Fault Detection of Temperature Sensor of a Heat Pump System

In this paper the influences on the performances of a heat pump are investigated when a temperature sensor failure occurs and possibilities to localise this type of failures are discussed. The heat pump under investigation is reversible air source type with vapour injection. Sensor failures are considered hidden ones because developed system controls can shade them and apparently heat pump is able to reach desired output. However, as it will be shown, such a malfunction has an effect on system performance and reliability. Two statistical methods are employed for temperature sensor failure detection of the considered heat pump: Principle Component Analysis (PCA) and Fuzzy Principle Component Analysis (FPCA). PCA is a widely used statistical technique for reducing data dimension based on linear transformation of original measured data and describes the significant variation in that data. PCA is vulnerable to the non-linearity in the systems and is very sensitive to noise in the original measured data, as any statistical method does. Therefore we propose the implementation of a non linear robust FPCA method using Gustafson-Kessel technique to divide the data into fuzzy clusters in order to diminish the noise in the original measurements and overbear the non-linearity in a heat pump system. We use these two methods to detect a failure in a temperature sensor signal and performance of the two methods will be presented and compared. Heat pump system experimental setup and test conditions will be described.