Multi-variable classification model for valve internal leakage based on acoustic emission time-frequency domain...

Area of Science:

• Engineering
• Acoustics
• Signal Processing

Background:

• Accurate detection of valve leaks is crucial for industrial safety and efficiency.
• Acoustic emission (AE) technology offers a non-intrusive method for monitoring valve integrity.
• Mathematical modeling of AE signals is a prerequisite for developing reliable AE-based detection systems.

Purpose of the Study:

• To develop a mathematical model for valve-internal-leakage acoustic-emission signals (VILAES).
• To establish a multi-variable classification model for relating VILAES characteristics to leakage rates under varying pressures.
• To assess the performance of the developed model against existing methods.

Main Methods:

• Preprocessing VILAES using a Butterworth bandpass filter (140 kHz-180 kHz) for liquid media.
• Calculating five VILAES characteristics: standard deviation, root mean square, wavelet packet entropy, peak standard-deviation probability density, and spectrum area.
• Employing a random forest classification model with six input parameters (pressure and five VILAES characteristics) to classify leak sizes (small, medium, large).

Main Results:

• VILAES characteristics were found to increase with increasing leakage rate.
• The random forest model demonstrated higher accuracy and shorter operating time compared to support-vector-machine methods.
• The combination of time-frequency characteristics and random forest provided an effective approach for VILAES modeling.

Conclusions:

• The developed mathematical model accurately characterizes VILAES.
• The random forest-based classification model is efficient and accurate for detecting and classifying valve leaks.
• This approach enhances the capability of acoustic emission technology for valve leak detection.