Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA


Virtual Testbed on Evaluating Automated Fault Detection and Diagnostic (AFDD) Algorithms for Common Faults of a Single Duct VAV System

Liping Wang, Majid Karami
Civil and Architectural Engineering, University of Wyoming, Laramie, WY

Abstract: Fault detection and diagnostics (FDD) are important for detecting and diagnosing faulty operations in HVAC systems. It is common that system performance fail to satisfy design expectations due to improper equipment installation, equipment degradation, sensor failure, or incorrectly configured control systems. However, few FDD technologies have been successfully implemented in actual buildings. One of main reasons is that false alarms and missed detection have been identified as common problems when AFDDs have been applied in nonexperimental facilities. This study proposed and created a virtual testbed to evaluate the developed automated fault detection and diagnostic (AFDD) with simulation data or measurement data before implementing AFDD to the HVAC system in field. The virtual AFDD testbed was developed under Matlab environment. Data from building simulation models and HVAC systems in existing buildings or lab experiments can be used to test AFDD algorithms through the virtual AFDD testbed. Via the virtual testbed, AFDD algorithms can communicate with building simulation models or existing HVAC systems for collecting data or adjusting system operation modes. Based on reports on fault diagnosis generated by AFDD algorithms, the tested results can be evaluated on the accuracy of diagnosis when tested faults are known. We used fault diagnosis indices including false alarms, missed detection and misdiagnoses for evaluating the AFDD algorithms. In this study, we demonstrated a case having a Dymola model for a single duct VAV system and AFDD algorithms running on the virtual tested. We have developed two AFDD algorithms for demonstration and testing: a fuzzy logic algorithm and a Naïve Bayesian Classifier algorithm. This study proposed a convenient and inexpensive method to test and evaluate AFDD algorithms on a virtual testbed for validation before deployment in field.
Pages: 2357 - 2362