Context – Aircraft engine maintenance, repair and overhaul (MRO) is crucial to ensuring aircraft reliability and passenger safety. Engines must be inspected on a regular basis to detect any defects at the earliest stage before they can propagate and cause any adverse outcome. Nonetheless, maintenance is a major contributor to aircraft accidents and incidents, and is responsible for flight delays and cancellations.

Issues – The visual inspection of engine blades, which is the area under examination in this thesis, is performed by human operators, who are unreliable, inconsistent, subjective, prone to error, and have different personal judgements based on their risk appetite. Furthermore, working under time pressure forces operators to balance between two conflicting aspects, namely safety and performance.

Need – The maintenance of aero engines is an intricate, time-consuming, highly repetitive and tedious process that is prone to slips, lapses and mistakes. This entails the risk of missing a critical defect during inspection. Hence, there was a need for more effective and efficient maintenance and inspection processes to reduce the turn-around times, maintenance costs, non-value adding tasks, human errors, and wrong-decision making, while improving the inspection quality and performance, i.e., increasing the inspection accuracy, consistency, reliability and productivity.

Approach – This work used multiple different methods including an ontology to link root causes and blade defects; Bowtie analysis to determine critical factors in the inspection process; eye tracking to extract the visual search paths and understand if and why a defect was missed; statistical methods including ANOVA, odds ratio and model building to evaluate the inspection performance (accuracy, consistency, and time) and influence factors; image processing techniques to develop a defect detection software; SWOT and weighted factor analysis to compare different inspection agents including image processing software, artificial intelligence software, 3D scanning, and human operators.

Results – A defect taxonomy was developed for blade defects. The subsequent Bowtie analysis showed that there are other important factors affecting visual inspection that are often neglected, such as management activities and training. As there was a lack of standardised Bowtie development, a systematic methodology was introduced applying 6M categorisation. Next, several factors affecting visual inspection were statistically analysed. Results show that, for example, cleanliness had a significant effect on defect detection rates. From the eye tracking recordings, the different search strategies and underlying cognitive processes were extracted and inspection errors identified. Experts in contrast, ap- plied a systematic search strategy, while novices’ gaze wandered across the blade in an unstructured way. The assessment of tactile perception showed higher inspection accuracies compared to sole visual inspection. The attribute agreement analysis revealed that human operators were highly inconsistent, and that the inspection system had low repeatability. An interesting result was that human operators outperformed both inspection software, while being less effective in their defect detection compared to the 3D scanner.

Findings – A risk framework was developed for visual inspection taking into account defect severity, criticality, and manifestation. In addition, a revision of the visual inspection framework was pro- posed based on the research findings. Furthermore, an inspection quality framework was introduced following the DMAIC structure.

Originality – This thesis makes the following novel contributions : Development of a blade defect taxonomy; Bowtie analysis for inspection activities; assessment of the human performance in blade inspection; development of a visual search framework; evaluation of technological inspection methods; and development of an operational decision support methodology.