Everything works wonderfullyIn Part 3 of Dr Michael Provost’s blog on ’23 lessons learned the hard way’, Mike shares five key lesson’s that will ensure that analytical techniques are relevant and that the output is communicated to the user in a way that leads to action. In particular he warns not to blindly apply big data techniques to Asset Management but use analytics that are based on sound technical and business logic.

This adds to the insights shared in his previous  two blogs:

Part 1: Business lessons

Part 2: People and Data

Part 3 Analysis and Visualisation

12. Keep it simple

The Asset Management literature is full of analysis methods that are poorly explained, steeped in obscure mathematics, lack clarity or obvious engineering relevance and seem to be aimed more at demonstrating the cleverness and academic credentials of the author(s) than enlightening humanity. The end result of any analysis has to be action, usually taken by someone with practical rather than academic intelligence. Techniques such as applying thermal paint to a component that changes colour when that component overheats are considerably cheaper and more quickly and easily understood by those in the workshop (who may not have much time to understand and fix problems) than more complex data gathering, transmittal and remote analysis. The smartest analysis or visualisation in the world is useless if nobody else understands and trusts it enough to act on it.

13. A physics-based asset model is a very powerful business and technical tool

It builds the foundations for full understanding of asset and business dynamics. Such a model (or set of models) improves communication within and between all interested parties both inside and outside your business and provides consistent and traceable predictions and baselines of asset, project and business performance. It also provides you with rapid assessments of how assets should behave in different environments and operational contexts and forms the basis of fast, consistent and accurate assessment of asset performance in the field. Physics-based asset models can support the application of many advanced analysis techniques that would not otherwise be practical and enable the optimisation of asset technical and business performance, profoundly improving the cost, speed, efficiency and effectiveness of asset development and in-service support.

14. If at all possible, compare all your asset measurements to a baseline, which ideally takes account of all known external drivers of the recorded values (e.g. load variation, ambient condition changes and other quantified effects)

There is always a baseline somewhere (from a number in someone’s head to a full physics-based asset model) against which measurements can be compared; find it and make it visible, so everybody can easily see what is good and what is bad. Residuals (the differences between measurements and baselines) are much easier to understand and analyse than raw measurements and provide order-of magnitude improvements in the ‘granularity’ of your analyses that significantly increase the timeliness and effectiveness of asset health and operational assessments.

15. A good measurement and/or analysis visualisation, tailored to the person you are talking to, will make all the difference

Some visualisations (such as time series, X-Y plots, bar and column charts, dashboards, alerts and interactive drill-down) will always be useful, while others (like mapping, statistical displays, reports and system synoptics) may find more specialised niches in the organisation or with customers. Use visualisation to persuade and excite; people rarely know what they want to see and how they want to interact with the data, but will provide enthusiastic feedback when you can show them examples of what can be done.

16. The appropriateness of the analysis is more important than the ‘bigness’ of the data

‘Big data’ is all the rage, with many commentators and IT consultants seeing the advent of massive unstructured databases, off-the shelf analytics and cheap ‘cloud’ storage and processing as panaceas for most Asset Management issues. While such approaches can work well in the ‘softer’ areas of retail, social science and financial asset management, there are more appropriate tools and thought processes that you can and should use for analysis of the performance and operation of physical assets. The use of ‘smart analytics’ to back-calculate what a good physics-based model of the asset could have told you gives a false sense of progress and potentially confuses failure signals with the noise of operational variation. ‘Black box analytics’ also make it too easy to ‘overfit’ data (producing an analysis that is not valid for new data when it arrives) and/or find spurious patterns or correlations in large datasets that don’t make logical sense. Asset Management must be based on sound technical and business logic; subcontracting the thinking to the latest IT hype can quickly lead you to expensive failure and loss of credibility.

Note: This series of blogs is based on material originally produced for the Society of Automotive Engineers (Jennions 2014): it has also been presented as a paper at the IET/IAM 2014 Asset Management Conference in London, UK (Provost 2014). It is reproduced with permission from both the SAE and the IET.