A Novel Machine Learning Technique for Fault Detection of Pressure Sensor

Area of Science:

• Engineering
• Instrumentation and Measurement
• Data Science

Background:

• Pressure transmitters are vital for process industry measurements.
• Sensing line integrity is crucial for accurate pressure transmitter output and reliability, especially in nuclear power.
• Sensing line blockage is a common failure mode impacting pressure transmitter performance.

Purpose of the Study:

• To propose a novel method for detecting sensing line blockages in pressure transmitters.
• To enhance the reliability and diagnostic accuracy of pressure transmitters through advanced signal analysis.

Main Methods:

• Developed a Trend Features in Time-Frequency (TFTF) domain method.
• Integrated multi-scale time series decomposition with time-domain and frequency-domain feature extraction.
• Utilized a sliding window algorithm and the XGBoost classifier for fault detection.

Main Results:

• The TFTF method demonstrated superior accuracy in detecting sensing line blockage faults.
• Experimental results confirmed the effectiveness of the proposed algorithm compared to traditional methods.
• The method successfully mitigated periodic component interference in diagnostic analysis.

Conclusions:

• The TFTF method offers a highly accurate and reliable solution for diagnosing pressure transmitter sensing line blockages.
• This approach significantly improves fault detection capabilities for critical industrial applications.
• The integration of signal decomposition and feature extraction provides a robust diagnostic framework.