Ferroelectret-based Hydrophone Employed in Oil Identification-A Machine Learning Approach.

Area of Science:

• Materials Science
• Electrical Engineering
• Acoustics

Background:

• Mineral oil quality is critical for electrical transformer performance and longevity.
• Traditional oil analysis methods can be time-consuming, costly, or susceptible to interference.
• Acoustic analysis presents a promising alternative for non-invasive and robust diagnostics.

Purpose of the Study:

• To develop and validate a novel acoustic-based method for assessing mineral oil quality in electrical transformers.
• To integrate a ferroelectric-based hydrophone and acoustic transducer for signal acquisition.
• To employ supervised machine learning for accurate oil classification.

Main Methods:

• Utilized a custom-built ferroelectric hydrophone and acoustic transducer to capture acoustic signals from four types of mineral oils.
• Collected three datasets with 180, 240, and 420 entries, respectively.
• Extracted 84 features from each dataset and applied two supervised machine learning classification approaches.

Main Results:

• The first classification approach achieved a classification error of less than 2% for distinguishing between four oil types.
• The second classification approach successfully classified the oils with 100% accuracy.
• The developed method demonstrated robustness and immunity to electromagnetic noise.

Conclusions:

• Acoustic analysis, particularly with ferroelectric hydrophones, is a highly effective and reliable technique for mineral oil quality diagnosis.
• The proposed machine learning-based approach offers a precise and efficient solution for ensuring the integrity of electrical transformers.
• This method provides a potentially low-cost and noise-immune alternative for transformer oil monitoring.