Even though big data may seem like a buzzword that has permeated almost every industry on the planet, this near limitless source of actionable information can drive unprecedented company growth and efficiency. Unfortunately, a large part of this potentially useful data is generated by industrial sensors and dumped — largely untapped — into vast data lakes. The noise created by the sheer amount of data we accrue everyday from thousands of sources makes it almost impossible for a human team to analyze. It’s a catch-22.
These hurdles can be easily applied to predictive maintenance in the aviation sector, which like many other industries around the world has not reached its full, data-enabled potential. However, with 10 times return on investment for aviation companies and a potential 70 to 75 percent reduction in airplane breakdowns, there is a clear case for automated predictive maintenance in the aviation industry.
One Piece At a Time
If we look at a solitary part of an aeroplane we can start to gauge the challenges that a manufacturer, a company, or MRO provider can have when aiming to gain effective equipment performance insights by using data generated by sensors. Take an average jet engine from a commercial airliner, which can be equipped with over 5,000 individual sensors, as an example. According to SAP business software, these engines can generate around 20 terabytes of data per hour. Taking a few steps back, an entire plane — such as the A380 superjumbo — can generate information on 200,000 different aspects of every flight, allowing it to create immense pools of data.
While the connectivity of planes is increasing, the data generated has already gone beyond what a human team of data scientists can handle and dipped into the realm of automated big data solutions.
The key for companies and providers within the industry is knowing how to fine tune the ear of data teams and implement rapidly evolving cognitive models to effectively analyze and harness this data.
But how does aviation leverage these data insights to reduce downtime of planes, decrease bottom line costs, and increase component life and efficiency? Is it down to finding the right team or the right solution? Beyond these, however, and as traditional models of data science continue to become outdated, it is important for companies to seek out the successful trends that will enable optimal use of their sensor data.
Diamond in the Rough, Identifying the Right Data Signal
The 2015 McKinsey Global Institute report points out the economic impact generated by the Industrial Internet of Things (IIoT) market, which incorporates machine learning and big data technology, to be $11.1 trillion by 2025. For aviation, this economic impact can be felt directly in the predictive maintenance sector, as leveraging this information and conversing with it to get actionable answers remains one of its biggest challenges.
According to SAP, 42 percent of delayed flights are caused primarily by airline processes, such as maintenance. When you take into account that a grounded plane can cost an airline $10,000 per hour, an efficient predictive maintenance process to reduce downtime for an aircraft can save millions of dollars every year.
The expanding connectivity of planes and the wave of data that this generates, has enabled the increased targeting of predictive analysis to certain areas, components, and systems to better inform the engineers on the ground. Right now, basically every part of a plane can tell crews on what needs replacing, or if something is faltering.
Ironically, this is also where the issues of information overload arise as thousands of data nodes can overwhelm teams and decision makers who look to streamline processes and implement accurate models.
But finding the true signal of a datapoint generated by machine sensors and hidden in a diverse pool of sources — all of which are constantly generating information at different times — has become impossible for a human team to process.
Even with the resources leading airlines have to store and analyze this information, much of the big data generated by sensors on a plane remains largely unexploited, according to the Financial Times. Paul Stein, chief scientific officer of Rolls-Royce, identified the industry’s lack of sufficient communications infrastructure to harvest and transmit the data as the industry’s leading short-term obstacle, the FT stated.
However, it’s not about constantly analyzing every line of data as it comes in. It is important to note that the terabytes gathered by millions of parts on a daily basis are not all necessarily useful.
At the 2014 "The Data Driven Business of Winning" summit, managing director of CMS Motor Sports Ltd., Mark Gallagher stated that Formula 1 teams efficiently analyze data to win races by identifying anomalies, reported TechRadar. "99 percent of the information we get, everything is fine [...] we're looking for the data that tells us there's a problem or that tells us there's an opportunity," Gallagher was quoted as saying.
Due to the degradation of monitored equipment and variations in sensory output, identifying these anomalies by a human team can be exceptionally complex. Furthermore, rules are hard to implement as the natural lifecycle of a component within an aircraft's ecosystem can be unpredictable.
To reduce the loss of valuable data, which can be timely and therefore perishable, data automation solutions for areas like predictive maintenance can be invaluable. More than just analyzing the data, being able to interact with the information and extrapolate particular anomalies that can offer differential perspectives on the state of a component, engine, or aircraft navigation board, can answer the important questions.
Monitoring Each Sensor in Isolation Doesn't Work
Components, whether in an aircraft or in a car, often fail due to numerous factors within the entire ecosystem. Because of this, monitoring just one sensor within an aircraft is unlikely to produce a complete dataset that depicts an accurate view of what is actually occurring.
What’s more, the manual effort required to combine a series of individually monitored sensors to successfully extrapolate alerts and filter critical signals from large amounts of data is very high. Not only is this method inefficient and expensive, but it also fails to successfully scale in the long run.
But just like finding the needle in the noise of data stack, this level of data generation falls into the realm beyond human teams due to the sheer volume of information and the ambiguity that the raw data can produce. However, through automated predictive maintenance, decision makers can successfully leverage part harmonization to gain a clearer overview of specific sensor insights. These help create accurate predictive models that can show the parts that are set to fail first and help schedule replacements, which in turn improves the management of part inventories.
Combining all these factors helps teams successfully implement complete and functional predictive maintenance, which according to an International Journal of Applied Mathematics and Informatics study, can decrease total maintenance costs by 30 percent and reduce stationary time of aircraft by up to 45 percent.
Additionally, the successful application of predictive maintenance can avoid the knock on effects that unscheduled emergency maintenance can have. Increased downtime generated by this spontaneous maintenance, for example, can have detrimental effects on customer views and company reputation.
Beyond Human-scale
Although data analysis as a process requires the input of human teams and professionals, the sphere of industrial data has already surpassed dated models that revolved around reports and charts.
It is here that the timely analysis of datasets is crucial. By having a machine do the job, for example, companies can see a problem within an aircraft before it occurs, a task that is now practically impossible for data science teams with the advent of increased component connectivity.
Furthermore, and not to definitively relegate humans to the corner, the mass shortage of data scientists simply requires the automation of certain processes. Plain and simple. Not only are these solutions able to collate a week’s (or a year’s) worth of data to build the most optimal model for the job, but they can facilitate the real time execution of decisions based on timely data.
For predictive maintenance in the aviation sector, the ability of these automated machine solutions to compare, contrast, and segment massive aircraft datasets for more accurate predictions is a near perfect fit. Unlike human teams that can take weeks to analyze just one segment of a dataset in order to accurately install predictive maintenance models, automated solutions can discover critical points in the data in mere seconds.
As industrial data continues to evolve and expand, predictive maintenance will require the agile analysis afforded by automated systems and retreat from human engagement sphere. For the aviation industry, the ability to execute practical business solutions in this area will require automated near real-time analysis and accurate interpretation of aircraft machine and sensor data conversations.
Sundeep Sanghavi is the co-founder and CEO of DataRPM, an industry pioneer in automating predictive maintenance. For more information visit datarpm.com.
