Predictive maintenance has been practiced in the aerospace industry for decades. The
critical safety concerns and value of the machinery has more than justified the cost of hiring teams of data scientists and developing new analytical techniques.
The ability to monitor things like vibration and torque levels associated with aircraft components have enabled engineers to develop a detailed understanding of their conditions, observe trends and unusual events in real-time, and (crucially) spot signs that a particular part is getting close to failing.
Instead of working to a predefined maintenance schedule or responding once a failure occurs, aircraft engineers are able to provide maintenance at precisely the right time, reducing the amount of time spent on unnecessary servicing while resulting in fewer components failing mid-flight and fewer aircraft falling from the sky.
And, of course, the benefits of this predictive-maintenance approach have been recognized in other environments where machine reliability is an important concern. Forward-thinking producers from automotive, consumer-goods manufacturing and even heavy industry have achieved notable successes by employing data scientists to monitor their most critical and expensive assets.
The ability to deploy these capabilities across all operational assets is only constrained by the high cost of recruiting and retaining skilled data scientists, and the time it takes to develop bespoke analytical models.
Traditional approaches to deploying predictive maintenance typically involve running a machine to failure, recording the data that occurs during that period and developing ways of spotting when similar patterns occur in the future. Identifying these crucial indicators and developing specific algorithms to spot them can be a long and laborious process.
Even when this preparatory work is done, monitoring the machines and analyzing the data they produce still requires manual intervention and interpretation from people with advanced analytical skills. Very few maintenance engineers are also skilled in data science and it can take years of intensive study to develop these skills. Engineers that become competent data scientists tend to migrate to better paid, full-time data science roles.
While applying predictive maintenance across an entire plant or manufacturing environment can deliver major improvements in machine uptime and maintenance efficiencies, it is uneconomic to scale up this kind of labor-intensive approach across thousands of machines.
We address this challenge by using artificial intelligence (AI) to automate predictive-maintenance analysis and ensure that the same analytical tools can be applied at low cost to any machine from any manufacturer. Organizations that already gathering data from their machines for incident logging or historical analysis can now, quite easily, deploy predictive maintenance at scale. Using existing data sets keeps the cost low and allows faster implementation. Win win.
Enterprises can begin assessing the condition of machines immediately using a series of automated algorithms to compare data the machines produce against known maintenance events. Crucially, the system then teaches itself to become more effective as it builds up a complete picture of each machine’s unique characteristics.
Whilst traditional forms of predictive-maintenance analysis would have required organizations to recruit more data scientists or retrain their engineers in a completely new set of skills, relying on AI to undertake monitoring and analysis of potentially thousands of machines requires only a slight shift in mindset and focus.
Automating predictive maintenance means that organizations employing data scientists are able to shift their focus from perhaps 10-20 critical assets to potentially thousands. They can leave the mundane day-to-day monitoring and analysis of machine conditions to a computer and focus their attention on the complex, systemic issues where they can add most value.
For organizations without data scientists, using AI enables engineering teams to access most of the benefits of data science without having to employ a single data scientist or spend years studying a completely new field. Instead of working to strictly defined maintenance schedules or responding to problems once they occur, engineers can look at a simple dashboard to see where and when their efforts should be best applied.
Using AI to undertake the heavy lifting of condition monitoring is more of a cultural shift than the acquisition of new skills, as it needs engineers to reconsider their existing ways of scheduling maintenance. The benefits, however, more than justify the change. Engineers that have already embraced this approach at scale have reduced their maintenance costs by 40 per cent, halved levels of machine downtime and delivered dramatic improvements in throughput, quality and margin.
Thanks to advances in AI, automation and analytics, predictive-maintenance techniques developed to transform aircraft maintenance and keep man safely in the air can now be deployed affordably and at scale in more down-to-Earth settings.
Dr. Simon Kampa is CEO and co-founder of Senseye.