Search archive          Sign up for our Newsletters          Aviation Jobs
Latest Aviation News  |  Industry & Technology  |  Air Transport  |  MRO & Support  |  Aircraft Interiors  |  Editorials  |  Events Calendar  |  About UsFR
 
Aviation News Predictive Maintenance : The magic of Data science

Predictive Maintenance : The magic of Data science

By David Renaud, Head of Data Science Capabilities at Assystem
12 APR 2017 | 866 words
Predictive Maintenance : The magic of Data science
David Renaud, Head of Data Science Capabilities at Assystem. All pictures © Assystem
For many years, the maintenance of everything from plant machinery to airliners has basically been a rush to identify and rectify faults, known as 'corrective maintenance'. With the advent of preventive maintenance and methods such as reliability-centered maintenance, the focus has shifted to anticipating faults and planning maintenance tasks ahead of time, based on the reliability characteristics of the asset in question.

Condition-based maintenance, where assets are monitored in order to spot the early signs of a fault, before it occurs, has been around for a number of years. Here, non-destructive testing techniques, such as acoustic emission, thermography and vibration analysis, are used to detect the onset of faults that could lead to downtime and repairs. However, they provide little or no visibility on the remaining lifespan of the part.

As well as finding the root cause of a fault, maintenance personnel need to know the risk of keeping the part in service, so they can change it at the right time -- not too early, which increases spare parts costs, nor too late, because an unexpected failure could be even more costly. They therefore need to measure this risk. In other words, determine the probability of a failure and how it increases with time.

To achieve the optimal balance in maintenance, parts need to be changed at just the right time. In other words, before a fault or failure occurs, thereby avoiding further damage and limiting the impact of system downtime and loss of production. Such 'preventive' repairs must also be scheduled to minimize disruption to production. Maintenance staff must therefore be able to calculate this optimum. Today, this is made possible with the arrival of predictive maintenance.

With all the instrumentation and sensors built into our factories and systems today, combined with the connectivity of the IoT, we now have access to a huge amount of potentially useful data for maintenance. However, it still needs to be extracted and interpreted. This is where data science comes in, providing a set of methods and tools for turning raw data into value-added, actionable information.

The data generated by the ERP system and machine control mechanisms in a factory, or the avionics suite on an airliner, for example, combined with maintenance reports, fault reports and external weather bulletins, are just some of the sources of information that can be usefully exploited. Data science helps us find correlations in these datasets, detect weak signals and develop predictive models.

Approaches are many and varied. Real-time anomaly detection algorithms can pick up the telltale signs of a fault in a huge mass of data. Text mining is used to extract information and statistical trends in maintenance reports, while machine learning allows us to make predictions based on representative datasets of past activity.



For many people, however, predictive maintenance is still a buzzword, conjuring vague images of data scientists as mathematically minded computer geniuses in white lab coats. Many companies talk about predictive maintenance, whereas what they are actually performing is conventional condition-based maintenance. In other words, tracking indicators and carrying out maintenance when a threshold value is reached. The real value of predictive maintenance is not widely understood and has yet to be exploited.

The predictive approach uses conventional deterministic methods to predict how fast a part is converging on this threshold value and, in turn, its remaining lifespan. It can be applied whenever we have known models of part or component ageing.

Today, however, data analytics enables us to implement predictive analysis even when we have no prior knowledge of the ageing process. It is based on machine learning methods, which are used to develop a mathematical model from a set of 'training' data representing the lifecycle of the system in question. Using this model, it is then possible to learn the characteristics of weak signals and calculate the probability of a fault occurring over a given timescale. Maintenance personnel are therefore able to define the level of risk they want to take and the criteria for replacing parts, based on this probabilistic approach.

Data science has its limitations, however, which need to be understood. The accuracy of this model is crucially important, because the right balance must be found between false positives (spurious alarms when there is no fault) and false negatives (actual faults which go undetected). In a typical maintenance environment, the number of actual faults is very low, compared to the number of healthy components. The risk, therefore, is that the inaccuracies inherent in these models will lead to a significant number of false positives, or false indications of faults. This, in turn, would have the opposite of the desired effect, increasing maintenance costs and eroding confidence in the approach.

To mitigate these limitations, data scientists, with their machine learning approaches, need to work closely with the company's reliability teams, utilizing their operational knowledge of equipment ageing processes to improve algorithm performance. This kind of operational approach allows us to implement predictive maintenance by cleverly applying the 'magic' of data science and adding real technical value for companies today.


 
They made this section possible
Top stories
09 DEC 2021
The world's first Airbus A320 freighter takes off The world's first Airbus A320 freighter takes off
We knew its first flight was near, but Elbe Flugzeugwerke (EFW) was able to keep the surprise until the end. The first Airbus A320P2F (Passenger ... Continue Reading
09 DEC 2021
NYCO signs a major strategic agreement with Air France to develop sustainable aircraft lubricants NYCO signs a major strategic agreement with Air France to develop sustainable aircraft lubricants
NYCO, the French specialist in aeronautical lubricants, has signed a strategic agreement with Air France to cooperate on engine oil field service evaluation and ... Continue Reading
09 DEC 2021
New long-term agreement between Safran and SIAEC on CFM International's LEAP engines New long-term agreement between Safran and SIAEC on CFM International's LEAP engines
Safran Aircraft Engines has just signed a new 10-year engine MRO contract with SIA Engineering Company (SIAEC). Under the agreement, Singapore Airlines' MRO division will provide ... Continue Reading
09 DEC 2021
The world's first Airbus A320 freighter takes off
09 DEC 2021
NYCO signs a major strategic agreement with Air France to develop sustainable aircraft lubricants
09 DEC 2021
New long-term agreement between Safran and SIAEC on CFM International's LEAP engines
09 DEC 2021
Boeing to add two 737-800BCF conversion lines at STAECO's facility in China
09 DEC 2021
Sabena Aerospace to take over several maintenance activities from Lufthansa Technik
Top stories
 
Latest News     Industry & Technology     Air Transport     MRO & Support     Aircraft Interiors     Editorials
© 2024 Le Journal de l'Aviation - All rights reserved