Wireless Patch Antenna Characterization for Live Health Monitoring Using Machine Learning

Area of Science:

• Electrical Engineering
• Materials Science
• Machine Learning

Background:

• Extreme environments like coal-fired power plants require robust temperature monitoring for boiler system health.
• Conventional methods (RFID, infrared, thermocouples) have limitations in these harsh conditions.
• Wireless, non-contact temperature sensing offers a promising alternative.

Purpose of the Study:

• To design and test wireless patch antennas for machine learning-aided temperature estimation.
• To develop an improved sensor calibration method addressing hysteresis.
• To demonstrate the advantages of this novel approach over existing technologies.

Main Methods:

• Wireless patch antennas were designed and tested for temperature sensing.
• A Vector Network Analyzer (VNA) was used for wireless interrogation and data acquisition.
• A Long Short-Term Memory (LSTM) model was employed for sensor calibration and temperature estimation, accounting for hysteresis.

Main Results:

• Sensors exhibited sensitivities ranging from 0.052 to 0.20 MHz/°C.
• The LSTM model achieved up to a 76% improvement in temperature estimation error compared to Linear Regression (LR).
• The wireless patch antenna system demonstrated effective non-contact temperature measurement.

Conclusions:

• Wireless patch antennas offer an innovative solution for temperature monitoring in extreme environments.
• Machine learning, specifically LSTM models, significantly enhances the accuracy of temperature estimation.
• This approach overcomes limitations of conventional temperature sensing methods.