HIGH PROFILE
AI analytics
Mining companies typically spend a massive 20-50% of
annual operating budgets on equipment repair and
maintenance, and reduced yield from downtime has an
even greater financial impact. The major cause of
downtime is that mines don’t have the information they
need to diagnose and prevent failures. Rithmik Solutions
created Asset Health Analyzer™ which uses AI to give
miners real actionable data insights. Paul Moore spoke to
Kevin Urbanski, Co-Founder and CTO
recognised the power of AI in extracting so much
more insight from available data which some
mines had been collecting for well over a decade.
There is so much untapped value in this data. We
set up Rithmik in 2018 with our core offering
being Asset Health Analyzer™ or AHA™. Behind
this is our analytics engine we call Rapid
Analytics Infrastructure or RAI which enables
extremely quick AI model testing and creation,
leveraging the maximum potential of cloud
computing. For example, some of the models we
have generated would have taken three months
on a single CPU; RAI utilised 3,300 virtual CPUs
and 11 terabytes of RAM to get a result in five
minutes.
A Rithmik Asset Health Analyzer™ drill down
dashboard that allows a user to dig into
equipment health and operation, efficiency and
failure modes
Q What is your background and what were your
reasons for setting up Rithmik?
A Before Rithmik I had already been in the mining
industry for ten years, starting out with what was
then called Matrikon, which went on to be
acquired by Honeywell. I started there when they
were introducing their proof of concepts for the
Matrikon Mobile Equipment Monitor solution; I
was the hardware guy, getting on new pieces of
equipment, figuring out how to get the data off,
writing new software drivers. After the Honeywell
acquisition I went to work for Teck Resources
where I spent three years working with mining
technology systems, including the Matrikon
MEM. I was working to get the value out of the
sensor data – the standard approach was and is
in the industry to set up so-called ‘user defined
events’ where you set thresholds on sensor
values from the equipment and when these are
exceeded you can trigger warnings and alarms.
But one of the challenges with this approach is
that due to the conditions in the mines being so
diverse with weather/temperature and
topography with slopes that can change daily,
the thresholds are tough to tune in because of
the changing mine environment. False positives
can be generated which can mean after a while
the events and alarms no longer get looked at
and therefore lose their meaning. This is the key
reason that we started Rithmik. With AI, systems
today have the ability to learn the multivariate
relationships between sensor data. We
Q What does all this mean in operational terms?
A What that means is that you are able to tell an
AI algorithm, this is a set of data (eg RPM, oil
temperature, oil pressure, coolant temperature)
and feed this into the algorithm and it is able to
find the interconnected relationships between all
the sensor values. With the AI knowing these
relationships you can get early prediction and
classification of failure modes as well as deep
insights into failure modes and inefficiencies,
enabling more efficient, smart maintenance
programs. These processes are the foundation of
automated maintenance scheduling, and
eventually — with augmentation from
operational data — maintenance and operations
orchestration. The future is exciting.
Q Can you give an example of a type of dataset
this works on?
A A good example is the relationship between
RPM and oil pressure. As RPM goes up, oil
pressure naturally goes up. But throughout that
range of RPM values, if you were to be setting
strict thresholds for user defined events, you
would have an infinite number of oil pressure
thresholds over that RPM range. The AI can say,
at 1,000 RPM, what is the normal oil pressure
range? If it is outside that range, the algorithm
will pick that out and indicate how much in psi
94 International Mining | SEPTEMBER 2020