Research and Professional Interests | 연구 분야
Click icons
for more information →
Autoencoder based building system fault detection using virtual sensors
Research objective
-
Detect sensor and equipment faults on building systems.
Research contents
-
Machine learned autoencoder has long term memory of normal operational condition.
-
Virtual sensors can be used as a autoencoder input to enhance hidden mathematical relations in auto encoder.
-
When faulty operational dataset is used as a input of trained autoencoder, reconstructed value (output) make difference from input value.
-
Fault detect using difference between autoencoder input and reconstructed value.
Research methodology
-
Autoencoder
-
Neural Network
<Autoencoder based fault detection using virtual sensors>
<Fault detection result (District heating system)>
Virtual sensing environment in building systems
Research objective
-
Development of a highly accurate virtual sensing environment in the building system.
Research contents
-
Autoencoder is a deep learning algorithm consisting of input layer and reconstruction of input layer(output layer).
-
Autoencoder can effectively learn the relationship between variables(both physical and virtual sensor) according to a learning algorithm that minimizes reconstruction errors between the input layer and the output layer.
-
The system-level virtual sensing environment derived from the autoencoder is implemented with high accuracy by its learning algorithm.
Research methodology
-
Autoencoder
-
Neural Network
-
MLP(Multiple Layer Perceptron)
<Autoencoder based virtual sensing framework>
<Autoencoder based pump frequency sensing result>
Environment virtual sensor
Research objective
-
Estimation and prediction of indoor environment using virtual sensor
Research methodology
-
Data driven estimation (Multi-layer perceptron and Autoencoder)
-
Model based Airflow Simulation
Research contents
-
Development of indoor environment virtual sensor through the use of external environment information and architectural / facility system information.
-
Estimation and prediction of indoor air quality and infiltration rate (pressure distribution) considering the mechanism of airflow in buildings
-
Real-time monitoring of building air quality and interlocking with indoor ventilation control
<Results of CO concentration measurement and prediction of underground parking lot>
<Results of building overall pressure distribution by virtual sensor>
Building airflow
Research objective
-
Airflow characteristics in high-rise residential buildings
-
Effective estimation method of leakage area in high-rise residential buildings
Research contents
-
Airflow analysis of the entire building considering the interaction between wind and stack effect.
-
Estimation of the building's leakage area through the theoretical formula (Thermal Draft Coefficient; TDC) and calibration through optimization techniques.
-
Calibration of leakage area through virtual sensing pressure distribution in a whole building.
Research methodology
-
Model based Airflow Simulation
-
Deterministic calibration using GA-based optimization
-
Statistical calibration using bayesian MCMC
<Building airflow analysis by calibration methods>
<Seasonal infiltration rate by floor/houseshold>
<Distribution of estimated leakage area of elevator door>
Thermal environment in building systems
Research objective
-
Diagnosis and evaluation of Thermal Resistance and Thermal transmittance of exterior walls.
Research contents
-
Convergence characteristics of thermal resistance and thermal transmittance using ISO 9869-1.
-
Thermal environment modeling and dynamic analysis.
-
Diagnosis of large-area thermal resistance and thermal transmittance using ISO 9869-2.
Research methodology
-
R studio
-
LORD(Logical R-Determination)
<In-situ measurement equipment>
<Thermal Discretization and Modeling>
<Prediction of heat flux and thermal performance analysis>
Building energy evaluation and optimization
Research objective
-
Predict filterable and condensable particulate matter from building energy consumption.
Research contents
-
The pattern of gas energy consumption in apartment buildings analyzed using dynamic time warping(DTW) hierarchical clustering method.
-
The building energy prediction model is developed using artificial neural network technology.
-
Filterable PM and condensable PM generated by the burning of LNG in buildings are calculated for each building use.
-
Based on the predicted particulate matter information, the dust dispersion simulation program is developed using the CALPUFF model.
Research methodology
-
Neural Network
-
Clustering
<Particulate matter dispersion simulation program process>
<Monthly gas energy consumption of apartment buildings>
<Gas consumption and particulate emmision per unit area according to building use>