Faraday Predictive

Contact: Geoff Walker <geoff.walker@faradaypredictive.com> and Will Boulton <will.boulton@artesis.co.uk>

ongoing discussion

Actually for our purposes we're not wholly focused on how good their data analysis is (though obviously if they manage to do a really great job, that would be excellent) - our thoughts were that as long as they've managed to design the system as a whole well, we could then spend some time refining or rebuilding the predictive fault diagnosis part, but what's really valuable for us would be a 'nice looking' user interface (maybe as a web app, if they felt that was most appropriate?) with capability to automatically provide reports, and a function to schedule maintenance in the future, and update a log to keep track of when known faults have happened; and the data stored logically in a database that they could design (depending on if we only gave them CSVs of the raw data). Perhaps one way to extend this into a more ambitious plan would be to ask for them to additionally have a reasonable looking phone app so that an engineer about to go to the site could get some idea about what was wrong with the compressor they were looking at (or make notes if they noticed something funny), whereas a manager on their PC could have an overview of what was happening across the whole plant?

A different option, though of a similar nature, could be to ask for the students to design a system such that their analysis of the data could be easily extended (by us) e.g. by uploading Python scripts to a repository and then getting results back in close to real time, with graphical results displayed in a friendly format - that would also be very valuable to us since and hopefully turn their task from just a data analysis task into one of overall design of the entire system.

One other thing we had in mind (though I'd need to discuss first with Geoff first) would be to provide a small bank of about 10 little DC motors (in a suitcase-sized box and with power supply etc already set up), and some sensors to go with them; if interfacing with that could provide a similar level of technical challenge compared to interfacing with the bike (in addition to using the sensor data to detect which motors are showing anomalies and display the results), then that could be very useful to us, and we'd be happy if the students wanted to test these motors to destruction, in which case there would be far less need for much insight into the mechanical systems as the problems introduced should be completely obvious.

Predictive Maintenance of Industrial Equipment

Faraday Predictive is a small local technology company which uses sensor data from electric motors to diagnose faults in rotating equipment (e.g. fans, pumps, compressors), and provide advice to clients about maintenance of their equipment. Useful advice contributes to the smooth running of this industrial equipment, potentially saving customers millions of pounds in some cases.

One of our customers operates in the gas industry. We have an expanding dataset (still in use) of thousands of records from 72 identical gas compressors, some of which have shown problems, that could be used to train machine learning models, rather than relying on expert analysis of each compressor. Your task is to deliver a system that can easily be used both by site engineers to alert them to problems, and allow them to provide feedback (alerting your system to when and where maintenance has occurred), and site managers, to schedule maintenance based on your models of machine health.

Potentially a project related to the company's work in condition monitoring of rotating machinery, for example using machine learning for failure prediction. However, this would have to be integrated into a larger product concept.

Alternatively, perhaps adapt The Headless Bicycle?