Succeeding at Predictive Maintenance at Scale
In our last post, we looked at the rise of predictive maintenance (PdM), and why so many vendors - and their customers - continue to get it wrong. Here we’ll explore how we’ve applied what we’ve learned about Predictive Maintenance over the years to everything we do today.
The combination of more than 30 years’ experience of working in the Aerospace industry, and over 150 person-years of research and development time spent exclusively on PdM has taught us a lot about deploying predictive maintenance technologies across various different sectors. And, as we continue to encounter different sectors and levels of customer maturity, we continue to learn. Above anything, we’ve learned that doing predictive maintenance well isn’t easy.
Importantly, we’ve learned that it’s not just about knowing how to do PdM well, but why it matters. We’ve learned, for instance, that the true value of PdM can be hard to explain, even if it’s delivering a significant return on investment. Implementing a PdM strategy is a move that shouldn’t be taken lightly - it affects a whole business, from the board to the shop floor. A degree of confidence in its success will be needed before embarking on a transformation of this scale.
Here, then, are a few key points that, in our experience, we believe organizations should bear in mind when setting out on a predictive maintenance journey.
Predictive Maintenance isn’t a typical Data Science use case
The more we speak to customers about their previous failed implementation attempts, the more certain we’ve become that PdM can’t be treated as a typical Big Data problem. This largely comes down to the fact that the rich context required for Big Data solutions to work is severely lacking in PdM. Many machine failure modes are subtle in nature, too, making it hard for established machine learning algorithms to accurately predict any discernible patterns.
It’s not unreasonable to expect accuracy levels of 90% and higher in a typical Data Science solution. Expecting similar accuracy from of a PdM solution, however, suggests a fundamental misunderstanding of how it works. Yes, using high quality, curated data in a laboratory environment will deliver results like this for specific machines and specific failure modes. But this doesn’t represent reality. Every machine and failure mode is unique, for one thing, and different sensor types will deliver data of different levels of accuracy. And that’s not to mention the dynamic nature of a factory floor, and the general lack of crucial contextual information.
Tough questions should be asked of anyone that asks generalized questions or makes specific claims regarding the accuracy of a PdM solution.
Know your audience
Another thing our experience has taught us is that busy maintenance teams typically have very little time at the start of their shifts to identify which of the many machines for which they’re responsible need their attention most. They certainly don’t want to spend that time studying graphs and sifting through raw data to uncover that information.
Simple, intuitive software design is therefore essential, providing maintenance professionals with the insight they require in an easy-to-read dashboard, and saving valuable time. However, while many vendors provide such dashboards, they can often be generic offerings, that don’t take a user’s unique workflows, preferences, and experience into consideration. Indeed, the knowledge and experience of these professionals is invaluable; harnessing it is crucial to the success of a PdM solution.
Senseye’s products are designed with all of this in mind. We aim for simplicity, seamlessly integrating analytics with a clear user interface. Rather than multiple charts and graphs, information is presented in the form of a list, sorted by the Attention Engine, a proprietary algorithm that uses machine data, maintenance data, and operator data to prioritize machines in need of attention.
Know where your users are on their journey
We’ve also learned that different companies - and, often, different areas within the same company - will at different stages on their predictive maintenance journey. One company, for example, may only be performing periodic, route-based condition monitoring checks. Another, however, will have combined robust, automated condition monitoring with a PdM solution for accurate predictions of their machines’ health and time-to-failure.
These examples are at opposite ends of a scale, of course, and most companies will sit somewhere between the two, the maturity level of both their data and cultural readiness increasing as a result of greater understanding by their management team and buy-in from their IT team. It’s important to know where a company sits on that scale, though, as each will need a different support and deployment package depending on its maturity level.
Years of experience have taught us a lot about what works and what doesn’t. In our next post, we’ll explore how all of this experience and understanding has made Senseye PdM the leading PdM product on the market. Until then, you can find more detail on the most important things we’ve learned about PdM in our white paper “Senseye in Depth: Why is Predictive Maintenance so hard?”.