Early AI Diagnostics at Westinghouse
Cond Na
SENSOR
OTHER
SENSORS
SENSOR
DIAGNOSTICS
Cond Na High
Cond Na SENSOR
MALFUNCTION
OTHER
VALIDATED
INTERPRETATIONS
VALIDATED
Cond Na High
PLANT
EQUIPMENT
MALFUNCTIONS
Figure 4: Schematic of diagnosis using the Condensate sodium sensor, showing sensor diagnosis and the method of using a
Paralt rule to reduce confidence in the validated Condensate sodium high. Dotted shapes and lines are used to represent many
hypotheses and rules.
caused a rule to be fired (executed) before
all of its input nodes were updated, and then
the rule would be fired multiple times. It was
later decided to organize the firing of rules
so that all the input to the rule had to have
been updated before the rule was fired.
As the Westinghouse engineers worked with
the system, it became clear that the
inference engine would be too slow to be
practical. It was decided to write an
inference engine in C to speed the process.
The original inference engine placed a rule
on the stack whenever one of its input nodes
was updated. Since the diagnostic systems
had several layers of hypotheses, this often
IIC Journal of Innovation
- 12 -