Development of a Rule Based Decision Support System for Pilots : Using Network Analysis, Label Propagation and Association Rule Mining

Sammanfattning: Maneuvering a fighter jet is not easy, and new methods for minimizing the risk of pilots experiencing information overload are constantly under development. Saab is currently looking into the possibilities of creating a decision support system (DSS) for pilots in situations where multiple independent system failures occur simultaneously. In today’s JAS 39 Gripen E such failures generate multiple simultaneous alerts for the pilots to act upon, with each alert proposing a set of actions for the pilot to handle the current situation. These actions are displayed on separate pages for each of the alerts, requiring the pilot to manually swap between pages of actions to get an overview of what to do. When pilots must use a significant part of their cognitive abilities to maneuver the aircraft in critical situations, a DSS is needed to apprehend the information and guidelines from the alerts more quickly. Our approach to developing such DSS is by representing each of the individual actions as nodes in a network, with links between nodes providing information about the relations between actions. The purpose of mapping the problem to become a networking problem is to use label propagation algorithms to divide the network into communities. With each node being assigned to a community and having the communities sorted on priority, getting a first draft of what a sorting algorithm can accomplish should be easy given a complex fault situation. We use rule mining to extract rules from the system as it is currently implemented to create a rule-based sorting within communities. The conclusion regarding label propagation algorithms is that although they are great for prototyping, propagation algorithms have no place in creating a DSS with high reliability. We also conclude that inconsistencies in the data that yield cycles in the created networks limit the sorting algorithms since there is no way of knowing where the cycle begins, hence no way of knowing how to sort actions being part of a cycle. Despite this, we argue that mined rules can be very helpful for engineers at Saab to be more consistent in their future work. Because of this, the conclusion is that our prototype of the DSS is better suited for the engineers working with providing pilots with the correct information than for the pilots directly. Lastly, we provide suggestions for future work, focusing much on possible ways of dealing with the cycles. Suppose Saab wants to pursue further research into whether creating a good enough DSS the way we propose is possible. In that case, some filter has to be developed to reduce the amount of information in complex fault situations.