Classification Techniques for Fault Detection and Diagnosis of an Air-Handling Unit.
Classification Techniques for Fault Detection and
Diagnosis of an Air-Handling Unit.
House, J. M.; Lee, W. Y.; Shin, D. R.
ASHRAE Transactions, Vol. 105, No. 1, 1987-1997, 1999.
heating; venting; air conditioning; classifications
The objective of this study is to demonstrate the
application of several classification techniques to the
problem of detecting and diagnosing faults in data
generated by a variable-air-volume air-handling unit
simulation model and to describe the strengths and
weaknesses of the techniques considered. Artificial
neural network classifiers, nearest neighbor
classifiers, nearest prototype classifiers, a rule-based
classifier, and a Bayes classifier are considered for
both fault detection and diagnostics. Based on the
performance of the classification techniques, the Bayes
classifier appears to be a good choice forfault
detection. It is a straightforward method that requires
limited memory and computational effort, and it
consistently yielded the lowest percentage of incorrect
diagnoses. For fault diagnosis, the rule-based method
is favored for classification problems such as the one
considered here, where the various classes of faulty
operation are well separated and can be distinguished by
a single dominant symptom or feature. Results also
indicate that the success or failure of classification
techniques hinges to a large degree on an ability to
separate different classes of operation in some feature
(temperature, pressure, etc.) space. Hence,
preprocessing of data to extract dominant features is as
important as the selection of the classifier.