual generation of semantically labeled 3D point clouds is an even more time-consuming task. To simplify the training data generation process for a wide range of domains, we have developed the BLAINDER add-on package for the open-source 3D modeling software Blender, which enables a largely automated generation of semantically annotated point-cloud data in virtual 3D environments. In this paper, we focus on classical depth-sensing techniques Light Detection and Ranging (LiDAR) and Sound Navigation and Ranging (Sonar). Within the BLAINDER add-on, different depth sensors can be loaded from presets, customized sensors can be implemented and different environmental conditions (e.g., influence of rain, dust) can be simulated. The semantically labeled data can be exported to various 2D and 3D formats and are thus optimized for different ML applications and visualizations. In addition, semantically labeled images can be exported using the rendering functionalities of Blender.

Evaluating the performance of complex systems, such as air traffic management (ATM), is a challenging task. When regarding aviation as a time-continuous system measured in value-discrete time series via performance indicators and certain metrics, it is important to use sufficiently targeted mathematical models within the analysis. A consistent identification of system dynamics at the evaluation level, without dealing with the actual physical events of the system, transforms the analysis of time series into a system identification process, which ensures control of an unknown (or only partially known) system. In this paper, the requirements for mathematical modeling are presented in the form of a step-by-step framework, which can be derived from the formal process model of ATM. The framework is applied to representative datasets based on former experiments and publications, for whose prediction of boarding times and classification of flight delays with machine learning (ML) the framework presented here was used. While the training process of neural networks was described in detail there, the paper shown here focuses on the control options and optimization possibilities based on the trained models. Overall, the discussed framework represents a strict guideline for addressing data and machine learning (ML)-based analysis and metaheuristic optimization in ATM.

Reliable and predictable ground operations are essential for 4D aircraft trajectories. Uncertainties in the airborne phase have significantly less impact on flight punctuality than deviations in aircraft ground operations. The ground trajectory of an aircraft primarily consists of the handling processes at the stand, defined as the aircraft turnaround, which are mainly controlled by operational experts. Only the aircraft boarding, which is on the critical path of the turnaround, is driven by the passengers' experience and willingness or ability to follow the proposed procedures. We propose a machine learning approach predict the boarding time. A validated boarding simulation provides data input for a recurrent neural network approach (discrete time series of boarding progress). In particular we use a Long Short-Term Memory model to learn the characteristic passenger behaviors over time.

In this paper we address the prediction of aircraft boarding using a machine learning approach. Reliable process predictions of aircraft turnaround are an important element to further increase the punctuality of airline operations. In this context, aircraft turnaround is mainly controlled by operational experts, but the critical aircraft boarding is driven by the passengers’ experience and willingness or ability to follow the proposed procedures. Thus, we used a developed complexity metric to evaluate the actual boarding progress and a machine learning approach to predict the final boarding time during running operations. A validated passenger boarding model is used to provide reliable aircraft status data, since no operational data are available today. These data are aggregated to a time-based complexity value and used as input for our recurrent neural network approach for predicting the boarding progress. In particular we use a Long Short-Term Memory model to learn the dynamical passenger behavior over time with regards to the given complexity metric.