IEEE Power & Energy Magazine - November/December 2020 - 94

During the design phase, system integrators develop the product
based on institutional or historical knowledge; however, the use cases
the system encounters in the field could lead to new fault indicators.
and can be useful in preventing future failures. Hence, we
recommend a culture where the reporting of near misses
is encouraged. More rigorous approaches involve 1) leveraging indicator-fault links established during the systemdesign phase and 2) combining with additional indicatorfault links from the analysis of operational data on fielded
systems. The identification of these links is an iterative process. During the design phase, system integrators develop
the product based on institutional or historical knowledge;
however, the use cases the system encounters in the field
could lead to new fault indicators. Thus, the predictive
maintenance approach should be scalable to adapt to new
"patterns" with minimal impact on the overall system cost
and availability. The following section elaborates on this
two-layer approach for identifying indicator-fault relationships during the design phase and based on the data analytics for fielded systems.

Identifying Indicator-Fault Relationships
During the Design Phase
It is expensive to retrofit a fielded system. Thus, the first step
during the design phase is to make a deliberate decision to
sense critical information and obtain as much data as possible to provide insight into various failure modes. Next, the
process requires
1) the creation of a comprehensive listing of recoverable
battery system faults and the linking of faults to leading indicators. This begins with thoughtful engineering consideration of the system design; however, the
designer should complement this consideration with a
collection of historical data from key developers, operators, and manufacturers.
2) the determination of whether indicators are already
being tracked through current BMSs, EMSs, or any
plant controllers.
3) the finalization of a list of indicators and criteria to
be monitored to reduce the field failures of BESS
equipment.
This is a beneficial process to leverage, although there
may be gaps when new failure modes are identified or the
process does not account for the design errors or field/environmental degradation that could lead to failures.
The industry has many well-established processes for system design, including various probabilistic risk assessment
(PRA) approaches (e.g., failure modes and effects analysis,
fault tree analysis, and so on) and systems-theoretic process
analysis (STPA). It is important to note the nuances from
94

ieee power & energy magazine

processes created from a system-safety perspective. These
processes are rooted in historical data, where the mechanism
of a past failure is identified to improve designs and prevent
a similar failure from occurring in the future. To contribute
substantially to predictive maintenance, however, the system-design process also needs to establish the monitoring
criteria that can be used in maintaining device operation.
PRA, built from a foundation of risk management, is
the most widely used safety engineering method. A PRA
approach identifies hazards, their deterministic causes and
consequences, and provides a method of describing uncertainty. The process enables the calculation of expected risk
(defined as the probability of an event multiplied by the relative severity of its consequences) so that a developer can
compare different design options. PRA uses fault and event
tree analysis to break a complicated system into subsystems
and components when there is insufficient data to directly
predict behavior. Risk is then increased or decreased based
on how failures in components and subsystems operate
together to generate accidents.
Additionally, failure modes and effects analysis (FMEA)
is a systematic procedure for assessing reliability and how
component failures can impact system safety. Developers
begin an FMEA by compiling a list of each component or
type of component in a system. Then they calculate the probability of each component failing in a variety of ways based
on historical data. Table 4 shows a brief list of typical FMEA
calculations for a BESS (adapted from an Electric Power
Research Institute report on ESS safety). The probability and
severity each receive a score of 1-10, with 10 corresponding to a more probable or severe event. Each failure mode is
linked to a hazard effect, consequence, method of prevention,
and method of detection. The identification of the detection
method lays the foundation for predictive maintenance. It is
apparent, however, that this conventional FMEA approaches
system design from a safety perspective (preventing catastrophic failure) rather than detecting faults while they are
still recoverable. The process creates a probability but does
not provide the leading indicators necessary to flag pending
failures of the areas. Still, these processes are beneficial in
understanding what areas to focus on when creating indicator-fault relationships.
A more recently developed design tool, STPA, views
a system as a collection of interacting control loops. Accidents happen when the component interactions in these
loops violate safety constraints. Unlike PRA-based tools, it
does not rely on any component failure rate data. Thus, this
november/december 2020

IEEE Power & Energy Magazine - November/December 2020

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