Speaking at MRO Management's annual PAM Europe Conference in Dublin today, Rory Johnstone, principal predictive maintenance engineer at Virgin Atlantic, outlined the complexities of applying predictive technologies across a long-haul fleet operating in a demanding environment.
With aircraft flying an average of 13.9 to 17.3 hours per day, maintenance opportunities are constrained by aircraft rotations, fleet configurations and the availability of replacement components.
Johnstone explained that even aircraft of the same type can have different components, failure characteristics and predictive maintenance requirements. Using the example of landing lights across Virgin Atlantic's A350 fleet, he demonstrated how variations in part numbers and component technologies can require different predictive models and spare parts.
Predicting a failure, he argued, offers limited operational value if the airline cannot act on that information.
To illustrate the dangers of relying too heavily on historical data, Johnstone turned to an unlikely comparison: the Rugby World Cup.
Examining the results of every men's Rugby World Cup final since 1987, he pointed out that the team leading at half-time had gone on to win every match. On that evidence alone, the result might appear inevitable before the second half had even begun.
But, he argued, "half-time represents only a snapshot. The game continues to evolve, just as aircraft reliability and operating conditions change throughout an aircraft's life."
Drawing on Virgin Atlantic's experience with the Boeing 787 and its early reliability challenges, Johnstone explained that predictive models must account for what happens between an initial condition and an eventual outcome, including engineering interventions that may have prevented a failure.
A central theme of his presentation was the need to retain human judgement within increasingly automated maintenance processes.
"AI is not the hammer to hit all nails with," Johnstone told delegates, emphasising that predictive technologies should "support engineering decisions rather than replace the expertise behind them."
Looking ahead, Johnstone suggested that predictive maintenance must move beyond processing ever-greater volumes of aircraft data towards providing engineers with clearer, more actionable information.
This includes combining onboard sensor and fault data with offboard information such as component maintenance histories, reliability trends and known failure modes.
Ultimately, he envisaged a future in which individual components communicate their own condition, degradation and remaining useful life, allowing "maintenance decisions to be based more directly on component health rather than aircraft operating hours alone."
The ambition, he suggested, should be to "shorten the journey from data to decision, enabling engineers to spend less time interpreting information and more time engineering."
