⚡ electrical engineering

Observability Core Vector Trajectories for Structural Health Monitoring: From Stable States to Structural Change

This paper extends the infrastructure-observability framework into the temporal domain using Observability Core Vectors (OCV) derived from Sentinel-1 data to characterize radar echo changes during various structural states, concluding that while OCV trajectories offer complementary monitoring information, they lack universal failure signatures and require independent validation for accurate structural interpretation.

Original author: Martin Hoffmann

Published 2026-11-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original author: Martin Hoffmann

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to understand the health of a massive bridge, a towering wind turbine, or a communications mast without ever touching it. Engineers have long relied on satellites equipped with radar to peer at these structures from space. Unlike a camera that takes a photograph, radar sends out radio waves that bounce off the ground and return to the satellite. By analyzing how these waves return, scientists can detect tiny shifts in position or changes in the surface texture of a structure. This method, known as interferometry, allows for the monitoring of infrastructure from hundreds of kilometers away, offering a way to spot problems before they become visible to the naked eye. The central question for researchers has always been: as a structure ages, gets repaired, or begins to fail, does its radar signature follow a predictable path? If a bridge is slowly weakening, does its radar "fingerprint" change in a specific, universal way that signals danger?

A new study by independent researcher Martin Hoffmann tackles this question by tracking the evolution of these radar signatures over time. Instead of looking for a single number that indicates health, the researcher developed a six-part description of how a structure appears to the radar. This description, called the Observability Core Vector, captures six distinct aspects of the radar return: how consistent the signal is over time, how long the structure remains visible, how much of the area is covered by the signal, how scattered or clumped that signal is, the total size of the signal, and how far the signal has shifted from its original position. By watching how these six numbers move together over months and years, the study creates a trajectory, or a path, that represents the changing radar appearance of the asset.

The researcher applied this method to a diverse collection of real-world structures, including a telecommunications mast in Espoo that has remained stable, a lattice tower and a concrete bridge that underwent controlled stress tests, a tower that was completely rebuilt, and several bridges that suffered documented failures or repairs. The goal was to see if these different physical events—stability, repair, reconstruction, and collapse—produced similar patterns in the radar data. If there were a universal sign of failure, one would expect all failing structures to show the same movement in their six-part radar description, perhaps all shrinking in size or all becoming more scattered in a specific sequence.

The findings, however, tell a different story. The study found that there is no single, universal path that leads to failure. When a structure is stable, its radar signature remains relatively constant, but when it changes, the direction of that change depends entirely on what is happening to the structure and how the satellite is looking at it. For instance, when a bridge was repaired, the radar signal covering the structure grew larger and more scattered. In contrast, when a different bridge collapsed, the signal covering it shrank and became less scattered. Even more telling, a structure undergoing a planned reconstruction showed a massive shift in its radar signature, moving far away from its previous state, just as a failing structure might. This means that a large change in the radar data does not automatically mean the structure is broken; it could simply mean it is being rebuilt or modified.

The research explicitly rules out the idea of a "universal failure signature." The study examined several documented failures, including the collapse of the Morandi Bridge and the Carola Bridge, and found that their radar signatures changed in completely different ways. One bridge showed a gradual shift in signal consistency before falling, while another showed a sudden change in the size and shape of the signal. A third structure, which experienced a progressive failure, showed so much variability in its data that no clear long-term trend could be identified. The study concludes that the radar signature is a sensitive record of change, but not a direct measure of structural health. A shift in the radar data tells us that something has changed in the way the structure reflects radio waves, but it does not tell us whether that change is good, bad, or neutral.

Ultimately, this work suggests that radar monitoring is a powerful tool for detecting that a change has occurred, but it cannot diagnose the cause on its own. The six-part radar description acts as a highly detailed log of an asset's appearance to the satellite, capturing everything from minor repairs to catastrophic failures. However, because these changes do not follow a single, predictable script, engineers cannot rely on a simple alarm system that triggers when the radar numbers cross a certain line. Instead, the radar data must be treated as a piece of a larger puzzle. To understand what the radar is seeing, it must be combined with other information, such as physical inspections or engineering models. The study confirms that while satellites can watch our infrastructure with incredible precision, interpreting what they see requires human judgment and context, as the path from a stable state to a broken one is as unique as the structure itself.

Technical Summary: Observability Core Vector Trajectories for Structural Health Monitoring

Problem Statement
While previous frameworks established that infrastructure states can be observed and distinguished via radar, a critical gap remained: how the radar representation of an asset evolves over time during stable operation, controlled transitions, reconstruction, repair, and failure. The paper addresses the question of what infrastructure "looks like" to radar as it undergoes physical changes. Specifically, it investigates whether distinct physical processes produce recurring, transferable trajectories in radar observability that could serve as universal signatures for structural health monitoring (SHM) or failure prediction.

Methodology
The study extends the infrastructure-observability framework into the temporal domain using the Observability Core Vector (OCV). The OCV is a six-dimensional vector representing the observable state of an asset at time $t$ :
$OCV(t) = [\gamma^2(t), P(t), D(t), F(t), A(t), S(t)]$

Dimensions:
- Cross-acquisition consistency: Coherence ( $\gamma^2$ ) and Persistence ( $P$ ).
- Echo-mask geometry: Density ( $D$ ), Fragmentation ( $F$ ), Area ( $A$ ), and Shift ( $S$ ).
Data Processing: The author utilizes Sentinel-1 Synthetic Aperture Radar (SAR) data. Unlike fixed-size image chips, the methodology employs asset- and channel-specific measurement support ( $W_{a,c}$ ). This is derived by projecting the 3D asset geometry ( $G^3D_a$ ) into native radar coordinates, accounting for slant-range shifts due to asset height and acquisition geometry.
Trajectory Analysis: An OCV trajectory ( $T_{OCV}$ ) is defined as the temporal evolution of the vector. The study analyzes assets across diverse scenarios: stable operation (Espoo), controlled state transitions (LUMO, Bautzen), reconstruction (KDLO), repair (Werder-Havel), and documented failures (Carola, Morandi, CTS).
Statistical Approach: The author employs Theil–Sen slopes with moving-block bootstrap confidence intervals for trend detection. They use State Discrimination Scores (SDS) and dimension-specific standardization ( $z$ -scores) to compare states. Crucially, they analyze "direction-independent departure" ( $R_B$ for behavior, $R_G$ for geometry) to quantify change magnitude without assuming a specific health direction.

Key Results
The analysis of diverse assets and physical processes yields several critical findings:

No Universal Failure Signature: The study finds no transferable life-cycle trajectory or universal failure signature. Different failure modes (Carola, Morandi, CTS) exhibit distinct temporal organizations. For instance, Carola showed pre-event changes in coherence and persistence followed by spatial reduction, whereas Morandi showed stronger pre-collapse spatial evolution, and CTS showed broad variability without a supported long-term coherence trend.
Context-Dependent Trajectories: The direction and magnitude of OCV changes are not consistent across physical processes.
- Repair vs. Failure: During the Werder-Havel repair, Area ( $A$ ) and Fragmentation ( $F$ ) increased. Similarly, in parts of the Morandi pre-collapse record, $A$ and $F$ increased. Conversely, after the Carola collapse, $A$ and $F$ decreased. Thus, an increase in these dimensions does not universally indicate deterioration, nor does a decrease indicate stability.
- Reconstruction: The KDLO reconstruction case showed a persistent, multidimensional departure from the reference state that did not return to the original baseline, indicating a regime change rather than a linear deterioration path.
Measurement Dependence: The observed trajectories are highly sensitive to the definition of measurement support and the specific radar acquisition geometry.
- Layer Sensitivity: In the CTS case, a negative coherence trend appeared on an alternative echo-mask layer but was absent on the primary registered series, demonstrating that inferred long-term evolution can be an artifact of the measurement layer.
- Geometry Sensitivity: The Haltern wind turbine case showed that major morphological changes alter the radar representation, and the interpretation of this change depends entirely on the defined measurement support.
Stable States: The Espoo telecommunications mast demonstrated bounded variation without supported long-term trends, confirming that stable assets maintain a consistent radar representation within a specific acquisition geometry.

Significance and Claims
The paper claims that the OCV successfully characterizes the temporal evolution of an asset's radar representation, providing complementary information for infrastructure monitoring. However, the author explicitly limits the scope of their claims regarding structural health:

Radar Representation vs. Physical Condition: The OCV describes how an asset appears to radar, not the physical origin of the change. The author states that "physical interpretation requires independent measurements or validated physical models."
Rejection of Universal Health Metrics: The study refutes the existence of a universal "health direction" in OCV space. Changes in dimensions like Area or Fragmentation do not inherently map to "healthy" or "unhealthy" states across different assets or processes.
State Discrimination: While the OCV can distinguish between specific, independently defined states (e.g., in the LUMO and Bautzen cases), these distinctions are asset-specific and channel-specific, not generalizable across the portfolio.

In conclusion, the paper establishes that while OCV trajectories are effective for tracking radar-observable changes, they do not constitute a direct measure of structural condition. The framework shifts the focus from seeking universal failure signatures to understanding the specific, context-dependent radar signatures of physical processes, requiring complementary data for structural interpretation.

More like this