Feature Optimization Method of Material Identification for Loose Particles Inside Sealed Relays.

Area of Science:

• Aerospace Engineering
• Materials Science
• Machine Learning

Background:

• Existing methods for identifying loose particles in sealed relays primarily focus on classification algorithms, often overlooking crucial material dataset features.
• Loose particles in sealed relays pose a risk to system reliability, necessitating accurate identification methods.

Purpose of the Study:

• To propose and validate a novel feature optimization method for enhancing the material identification of loose particles within sealed relays.
• To address challenges in material datasets, including missing values and uneven data distribution, to improve identification accuracy.

Main Methods:

• Data preprocessing techniques were employed to handle missing values, with direct-discarding identified as optimal via Random-Forest classifier (RF classifier) accuracy comparison.
• Min-max standardization was selected as the optimal method for addressing uneven data distribution, validated by identification accuracy.
• An innovative multi-index-fusion feature selection method was developed and tested for its effectiveness.

Main Results:

• The proposed feature optimization method improved the RF classifier's identification accuracy from 59.63% to 63.60% on the primary dataset.
• Testing on ten material verification datasets showed a significant average accuracy improvement of 3.01% for the RF classifier.
• The achieved accuracy represents a record high for loose particle material identification in aerospace engineering.

Conclusions:

• The developed feature optimization method substantially enhances loose particle material identification accuracy in sealed relays.
• This research contributes significantly to loose particle detection and material identification, with direct implications for aerospace system reliability.
• The feature optimization techniques are theoretically applicable to broader machine learning applications beyond aerospace engineering.