Performance characteristics of a R410A residential unitary split heat pump equipped with a thermostatic expansion valve (TXV) were investigated in the cooling mode under no-fault and faulty conditions. An automated method of steady-state detection was developed to produce consistent collection of data for all tests. The no-fault test measurements were used to develop a multivariate polynomial reference model for those system features (temperatures) that varied the most when a single fault was imposed. The rule-based chart method of fault detection and diagnosis presented in this work requires knowledge of the variation of system features at steady-state and during transient operation. Knowledge of the transient variation of the various features is necessary to establish the size of the moving window used by the steady-state detector, which is a key part of our FDD method. Once the steady-state detector indicates that the important FDD features are steady, the difference in the moving window mean and the no-fault reference model values, feature residuals, are calculated using the no-fault reference model. A feature residual may have one of three values; positive or negative or neutral. The calculation of the neutral threshold value, ε, involves selecting a confidence level (for example 99 %) to avoid a false alarm. For a given confidence level, the calculation of the appropriate confidence interval involves determining the appropriate variances (uncertainty) associated with steady-state measurement variations, modeling, and lack of measurement repeatability. The techniques discussed are applied to a residential heat pump in the cooling mode with our results discussed herein.
Citation: Special Publication (NIST SP) - 1087Report Number:
NIST Pub Series: Special Publication (NIST SP)
Pub Type: NIST Pubs
fault detection and diagnosis, heat pump, polynomial reference model, probability distribution function, rule based chart