Unplanned downtime is still one of the predominant
costs in a manufacturing operation
despite the advances available in predictive
and condition-based maintenance technologies.
Even though the focus of proactive maintenance has
been to catch symptoms of failure early, infer the
probable causes from the symptoms, and take proactive
actions to prevent the failure, only limited success
has been achieved compared to the previous generation
of predictive maintenance approaches.
This is mainly due to the treatment of the asset in
question as a “black box.” Without the knowledge of
the internal state of the asset and state transitions
that take place between the sub-processes of the
asset during operation, it is not possible to exhaustively
infer the real cause and/or potential for a failure.
In the past, even if a failure was avoided because
a symptom was treated via predictive maintenance,
the actual cause was not addressed.
The reality is that an asset will fail unexpectedly
when some failure-inducing condition in its
past operational behavior has gone undetected.
That’s why detection of sub-optimality is required
not only to avoid unexpected failures, but also to
achieve superior performance.
This is where the idea of continuous asset optimization
enters the picture. With the advent of
Industry 4.0 technologies, it is now possible to
reasonably strive towards dynamically achieving
asset optimality given the current capability to
generate meaningful and contextual information
on asset performance in real time.
Traditional Big Data analytics apply deep learning
for strategic optimization to enable decisions
that will have a significant impact in a month,
six months, or a year. This requires a plethora
of past data to identify hidden patterns of asset
performance. Continuous asset optimization differs
in that it can achieve asset optimization in
a relatively immediate future time frame, such
as the remaining period in a shift, the next shift,
or the next day. And it can do this without the
need for a significant amount of historical data,
making it more tactical and operational. This realtime
approach to asset optimization also ensures
that the behavior of each component/sub-process
associated with every critical asset is understood
within its wider operational context. By modelling
(based on first principles physics) the interaction
between the behaviors of components of a process/
asset, future behaviors can be predicted and
optimized using artificial intelligence techniques
and operating costs can be significantly reduced.
Such continuous and practical optimality in
assets can be achieved in a standard and systematic
manner via digital twins. Digital twins have been
around for almost a decade now, but their applications
have evolved since their first definitions.
The Digital Twin framework shown comprises
four layers: operations, performance, optimization,
and prediction. The operations layer connects
to the assets in the plant and collects real
time data from the assets and sub-processes. The
performance layer computes the performance of
the assets from the collected data. The optimization
layer applies the data from the operations
layer and the key performance indicators computed
in the performance layer to models of specific
sub-processes of the assets. The individual
optimizations computed by the model are then
fed as inputs to a predictive layer which reconciles
the sub-process optimizations to arrive at an
overall asset/process optimization. Communication
between the layers happens opportunistically
and dynamically to ensure that the assets are
continuously optimized. That’s how a significant
component of the digital twin can be achieved
using a minimalist network infrastructure, edge
computing, IoT hardware, and intelligent firmware/
application software, keeping overall implementation
costs within affordable levels.
