IEEE Power & Energy Magazine - November/December 2020 - 93

The nascent state of the industry also means reduced data
on fault patterns, especially due to limited public knowledge
exchange, making data analytics more difficult.
✔✔ The monitoring and diagnostics center at American Elec-

tric Power identified warning signs of failure and initiated
repair work of a gas turbine blade before breakdown, resulting in savings of approximately US$19 million.
✔✔ Duke Energy used predictive analytics for the early
detection of a crack in a turbine rotor. This resulted in
savings of over US$7.5 million.
✔✔ Southern Company applied predictive analytics to power station models to decrease unexpected maintenance
and maintain data quality reliability, resulting in savings of approximately US$4.5 million.
✔✔ Many wind turbine operators now use predictive analytics to monitor the health of gearboxes. The cost of gearbox failure can be upward of US$350,000 per incident.
Despite these and other successes, both business and
technical challenges hinder a broader adoption of predictive
maintenance in BESSs. From a business perspective, the
energy storage industry is relatively young. Thus, business
priorities and budgets do not always motivate investment in
"soft" engineering methods such as data analytics and artificial intelligence-based services. The nascent state of the
industry also means reduced data on fault patterns, especially
due to limited public knowledge exchange, making data analytics more difficult. Furthermore, the pressure to keep the
cost per kilowatt hour as low as possible means less investment in the sensors and infrastructure needed to process
large volumes of data.
In terms of technical challenges, predictive maintenance
techniques tend to be used for mechanical systems where
factors such as "wear and tear" can be readily measured
and monitored. For electronic systems, predictive practices
may be more difficult to apply. Rather than wear out, electronic components are more likely to have a binary failure
profile. An issue with one component may manifest itself
downstream and result in the failure of another component,
requiring data collection from multiple points to identify a
pending failure.
Implementing predictive monitoring in conventional BESS
hardware is also difficult due to limitations in communication channel availability and the processing power of battery/
energy management devices. In a BESS, predictive monitoring would involve processing data from battery racks and
the overall system to identify failure indicators. Ideally,
the BMS of an energy storage device should have the ability to assist in this area; however, not all current BMSs are
designed to recognize faults occurring outside of the immediate impacts on the battery itself (e.g., cells and modules)
november/december 2020

and may not have the throughput to process all the data. Predictive analysis must also depend on data from the EMS to
understand the system's behavior. Current EMSs are often
intended for dispatching/controlling multiple grid resources
and do not include the necessary monitoring and safety functions to manage single or multiple BESSs. EMSs often lack
direct communication with BMSs, and any fault detection by
the BMS may not get communicated to the EMS, limiting
preventive actions from system operators. Current standards
have not addressed this issue.
A properly designed monitoring approach for operational
ESSs will create indicators that can provide characteristics
such as those listed in Table 3. The overall goal is clear:
identify the indicators of potential faults and preemptively
intervene on an operational ESS without making the intervention itself a problem. However, the links and causal relationships between faults and indicators and the potential
of those early faults to lead to larger faults are not readily apparent at this early stage of the BESS industry. Ultimately, stakeholders must establish a methodology for identifying the indicator-fault relationships that can be tracked
and monitored in these systems.

Creating a Predictive Maintenance
Approach for BESSs
The sophistication of approaches for identifying useful
"flags" or fault indicators has evolved substantially. In the
most basic, reactive approach, these indicators are based on
near misses reported by employees. All data are significant
table 3. The key characteristics of indicators
for predictive monitoring.
Element

Description

Time
element

* Days of warning rather than minutes or hours
* Not all faults will have long lead times, but
anything that can extend the timeframe can
minimize destructive failures

Actionable
warnings

* Point to the components that need to be
replaced
* Allow time for the examination of areas
causing the warning to occur

Recoverable * Safety measures intended to prevent
actions
catastrophic failure and threats to human
safety can ultimately destroy the unit
(unrecoverable)
* Recoverable action must have minimal
impact on the system
ieee power & energy magazine

IEEE Power & Energy Magazine - November/December 2020

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2020