Now that you have decided on the machinery you want to use for predictive maintenance, and the way to do it, it is time to connect it up and provision your IoT platform.
The selection of the IoT platform is a difficult one, as there are many out there (more than 300 and still counting). Doing your research by going to a number of conferences, talking to some analysts in the field of IoT, and reading their reports can help.
Make sure you lead with your use case. The reason is, some platforms are more use-case centric for smart homes or connected customers. So if you are going after an industrial use case, that could cause you some issues. If you are aware of some important future requirements, then that is great. Why should you look toward your future needs? Simple! Nearly all platforms will be able to cover your basic needs, but many of those platforms are similar to liquid concrete. Once you step your foot in, it will be painful to transition to another one if your future needs require you to do so.
Luckily, many vendors allow trials.
So let’s presume that in our case, your team selected a platform that:
- Allowed you create the device connectivity and to take your industrial vibration measuring devices under management
- Had some customer use cases which gave you the confidence that your predictive cases were able to be achieved
- Had some interesting features on the horizon when it came to real-time analytics and real-time visualization going forward
It is now time to set up the environment and connect the first machine (keep in mind we are using an example; in practice, it could be any “thing” or device. As you probably will have more than one machine—and are very likely to have more and different sensors in the future—it is a good habit to create a tree structure to achieve a form of hierarchy. This will allow you to efficiently filter and browse through information in the future.
Let’s presume that the devices are rolled out and connected, and the tree of sensors is created and data is flowing in. What is the next step? Well, probably, it is getting a routine for data collection (don’t think hours, think days and weeks) and starting to look for trends. Can you see any trends in the data you are collecting? Are the values steady, or do you see some up or down trends?
Now the experts and the data scientist can come in and start to work closely together. What do these trends mean? Can we correlate machine failure to these trends? How strong is the correlation, and can we identify a root cause and a cure? For example, an increase in acceleration might relate to a ball bearing and the cure might be either more frequent greasing or a replacement.
The problem often encountered is that the understanding of how cause and effect are related is limited, so give your team a grace period while learning. Splitting the project into two is a very sensible strategy; part one is to get data from the sources and visualize it in meaningful ways (that project often goes under the name “condition-based monitoring”) and part two (only when that phase is mastered) is to implement a follow-up project to drive the actionability—through, for example, predictive maintenance programs.
Phase 3: Embedding in the Organization
Until now, I have left two main aspects out of this article so as not to confuse. But, when you get to the actionability stage, it is inevitable that you have to start thinking about these: The first is integration and the other is process and workflow.
Integration relates to two aspects:
- How do you gather data from other sources?
- What do you need to enrich the decisioning?
Think about data such as previous maintenance schedules or production-related numbers. Having this kind of data will help to create a sense of context, which will make it easier for the operators to make informed decisions. The other aspect is to expose the intelligence gathered through the new systems to other systems and parties.
If the number of sources to connect with are limited, you might want to go for point-to-point integration, but for future-proofing, it might be better to implement some SOA architecture principles—such as an enterprise service bus with API management capabilities. This will allow you to not only onboard new data sources quickly but also support new channels such as mobile application and direct integration into partner solutions reliably and securely.