This section introduces the top layer of our pattern identification process. Our hierarchical
classification, from bottom to top, mirrors the structure of natural language, with
higher level representing more abstract or qualitative concepts. For instance, the patterns
introduced earlier help define a controller workload based on factors like the number of
aircraft under the controller's supervision or the number of instructions issued. However, such a
description remains predominantly quantitative and doesn't capture the actual workload,
i.e., what the controller truly handles. The challenge lies in quantifying the cognitive
workload—the number of potential issues a controller manages simultaneously.
Another facet of performance is efficiency, which necessitates an understanding of the
strategies employed by the controller. Lastly, the concept of operational risk
evaluates the "difficulty" of a situation, considering what is at stake or how much the situation could
deteriorate.
Below, we delve into how these concepts are captured through specific patterns, relying on lower-level
patterns. We initiate with a general proximity pattern, versatile in various applications.
Before delving into the core topic, it's essential to provide context. The lower-level patterns introduced earlier can be likened to the words in our natural language, and now we progress to forming sentences, wherein we manipulate words to express ideas or concepts. The proximity pattern is distinct; it serves as a Swiss Knife—a mini-toolbox with diverse applications, as elaborated further. This pattern describes "how close" a pair of aircraft has been over time, extracted from air surveillance data. Assume we have a pair of aircraft trajectories as depicted in the left figure. To determine proximity, three parameters are defined, from which rules can be easily derived.
The steps are as follows: Firstly,
. Secondly, from any identified controller's instruction,
horizontal scenario.
Thirdly, estimate the minimal horizontal separation for both
(denoted
$sep_{cur}$) and for
(denoted
$sep_{scen}$). By minimal horizontal separation, we mean: how close the pair of
aircraft has been over time.
Finally, determine a time anticipation parameter (denoted $t_{anticip}$) as the duration
between the controller's instruction and the crossing of trajectories, as
.
The intuitive meanings of our three parameters are as follows: $sep_{cur}$ represents how close
the aircraft pair has been in reality, while $sep_{scen}$ represents how close it would have been
if the controller had not intervened. Therefore, if $sep_{scen}$ is "small" and
$sep_{scen}$ is "significantly greater than $sep_{cur}$," we are in a situation where
a potentially conflicting pair of aircraft has been separated by the controller.
The third parameter, $t_{anticip}$, offers further evidence toward conflict avoidance by the
controller. If $t_{anticip}$ is very large (e.g., half an hour), the controller's instruction is
likely unrelated to the conflicting pair, as controllers don't anticipate conflicts to that extent. So,
$t_{anticip}$ must remain within reasonable bounds (e.g., no greater than 20 minutes), providing an
additional rule for identifying conflict avoidance situations. Furthermore, $t_{anticip}$ offers insight
into the urgency of the controller's intervention, representing the time remaining before the
aircraft pair crosses.
Without delving into additional details (omitting the vertical profile, which also requires
intersection), the key idea is that it's possible to extract conflict avoidance interventions
from air surveillance data files using this proximity pattern. From this pattern, various
applications can be derived, detailed below.
Once conflicting aircraft pairs are identified through the proximity pattern, additional insights can be
gained. For instance, the parameter $sep_{scen}$ describes how close the aircraft pair is planned to
pass, offering insight into the severity of the conflict. Another notion is urgency
(already captured with $t_{anticip}$), describing "how quickly the controller has to intervene," and
this can be captured by estimating the available duration from when the aircraft pair enters the
controller's position to when the two aircraft will cross.
Similarly, the controller's performance can be assessed by the number of instructions issued to resolve
conflicts (given that a single instruction can resolve multiple conflicts). If the controller deviates
an aircraft from its intended route (as illustrated
here), an additional instruction is required to return the aircraft to its route, adding
workload and fuel consumption.
As seen, numerous applications are possible, depending on whether the focus is on performance or risk.
The proximity pattern presented earlier, based on the horizontal intended route scenario, can be replicated with the vertical optimal profile scenario. This replication provides insights into optimizing fuel consumption.
Let's take an example of an aircraft receiving an early descent instruction, as depicted in the left figure. The vertical optimal profile scenario is showcased . Employing the same reasoning as in the preceding proximity pattern, we identify aircraft that didn't conflict with the illustrated blue aircraft but would have conflicted if the blue aircraft had adhered to its optimal profile, as depicted .
Subsequently, upon recognizing these potential conflicts, we explore modifications (e.g., procedural design changes, adding constraints) to prevent conflicts. The revised "conflict-free" optimal descent profile then contributes to optimizing the fuel cost for the blue aircraft.
The examples presented above demonstrate how to construct higher-level patterns capturing additional
costs for both controllers and airlines. The objective is to design "win-win" modifications in air
operations, achievable at all levels—from airspace to procedure design and controller working methods.
We contend that these modifications can be implemented once a sufficient understanding of the system is
acquired, encompassing where extra costs manifest. Our software solution incorporates aids for
visualizing extra costs and potential improvements, facilitating an interactive approach. Analogously,
using our methodology is akin to playing chess with the assistance of a computer engine; the
engine aids in computing possible lines, but you retain control of the final strategy, choosing
the most suitable line among several candidates.
This approach, we believe, maximizes the benefits of software and tools based on the AI paradigm.