The Path from Data to Actionable Information as a Driver for the Industrial Ecosystem
issued, engineers dispatched and processes
relocated.
ecosystems to understand the impact and
the outcomes of the processes. Predictive
analytics also requires a statistically
significant amount of history to correlate the
execution with the expectations and
interpret feedback from the operators who
provide comments about the outcome. This
determines the relationships and training
sets to construct the statistical models for
classification at the earlier stages with
machine learning or deep learning.
Standards that are useful for ecosystem
integration include ISA-95 16 or its
implementation in XML by MESA, B2MML 17 .
In Figure 4, ISA-95 presents a layering system
that provides a logical separation of
functionality for industrial manufacturing
processes. 18 From a data perspective, the
first four stages map to layers 0 through 2
and the ecosystem map to layers 3 and 4.
Predictive analytics can also provide
machine health-related events to remove
machines from certain activities before they
result in delays and loss of revenue. For
predictive analytics to be effective, there
must be adequate context to understand
how the information relates to the
equipment, process and business, as well as
the intended results.
ISA-95 provides semantic information
describing the requirements, resources,
personnel and delivery of the job or order.
When combined with IIoT data, ISA-95
enables dynamic feedback to verify that the
intended process outcome matches the
execution and enables increased stability
and performance by informing design,
engineering and planning.
Predictive models are often used with
simulations to create what is now being
called digital twins or surrogates. The
predictive
models
are
commonly
constructed using first principle engineering
models (if standards are used, they are
provided in SysML) to describe the expected
behavior. A digital twin also represents a
process or a product and a piece of
equipment. IIoT data is used to refine the
first principles models based on actual
observations.
P REDICTIVE
The previous stages provide information
that is reactive to situations that have
occurred but are not attempting to look into
the future and predict outcomes or prevent
problems before they occur. This stage
begins to build the analytical models that will
look into the future and extend the time
horizon for problem avoidance.
Predictive
analytics
requires
an
understanding of the cause and effect
related to the semantic and enriched data
when combined with the business
Predictive models can also be deployed in
the enrichment stage local to the
16 https://isa-95.com
17 http://www.mesa.org/en/B2MML.asp
18 Brandl, Dennis. 2008. "T061_isa95-04.pdf." 05 19. https://apsom.org/docs/T061_isa95-04.pdf.
- 30 -
June 2018