cs.LGOct 21, 2026

The Path to NDE 4.0: A Review of Algorithms, Techniques, and Strategies for Autonomous Inspection

Authors: Ashiqur Rahmana, Venkata Devesh Reddy Seethi, Zachary Kral, Austin Yunker, Rajkumar Kettimuthu, Hamed Alhoori

Organizations: Northern Illinois University, DeKalb, 60115, IL, USA · Spirit AeroSystems (Boeing), Wichita, 67210, KS, USA · Argonne National Laboratory, Lemont, 60439, IL, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Non-Destructive Inspection (NDI) is a cornerstone of modern quality assurance, critical for ensuring the safety and reliability of infrastructure and manufactured components without causing damage. This paper presents a comprehensive review of the current NDI landscape, analyzing the field through four distinct lenses, namely, computational algorithms, physical techniques, industrial applications, and research objectives. We synthesize a broad range of academic literature to illustrate the rapid evolution from manual, analog interpretation to a digitized, data-centric paradigm known as NDE 4.0. Our analysis highlights the dominance of ultrasound and visual inspection as primary physical techniques, while revealing a significant shift toward deep learning and generative models for automated defect recognition. Despite these technological strides, the industry faces critical bottlenecks, notably the scarcity of standardized open-access datasets and the "black box" nature of AI models, which hinders widespread adoption. By charting these developments and identifying key research gaps, such as the need for explainable AI (XAI) and standardized digital twin protocols, this review offers a strategic roadmap for future research. It concludes that the future of NDI research should focus on the successful integration of intelligent, autonomous fusion of inspection data with advanced sensor technologies into asset lifecycle management.

Figures & tables

Figure 1: Progress of NDI over time.
Research AreaCategoryNo. of Papers
AlgorithmsComputer Vision and Image Processing16
Machine Learning8
Deep Learning10
Anomaly Detection4
 Generative Network4
NDI TechniquesUltrasound22
Table 1: Overview of the reviewed papers.
Figure 2: Qualitative comparison of NDI algorithms.
PaperCVIPMLDLADGM
Sezgin and Sankur [63]     
Huang et al. [34]     
Brosnan and Sun [12]     
Dolinko [18]     
Memmolo et al. [48]     
Yousefi et al. [79]     
Table 2: Domains of the research papers discussed in Section 3 . Domain labels are CVIP = Computer Vision and Image Processing, ML = Machine Learning, DL = Deep Learning, AD = Anomaly Detection, and GM = Generative Models.
Figure 3: Effectiveness of different NDI techniques.
PaperURXITLFVIEMF
Cawley [13]     
Cawley [14]     
Lange [40]     
Ould Naffa et al. [56]     
Han et al. [29]     
Huang et al. [34]     
Table 3: Domains of the research papers discussed in Section 4 . Domain labels are U = Ultrasound, RX = Radiography and X-Ray, ITLF = Infrared, Thermal, and Low Frequency, VI = Visual Inspection, and EMF = Electromagnetic Field.
PaperCMOMFSCnMFAMSHM
Nemet et al. [53]      
Ould Naffa et al. [56]      
Brosnan and Sun [12]      
Diamanti et al. [17]      
Farrar and Worden [23]      
Achenbach [3]      
Table 4: Domains of the research papers discussed in Section 5 . Domain labels are CM = Composite materials, OMFS = Organic Materials and Food Safety, CnM = Construction Materials, FA = Fragile Artifacts, M = Medical, and SHM = Structural Health Monitoring.
PaperRPNMEM
Lange [40]   
Sezgin and Sankur [63]   
Huang et al. [34]   
Růžek et al. [62]   
Farrar and Worden [23]   
Achenbach [3]   
Table 5: Domains of the research papers discussed in Section 6 . Domain labels are R = Review, PNM = Proposal of New Method, and EM = Evaluating Methods.
AlgorithmBenefitsPossible improvements
Computer Vision and Image Processing- Automation of visual inspection [ 12 , 66 , 45 ] . - Better speed, consistency, and objectivity compared to human inspector [ 12 , 45 ] . - Understandability and interpretability for human inspectors [ 34 , 66 ] .- The algorithms depend on proper illumination and other image acquisition environments [ 12 ] . - Noise reduction in images from processed signals [ 34 ] . - Computational cost is quite high [ 79 ] . - No general model for all different defects [ 45 ] . - High cost of equipment capable of handling real-world noise [ 48 ] .
Machine Learning- Ability to make predictions based on limited data [ 50 , 25 ] . - Processing large amount of data, localization of damage, automated damage detection [ 25 , 71 ] . - Increased accuracy compared to other automated techniques [ 71 ] .- Lack of quality training data [ 25 , 71 ] . - Models depend on pre-processing the data before training [ 71 ] . - There is a high computational cost for training with a large amount of data [ 71 ] . - Interpretability of the models is low [ 43 ] . - Model accuracy depends on appropriate data selection for training [ 68 ] .
Deep Learning- Higher accuracy and performance compared to traditional models [ 79 , 4 , 44 , 64 , 58 , 43 ] . - Are capable of detecting the severity of the defects [ 64 ] . - Noise tolerant [ 58 ] .- Requires more training data [ 43 ] . - Requires higher computational resources and training time [ 43 ] . - Model accuracy depends on appropriate hyperparameter tuning [ 68 ] .
Anomaly Detection- High accuracy with a limited amount of labeled data [ 4 , 11 ] . - Low computational cost [ 11 ] . - Noise tolerant [ 58 ] .- Lack of labeled data and benchmark [ 7 ] .
Generative Model- Synthetic data addressing the scarcity of training data [ 4 ] .- Performance is highly dependent on the generated loss function and can be difficult to generalize for a large dataset [ 22 ] .
Table 6: Benefits and possible improvements for different algorithms in NDI.
TechnologyBenefitsLimitations
Ultrasound- Ability to detect subsurface defects [ 20 ] . - Works on a variety of materials and arbitrary shapes in great detail [ 74 , 21 ] . - Internal access is not necessary [ 75 ] .- Requires the material to be sound conductive [ 20 ] . - Possible to miss micro defects [ 55 , 27 ] . - Requires processing and analyzing signals and depends on the expertise of the inspector [ 27 ] . - Sensitive to environment [ 75 ] .
Radiography and X-Ray- Ability to detect subsurface defects [ 75 , 27 ] . - Works on variety of materials [ 75 , 55 , 27 ] .- Accuracy drops for complex structures [ 75 ] . - Expensive technology [ 27 ] . - The process is slow [ 55 ] . - Access to both sides of the material is required [ 75 , 55 , 27 ] . - Safety hazard for human inspectors [ 75 , 55 , 27 ] .
Infrared, Thermal, and Low Frequency- Dry surface contact is adequate for low-frequency NDI techniques [ 13 ] . - Internal access is not necessary [ 75 ] . - Cost effective [ 75 ] .- Defect shape and depth are not precise [ 75 , 55 ] .
Visual Inspection- Simple, fast, and low cost [ 75 , 55 ] . - Non-contact testing [ 55 ] .- Usually works only for surface defect detections [ 75 , 55 , 27 ] . - Micro-defects are usually ignored [ 75 ] . - Detection is not very reliable and subjective [ 62 , 75 ] .
Electromagnetic Field- Usually, dry-contact testing, making it a useful technique for many conductive materials [ 40 ] . - Detects subsurface defect location and shape in ferromagnetic materials [ 38 ] .- Requires conductive material [ 55 ] . - Sensitive to electromagnetic interference [ 65 ] .
Table 7: Benefits and limitations of different technologies in NDI.
Application areasBenefitsChallenges
Composite Materials- Detecting small surface-level and subsurface defects, cracks, porosity, delamination, and others requires NDI [ 17 , 26 , 64 ] . - Different NDI techniques are available to identify defects in materials with complex shapes [ 32 ] .- Most NDI techniques require additional processing of signals and data [ 32 ] . - Requires balancing between cost and efficiency [ 17 , 26 ] .
Organic Material and Food Safety- Ensures speed, consistency, and objectivity of inspections [ 12 ] . - Real-time damage detection, including internal defects [ 45 ] .- Quality data to train automated models and interoperability is a challenge [ 43 ] .
Construction Material- Different NDI methods offer on-site inspection where access to only one side is sufficient [ 56 ] .- Varying penetration depth and difference in materials make it challenging to have a universal inspection method [ 56 , 16 ] .
Fragile Artifacts- NDI is essential to evaluate the strength, weathering, and presence of defects in fragile artifacts without damaging or degrading the artistic values [ 47 , 50 ] .- On-site inspections are vulnerable to the environment and require special equipment, which can be expensive [ 48 ] .
Medical- Higher throughput of medical scanning equipment can be useful where high resolution is not needed [ 19 ] .- Low penetration depth due to low voltage [ 19 ] .
Structural Health Monitoring- Safety and structural integrity of buildings after natural events like earthquakes, or routine check of infrastructure integrity requires NDI techniques [ 23 ] .- Data interoperability, penetration depth, and cost are major challenges [ 39 ] . - Research in modern eco-friendly sensors adaptive to the environment is necessary [ 31 ] .
Table 8: Benefits and challenges of application areas in NDI.
Research AreaIdentified Gap / LimitationFuture Research Idea
Deep Learning AlgorithmsModels act as “black boxes,” lacking transparency in how defect classifications are derived, which hinders adoption in safety-critical sectors.Development of Explainable AI (XAI) frameworks specifically for NDI that provide visual or feature-based justification for defect classification.
Data AvailabilitySevere scarcity of labeled industrial defect data, and reliance on synthetic data that may not perfectly mimic real-world physics.Investigation into physics-informed neural networks (PINNs) and federated learning to train robust models without sharing proprietary data.
Sensing TechniquesSingle-modality inspections have inherent physical blind spots regarding depth or material type.Development of real-time multi-modal sensor fusion algorithms that combine surface and subsurface data.
Structural Health Monitoring (SHM)Sensors in harsh environments suffer from power limitations and degradation over time, and the data transmission bandwidth is limited.Research into self-powered, eco-friendly piezoelectric sensors and edge-computing algorithms that process data on-chip to reduce transmission load.
NDE 4.0 EcosystemLack of standardized protocols for integrating NDI data into digital twins, restricting the data siloed in inspection reports.Creation of universal protocols that allow automated NDI results to dynamically update the structural simulation of a digital twin.
Table 9: Research gaps and future research opportunities in NDI.
Figure 4: Strategic research roadmap illustrating the transition from current limitations to NDE 4.0 opportunities.
More in this topic

Non-Destructive Testing

See all 12 papers

Explore similar work

CardsList
Mar 4, 2026

Autonomous Aerial Non-Destructive Testing: Ultrasound Inspection with a Commercial Quadrotor in an Unstructured Environment

This work presents an integrated control and software architecture that enables arguably the first fully autonomous, contact-based non-destructive testing (NDT) using a commercial multirotor originally restricted to remotely-piloted operations. To allow autonomous operation with an off-the-shelf platform, we developed a real-time framework that interfaces directly with its onboard sensor suite. The architecture features a multi-rate control scheme: low-level control is executed at 200 Hz, force estimation at 100 Hz, while an admittance filter and trajectory planner operate at 50 Hz, ultimately supplying acceleration and yaw rate commands to the internal flight controller. We validate the system through physical experiments on a Flyability Elios 3 quadrotor equipped with an ultrasound payload. Relying exclusively on onboard sensing, the vehicle successfully performs autonomous NDT measurements within an unstructured, industrial-like environment. This work demonstrates the viability of retrofitting off-the-shelf platforms for autonomous physical interaction, paving the way for safe, contact-based inspection of hazardous and confined infrastructure.

Ruben Veenstra, Barbara Bazzana, Sander Smits +1

Robotics and Mechatronics lab, Faculty of Electrical Engineering, Mathematics & Computer Science, University of Twente, Enschede, The Netherlands. · Department of Computer, Control and Management Engineering, Sapienza University of Rome, 00185 Rome, Italy.

Robotic InspectionIndustrial InspectionQuadrotor ControlAerial Manipulation
AskPDF
Aug 29, 2026

An Integrated Vision-and-Language Pretraining (VLP) and Visual Question Answering (VQA) model to Automate Nondestructive Evaluation Image Analysis

An AI-based approach called ChatNDE Figure to Caption is introduced, which aims to automate the interpretation of NDE images using deep learning and natural language processing (NLP). A Vision-and-Language Pretraining (VLP) strategy is developed to help the model learn how to connect visual features with meaningful language. Basically, we built a large NDE image dataset, trained the model using annotated examples, and then evaluated how well it performed using BLEU scores to compare its output to expert written descriptions. So, the system combines a ResNet50 model to extract important features from the images and a GPT2 language model to turn those features into natural sounding text. Even though the accuracy of the model has been low the generated caption results have been solid so far, the captions were shorter but mentioned some important features of images what human experts would say, which shows the model is learning to pick up on key details. Also, a Visual Question Answering (VQA) model is used as part of the system. VQA models are designed to take an image and a question about that image (like Is there a crack? or Where is the defect located?) and generate a useful answer. By adding this layer, the platform will not just describe what it sees, it can also respond to specific questions, making it even more interactive and helpful for inspectors in the field. This whole approach is a big step toward speeding up NDE workflows, reducing human error, and making the technology more accessible.

Mehrdad Shafiei Dizaji, Hoda Azari

Federal Highway Administration, Turner-Fairbank Highway Research Center, McLean, VA 22101

Non-Destructive TestingImage CaptioningVision-Language Pretraining
AskPDF
Aug 25, 2026

Lowering the Barrier to AI-Driven Inspection: A No-Code Workflow for Automated Structural Defect Detection

Structural health monitoring (SHM) is essential in modern engineering, providing data for condition-based maintenance, lifecycle assessment, and predictive decision-making. Traditionally, SHM relied on visual inspection to detect defects such as cracks and deformations. Early computer vision (CV) methods, including thresholding, edge detection, and handcrafted features, aimed to automate this process but were highly sensitive to noise, imaging variations, and multiscale defects, limiting their reliability. Recent advances in machine learning, particularly convolutional neural networks (CNNs) and You Only Look Once (YOLO), have improved defect detection accuracy and enabled real-time analysis. However, adoption in SHM remains limited due to technical barriers such as data labeling, model training, and deployment, which typically require programming expertise. To address this gap, we introduce YOLOEZ, an open-source, GUI-based tool for end-to-end YOLO model application. YOLOEZ integrates data labeling, training, and inference into a single interface, enabling high-performance model development without code while supporting reproducible workflows. Evaluation against existing software and classical image processing demonstrates that YOLOEZ not only outperforms traditional methods across most detection metrics, but also lowers adoption barriers present in other modern CV tools. By combining accuracy with accessibility, YOLOEZ facilitates wider use of AI-driven monitoring for predictive maintenance, digital twins, and intelligent structural systems.

Michael Holm, Tanner McElroy, Xinghang Zhang +1

Purdue University, West Lafayette, IN

Fault DetectionStructural Health Monitoring
AskPDF