312041 Evaluation of Dynamic Ica-Based Sensor Fault Validation Method of Indoor Air Pollutants Data On Energy Consumption of Ventilation System

Tuesday, November 5, 2013: 2:30 PM
Union Square 10 (Hilton)
MinJeong Kim, Dept. of Environmental Engineering, Kyung Hee University, Yongin-si, Gyeonggi-do, South Korea, JeongTai Kim, Dept. of Archi. Engineering, Kyung Hee Unversity, Yongin-si, Gyeonggi-do, South Korea and ChangKyoo Yoo, Department of Applied Environmental Science, Kyung Hee University, Yongin-Si, South Korea

Millions of people spend a considerable amount of time in indoor building spaces such as office, classroom, and subway station. However, due to overcrowding and inadequate ventilation system, various types of indoor air pollutants accumulate in the indoor building, specially underground spaces. Therefore, to ensure people's good health, an accurate monitoring is necessary for regulating indoor air quality (IAQ) in the underground spaces.

Sensors are important components of the IAQ monitoring, since they provide data that are required for continuous monitoring of the IAQ. However, these sensors suffer reliability problems (namely, sensor fault) due to extended usage or hostile environment where the sensors are installed. The sensor fault provides incorrect information while the IAQ monitoring in the underground spaces. Therefore, sensor fault identification and reconstruction (namely, sensor fault validation) are necessary for accurate IAQ monitoring.

Some studies on IAQ ventilation systems have reported that an amount of energy consumption of the ventilation system is influenced by the sensor reliability. Once the IAQ sensor is deteriorated due to the precision degradation, the IAQ level measured from the faulty sensor is lower than the actual indoor air pollutants in the building space. Then, the ventilation system misjudges that higher power is needed to supply more amount of fresh air than the required, and the energy consumption of ventilation system is increased. It indicates that a consequence of sensor fault on the energy consumption of ventilation system is critical. Therefore, a systematic sensor fault validation method is required to satisfy the energy efficiency as well as people's comfort in the indoor building spaces.

(a)

(b)

Fig. 1. (a) Hostile environment of the underground space and deteriorated sensor, and (b) influence of sensor fault on the energy consumption of IAQ ventilation system

This study is carried out 1) to identify and reconstruct the faulty sensor of IAQ measurement and 2) to evaluate the effect of sensor fault validation on the energy consumption of ventilation system at the underground subway station in Seoul Metro, South Korea. The IAQ variables measured from the subway station have two characteristics of non-Gaussian distribution and auto-correlation. In this study, a sensor fault validation method based on dynamic independent component analysis (DICA) is proposed to take the dynamic non-Gaussian behavior of the IAQ variables into account. DICA is an optimal technique which can extract essential independent factors from the dynamic non-Gaussian distributed system.

Fig. 2 shows a conceptual framework of the proposed method to validate IAQ sensor's reliability and its reconstruction as well as to predict the energy consumption of the ventilation system. It consists of three parts: 1) to develop the DICA-based sensor fault validation model using IAQ data under normal sensing condition, 2) to apply the developed model to new IAQ data under faulty sensing condition, and 3) to estimate the energy consumption of ventilation system using the faulty and validated IAQ data.

Fig. 2. Proposed framework of validation of IAQ sensor fault and estimation of energy consumption of ventilation system depending on the sensor reliability

To detect whether sensor fault occurs or not, squared prediction error (SPE) which is typical statistics in DICA for detecting abnormal condition is used. If the sensor fails, the normal correlation inside the DICA model is broken, and then the SPE increases significantly. Thus, the occurrence of sensor faults is detected by comparing the SPE values with its threshold. Then, the fault identification is carried out using DICA-based sensor validity index (namely, DI-SVI). The fault identification finds out the source of sensor faults from the large number of sensors and the time-variant characteristics of fault. Finally, DICA-based reconstruction algorithm is used to reconstruct the faulty sensor to normal.

To evaluate the influence of sensor fault validation on the energy consumption of ventilation system, an IAQ ventilation system model developed by Liu et al. [Energy and Buildings, 2013 (accepted)] is used. Fan speed (revolutions per minute, RPM) of the ventilation system is estimated using the faulty and reconstructed IAQ data, respectively. Then, the energy consumption of ventilation system is calculated based on the relation between the fan speed and its energy consumption:

The first plots in Fig. 3(a) and (b) show the SPE plots of DPCA and DICA models to detect the sensor fault. Note that the faulty sensor signal was introduced to samples 60 to 110 of particulate matter less than 10mm (PM10) sensor. To demonstrate the superiority of the DICA-based method over conventional method, the results obtained using DICA and dynamic principal component analysis (DPCA) based methods were compared. After the faulty signal was introduced, both SPE values increase significantly. Note that the DICA-based SPE detects exactly the occurrence of sensor fault compared to the DPCA. It highlights that the DICA-based method has better detection accuracy when detecting the dynamic non-Gaussian distributed fault.

                      

(a)

                        

(b)

Fig. 3. Detection of sensor fault and reconstruction using (a) DPCA-based SPE and reconstruction algorithm, and (b) DICA-based SPE and reconstruction algorithm

The reconstruction of faulty sensor using the reconstruction algorithm based on DPCA and DICA methods are shown in the second plots of Fig. 3(a) and (b). Table 1 lists errors of the reconstructed PM10 data using both methods. The reconstructed values in both figures are consistent with the normal measurements, where the reconstruction error of the DICA-based method is lower than that of DPCA-based method. This result illustrates that the proposed method can capture the system dynamics of non-Gaussian IAQ data. In addition, it provides a way to effectively reconstruct PM10 faulty value where the reconstructed IAQ can be used in the IAQ ventilation system instead of the faulty data.

Table 1. Errors of the reconstructed PM10 data using DICA- and DPCA-based methods

DICA-based method

DPCA-based method

Reconstruction error

19.54

22.96

Table 2 lists the amount of IAQ ventilation system energy consumption with the faulty and reconstructed IAQ data. Note that the reconstructed PM10 data using DICA-based method show a better reconstruction performance and lower ventilation energy consumption compared to the DPCA-based reconstruction as well as faulty data. Once the faulty sensor is reconstructed, 307 kWh of the ventilation energy is reduced. It highlights that the accurate reconstruction of faulty sensor influences on the energy consumption of IAQ ventilation system directly.

The result of this study showed that the proposed method could improve the IAQ monitoring performance as well as reduce the energy consumption of the IAQ ventilation system in the underground subway station or buildings.

Table 2. Comparison of ventilation system energy consumption with the faulty and reconstructed IAQ data

(kWh)

Faulty PM10 measurement

Reconstructed PM10 data

Ventilation system energy consumption

1430.1

1123.9

ACKNOWLEDGEMENTS: This work was supported by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korean government (MEST) (KRF-2012-001400) and also by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (No. 2008-0061908).


Extended Abstract: File Not Uploaded
See more of this Session: Process Monitoring and Fault Detection I
See more of this Group/Topical: Computing and Systems Technology Division