Active Alarm Processing in a Nuclear Power Plant
Gwi-sook Janga , Jong-yong Keuma, Je-Yun Parka, Young-Kuk Kimb
a
I&C HFE Department, Korea Atomic Energy Research Institute, 1045, Deadeokdaero, Daejeon 305-353,
Republic of Korea
TEL : +82(42)868-8626, FAX : +82(42)861-9618,
b
Department of Computer Science & Engineering, Chungnam National University, Daejeon, 305-764,
Republic of Korea
E-mail: (gsjang, jykeum,jypark)@kaeri.re.kra, ykim@cnu.ac.kr b
Abstract- An Advanced Alarm System (AAS) of a NPP (Nuclear
Power Plant) is primarily a digital system employing advanced
alarm process logics and a VDU (Visual Display Unit) based
control and display for the alarms. The role of the AAS is to
provide the information necessary to safely shutdown the reactor
under all plant conditions, to monitor the plant parameters
approaching or exceeding the operating limits and to minimize
the number of alarms, and to group and prioritize the alarms.
The secondary function of the alarm system is to serve as an
alarm and event log. This paper proposes a new approach to an
active alarm processing in a NPP by using active database and
data mining techniques.
Keywords-active database; alarm processing; data mining
I. INTRODUCTION
An AAS (Advanced Alarm System) of NPP(Nuclear
Power Plant) is to assist the operator to monitor the systems
and processes and to take the necessary actions required to
preserve the normal operating conditions. The AAS must only
present meaningful information, and must not flood the
operator with miscellaneous data that he/she does not have to
act upon. So, irrelevant or unnecessary alarms have to be
eliminated and less important alarms are suppressed [1].
The AAS is a database driven information system. The
need for an automated extraction of useful knowledge from
huge amounts of data has become widely recognized. To
efficiently manage data that varies rapidly over time and
process event driven transactions, AAS requires database
systems that support an active rule processing.
The active behavior of a system is generally specified by
means of rules, which describe the situations to be monitored
and the reactions of the system when these situations are
encountered. In general form, a rule consists of a rule event, a
condition and an action, known as Event-Condition-Action
rules or ECA rules.
Data mining in databases is the efficient discovery of
previously unknown patterns in large databases and is
emerging as a major application area for databases.
This paper proposes an active alarm processing method in
a NPP by using active database and data mining techniques.
978-1-4244-2972-1/08/$25.00 ©2008 IEEE
Authorized licensed use limited to: RNS Institute of Technology. Downloaded on February 9, 2009 at 00:39 from IEEE Xplore. Restrictions apply.
In this method, we can acquire alarm reduction rules by using
data mining techniques and these rules can be automatically
triggered by an active database. And rules are continuously
mined at a desired frequency.
II. ALARM REDUCTION USING ACTIVE
DATABASE
A. Functions of AAS of NPP
An alarm system of a NPP is a basic operator support
system for managing abnormal situations and it has the
following two functions [2].
The primary function of the alarm system is to warn the
operator about a situation that is not normal. Each alarm
should alert, inform and guide the operator.
The secondary function of the alarm system is to serve as
an alarm and event log. The log functions supports the
operator’s need to analyses the event that have lead to the
current or previous process conditions. The alarm log should
be used for:
- Analysis of incidents
- Optimizing plant operation
The alarm log should be flexible and contain also events,
suppressed alarms and other pieces of information that were
not presented in main alarm list, but which could be useful for
offline investigation of incidents. Alarm log information
should be used for monitoring and improving the alarm
system performance. The general structure of AAP(Advanced
Alarm Processing) is shown in Fig.1.
B. Active Database
Active database technology can, therfore, enable a
system to have a reactive processing and autonomous
response to an event that occurs inside or outside a system.
Many applications, including classical ones like an inventory
control, would run much more efficiently. This would free the
applications of the need to continously monitor relevant
events and as result, the application code would be less
complex and more manageable and reliable.
1
 A fault occurring in a certain part of the plant process
propagates through the process following a cause-
consequential relationship. As a result of the fault
propagation, an alarm occurring in a part of the process may
cause another alarm in another part to be activated. (See
TABLE III)

TABLE III
EVENTS DEFINED FOR A CAUSE-CONSEQUENCE RELATIONSHIP
Event of Interest What Triggers the event Action to be taken
E1 MFW Pump Trigger C1 Trigger C1
Low Suction Flow
E2 Heater Drain Tank Trigger C1
Low Level
C1 E2, followed by Consequence alarm is
E1(sequence) suppressed

III. ALARM REDUCTION USING ACTIVE DATA

Fig.1. A structure of an AAP MINING

C. Adapting Active Rules for AAP With the help of data mining techniques, clear
relationships or general regularities of a dynamic alarm
Alarm processing techniques were developed to support processing system could be discovered and the conditions and
operators in coping with the volume of alarms, to identify actions of rule could be specified at a higher concept level. In
which are significant, and to reduce the need for operators to this paper, data mining is introduced into active rule
infer plant conditions. This section describes the test case extraction for the alarm reduction in NPP. And alarm data is
used to demonstrate the AAP by using the concepts of continuously mined at a desired frequency.
composite events.
A. Rule Extraction using Data Mining
z Equipment-State Dependency
The alarms activated as a consequence of an equipment- Rule extraction is performed during the analysis phase of
state change (e.g., pump tripped) are lowered in priority. (See active alarm processing applications. Rule extraction may
TABLE I) cover requirements analysis and the first steps of a conceptual
database design as well. The results of rule extraction are the
TABLE I rule requirements. Rule extraction consists of a sampling, an
EVNETS DEFINED FOR AN EQUIPMENT STATE DEPENDENCY exploration and a modeling phase.
Event of Interest What Triggers the event Action to be taken A mining system for a rule extraction receives the alarm
E1 Pump Trip Trigger C1 data from a database and establishes useful rules. Association
E2 Low Flow Trigger C1 rule miner can find a correlation among the attributes in a
C1 E1 and E2 Low flow alarm is
record, although a frequent episode miner searches for event
eliminated
patterns in the records.
In addition, a clustering analysis provides a data
z Multi-Setpoint Relationship abstraction from the underlying structure. And it groups data
The lower level alarms are reduced in priority when the objects into clusters such that the objects which belong to the
higher level alarms also are activated. (See TABLE II) same cluster are similar, while those belonging to a different
cluster are dissimilar.
TABLE II
Because we consider the characteristics of the alarm data,
EVENTS DEFINED FOR A MULTI-SETPOINT RELATIONSHIP
we improved the existing data mining algorithms to create
Event of Interest What Triggers the event Action to be taken
candidate item sets that include only the interesting attributes.
E1 PZR Pressure Hi Trigger C1
B. Active Rule Analysis and Rule Update using Data Mining
E2 PZR Press Hi-Hi Trigger C1
C1 E2, followed by Low alarm is
The rule controller analyses input stream data for alarm
E1(sequence) suppressed
based on data mining algorithms, updates ECA rules, and
controls event, condition and action of rules. The rule
z Cause-Consequence Relationship

2
Authorized licensed use limited to: RNS Institute of Technology. Downloaded on February 9, 2009 at 00:39 from IEEE Xplore. Restrictions apply.
controller consists of programs for establishing association Composite
Plant Composite Condition
rules, frequent episodes and a clustering. Association rule Operation Event
Event
Condition
Evaluator
Detector Evaluator
miner can find a correlation among the attributes in a record. Data Detector

The frequent episodes miner is used to search for a series of Sampling

Clock
Sampling Event Invoked
event sequences for frequently occurring episodes. An Detector Detected

episode is defined by a sequence of specific events that occurs Exploration Signal Event
Exploration
frequently, and events composing a sequence are closely External
Association Rule

related with one another. Here, exploring is done to find all Event Clustering Action
Action
Detector Executer
the frequent episodes in a set of time windows of a user- Association Rule
Signal
Event Signal Event Frequent Episode
Executer

defined size. A sequence pattern and an episode are similar in Clustering Rule Controller
that they explore patterns in sequence. ECA
Frequent Episode Rule ECA
Base Execute Application
Modeling
C. Process of the Detection of Event Operation

This paper proposed an event detection process which is

Pre-Processing Application
the most important in the whole S/W architecture of an APP
using the data mining and active database techniques. The Rule Extraction Active Rule Analysis / Active Rule Update
models for these rules are assigned to a rule controller and are Fig.2. Process of the Detection of Event
used for the basis of a determination of a signal event input.
The rule execution model is based on events detection. The TABLE IV
steps enumerated in Fig.2 are elaborated as follows: DESIGN CONSIDERATIONS AND THE SOLUTIONS
Considerations Solutions
Rule Modeling -Acquiring of operation data for NPP simulator
- When a signal (clock, application or user) causes a -Analysis classification, association, clustering
primitive event to trigger and the event is detected by a and sequence pattern
rule manager, a message is sent to the rule controller Rule V&V IDEA methodology
Primitive Event Time, external event
component. Composite Event Conjunction, disjunction, sequence, negation,
times,
- The corresponding condition and actions are checked E-C Coupling Independent mode
against the ECA Base by the rule controller component. C-A Coupling Independent mode
Execution Model TMO based Micro-kernel
If there are more than two rules to be fired, the rule Rule Termination Triggering graph
controller uses their priorities to decide which of the Recursive Rule No recursive rule execution
rules should be fired. The corresponding condition and
action are sent to the condition evaluator and action IV. CONCLUSION
executor components respectively. This work is done This paper proposed a new approach to an advanced
inside the rule controller component. alarm processing by using Event-Condition-Action(ECA)
rules that can be automatically triggered by an active database
- If the condition is satisfied, the action is fired by the and a knowledge acquisition method using active data mining
action executor. techniques to obtain the relevant alarm reduction rules. Also
this paper proposed design considerations and solutions for
- After the action execution, the composite event detector the ADS for AAP and the S/W structure for process of the
will be invoked to check if this event can cause any detection of an event. We will implement and test a prototype
composite events to trigger. This is done by the rule for the ADS for the AAP that captures the main elements of
controller component the ADS for the AAP in the future.

D. Design Considerations of Active Database Systems (ADS) REFERENCES

for the AAP [1] C.K.Lee, S.Hur, “Development of a new indicator and alarm
system in NPPs,” American Nuclear Society, 1996.
This section describes the design considerations of the
ADS for the AAP, and it also describes the solution plan for [2] Nureg-0700, Rev.2, “Human-system interface design review
each consideration. (See TABLE IV) guidelines,” U.S. Nuclear Regulatory Commission , 2002.

3
Authorized licensed use limited to: RNS Institute of Technology. Downloaded on February 9, 2009 at 00:39 from IEEE Xplore. Restrictions apply.