Contact Failure Identification in Multilayered Media via Artificial Neural Networks and Autoencoders.

Area of Science:

• Materials Science
• Artificial Intelligence
• Heat Transfer

Background:

• Defect detection in material interfaces is critical for manufacturing and failure diagnosis.
• Inverse heat conduction problems are often used to model these defect estimations.
• Thermography offers a non-destructive method for surface temperature measurements.

Purpose of the Study:

• To develop an artificial neural network model for estimating defect positioning at material interfaces.
• To utilize inverse heat conduction principles and thermography data for defect analysis.
• To assess the accuracy and computational efficiency of the proposed neural network approach.

Main Methods:

• An artificial neural network was employed, modeled as an inverse heat conduction problem.
• An autoencoder was used for dimensionality reduction of transient 1D thermography data.
• A fully connected multilayer perceptron processed the reduced data to predict thermal-physical properties and defect locations.

Main Results:

• The artificial neural network demonstrated good accuracy in estimating defect positioning.
• The approach exhibited a low computational cost, making it efficient for practical applications.
• The network showed generalization capabilities across varying noise levels in training data.

Conclusions:

• The proposed artificial neural network approach is a promising method for defect detection in material interfaces.
• This technique offers an accurate and computationally inexpensive solution for quality control and failure diagnosis.
• The model's ability to handle noisy data enhances its robustness for real-world experimental applications.