Quantifying the Severity and Dimensions of Structural Defects

29 concepts26 papers1 proposal

In the fields of structural health monitoring (SHM) and non-destructive testing (NDT), a fundamental shift is occurring. For years, the primary goal was binary detection: answering the simple question, "Is there a defect?" With the advent of modern deep learning and highly sensitive sensors, this question has largely been solved. We can reliably flag a crack in an aircraft wing, a leak in a pipeline, or an anomaly on a power line.

However, knowing a defect exists is no longer enough. To make informed maintenance decisions, engineers must answer a much harder question: "How bad is it?" Moving from qualitative detection to quantitative severity assessment—measuring the exact length, depth, volume, or structural impact of a defect—remains one of the most persistent bottlenecks across engineering disciplines.

Understanding why this problem is so challenging, and how researchers are attempting to solve it, requires looking at the physical and mathematical limitations of our current diagnostic tools.


The Spatial Blindspot of Computer Vision

Computer vision models have revolutionized automated inspection. Frameworks like YOLO and Vision Transformers excel at scanning thousands of images to localize and classify anomalies. Yet, when applied to damage assessment, these models encounter a fundamental limitation: they treat a three-dimensional physical degradation as a two-dimensional bounding box.

For instance, Wang et al. (Insulator Defect Detection Based on ML-YOLOv5 Algorithm) and He et al. (Insulator Defect Detection Based on YOLOv8s-SwinT) demonstrate that while optimized deep learning architectures can easily spot tiny defects on high-voltage power line insulators from drone footage, they struggle to accurately quantify the precise severity of multiple co-occurring defects.

The core of the issue is that visual features often saturate. A surface scratch and a deep structural crack can look identical to a camera under certain lighting conditions. As Tian et al. (Wind Turbine Blade Defect Recognition Method Based on Large-Vision-Model Transfer Learning) note in their work on wind turbine blades, large vision models are highly effective at localization and classification, but they lack the capability to perform quantitative defect analysis, such as mapping out the exact geometric shape and physical extent of the damage. Without this geometric data, an AI can tell you that a blade is damaged, but it cannot tell you if the blade is about to fail. This diagnostic limitation is echoed by Tran et al. (CarcassFormer: an end-to-end transformer-based framework for simultaneous localization, segmentation and classification of poultry carcass defect), who point out that even advanced end-to-end segmentation transformers often fail to resolve the fine-grained diagnostic details necessary to differentiate complex defect characteristics.


The Inverse Problem in Wave Propagation

To look beneath the surface, researchers rely on stress waves—such as guided waves, Lamb waves, or Rayleigh waves. When an ultrasonic wave travels through a material, any internal defect will scatter, reflect, or attenuate the signal.

In theory, the altered wave carries the signature of the defect. In practice, reconstructing the defect's exact dimensions from that signal is a highly ill-posed inverse problem. Multiple distinct defect geometries can produce identical signal attenuations at a distant sensor.

Qualitative Detection (Easy):
[Wave Source] ------> (Defect Exists) ------> [Sensor: Signal Attenuated]

Quantitative Sizing (Hard):
                      /? (Deep & Narrow?)
[Sensor: Attenuation] ---? (Shallow & Wide?) ---> Reconstruct Exact Geometry
                      \? (Internal Void?)

Bao et al. (A Density Clustering RAPID Based on an Array-Compensated Damage Index for Quantitative Damage Diagnosis) address this by using array-compensated damage indices to improve localization, yet they explicitly state that directly estimating physical dimensions like crack length or width remains an open challenge. Similarly, Zhou et al. (Development of a Meander-Coil-Type Dual Magnetic Group Circumferential Magnetostrictive Guided Wave Transducer for Detecting Small Defects Hidden behind Support Structures) succeeded in engineering highly sensitive transducers capable of detecting tiny hidden pipeline defects, but noted that precise quantitative defect sizing remains underdeveloped.

Even when extracting advanced signal features, such as the instantaneous phase and marginal spectrum analyzed by Xu et al. (Application of Instantaneous Parameter Characteristic in Active Lamb Wave Based Monitoring of Plate Structural Health), or optimizing ultrasonic parameters for Rayleigh waves as seen in the work of Im et al. (NDE Characterization of Surface Defects on Piston Rods in Shock Absorbers Using Rayleigh Waves), the system can reliably flag the presence of a fault, but lacks the automated algorithms required to accurately calculate its physical size.


Environmental Noise and the Localization-Severity Trade-off

Another reason quantification fails in real-world scenarios is environmental noise. In a pristine laboratory, a sensor's response correlates cleanly with damage size. In the field, temperature swings, structural friction, and background vibrations mask the true scale of a defect.

This creates a frustrating trade-off between localizing a fault and sizing it. Consider fluid networks: Bastin et al. (Leak detection in open water channels) demonstrate that CUSUM-based algorithms can quickly identify the presence of a leak by monitoring water levels, but cannot isolate its size or location due to natural water variations. When trying to pinpoint the location, Weyer et al. (Methods for the localization of a leak in open water channels) discovered that the problem is inherently difficult because minor uncertainties in channel friction or leak start times cause massive errors in spatial estimation.

Conversely, in underwater pipelines, Xie et al. (Time-Frequency Distribution Map-Based Convolutional Neural Network (CNN) Model for Underwater Pipeline Leakage Detection Using Acoustic Signals) built an acoustic time-frequency model that successfully classifies leak severity, but lacks the ability to pinpoint the physical location of the leak along the pipe.

This inability to simultaneously resolve both metrics is also prevalent in mechanical systems. The authors of Fault Diagnosis of Strong Noise in Rolling Bearings based on EEMD-RRSD show that while advanced signal processing can isolate subtle mechanical fault signatures from intense background noise, these methods operate purely as binary detectors; they cannot quantify the true extent of the damage or estimate the component's remaining useful life.


Bridging the Gap: From Indicators to Integrity

To overcome these limitations, modern research is moving away from superficial signal features and toward deep physical, chemical, and structural context. True quantification requires mapping a sensor's raw output to an underlying physical property.

Ultimately, whether assessing large-scale civil infrastructure—such as the full-parameter inversion models for building subsidence by Liu et al. (TDFPI: A Three-Dimensional and Full Parameter Inversion Model and Its Application for Building Damage Assessment in Guotun Coal Mining Areas, Shandong, China) and the analytical deflection theories for arch bridges by Zhou et al. (Research on Damage Identification of Arch Bridges Based on Deflection Influence Line Analytical Theory)—or micro-scale material defects, the trajectory of the field is clear. The future of structural health monitoring relies on replacing qualitative indicators with rigorous, physics-backed quantitative metrics that measure not just the presence of damage, but its true physical scale.

Key Foundational Work

Diagnosis of Evaporative Leaks and Sensor Faults in a Vehicle Fuel System (2001) — contributed 1 concept elements to this direction.
Leak detection in open water channels (2008) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_quantitative_damage_assessment_13a0ef08,
  title = {quantitative damage assessment},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/quantitative-damage-assessment/rdr_13a0ef080e7449379bdcc17b58392653},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_quantitative_damage_assessment_13a0ef08_v1,
  title = {quantitative damage assessment},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/quantitative-damage-assessment/rdr_13a0ef080e7449379bdcc17b58392653/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake