Engineering a solution

Engineering a solution

It would be a grave mistake if the future of predictive machine maintenance was led solely by data science, warns Professor Andrew Ball

The proverbial advice is that we should be careful what we wish for – just in case life has a nasty trick up its sleeve. I am beginning to see more than a grain of truth in that old saying.
My wishes are the same as ever. I am a diagnostic engineer, specialising in condition monitoring, and since my first degree in 1988 I have worked and researched on a global scale. My subject area is crucial to the performance and efficiency of many sectors and will be more important than ever as we embark on Industry 4.0, a fundamental tenet of which is that the downtime of manufacturing facilities must at least be minimised if not eliminated.
But for most of my career, machine maintenance – however technologically advanced – has been synonymous with the word “breakdown”, it has struggled to shed the misleading stereotype of a bloke in oil-stained overalls brandishing a big spanner.
I am not being snobbish. There are situations in life where a bloke brandishing a big spanner is precisely what you need – and I should know because in my spare time I run a successful sheep farm, which comes with an eclectic array of ageing agricultural equipment. But for the best part of 30 years, by which time I had become a university professor running a cutting edge diagnostic engineering research facility and forging collaborations with colleagues around the world, I have longed for the day when the old, oily rag stereotypes would fade away from world class industrial maintenance.
A few years ago, it looked as though it was to happen, and I said as much at several conferences around the world. Condition monitoring had a rosy future and a burgeoning reputation, and I could provide plenty of evidence to back up my optimism. For example, the influential research organisation MarketsandMarkets issued a detailed report that examined the machine condition monitoring sector and forecasted exponential growth in the 2020s, to be “driven by the advent of secure cloud platforms used in condition monitoring, increasing disposition toward predictive maintenance, and high demand from emerging applications”.
The report predicts that all sectors will see growth in CM investment, both hardware and software, and most notably that the vibration monitoring segment will see the greatest increase in spend during the forecast period.
My interpretation of this newly found limelight was that professional maintenance engineers – the people who know how precisely how to operate and maintain state of the art manufacturing plant such that its reliability is maximised – would have a central and esteemed place in Industry 4.0 and beyond.
But then I began to notice a new and worrying trend – a trend which threatens to relegate maintenance engineers in the industrial hierarchy as they are elbowed aside by data scientists.

I spoke about this at the 2019 Condition Monitoring and Diagnostic Engineering Management Conference (COMADEM), held at the University of Huddersfield.
There is no doubt that data science has a vastly important role in many sectors of modern life, including manufacturing industry. I have highly valued academic colleagues who work in this field, and it’s true to say that I myself have extensively and successfully used data driven approaches to the automated detection of abnormalities in complex plant. But now I am reading articles and web posts which claim that the future of predictive maintenance will be led not by engineers but by data scientists.
At recent conferences I have heard speakers talk about purely data-driven approaches to predictive maintenance, with no conception of the engineering behind it. I find this scary, not because of any threat to my professional self-esteem, nor because I am precious about my field, but because if there is one thing I have learned over many years about high performance in machinery fault diagnosis and prognosis, it is that engineering know-how and context are absolutely critical.
So, we need to be very careful. Predictive maintenance is clearly coming into the limelight at long last and it is quite rightly seen as being of central importance to the future of advanced manufacturing. But it is also clear that predictive maintenance – while it can be aided by data science – is an engineering discipline that must be led by professional engineers.
To prove my point, let me leave the last word to… a data scientist! At the University of Huddersfield, I have colleagues who are carrying out highly innovative work in the field of big data and AI, but they also accept that engineering knowledge will remain vital for accurate diagnosis and prognosis of machinery faults.
For example, my colleague Dr Simon Parkinson, a Reader in Cyber Security, comments: “Data science techniques are proving very useful in engineering. However, their utility without engineering knowledge is limited as generalised computer software models do not have practical knowledge and experience, which is required to add valuable meaning and understand unexpected situations. I would argue that the importance of practical knowledge and experience is strengthening with the increasing use of data science.

Andrew BALL is Pro Vice-Chancellor for Research and Enterprise and a Professor of Diagnostic Engineering at the University of Huddersfield.  He’s a world-renowned expert in the field of diagnostic engineering and has authored hundreds of technical and professional publications.