Coating Thickness Estimation Using a CNN-Enhanced Ultrasound Echo-Based Deconvolution.

Area of Science:

• Materials Science
• Acoustics
• Non-Destructive Testing

Background:

• Coating degradation monitoring is vital for offshore structural integrity and corrosion prevention.
• Accurate coating thickness estimation is crucial for assessing protective layer performance.
• Traditional ultrasound pulse-echo methods struggle with overlapping echoes from closely spaced interfaces, hindering precise thickness measurement.

Purpose of the Study:

• To develop an advanced ultrasound-based methodology for accurate coating thickness estimation.
• To overcome limitations of conventional methods in resolving closely spaced acoustic interfaces.
• To enable reliable non-destructive monitoring of coating integrity in demanding environments.

Main Methods:

• A novel deconvolution model was developed using two consecutive backwall echoes to isolate propagation path information.
• A 1D convolutional neural network (1D-CNN) was employed for enhanced detection of coating thickness within the reflectivity function.
• Synthetic signals were generated using finite-difference time-domain (FDTD) simulations (k-Wave MATLAB toolbox) for training the 1D-CNN model.

Main Results:

• The methodology successfully estimated front-side coating thickness (60μm–740μm) on steel substrates under varied conditions.
• The minimum detectable thickness was approximately λ/5 for an 8 MHz ultrasonic transducer.
• The 1D-CNN model achieved an accuracy of approximately 8μm on synthetic data.

Conclusions:

• The proposed ultrasound echo-based approach offers a robust solution for coating thickness estimation.
• The method demonstrates significant potential for reliable monitoring of coating thickness changes in real-world applications.
• This technique enhances non-destructive testing capabilities for offshore industries and beyond.