Towards E-Maintenance: An Exploration Approach for Aircraft Maintenance Data

Introduction

Achieving safety during flight the mission is the highest priority of any aviation industry. As a result, the aviation industry usually tries to adopt schedules of maintenance activities, to optimize the operational lifetime of the aircraft, and minimize incidences of mechanical failures. A significant attention is usually focused on maintenance-related accidents, which according to a recent IATA (International Air Transport Association) document, 15% of reported aircraft accidents are attributed to maintenance-related factors (Chen et al. 2017). Additional factors influencing aircraft accidents includes meteorology, and human error related factors, which are largely mentioned as a leading factor contributor to accident events. It is important to note that the report highlights a close correlation between maintenance-errors and equipment failure, which lead to accident events. Additional risk factors, which influences how maintenance interventions are carried out, includes the level and quality of compliance with standard operating procedures (SOP) while performing maintenance activities, training of maintenance crew and work-related factors such as fatigue. Furthermore, maintenance-related human errors tend to be influenced by communication challenges, work pressure, among other factors. To address these aspects, which influence the quality of maintenance activities for aircraft systems, IATA places considerable efforts developing and enforcing guidelines for maintaining aircraft systems, while emphasizing quality of the activities and compliance to standardized rules for maintenance.

Since aircrafts are equipped with sensors attached to the critical components of interest, often these sensors are used to record different attributes related to the aircraft condition. Consequently, this sensor information is useful for assisting maintenance technicians to monitor the performance of the critical components and derive diagnostic information to aid root cause analysis and plan maintenance activities. For example, sensors monitor attributes such as vibration signals and generate fault messages or error warnings. Once analyzed, the information aggregated from the sensors forms the basis of decision making on aspects such as planning maintenance interventions. Moreover, the sensor information is useful to support airlines to better allocate maintenance resources, such as technicians.

However, to a large extent, sensors and maintenance information collected from aircraft systems are not sufficiently used in decision making for maintenance. Rather, aircraft maintenance largely relies on standardized maintenance manuals, with insufficient attention focusing on maintenance data collected from aircraft sensors or repair information recorded by maintenance technicians. Some challenges limiting use of, especially repair information recorded by maintenance crew, pilots or quality assurance staff, is the textural form of the information. These limits use of traditional statistical analysis tools such as correlation analysis, which means that maintenance technicians cannot fully exploit maintenance records for decision support. This chapter proposes a data mining approach, which uses visual exploration, text mining, and association rule mining for deriving insights on patterns and correlation, which are useful for maintenance decision support, but embedded in data. Such patterns are useful for supporting decisions on aspects such as failure diagnosis and identifying the focal root causes of equipment failure. Importantly, the approach uses predictive analytics to identify critical failure patterns embedded in maintenance data and is useful for assisting maintenance planning and allocation of often scarce maintenance resources.