Maintenance is one of the most critical points in business. It prevents operational hiccoughs and additional damage to valuable equipment and - at the end of the day - everything it does leads to cost savings. Historically, maintenance went through different stages of development: from a simple “fix it when failure occurs” approach called reactive maintenance, through scheduled maintenance called preventive maintenance and maintenance based on monitoring the condition of the assets, called condition-based maintenance, to the more evolved form of predictive maintenance, which is based on creating models using the data gathered in the previous stage.
The latest approach is about getting results from the field using IoT, predict failures based on machine learning using data collected from various sources - we particularly value sensor, asset and ticket management data sources.
Use case/problem description
In this case, we have a company doing maintenance on wind turbines which produce electricity. These are located in remote areas on various locations and are grouped into so-called wind turbine farms. The problem is, that these wind turbines are tall objects and attract lightning strikes during stormy weather which may cause damage.
When a storm passes by, a maintenance team is dispatched to remote areas to inspect whether any of the wind turbine’s blades were damaged. This is necessary to prevent further damage since damage to blade material can also lead to the loss of the blade. You can imagine that it is very hard to stop and restart the wind turbine in harsh northern cold climates. And they need to be stopped so each blade can be inspected in detail.
The problem we face is:
- we always need to visit the wind farm location (lightning strike occurred or not),
- we need to check each wind turbine if any of the blades were damaged during the storm, and
- we need to stop the turbine to perform a thorough inspection of the blades.
Solution to the problem
To optimize maintenance, we need some data and that’s where IoT comes in handy. IoT is very popular these days, but as we know it is not new. It can be traced back to the era when we called such solutions embedded projects. Now, let’s see what are the challenges that we face in this case:
- Wind turbine blades are movable parts, so no physical connection is applicable
- These devices need to be self-powered
- We need wireless communication to gather data from each blade
- We need to pass this data to the cloud for real-time alerts and place data in deep storage for further analysis
In this case, we are installing a lightning strike sensor in each blade and assign it to a particular wind turbine and wind turbine farm in our asset management module. Each asset is equipped with the data required for easy identification, such as GPS location. This way, we know where the event occurred as it is shown on a map on our dashboard module.
Technically, LoRa WAN-based communication is used to get the data from sensors to on-site LoRa WAN-based gateway, which then passes the data to the cloud solution using a wired or mobile connection.
When a lightning strike occurs (stage 1 in the image), the sensor detects the strike, measures the current passed through an individual blade (A) and provides some additional data gathered by the sensor. It passes this event’s data to the on-site gateway (stage 2 in the image). The data is then aggregated in the cloud (stage 3 in the image) where we perform rule-based actions about alerting (stage 4 in the image) resulting in an SMS and/or e-mail notification.
With this info, the maintenance team knows exactly which remote wind turbine farm location they need to visit (stage 5 in the image) and knows exactly which wind turbines to check (stage 6 in the image) and finally, exactly which blade on the wind turbine needs to be checked. This way, we can optimize maintenance and consequently reduce cost.
To take maintenance another step forward, we involve some advanced techniques based on machine learning. If we record the events and their details, we can now combine this data with the maintenance reports (stage 7 in the image) about the actual condition of the blade’s material after the event - we can use it to learn/find (stage 8 in the image) some patterns which lead to a prediction (stage 9 in the image) when something is most likely to break. This data can then be used for further maintenance optimizations and planning resulting in savings (stage 10 in the image).
As we can see, with the IoT and Machine Learning approach we can do marvelous things, this is just one of the examples in the Utility area of how to optimize the operation, Maintenance is everywhere. This is an operational solution and with technology covering the whole IoT stack, from gathering sensor data through Gateway, secure communication with cloud using one of the IoT protocols (MQTT in this case), storing data in the cloud, alerting/overseeing the events in the cloud solution, and as a cherry on top, getting value out of the gathered data using Machine Learning capabilities.
We achieved maintenance optimization in:
- Deciding do we need to go to remote area
- Stopping only the required wind turbines (minimize impact on production)
- Use advance techniques for better prioritization and planning