Signal processing approaches withing the TFM (Total Focusing Method) family, such as SAFT, and TFM are becoming standard in the nondestructive testing industry, as they generally give better image quality than conventional phased array ultrasound. Since all TFM methods are time-based approaches, they provide excellent results, assuming the geometry and the acoustic properties of the material are well known. It is often the case where the characteristics of the part being tested, specifically the surface geometry, are not well known, or flat (non-parallel to the transmitting aperture).
In this presentation, we propose an adaptive approach of the total focusing method using both FMC (Full Matrix Capture) and PWI (Plane Wave Imaging) data acquisition techniques, known as ATFM and APWI, respectively. The adaptive approach allows an ultrasonic compensation for a complex, or non-flat, specimen shape. ATFM and APWI only require a single data acquisition cycle for a single image and do not need multiple acquisitions to detect the profile, i.e., the same acquisition cycle for the specimen imaging also provides the information for the adaptive compensation.
We show examples from the inspection of non-flat steel parts containing various flaws, inspected with a 128-element probe. Several ATFM and APWI acquisitions schemes are compared in terms of image quality (resolution, contrast) and computation time. We show that with these adaptive approaches, the agreement between the actual specimen profile and the adaptively estimated profile is coherent, and moreover the flaw positioning is correct.

Total-focusing imaging of coarse-grained materials suffers from low imaging signal-to-noise ratio (SNR) and massive data volume, which limits its practical engineering applications. To address these issues, this study proposes a total-focusing imaging method combining noise suppression and compressed sensing (CS) sparse arrays. Taking a 76 mm-thick forged austenitic stainless steel weld in the nuclear power industry as the research object, a dual-crystal matrix array probe with a frequency of 2.25 MHz and an element arrangement of 16×4 was used for full matrix capture (FMC). In terms of signal noise reduction, aiming at the problem that noise reduction methods such as wavelet packet decomposition(WPD) are difficult to effectively separate structural noise, a noise reduction method based on dual-crystal matrix array vector coherence and variational mode decomposition (VMD) was proposed. According to the dual-module structure of the probe’s 2D array, a delay-and-sum calculation rule based on the 3D acoustic beam propagation path was established. The instantaneous phase of the signal was extracted to construct a vector coherence factor (VCF), and the imaging data matrix was reconstructed using VCF. Meanwhile, the grey wolf optimizer (GWO) was employed to optimize the penalty factor coefficient and decomposition level of VMD for FMC signals. The full matrix data was reconstructed based on the spectral energy distribution and cross-correlation coefficient of modal components, and total-focusing imaging was performed on the denoised FMC signals. The results show that compared with the original total-focusing imaging results, the imaging SNR is improved by 3.00–6.37 dB and 4.69–6.85 dB after applying VCF imaging and VMD decomposition reconstruction, respectively. In terms of data volume compression, to solve the problems of long CS reconstruction time and insufficient data sparsity, a joint strategy of array sparsity and CS optimized by machine learning was proposed. The sparse array arrangement was optimized using the genetic algorithm (GA) with the pointing function’s sidelobe level and main lobe width as objectives. Compared with the full array, the image array performance index (API) of the 25% GA-sparse array imaging only increases by 1.26, the imaging time is reduced from 512.93 s to 56.63 s, and the detection efficiency is improved by nearly 9 times. Meanwhile, three typical deep learning network models—1D-CNN, GoogLeNet, and sparse autoencoder—were optimized for signal reconstruction. Taking the 1D-CNN model as an example, key parameters such as the maximum number of iterations, minimum branches, initial learning rate, and optimizer were optimized. Compared with the traditional Basis Pursuit (BP) reconstruction algorithm, the optimized deep learning model reduces the percent root mean square deviation (PRD) by 9%–12%, improves the reconstruction speed by approximately 406 times, and the compressed data of the transmitted sparse array is only 5% of the complete full matrix data.
The proposed method can effectively improve the image quality, imaging efficiency, and engineering application value of total-focusing imaging for coarse-grained materials.

Boiler and thermal power generation equipment play a crucial role in stable electricity supply and grid reliability. With electricity demand expected to rise and pressure to improve efficiency and decarbonize intensifying, upgrading existing assets and operating them reliably and efficiently are essential. Regular inspection and refurbishment work—tailored to equipment damage modes—remain critically important.

Ultrasonic testing (UT), a nondestructive inspection technique, is widely used to assess critical areas such as weld interiors. Advances in digital signal and image processing have expanded UT use from manufacturing inspections to maintenance during periodic outages. Maintenance inspections require efficient, objective testing that accounts for equipment structure and damage characteristics; advanced UT methods such as phased-array UT and FMC/TFM (Full Matrix Capture/Total Focusing Method) are increasingly used. Below are two field-proven examples of cutting-edge ultrasonic inspection technologies applied to boiler equipment.

Visualizing weld geometry with adaptive UT
Radiographic testing (RT) alternatives using ultrasonic methods are gaining adoption. We have developed a novel RT-replacement UT technique that combines phased-array UT with adaptive UT. Adaptive UT employs a soft gel couplant and an array probe to directly introduce ultrasound into uneven surfaces such as weld reinforcements, then uses FMC/TFM image processing to visualize internal features. In this approach, oblique-angle phased-array UT provides high-sensitivity detection of internal weld flaws, while adaptive UT enables quantification of previously hard-to-evaluate weld geometry and backwall conditions (e.g., depressions, sagging). Applied to tens of thousands of welds to date, this method has significantly shortened outage schedules and, by visualizing weld quality, contributed to optimization of welding processes.

Quantifying flaw height with TOFD-TFM
Recently, increased load-cycling operation in thermal power plants has raised concerns about accelerated damage initiation and progression, increasing demand to quantify flaw height and growth rates nondestructively for rational maintenance planning. Conventional ultrasonic testing struggles to ensure accurate flaw-height quantification because of variability in flaw orientation and tilt. In contrast, TOFD (Time of Flight Diffraction) is known for robust flaw-height assessment by measuring diffracted waves generated at flaw tips. We developed a TOFD-TFM ultrasonic imaging system that combines TOFD with multi-channel signal acquisition and TFM image processing. The system acquires full-waveform data from multiple array-element configurations, isolates diffracted waves from flaw tips, and objectively quantifies flaw height on images with high accuracy. Real-time imaging enables tracking of indications in sync with sensor scanning, facilitating discrimination of weak signals; during field inspections the system performs wide-area screening with continuous data capture while simultaneously locating and quantifying maximum flaw height. Recent increases in field deployments demonstrate the technology’s contribution to data-driven, cost-effective maintenance planning tailored to plant operation needs.

Phase Coherence Imaging (PCI), a post-processing technique derived from Full Matrix Capture (FMC) and Total Focusing Method (TFM), uses instantaneous phase information rather than amplitude to significantly enhance the visibility of small defect indications (crack tips, inclusions, porosity). This paper presents a large-scale comparative field study involving more than 150 meters of girth welds in a 48-inch pipeline construction project. All welds were inspected using an approved AUT procedure combining Phased Array Ultrasonic Testing (PAUT) and Time-of-Flight Diffraction (TOFD). In parallel, the same welds were scanned with FMC data sets, from which both conventional TFM and PCI images were reconstructed offline. Detection performance, sizing accuracy, signal-to-noise ratio, and probability of detection (POD) for critical indications were systematically compared between the qualified PAUT/TOFD method and the emerging TFM/PCI approach. Acquisition and post-processing times were also evaluated to assess practical deployability. Results demonstrate that PCI considerably improves the detectability and characterization of small volumetric and planar defects while maintaining or reducing overall inspection time, confirming its potential as a complementary or future replacement technique in high-consequence pipeline welding projects.