Master's Thesis Yuchen Jia
Automized commissioning and functioning tests for building automation systemsCopyright: EBC
Commissioning of building service systems with associated automation systems is important for functionality verification and performance evaluation. Especially for building HVAC system, a rudimentary commissioning and functioning test will lead to suboptimal behavior and even faulty equipment. Hence, an automated commissioning methodology is developed. The objective is performing developed procedures upon two representative subsystems and hence validating the methodology regarding feasibility. To automate the main step in the commissioning cycle i.e. functional performance test, a data driven method is adopted. Compared to first principles-based approach, this method requires a minimum of engineering-related information. By means of top-down strategy, a single HVAC subsystem is chosen and a corresponding black-box-like Multi-Input Single-Output (MISO) model is built. As the output relates to a specific set point variable, inputs consists of several process variables that might be changeable due to set point variation. Moreover, a multi-layer perceptron is introduced to estimate mathematical cause-effect correlations between involved variables. With the Python-based package Keras, the backpropagation neural network (BPNN) algorithm is used for data training. The concrete commissioning procedures are comprised of four main steps: preliminary data management, signal mapping test, training data acquisition and system identification. Instead of the real BAS, a simulation model serves as testing environments. As a result, the commissioning concept is implemented in a simplified simulation model of a test hall. For each tested technical system, three testing points are chosen to be validated in signal mapping test. Furthermore, 80 data examples are extracted from simulation to be trained in neural network regressor. Consequently, signal mapping tests indicate the fluctuation of controlled temperatures in heating circuits, and the obtained neural network-based cause-effect relationships reveal a high goodness of fit over 0.99. Due to proven feasibility, the commissioning method will be further optimized and implemented in real devices in the future.