Dynamic Calibration Method of Sensor Drift Fault in HVAC System Based on Bayesian Inference

Area of Science:

• Building Systems Engineering
• Control Systems
• Data Science

Background:

• Sensor drift faults degrade HVAC system performance and require effective calibration.
• Bayesian inference (BI) is a popular method but struggles with time-varying drift faults.
• Existing methods primarily address sensor bias, not dynamic drift.

Purpose of the Study:

• To develop a dynamic calibration method for HVAC sensor drift faults using Bayesian inference.
• To evaluate the performance of this method using a chilled water supply temperature sensor example.
• To improve the accuracy of drift fault calibration in HVAC systems.

Main Methods:

• A novel dynamic calibration method based on Bayesian inference (BI) was developed.
• The Exponentially Weighted Moving-Average (EWMA) method was integrated for enhanced detection.
• The method was tested on a chilled water supply temperature sensor in an HVAC chiller.

Main Results:

• The combined EWMA and BI dynamic calibration method effectively calibrated sensor drift faults.
• The Mean Absolute Percentage Error (MAPE) between calibrated and normal data was reduced to below 5%.
• The proposed method demonstrated high accuracy in addressing time-varying drift issues.

Conclusions:

• The dynamic BI calibration method, enhanced by EWMA, significantly improves drift fault calibration accuracy in HVAC systems.
• This approach offers a robust solution for maintaining HVAC operational integrity.
• The findings suggest a practical and effective strategy for sensor fault management in buildings.