Periodic Reporting for period 2 - SafeOPS (From Prediction to Decision Support - Strengthening Safe and Scalable ATM Services through Automated Risk Analytics based on Operational Data from Aviation Stakeholders)
Período documentado: 2022-01-01 hasta 2022-12-31
The next generation Air Traffic Management systems are pushed more and more towards digitization, fueled by the demands to increase capacity and cost-efficiency while also increasing the already high safety and resilience levels. SafeOPS proposed and investigated a decision support tool for Air Traffic Controllers, which provides real time, predictive risk information on the likelihood of approaches to perform a go-around. The underlying idea is that Air Traffic Controllers incorporate this predictive information in the handling of departures and approaches, especially in high traffic situations.
As of today, Air Traffic Controllers recognize the onset of go-arounds through pilots' communication and from observing the aircraft visually or via radar. Go-arounds are a standard and well-established flight procedure, for flight crews as well as for Air Traffic Controllers, despite the relatively low likelihood. Strategies to handle go-arounds however are of reactive nature, meaning they become active once the Air traffic Controller identifies the ongoing go-around. Especially in high traffic congestions, handling a go-around becomes complex, since knock-on effects like separation infringement or wake turbulence challenges with preceding departures can arise. Air Traffic Controllers are trained for such situations, nevertheless, resolving such situations, while maintaining safe separation, increases the workload of the Air Traffic Controller, as well as the Flight Crew.
The idea when providing predictive risk information in this scenario is to thereby improve the situational awareness and thus the decision-making of Air Traffic Controllers, by enabling a proactive approach in handling go-arounds. This shift from reactive to proactive go-around handling could avoid the described knock-on effects, possibly triggered by go-arounds, yielding a positive impact on safety and resilience but also capacity of Air Traffic Management.
The question addressed by SafeOPS was, how predictive tools, and the inherently probabilistic nature of their outputs, will change the approach and departure handling of Tower Controllers. Can predictive tools increase the safety and cost-efficiency and can the resilience of the system be maintained or further improved.
The main objectives of this project towards answering these questions were, in the scope of the proposed go-around handling context, to:
1. develop an AI/ML tool for go-around predictions and explore it in terms of achievable performance metrics as well as explainability,
2. enhance a risk assessment method, such that it can cope with the introduced AI/ML component, and
3. investigate the AI/ML based decision support solution for ATM, and evaluate the effects on capacity, safety and resilience of the ATM operation.
In conclusion, SafeOPS found that these proactive solutions benefit the safety and resilience of the ATM system in complex go-around situations, by providing the Air Traffic Controllers with more time and better information for the necessary coordinative actions, which have to be taken in the event of a go-around. On the contrary, the proactive tactics can reduce capacity/efficiency, in case of false prediction, but only in an amount, which is negligible, compared to the foreseen overall increase of capacity. Thus, predictive risk information can be used as a decision support between a reactive and proactive approach to handle go-arounds, especially in high traffic situations.
As of today, Air Traffic Controllers recognize the onset of go-arounds through pilots' communication and from observing the aircraft visually or via radar. Go-arounds are a standard and well-established flight procedure, for flight crews as well as for Air Traffic Controllers, despite the relatively low likelihood. Strategies to handle go-arounds however are of reactive nature, meaning they become active once the Air traffic Controller identifies the ongoing go-around. Especially in high traffic congestions, handling a go-around becomes complex, since knock-on effects like separation infringement or wake turbulence challenges with preceding departures can arise. Air Traffic Controllers are trained for such situations, nevertheless, resolving such situations, while maintaining safe separation, increases the workload of the Air Traffic Controller, as well as the Flight Crew.
The idea when providing predictive risk information in this scenario is to thereby improve the situational awareness and thus the decision-making of Air Traffic Controllers, by enabling a proactive approach in handling go-arounds. This shift from reactive to proactive go-around handling could avoid the described knock-on effects, possibly triggered by go-arounds, yielding a positive impact on safety and resilience but also capacity of Air Traffic Management.
The question addressed by SafeOPS was, how predictive tools, and the inherently probabilistic nature of their outputs, will change the approach and departure handling of Tower Controllers. Can predictive tools increase the safety and cost-efficiency and can the resilience of the system be maintained or further improved.
The main objectives of this project towards answering these questions were, in the scope of the proposed go-around handling context, to:
1. develop an AI/ML tool for go-around predictions and explore it in terms of achievable performance metrics as well as explainability,
2. enhance a risk assessment method, such that it can cope with the introduced AI/ML component, and
3. investigate the AI/ML based decision support solution for ATM, and evaluate the effects on capacity, safety and resilience of the ATM operation.
In conclusion, SafeOPS found that these proactive solutions benefit the safety and resilience of the ATM system in complex go-around situations, by providing the Air Traffic Controllers with more time and better information for the necessary coordinative actions, which have to be taken in the event of a go-around. On the contrary, the proactive tactics can reduce capacity/efficiency, in case of false prediction, but only in an amount, which is negligible, compared to the foreseen overall increase of capacity. Thus, predictive risk information can be used as a decision support between a reactive and proactive approach to handle go-arounds, especially in high traffic situations.
How big data and artificial intelligence based decision support systems could impact daily air traffic operations has not been explored yet. Over the course of the project, the SafeOPS team held several workshops together with air traffic controllers from two major European hubs to elaborate this question in the context of go-around handling. Oriented on the three main objectives, SafeOPS performed research on three important aspects:
1. the underlying machine-learning technology,
2. the risk of incorporating machine-learning algorithms in Air Traffic Management,
3. how controllers can use the envisioned solution in their daily work and what effect the solution has on the safety, resilience and capacity.
Regarding the first aspect, SafeOPS produced two deliverables, describing the developed IT-infrastructure for gathering, cleaning, and fusing several aviation related data sources in an automated way. Based on the thereby generated data set, a benchmark study for several machine learning algorithms, regarding their suitability as go-around predictors, was performed. Based thereon, a go-around prediction prototype was developed, which was used as input to the remaining two aspects. This work was also published in a paper at DASC Conference 2022.
Regarding the second aspect, SafeOPS adapted Eurocontrol’s Accident Incident Model Risk Framework, to incorporate the envisioned machine learning tool and the uncertainty it introduces to Air Traffic Management. This work also reflects the human factors aspect and was documented in two deliverables.
Regarding the last aspect, and based on the results of the first two aspects, SafeOPS developed an initial Concept of Operation for the envisioned decision support tool. To investigate the impact of the tool on the operation, a low-fidelity simulation was developed and performed with Air Traffic Controllers. This work was presented at the EASN Conference 2022 and in the relevant Eurocontrol as well as EASA committees.
1. the underlying machine-learning technology,
2. the risk of incorporating machine-learning algorithms in Air Traffic Management,
3. how controllers can use the envisioned solution in their daily work and what effect the solution has on the safety, resilience and capacity.
Regarding the first aspect, SafeOPS produced two deliverables, describing the developed IT-infrastructure for gathering, cleaning, and fusing several aviation related data sources in an automated way. Based on the thereby generated data set, a benchmark study for several machine learning algorithms, regarding their suitability as go-around predictors, was performed. Based thereon, a go-around prediction prototype was developed, which was used as input to the remaining two aspects. This work was also published in a paper at DASC Conference 2022.
Regarding the second aspect, SafeOPS adapted Eurocontrol’s Accident Incident Model Risk Framework, to incorporate the envisioned machine learning tool and the uncertainty it introduces to Air Traffic Management. This work also reflects the human factors aspect and was documented in two deliverables.
Regarding the last aspect, and based on the results of the first two aspects, SafeOPS developed an initial Concept of Operation for the envisioned decision support tool. To investigate the impact of the tool on the operation, a low-fidelity simulation was developed and performed with Air Traffic Controllers. This work was presented at the EASN Conference 2022 and in the relevant Eurocontrol as well as EASA committees.
There has already been, and are still ongoing, various research activities within the SESAR JU on advanced automation and the use of data analysis solutions in the Air Traffic Control environment. Also other initiatives, such as Eurocontrol's System-Wide Information Management (SWIM), aim to reveal the potential, incorporated in collected data and the need to further support the decision taking of air traffic controllers.
SafeOPS will progressed the state of the art, complementing existing and past projects, and providing new insights and analysis on how the increase in digitization and automation could impact the safety and resilience in the ATM industry. We find that, although extensive research is dedicated to new tools, algorithms and interfaces, the management of the inherent but mathematically measured uncertainty that AI/ML tools provide, will help to progress on how this new support tools affect the decision making processes of ATCO's.
SafeOPS increased the body of knowledge of safety and resilience in the context of new digitalization tools supporting ATM. Particularly, SafeOPS contributed by developing an operational risk framework that leverages the existing research on machine learning. Outcomes of analytical methods, including predictions, are integrated into the operational environments. Implications, benefits, and disadvantages associated with the outcomes were carefully analyzed. The close interaction between aviation stakeholders, including an ANSP, two airlines, and research institutions, with respect to applied data science was a novelty in the research field of ATM.
SafeOPS will progressed the state of the art, complementing existing and past projects, and providing new insights and analysis on how the increase in digitization and automation could impact the safety and resilience in the ATM industry. We find that, although extensive research is dedicated to new tools, algorithms and interfaces, the management of the inherent but mathematically measured uncertainty that AI/ML tools provide, will help to progress on how this new support tools affect the decision making processes of ATCO's.
SafeOPS increased the body of knowledge of safety and resilience in the context of new digitalization tools supporting ATM. Particularly, SafeOPS contributed by developing an operational risk framework that leverages the existing research on machine learning. Outcomes of analytical methods, including predictions, are integrated into the operational environments. Implications, benefits, and disadvantages associated with the outcomes were carefully analyzed. The close interaction between aviation stakeholders, including an ANSP, two airlines, and research institutions, with respect to applied data science was a novelty in the research field of ATM.