Intelligent Event Notification based on Active Database and Fuzzy
Triggers
by Antoni Wolski
A new breed of active databases incorporates soft reasoning about
the data state. The technology may be utilized to help the operators
of large industrial installations to detect complex and, potentially,
hazardous conditions in a process. The RapidBase-S intelligent
database system incorporating history database, temporal triggers
and fuzzy rules was developed at VTT Information Technology (Espoo,
Finland).
In industrial process management systems like those used, for
example, in control rooms of power stations and chemical plants,
vast amounts of measurement data are being collected and analysed.
The sheer magnitude of the data flux - reaching thousands of measurements
per second - is a challenge for a real-time process management
system. If, additionally, condition testing or reasoning has to
be performed on the data, most existing techniques do not perform
adequately in real-life situations. In traditional rule-based
system, large data sets, especially when they are formed as time
series cause performance problems. Utilizing traditional databases
is to no avail because they are too slow for the data acquisition,
do not accommodate time series, and require the management applications
to poll the database continuously, leading to performance problems
again.
The solution we are proposing in our SENSE project is an intelligent
active database system built to handle efficiently time series
data and to reason therefrom. Instead of having rules outside
of the database, we are putting them into the database in the
form of crisp and fuzzy triggers. Because rules are precompiled
at the moment of creation, the rule system can run the rules and
access the process data with unprecedented speed, without the
need to cross the external database interface boundary. For example,
in our current implementation we are able run even a thousand
database-initiated fuzzy rule set executions per second, in a
typical Intel-based PC equipment.
We have developed a system called RapidBase-S which contains all
the necessary components of an intelligent active database. Its
central part is a main-memory-based database engine supporting
an extended relational-time-series data model. The model is based
on the traditional relational data model, with the addition of
a new column type: a history column. A value of a history column
is a time series of history records, as illustrated in the figure.
Thus any attribute or a set of attributes of a real-life object
may be defined as having only a current (snapshot) value or a
temporal (time series) value.
A model based on the relational model has the advantage that the
data can be naturally manipulated through the current view which
corresponds to a traditional relational table showing only the
latest data. Standard SQL statements may be used to do that. For
example, in order to add a new record to a time series, only a
standard UPDATE statement is required. It changes the state of
the current view according to the SQL semantics but, in the same
time, it has an effect of appending a record to a history.
The time travel through a history is possible using extended syntax.
The trigger engine is responsible for the active behaviour of
the system. Various triggers may be defined at run-time, including
temporal ones. For example, it is possible to define the rule:
”when the value of the data point A has been above the alarm limit
L for more then 10 minutes, then do ... “
More precisely, trigger definitions are expressed in the form
of CREATE TRIGGER statements of an SQL-based definition language.
In a true spirit of contemporary databases, triggers may be defined
and modified at run-time.
The basic form of a trigger includes a regular Boolean crisp predicate
evaluated over the whole database. We have also introduced several
ways to introduce reasoning to trigger condition evaluation. Fuzzy
rule sets are used as a basic unit of imprecise reasoning. For
example, using fuzzy rules, fuzzy set quantifiers and fuzzy temporal
restrictors, it is possible to express a rule like this: ”when
most pump motors have been slightly overheated for few minutes,
then do ... “
Several pilot trials of RapidBase-S are under preparation. They
involve industries like cellulose pulp production industrial drive
systems and telecommunications systems.
More information about this and other projects on industrial database
technology is at: http://www.vtt.fi/tte /projects/industrialdb/
Please contact:
Antoni Wolski - VTT Information Technology
Tel: +358 9 456 6012
E-mail: Antoni.Wolski@vtt.fi
