U-Net CNN for ASSESS Defect Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Nondestructive evaluation (NDE) techniques, such as Acoustic Steady-State Excitation Spatial Spectroscopy (ASSESS), face challenges with increased processing time for larger structures, limited accuracy and spatial resolution for smaller defects, and inaccuracies near the edges of examined structures.
Innovation Solution
Employing a deep learning system, like a U-Net style convolutional neural network (CNN), to perform semantic segmentation on simulated ultrasonic wavefield images, enabling improved processing speed and spatial resolution, and using transfer learning on an augmented wavefield dataset to localize and characterize defects in various materials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatial Fourier transform is applied for defect analysis, then processing can be performed, but inaccuracies occur in the vicinity of edges of structures
Solution Approach 1:
The patent extracts and removes edge regions from the analysis domain by applying windowing functions that suppress edge effects in the spatial Fourier transform. This isolates the problematic edge regions and prevents them from contaminating the defect detection results in the interior regions of the structure.
Solution Approach 2:
The patent introduces windowing functions as intermediary elements between the spatial domain data and the frequency domain transformation. These windowing functions act as mediators that smooth transitions at edges and reduce the harmful edge effects that would otherwise propagate through the Fourier transform process.
2Productivity
If ASSESS technique is used for full-field inspection, then inspection speed is improved, but spatial resolution is limited for smaller defects
Solution Approach 1:
The patent applies local quality enhancement by using spatial filtering and wavefield processing techniques that adapt to local regions of the structure. Different processing parameters and filter characteristics are applied to different spatial locations, allowing high spatial resolution for small defects while maintaining overall inspection speed across the full field.
Solution Approach 2:
The patent transitions from analyzing only the spatial domain to incorporating the frequency domain through spatial Fourier transform. This adds a frequency dimension to the analysis, enabling resolution of finer spatial details that are not visible in the original spatial domain data, thereby improving spatial resolution without sacrificing inspection speed.
3Area of stationary object
If ASSESS technique is applied to larger structures, then full-field coverage is achieved, but processing time increases
Solution Approach 1:
The patent divides the large structure into smaller sub-regions or segments that can be processed independently and in parallel. The spatial Fourier transform and defect analysis are performed on these smaller segments, significantly reducing the computational burden and processing time while still achieving full-field coverage by combining the results from all segments.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The deep learning system enhances defect detection accuracy and reduces processing time by providing sharper and more reliable defect boundary identification, overcoming edge-related inaccuracies and improving detection of smaller defects.
Implementation Method 1
select-tones of ultrasonic excitation are applied to the surface of a structure by a piezoelectric transducer (PZT)
Implementation Method 2
a scanning laser Doppler vibrometer (LDV) obtains the steady-state, surface response of the structure to the excitation
Data Source
AI summary
A method and system employing deep learning techniques improves processing speed and spatial resolution of acoustic wavenumber spectroscopy (ASSESS) techniques by performing semantic segmentation on simulated ultrasonic wavefield images of a steady-state, select-tone excitation of a structural or mechanical component. One or more embodiments may employ a convolutional neural network (CNN), pre-trained on openly-available datasets, and trained by transfer learning on an augmented wavefield dataset, to localize and characterize defects or damage from inspection measurements of components.