Entropy sucks. But industrial predictive maintenance can help it suck a little less for factories, oil rigs, aircraft engines, and even data centers. The key is to leverage the Internet of Things (IoT) and machine learning to help companies “accurately determine when a manufacturing plant, machine, component, or part is likely to fail, and thus needs to be replaced.”
That, in a nutshell, is the point behind a fascinating new Google Cloud blog post by Prashant Dhingra, Machine Learning Lead, Advanced Solutions Lab, laying out “A strategy for implementing industrial predictive maintenance.”
Pitched as part one of a three-part series, Dhingra explains “how predictive maintenance can function to reduce downtime, reduce maintenance costs and improve operational effectiveness and safety, by identifying impending failures before they occur.”
Entropy = Everything degrades over time
This is a big deal, because as Dhingra points out, “machines, batteries, and other mechanical and electrical components … degrade with time and use. In the long run, every piece of manufacturing equipment will reach the end of its useful life.” (Sure sounds like entropy to me, even if Dhingra doesn’t use the term.)
Once you accept entropy — and you really don’t have much choice about that — the question is how to deal with it. Unfortunately, it isn’t easy to predict when a given piece of equipment is going to give up the ghost. As Dhingra points out, companies have tried everything, from schedule-based maintenance to condition-based maintenance to figure out when repairs or replacements are required. The problem is that to minimize expensive failures in the field, you need to rely on worst-case lifespan estimates, which often means you’re replacing things that could have continued to function normally for a long time—and that wastes time and money.
By combining technologies such as big data, cloud computing, machine learning, edge computing, and the IoT, Dhingra says, companies can use IoT sensors to collect multiple parameters from highly instrumented equipment, send it to the cloud, and employ machine learning to apply predictive analytics and identify failure patterns.
If done right, Dhingra says, the result is the ability to get better answers to such questions as:
- Will this equipment fail?
- What is the remaining life of the equipment?
- Is there an anomaly in the equipment behavior?
- How can the equipment settings be optimized? (Interestingly, Dhingra cites a data center case study to illustrate this example.)
Predictive maintenance: It’s all about the dataset
So, how do you do it right? According to Dhingra, “The most important requirement to build a predictive maintenance solution is to have the right dataset. It is ideal to have a dataset that shows identifiable equipment degradation.” He says IoT data about temperature, vibration, sound, and voltage produced by machinery are the most valuable for building a predictive maintenance model.
Combined with metadata about the equipment involved (make, model, revision, etc.), usage history, and maintenance history can make equipment last longer, reduce unscheduled downtime, save money, and even enable new business models. For example, equipment built to facilitate predictive maintenance would have a significant competitive advantage and could command a market premium.
That’s not exactly reversing entropy, but it still sounds pretty good.
