The latest sensor technologies and AI-algorithms support aircraft maintenance processes to become more efficient. Sensor data and AI are used to monitor the health of aircraft systems and trigger alerts before they fail. This is referred to as condition-based maintenance: maintenance is done only when the health of the systems really needs it and not at fixed intervals, as done traditionally. Exploitation of systems thus can be done as efficient as possible. This results in a reduction in maintenance costs, a reduction in unscheduled aircraft maintenance events, and an increase in aircraft availability. But how can we prove that condition-based aircraft maintenance is efficient and safe? One first has to capture condition-based aircraft maintenance into a model. We discuss this in our recently published paper:
Lee, J., & Mitici, M. (2020). An integrated assessment of safety and efficiency of aircraft maintenance strategies using agent-based modelling and stochastic Petri nets. Reliability Engineering & System Safety, 2020, 107052. https://doi.org/10.1016/j.ress.2020.107052
Here are the 5 steps to model and analyze the efficiency and safety of condition-based aircraft maintenance.
Step 1: Understand the maintenance process of aircraft
That’s easier said than done. Aircraft maintenance is a complex process, involving many stakeholders such as mechanics, maintenance planners, flight crews, etc. A good understanding of the maintenance process is the first step in our research. We combined desk research (regulations, maintenance handbooks) with field research. To understand the practice of aircraft maintenance, we interviewed several experts and maintenance engineers.
Step 2: Identify and model the key agents in the aircraft maintenance process
A characteristic of aircraft maintenance is that it is a collaboration of multiple people and computer systems, each with their own roles and responsibilities. Therefore, we model the individual entities as agents using an Agent Based Modelling approach. We identified five key agents: 1) aircraft, 2) task generating team, 3) task planning team, 4) mechanics, and 5) flight crew.
Step 3: Formalize the agents and interactions in the aircraft maintenance process
Now that we have distinguished the key agents of the aircraft maintenance process, we have to define the states, actions of these agents and the interactions between them. We use Stochastically and Dynamically Coloured Petri Net (SDCPN) to formalize the model. For instance, the degradation level (state) of a system is changed based on a stochastic Gamma process (action). The Task Generating Team generates a maintenance task (action) when the system is degraded, sends the task to the mechanics (interaction), and the mechanics perform the task (action).
Step 4: Simulate the agent-based model of aircraft maintenance
Now that we have a formal model of the aircraft maintenance process, we conduct Monte Carlo simulations. With the help of simulations, we analyze the efficiency and safety of different maintenance strategies. Is the traditional maintenance process efficient when performing inspections and replacements at fixed time intervals? What about condition-based maintenance which uses sensor data and AI? Using simulations we can compare different maintenance strategies.
Step 5: Compare fixed timed interval maintenance with condition-based maintenance
To validate our approach we conduct a case study for the maintenance of aircraft landing gear brakes, based on operational data of a wide-body aircraft. A brake disc should be replaced before it has deteriorated significantly. Under current maintenance strategies, the brakes are inspected and replaced at fixed time intervals. Under condition-based maintenance, inspections of the brake disc are done only after a sensor alert is triggered or based on predicted Remained Useful Life (RUL) for the brakes.
Results of simulating traditional vs condition-based aircraft maintenance
The simulation results show that condition-based maintenance using RUL estimates for brakes improves the efficiency while attaining a similar level of safety, when compared to traditional, fixed-interval maintenance strategies. Moreover, our approach is generic and it can be applied for other components and maintenance strategies.
If you want to know more about this research please go to here.
About the authors:
Juseong Lee is a PhD candidate at the Air Transport and Operations Section, Faculty of Aerospace Engineering, Delft University of Technology. He is participating H2020 ReMAP project. His research interest includes simulation-based assessment and optimization, in the domain of aircraft maintenance. He holds an MSc degree in aerospace engineering from Korea Advanced Institute of Science and Technology (KAIST).
Mihaela Mitici is an assistant professor in the Air Transport & Operations section, Faculty of Aerospace Engineering, TU Delft. She specializes in Operations Research, with a focus on stochastic processes, decision-making under uncertainty, applied probability theory. Her main application domains are predictive aircraft maintenance, airport operations, and operations of urban air mobility.
Floris works as a research coordinator at KLM Royal Dutch Airlines. As part of the H2020 ReMAP project, Floris is involved the field of prognostics, maintenance decision support and CBM evaluation. He holds an MSc degree in Aerospace Engineering from TU Delft.