Dr. Diganta Das (CALCE) was the first speaker and his
topic entitled “Is Your Reliability Prediction Reliable?” Firstly, Dr. Das
introduced Reliability and Predicted Reliability. Reliability is the probability that an item
will perform its intended function for a specified interval under stated
conditions. Predicted Reliability is for
the stated conditions of use, and taking into account the design of an item,
the reliability computed from the observed, assessed, or extrapolated reliabilities
of its parts, interconnects and interfaces.
Then Dr. Das explained the reasons to perform
reliability prediction that could understand the risks and take different
actions. However, such prediction
usually only consider as a contractual requirement or regulatory
obligation. He also briefed different
method for prediction including reliability prediction handbooks, reliability
testing, use of field data and physics of failure, etc.
After that he briefed a timeline for MIL-HDBK-217
since 1940s. And then he described the
change of transistors number and features from 1970s to 2010s. MIL-HDBK-217F,
Notice 2 was issued in 1995.
However, most prediction form various handbook and
different methodologies had large deviation.
Problems included incorrectly assume constant failure rates, no
information on potential failure modes, mechanisms and sites. Finally, Dr. Diganta Das introduced IEEE Standards
1413 and 1413.1 which provides a framework for reliability prediction of
hardware that is a tools for evaluation of reliability predictions.
Dr. Michael H. Azarian (CALCE) was the second speaker
and his presentation named “Prognostics and Health Management enabled
Maintenance”. Firstly, he introduced health progress timeline from “Start of
product life cycle” to “Early incipient fault” to “Component or sub-system
failure” and to “System failure”.
And then Dr. Azarian introduced four maintenance
strategies and they were “Corrective Maintenance”, “Scheduled Maintenance”, “Condition-based
Maintenance” and “Predictive Maintenance”. Then he explained it using
automotive health monitoring sample.
After that he briefed Prognostics and Health
Management (PHM) that permits the evaluation of system’s reliability in its
actual life-cycle conditions. Then PHM process cycle was mentioned. He raised different possible PHM application
levels and they were Level 0 – die, Level 1 – component, Level 2 – assembly,
Level 3 – module, Level 4 – system, and Level 5 – system of systems.
Finally, Dr. Azarian discussed two Prognostic
Approaches that were Physics-of-Failure (PoF) Approach and Data-Driven
Approach. The PoF-Based PHM methodology
was showed in the following diagram.
Data Driven PHM was discussed. Machine learning based on statistical methods
is well-suited for PHM because it is capable of actively learning about the system
and its dynamics, faults, and failures. Lastly,
Dr. Azarian concluded that both approaches are used to evaluate the health of a
system, to predict its remaining useful life, and to implement risk-mitigating
actions such as preventative maintenance.
The third speaker was Ir. Wilson Kwok (Head of
Technical Services, The Hongkong Electric Co., Ltd.) and his presentation topic
was “Application of AI in the Reliability and Security of Power Grids”. In the
beginning, Ir. Kwok said the meaning of smart cities is to use of connected technologies
to improve efficiency and quality of life in a sustainable way.
Then Ir. Kwok compared smart cities and human body
that both key elements included brain (decision making), senses (CCTV) and body
(ICT). He also shared some application such as hot spot on apparatus, water
dripping and flooding.
And then he introduced their Intelligent Volt-VAR
Regulation (IVAR) System Architecture and Software Design
After that he briefed the history of medium voltage
(MV) cable diagnostic assessment since 1980.
At the end, Ir. Wilson Kwok introduced that they adopt
AI methodology to process and analyze data and review the causes of faults for
cable life assessment.
The last speaker was Dr Siqi Bu (Associate Professor,
PolyU) and his title named “Data-Driven Techniques in enhancing Reliability and
Security of Power Networks”. Firstly Dr. Bu briefed the background of power industry
transformation as 4D including Decarbonization, Digitization, Decentralization
and Deregulation.
Then Dr. Bu briefed the predictive maintenance of
power equipment. The future predictive
maintenance (PdM) would be AI-based, Data-driven approach.
After that Dr. Bu mentioned tools of Situation Awareness
(SA) that modern power systems were being treated as comprehensive cyber
physical systems (CPS). He proposed an Enhanced Situation Awareness Based on
Random Matrix and Deep Learning.
Finally, Dr. Siqi Bu discussed CNN-LSTM model to
achieve the collaboratively spatiotemporal data mining. Technically, CNN module is to extract and learn
spatially correlated features and LSTM module is to handle temporal correlation
of the extracted feature. Lastly he concluded that the very fast post-fault
actions (PFAs) could be implemented by the system operators to avoid any
unstable operational status of the electric power system.
Q&A Session
At the end, Prof. Winco Yung (Centre Director, CAiRS)
gave closing remark. He said we needed
data and knowledge domains from different industries for employing AI in reliability
modeling to enhance safety level and CAiRS had different experts and partners
to achieve this goal.
