Sensor fault diagnosis and validation of building air conditioning systems

Abstract

This thesis develops the methodology for the automated detection, diagnosis and evaluation (FDD&E) of soft sensor faults in building Heating, Ventilation, and Air Conditioning (HVAC) systems. Literature survey shows that soft sensor fault, particularly, existing and/or slowly developing bias or drift, has not been properly addressed. Existing methods of sensor FDD in the field of HVAC and other disciplines are reviewed. Techniques based on the commonly used model-based approach is found inappropriate to deal with slowly developing soft sensor faults, due to its incapability to distinguish the sensor faults from system and/or component performance degradations and the difficulty to obtain reliable models. The sensor FDD&E strategy described in this thesis is based on fundamental physical laws - the law of energy conservation and the law of mass conservation. Residual of the conservation constraint (balance) in steady state is used as an index to detect existence of sensor biases. Diagnosis and evaluation are achieved by estimating the values of the biases such that the sum(s) of the squares of corrected balance residuals is minimised. Since the validity of the balance constraints are not in doubt and independent of system/component conditions, such a law-based sensor FDD&E strategy can accommodate system/component performances degradations and does not require system/component models. Application of the law-based strategy to the FDD&E of fixed biases on the temperature sensors and flow meters in primary-secondary chilling plant is described in this thesis. Basic (sequential) schemes and algorithms are developed for the sensors in chilled water network. A scheme based on a Genetic Algorithm is developed to improve the robustness of the basic schemes. A correlation cancellation method is developed for the cooling water sensors associated with chillers. The method uses a characteristic quantity derived in this thesis to detect and diagnose cooling water flow meter bias (or to estimate constant cooling water flow rate) without knowing whether the associated temperature sensors and power meter are biased or not. A prototype of computer software implementing the developed sensor FDD&E schemes is described. The window-based software is developed aiming at providing a convenient tool for commissioning or re-commissioning the sensors in the chilling plant. A study of the FDD&E scheme and algorithms for a time-varying sensor fault is conducted. The objective is to investigate and extend the capability of the law-based strategy in tracing time-varying sensor biases. A method of obtaining the time-varying sensor bias estimators by simply converting the counterparts for fixed sensor biases is developed. The FDD&E strategy and schemes are validated in tests using dynamic simulation data and in a case study in an existing building HVAC system. The simulation tests showed that both fixed and time-varying biases of the flow meters and differential temperature sensors can be correctly identified and estimated. Absolute values of the individual temperature sensors can be estimated, given that a common reference temperature sensor is reliable or its bias value is known. In the field case study, three biases of 1.5, -1, 1.5 degree Celsius were introduced respectively into three temperature sensors in the chilling plant. Based on the data recorded in the BMS central station, the introduced faults were successfully identified and estimated. The estimates of the individual chiller chilled water flow meter biases and the constant cooling water flow rates were compared with commissioning records.