Small Scale On-line Diagnostics for an Hvac

This article describes a simple probabilistic diagnostic system for a fan-powered mixing box. Several types of electrical and damper failures are diagnosed. Empirical data is available for modeling the normal operation of the mixing box, but no empirical data is available for failure states and models of failure states must be constructed from expert knowledge. The distribution over operational states (normal or otherwise) is derived from generative models via Bayes' rule. A general misclassiication cost matrix is speciied, and the risk of outputting various diagnostic messages is calculated ; the message with least risk is output. The results of laboratory tests of this diagnostic scheme on three mixing boxes are described.