The Internet of Things (IoT) is full of promise for remote monitoring. This is critical to modern businesses – not just because customer expectations are higher, but they are tending to provide services rather than hardware. Rolls-Royce famously did this with their jet engines, selling thrust rather than metal. This has accounting advantages for the users, but crucially allows Rolls to take care of the fleet of engines and upgrade, test and maintain their hardware in a very efficient way. Key to this is remote condition monitoring – the investment needed to make it possible can be spread over many airlines and so be worth the cost.
This ability to remotely monitor everything from turbines to photocopiers is now being democratised – many hardware and service vendors are stepping in with a raft of solutions (Industry 4.0 / Smart Factory). They all have to make a critical compromise – the amount of ‘intelligence’ present in the sensor. A ‘dumb’ sensor just pipes all the data it’s collecting to a server farm somewhere where it can be analysed, whereas a ‘smart’ sensor is able to just put the critical bits of information onto the network.
The drawback with a ‘dumb’ sensor is that there can be an awful lot of data to shift. Imagine a yarn factory in the developing world – there may be 5000 machines with a single 3G modem giving Internet connectivity. Or, it could be a steelworks where wireless signals don’t propagate and there’s no prospect of adding Ethernet cables. So, why doesn’t everyone just fit smart sensors?
The answer is that it’s hard to program a smart sensor to recognise what is important, and what to send over the network. Imagine a car engine – a subtle knocking noise that indicates that the big end bearings are failing can’t be spotted by a simple threshold detector. Or an unexpected vibration from a wind turbine which only happens at particular wind speeds can’t be caught by an off the shelf filter. What’s needed is a sensor which ‘self learns’ what to send over the network by looking for things that change.
An example – mp3 compression changed the music industry forever. This wasn’t through any audiophile, quality centred advantage, but it allowed music to be compressed to a point where it could be easily sent over the networks of the day and stored on solid state memory cards. In fact, it’s very similar to the solution needed for a vibration sensing system – it keeps the information which matters and filters out the data which doesn’t.
We see the future of sensing as being a split between ‘edge’ and ‘centre’ intelligence. The edge (the sensor) is able to decide on the fly what matters and what doesn’t, and send only the most relevant information. The centre (the servers) are able to steer the sensors to report different aspects of their environment, and critically, build up a picture of ‘what’s normal’ across thousands of machines. In this way, the sensors form a linked police force across the entire user base, reporting on what’s different, what’s changed and what’s unexpected.
In this way, maintenance and repair can be directed to where it’s about to be needed. Imagine a day when the boiler repair technician appears with the right spare part just before your boiler breaks down…