Predictive Energy Maintenance with Machine Learning
By Quentin Griffiths
How good would it be to fix an issue long before it even occurred?
Infrastructure companies the world over own and maintain a vast number of assets deployed over large geographical areas. Electricity transmission and distribution (lines) companies are a good example of this with the deployment of lines, transformers, switch gear, relays, and so on over entire states or countries.
With such a large asset portfolio comes a huge maintenance overhead, a wide skill set requirement and active maintenance schedules to keep a system running. Couple this with long asset life cycles (min 10 years) and aging infrastructure, you get a maintenance nightmare.
So, how can we move from reactive maintenance to predictive maintenance?
Taking the energy industry as an example, we can look at ways in which predictive maintenance (PdM) can be achieved.
Firstly, what is Predictive Maintenance? Predictive Maintenance is an action that is taken to maintain an asset based on a change in its performance or operating condition, returning the asset to normal operation and therefore reducing the likelihood of failure.
Lets take a transformer as an example. When a transformer is first installed, a range of low cost sensors can be attached to the transformer to monitor its temperature, vibration, oil levels, oil temperature, sound and perhaps CT clamps on incoming/ outgoing lines.
Using connected sensors (#iot) and feeding this data straight into a real time machine learning algorithm, you can create a model of the 'normal operation' of the transformer.
Continually monitoring these sensors will highlight the different operating modes of the transformer as external temperatures/ environmental conditions change and will train the algorithm to understand the modes of operation over time.
As the algorithm learns how to model the transformer, it can then highlight any changes in the operation/ performance of the transformer caused by non environmental factors such as insulation breakdowns, oil issues or over heating.
With a real time view on these performance changes, the system can alert owners to an issue with an asset, allowing them to review and dispatch a maintenance crew. The maintenance crew may then perform more in depth analysis using high sensitivity equipment to evaluate the cause of the issue and act accordingly.
Overtime, as the system improves the accuracy of the PdM will improve, leading to a more predictive rather than reactive maintenance operating model.
Cool...Is this actually achievable?
All of this might seem a little vague and idealistic, but in reality it is very achievable with the technology available today. A wide range of low cost sensors are readily available on the market today!
Couple this with real time cloud computing and a strong Data Science partner like YouDo, predictive maintenance is achievable.
How could this affect your business?