State-of-the-art of cognitive computing algorithms
In the first six month of the project the consortium carried out a state of the art of cognitive computing algorithms. The state of the art consists in a review of machine learning and cognitive computing algorithms and their application in several domains, from healthcare to automotive, education and of course, aerospace. The review included the most used as well as the most promising techniques to solve artificial-intelligence tasks for each domain.
This task was the basis to explore and start defining a set of use cases which an AI based virtual assistant would be enable and support Single Pilot Operations.
HARVIS Artificial Intelligence Virtual Assistant Use Cases
The Virtual Assistance concepts and use cases have been performed by the consortium following an iterative selection process, with the contributions of different internal and external experts (Safety, Human Factors, Operational and technical experts).
The initial set of eight Use Cases has been discussed and detailed in feedback sessions and workshops. From this work resulted two Use Cases that have been selected based on their potential to greatly support pilots .
HARVIS Concepts and use cases generation and selection process
During the approach the workload is high in the cockpit because the goal is to stabilize the A/C for the landing. If not stabilized, pilots will have to perform a go-around. In standard operations, the workload is usually very high for both pilots. The PF is focus on flying the A/C maintaining it on the flight path with adapted flight parameters and calling configuration changes while the PM is monitoring what the Pilot Flying is doing and puts efforts in keeping a good situation awareness. In single pilot operations, i.e. without the PM, the PF could have difficulties to multitask and could be subjected to cognitive tunneling.
The assistant will be trained to provide real-time support on the decision-making on the need to perform a go-around or land. The algorithm training will be based on the expert pilot classification of several realistic non-stabilized approach operational scenarios (collected via on-line survey platform).
Non-Stabilized Approach Assistant Concept Summary
This situation is what is called an abnormal situation that pilots are rarely facing. In this timely constraint situation, workload and stress tend to increase, in standard operations and even more in Single pilot operations. The objective of pilots in this situation is to mitigate the consequences of the potential system failure while gathering all the necessary information to prepare a possible emergency landing.
The assistant aim is to support the pilot in anticipating the possible variations in the arrival routes and providing the most likely arrival trajectories that the ATC would suggest.
Deep Learning model based on Multi-Layer Perceptron capable of determining the approach trajectory of aircraft flying prior to the expected landing time, using only METAR information.
Aircraft Dynamic Rerouting Assistant Concept Summary