wilamowski-b-m-irwin-j-d-industrial-communication-systems-2011

68-4 Industrial Communication Systems
that are hierarchical as well, but nevertheless strongly interconnected with each other (Figure 68.1).
Observing the behavior within one experiment cannot lead to a complete description of the system. This
is the wrong approach, even if it looks stunningly simple. To give an analogy: biologists do not explain
flowers by visual features any more, but they look for functional units by considering the flower as a
“process.” Similarly [Dav97], trying to describe merely the behavior of a communication protocol under
different conditions is a lost cause, since it will never cover the complete abilities of a communication
protocol. But it is possible to develop a functional model of a process [HJ97] (where we use the word
“process” to refer to both technical and biological occurrences) and understand its behavior based on
this model. If we want to build robots or other systems that are intelligent in a more human-like way,
we have to get away from behavioral description and employ functional development instead. While
this is well understood in computer and communications engineering, some psychological schools like
behaviorists have a slightly different view of the matter.
68.4 automated Methods for Sensor and Actuator Systems
Today’s building sensor and control systems are primarily based upon the processing of sensor information
using predefined rules. The user or operator defines, e.g., the range of valid temperatures for a room
by a rule—when the temperature value in that room is out of range (e.g. caused by a defect), the system
reacts (e.g., with an error message). More complicated diagnostics require an experienced operator who
can observe and interpret real-time sensor values. However, as systems become larger, are deployed in a
wider variety of environments, and are targeted at technically less-sophisticated users, both possibilities
(rule-based systems and expert users) become problematic. The control system would require comprehensive
prior knowledge of possible operating conditions, ranges of values and error conditions. This knowledge
may not be readily available, and will be difficult for an unsophisticated user to input. It is impractical
for experienced operators to directly observe large systems, and naive users can not interpret sensor values.
A solution to this problem is to automatically recognize error conditions specific to a given sensor
[HLS99], actuator, or system without the need of preprogrammed error conditions, user-entered parameters,
or experienced operators. The system should observe sensor and actuator data over time [JGJS99],
construct a model of “normality” [PL03], and issue error alerts when sensor or actuator values vary from
normal. The result would be a system that can recognize sensor errors or abnormal sensor or actuator
readings, with minimal manual configuration of the system [SH03]. Further, if sensor readings vary or
drift over time, the system could automatically adapt itself to the new “normal” conditions, adjusting its
error criteria accordingly.
68.5 the Diagnostic System
For illustration, a diagnostic system utilizing statistical methods (BASE [SBR05]) is compared to a standard
building automation system (BAS). The BAS consists of a number of sensors and actuators connected
by the LonWorks fieldbus (LON) [LDS01]. It offers a visual interface using a management information
base (MIB) for retrieving and manipulating system parameters and for the visualization of system malfunctions.
The diagnostic system BASE is based on statistical “generative” models (SGMs). The goal of the
system is to automatically detect sensor errors in a running automation system [SBR05,FS94]. It does so
by observing the data flowing through the system and thus learns about the behavior of the automation
system. The diagnostic system builds a model of the sensor data in the underlying automation system,
based on the data flow. From the optimized model, the diagnostic system can identify abnormal sensor
and actuator values. The diagnostic system can either analyze historical data, or directly access live data.
We use a set of statistical generative models to represent knowledge about the automation system.
A statistical generative model takes inputs like a sensor value, a status indicator, time of day, etc., and
returns a probability between zero and one.
© 2011 by Taylor and Francis Group, LLC