ON THE IMPLEMENTATION OF PREDICTIVE MAINTENANCE
Flying is an already time-consuming aspect of life. Factor in variable delays and it can quickly become a nightmare for most. While weather delays are quite hard to avoid, nearly 30 percent of air traffic delays result from unplanned maintenance. The result: higher airlines expenses due to compensations to unhappy customers stuck in airports, overtime maintenance crew costs and more flight delays. Already deployed by some of the more data savvy airlines, predictive maintenance can save the headache of an aircraft stuck on the ramp unexpectedly by better managing data from aircraft health monitoring sensors.
By granting access to historical and real-time data and knowing an aircraft’s current technical condition through alerts, notifications, and reports, technicians can spot issues pointing at possible malfunctions and replace parts proactively. In turn, executives and team leads can foster an efficiency culture, more accurately anticipate next big maintenance operations, reduce costs on tool and part inventory…
However, these solutions need to be tailored to the airline’s operations. Here we share the three stages incremental approach to fully benefit from predictive maintenance practices:
OFFLINE FORECASTING
Mining historical data allows producing a timeline of the life of any given component, mapping the necessary maintenance steps to keep the aircraft flying uninterrupted. Lack of historic data == lack of forecasting.
OFFLINE MODELING-FORECASTING
Nowadays, computational power is easily scalable. This means we can model almost anything from atomic scale to entire flying aircrafts. The models can then be used to estimate the failure of the components. Then, populating a maintenance planning tool will produce the optimal maintenance schedule. Little control from the airline, manufacturer has the main control of this.
REAL TIME PREDICTIVE MAINTENANCE: MONITORING-MODELING-FORECASTING
Collecting data from sensors allows inferring behavioural patterns from any component of interest of the aircraft, looking for anomalies in the data and flagging them in real time, automatically to the operators. Trade off: Bad sensors can easily become an increasingly worrying issue.
This is not an either/or approach, it is an incremental process. Understand the problem, structure the data properly, and then produce the right solution offline must be the first step. Then, any kind of machine learning to correlate the three knowledge streams can be fructuous in maintenance planning.
In addition, a fully connected solution employing edge devices could then improve the crew response time if a technical problem did occur. And the result of this? Not only a proactive (not reactive) company, but happier customers, which are likely returning customers.
