Two incidents are in the news on pieces of equipment
breaking down at huge cost. In this world of “Big Data”, I really wonder how such
incidents are still so prevalent. However, this post is not just predictive
maintenance; what is interesting is that the 2 cases Boeing+Rolls Royce and
SMRT+LTA respectively illustrate 2 different worlds and how 2 different
approaches should be used to tackle them. Still I would have thought lessons
would have been learnt by now. Come on people, predictive maintenance is not rocket
science, neither should it be that expensive (the post script briefly addresses
the latter issue).
First the case of Boeing+Rolls Royce. ANA, the Japanese
airliner, is the largest operator of dreamliners; with 47 (https://en.wikipedia.org/wiki/Boeing_787_Dreamliner#Operators).
However, ANA was planning to ground its whole fleet of dreamliners (https://www.theguardian.com/business/2016/sep/01/engine-failures-on-787-dreamliners-prompt-ana-to-refit-entire-fleet)
and to replace around 100 Rolls Royce Engines over cracking and corrosion of
engine blades. This issue has come to a head in the last couple of weeks (http://247wallst.com/aerospace-defense/2016/08/29/boeing-787-engine-troubles-ground-ana-flights/).
Rolls Royce will conduct the replacement over 3 years,
apparently over the course of normal maintenance (see this article from the Financial Times http://www.ft.com/cms/s/0/be030804-6ec7-11e6-9ac1-1055824ca907.html#axzz4IzJ9bn3a),
and the number of flights ANA has cancelled is not as high as originally
feared. But ironically, the retrofitting at ANA will take so long precisely
because they are such good dreamliner customers.
What does that have to do with predictive maintenance? Well
I am sure that Rolls Royce would have tested its blades thoroughly before
delivering them for use in commercial aircraft; the responsibility is huge and
I have no doubt Rolls Royce and Boeing are serious about safety. But sensors
recording every aspect of a complex piece of engineering such as an aeroplane
in flight, including the external sensors recording the conditions of the
flight show a very different picture of the performance of the aircraft
compared to the tests.
I do not know enough about aeroplane engines, but I do
expect that if the blades are deteriorating, this should have an effect on fuel
consumption, and proper models monitoring health indicators such as changes in
fuel consumption given the conditions the aeroplane is operating in should have
rung some alarm bells.
It is not necessary to visually see cracks, they impede performance
and this should have been captured and isolated (especially when one of the
selling points of the aeroplane is precisely fuel-efficiency (http://travel.cnn.com/explorations/life/whats-so-special-about-Boeing-dreamliner-766616/)
It is even more so in the case of Rolls Royce where the
aeroplanes using the engines are flying in vastly differing conditions. While individual
airlines might have a smaller dataset, Rolls Royce should be providing
predictive maintenance based on the performance of the engines worldwide, a bit
like swift should be monitoring attacks against the system worldwide, something
individual members would not be able to do as efficiently.( http://www.bloomberg.com/news/articles/2016-05-26/swift-hack-probe-expands-to-up-to-dozen-banks-beyond-bangladesh).
This ability to pool data and run analysis and predictive maintenance global data it is
something I would expect to be a value-added service from someone like Rolls
Royce. (But then again, I would also have expected Boeing to be able to track
its aeroplanes live, so we do not have issues like MH370 that is still missing http://www.bbc.com/news/world-asia-pacific-26544554
, but we have to wait until 2021 http://www.theverge.com/2016/3/9/11184544/un-flight-tracking-system-malaysia-airlines-mh370).
A very simple example where a central body collects data to benefit all members
is the case of the Credit Bureau in Singapore.
The second case of the lack of predictive maintenance is from
SMRT+LTA in Singapore; today is the 4th day of interrupted train
systems (http://www.msn.com/en-sg/news/other/delays-hit-circle-line-for-the-4th-day-in-a-row-smrt-taking-manual-control-of-trains/ar-AAilpv9?li=AAaGkVj&ocid=spartandhp).
This time, the issues doesn’t seem to be directly related to rails - so said a
$10m inquiry (http://news.asiaone.com/news/transport/analysing-train-breakdowns-line-line)-
or the trains themselves (http://theindependent.sg/smrt-secretly-shipping-35-prc-made-trains-back-to-china-for-repairs/),
but with the communication between the control and the individual trains – the trains
affected are driverless – every time the train loses contact with the control,
it applies emergency braked, a safety feature. For now, the trains will be
manually driven; man saves machine...
SMRT+LTA are saying that the issue is taking so long
to resolve because they can only do checks and repairs when they system is shut
down.
How could predictive maintenance have helped? Again, sensors
are the answer.
I assume that for driverless vehicles, it is critical to
have sensors that are constantly feeding information into the system so it can
perform safely. Even before we look at what the signals are, the strength of
the signals should be crucial.
May be there was some over confidence since you can surf at
speeds of up to 1Gbps (http://news.asiaone.com/news/singapore/hetnet-trial-mrt-stations-seamless-speed-surfing),
but simple monitoring of the quality and variability of the signal should be
the basic information required to understand the reliability of the service.
It’s not just predictive maintenance, I think it is crucial
to have an understanding of the systems and a ranking of the types of issues
that can impact the system, and the collection of data should be done
accordingly. You do not simply take “all the data” and “throw it in the machine”
and expect to “get all the answers”, especially if you keep having new issues.
The SMRT and LTA have engineers; this human knowledge that
they have should be put to good use to design proper monitoring and predictive
maintenance so such issues don’t just keep on popping up over and over again.
Catch them early so it is more feasible to intervene during down time, all at a
minimal cost to the economy.
PS
The idea is that such technologies are terribly expensive.
Actually they should not be; it’s all in how the solution is designed. One huge
cost is driven by the need to “real-time”. How important is it for the
information to be real time? I have had a few clients who wanted information
real time, but who would only take action the next day, so effectively it doesn’t
matter in terms of the timing of action taken if the information arrives by
batch at end of day, or in real time. But it’s cool to be able to show stuff in
real time.
However, emergency activity that requires instant reactions
should be on-board, for example the train applying emergency break when signal
is lost. Predictive maintenance should alert you way in advance that you are likely
to be having signal issues, so you have the time to solve the problem before it
occurs.
Also, many of the sensors are already in place, it’s just
that the systems were not designed to store the measurements. That need not be
a total replacement, but a retro-fit could do in a pinch. And to add to this,
storage is getting cheaper and cheaper.
Designing systems to capture the data needed for predictive
maintenance (or any other type of analytics for that matter) is a huge topic.
But for today, I’ll just say that it’s usually less expensive than thought if
the focus is on the minimum requirements that would deliver the results
required.
No comments:
Post a Comment