Mechanical and electrical equipment is sure to break down at some point. When such equipment breaks down, it is usually repaired to bring the equipment back to its working condition. Repairing equipment after a breakdown is what we call “reactionary maintenance”, and is how maintenance has been conducted for hundreds of years. This way of doing maintenance has a few drawbacks, however: if a breakdown happens in an industrial production line, then the breakdown can cause a great loss in revenue due to lost production time.
To get around this drawback, companies have recently adopted what we call “preventive maintenance”, where equipment is replaced according to a schedule. In this way, the companies can ensure that most of the time, their equipment will work without breakdowns. This approach naturally has the drawback that perfectly working equipment will sometimes be replaced, which can be both costly and wasteful.
The latest development is called “predictive maintenance”, where data from sensors and the environment is used to predict when equipment will break down. This information can then be used to repair equipment before it breaks down, without wasting perfectly working equipment.
At this time, predictive maintenance is mostly applied by sending data collected from sensors over great distances to an advanced computer. This in itself has security and privacy concerns, as the data being sent over a network can be intercepted and used for malicious purposes. Furthermore, sending data through a network requires a large amount of power compared to regular sensor operations. It would therefore be preferable, both for security and privacy, but also to limit energy usage, to compute part or all of the predictive maintenance on the sensor devices themselves.
Therefore, the focus of this PhD project is to develop algorithms and systems to plan and support predictive maintenance of sensor deployments and industrial equipment in real-time where at least part of the predictive maintenance is computed on the sensor devices. The project will also investigate how the generated predictions can be used with scheduling techniques to generate optimised maintenance schedules.