Sensor Fault Detection and Diagnosis Method for AHU Using 1-D CNN and Clustering Analysis

Area of Science:

• Building Systems Engineering
• Artificial Intelligence in HVAC
• Sensor Fault Detection

Background:

• Accurate sensor data is critical for effective control of Air Handling Units (AHUs).
• Sensor faults in the supply air temperature (Tsup) control loop can lead to system inefficiency and poor indoor air quality.
• Existing fault detection and diagnosis (FDD) methods may lack accuracy or robustness in real-world conditions.

Purpose of the Study:

• To develop and validate an advanced FDD method for Tsup sensors in AHUs.
• To leverage deep learning and clustering techniques for improved fault identification.
• To enhance the reliability and performance of AHU control systems.

Main Methods:

• Utilized a one-dimensional convolutional neural network (1-D CNN) for feature extraction from raw sensor data.
• Applied WaveCluster clustering analysis to categorize extracted features and identify anomalies.
• Implemented a T c acquittal procedure to refine fault diagnosis accuracy.

Main Results:

• The proposed FDD method successfully detected and diagnosed abrupt sensor faults in the Tsup control loop.
• The method demonstrated high efficiency, with low false alarm and missing diagnosis ratios.
• The approach exhibited good robustness against noise within a 6 dBm to 13 dBm range.

Conclusions:

• The combined 1-D CNN and WaveCluster approach provides an effective solution for Tsup sensor fault detection and diagnosis in AHUs.
• The method offers a reliable and robust means to maintain optimal AHU performance.
• This FDD technique contributes to improved building energy efficiency and occupant comfort.