LITF-PA-2026-040 · Infrastructure / Distributed Fiber Sensing

System and Method for Continuous Non-Contact Assessment of Asphalt Pavement Structural Condition Using Traffic-Induced Vibration Transfer Function Analysis from Distributed Acoustic Sensing on Existing Telecommunications Fiber Optic Infrastructure with Recurrent Neural Network Degradation Modeling

Defensive Prior Art Disclosure
Cross-section view of a highway showing buried fiber optic cable beneath asphalt layers with vibration wave visualizations propagating from traffic down through the pavement structure to the sensing fiber
⚖️ Prior Art Notice: This document is published as defensive prior art under 35 U.S.C. § 102(a)(1). The inventions described herein are dedicated to the public domain as of the publication date above. This disclosure is intended to prevent the patenting of these concepts by any party.

Abstract

Disclosed is a system and method for continuously assessing the structural condition of asphalt pavement using traffic-induced vibration signatures captured by distributed acoustic sensing (DAS) interrogators connected to existing buried telecommunications fiber optic cables running parallel to or beneath roadways. DAS technology converts standard single-mode optical fiber into a continuous array of thousands of vibration sensors by injecting coherent laser pulses and analyzing Rayleigh backscatter phase changes caused by strain perturbations along the fiber. When vehicles traverse pavement above or adjacent to a buried telecom fiber, the dynamic axle loads generate stress waves that propagate through the pavement structure (surface course, binder course, base, subbase, subgrade) before reaching the fiber at depths of 0.5 to 2.0 meters. The vibration transfer function between the pavement surface (where the load is applied) and the fiber (where the signal is measured) encodes the mechanical properties of the intervening pavement layers: stiffness, damping ratio, layer thickness, and interlayer bond condition. As pavement degrades through fatigue cracking, rutting, moisture infiltration, and aggregate breakdown, these mechanical properties change in characteristic ways that alter the measured transfer function. The system applies a recurrent neural network (LSTM architecture) trained on paired DAS measurements and ground-truth pavement condition data (Pavement Condition Index scores from manual surveys, Falling Weight Deflectometer measurements, and ground-penetrating radar profiles) to estimate the current structural condition of every meter of pavement along the fiber route and to predict remaining useful life. By leveraging the approximately 1.5 million route-miles of buried fiber optic cable in the United States, much of it installed in highway rights-of-way, the system enables continuous, network-scale pavement monitoring at zero incremental sensor hardware cost, replacing the current practice of periodic manual surveys conducted at 1- to 5-year intervals.

Field of the Invention

This invention relates to pavement engineering and infrastructure health monitoring, specifically to the non-contact assessment of asphalt pavement structural condition using vibration measurements from distributed acoustic sensing systems operating on existing telecommunications fiber optic cables, with machine learning models for condition estimation and remaining life prediction.

Background

The United States maintains approximately 4.19 million miles of paved roads, of which roughly 94% are surfaced with asphalt (FHWA Highway Statistics 2022). The American Society of Civil Engineers 2025 Infrastructure Report Card rated US roads at D+, estimating a $786 billion backlog in road maintenance and rehabilitation. Pavement deterioration imposes direct costs on vehicle operators through increased fuel consumption ($0.03-0.05 per mile on rough roads per Chatti and Zaabar, Transportation Research Record 2012), accelerated tire and suspension wear, and crash risk from rutting, potholes, and surface irregularities. The TRIP National Transportation Research Group estimated that poor road conditions cost US drivers $130 billion annually in vehicle operating costs.

Current pavement condition assessment relies on several approaches, each with significant limitations:

Meanwhile, the US telecommunications infrastructure includes approximately 1.5 million route-miles of buried fiber optic cable (FCC Broadband Map, 2024), a substantial fraction of which is installed within highway rights-of-way. The FHWA Utility Accommodation Policy permits utility installations within highway rights-of-way, and telecommunications companies routinely bury fiber in highway shoulders, medians, and adjacent rights-of-way at depths of 0.5-2.0 meters. Major fiber routes follow Interstate highways and US routes, with spur fiber extending along state routes and urban arterials to reach distribution nodes. The result is a vast, dense network of single-mode optical fiber already in the ground alongside the road network.

Distributed Acoustic Sensing has matured from a laboratory curiosity to a commercially deployed sensing technology. DAS interrogators manufactured by OptaSense (a Luna Innovations company), Silixa, Fotech Solutions (a bp subsidiary), and AP Sensing inject coherent laser pulses into standard single-mode fiber and analyze the phase of Rayleigh backscattered light to detect strain perturbations along the fiber with spatial resolution of 1-10 meters, temporal sampling rates of 1-100 kHz, and strain sensitivity of 0.1-1 nanostrain/√Hz. DAS has been deployed commercially for pipeline leak detection (Mestayer et al., Proceedings of SPIE 2016), seismic monitoring (Lindsey et al., Nature Communications 2019), perimeter security, railway monitoring, and vertical seismic profiling in oil and gas wells. The critical insight is that DAS measures vibrations at every point along the fiber simultaneously, converting kilometers of existing telecom fiber into a continuous vibration sensor array without any physical modification to the fiber or its installation.

The gap in the art is a complete system that: (a) repurposes existing buried telecommunications fiber optic cable as a distributed vibration sensor for pavement monitoring, (b) extracts pavement structural condition indicators from the vibration transfer function between traffic-induced surface loads and the DAS-measured response at fiber depth, (c) applies machine learning models to estimate pavement condition indices and remaining useful life from the time-varying transfer function features, and (d) operates continuously and autonomously, providing daily updates on pavement condition for every meter of road along the fiber route at zero incremental sensor cost.

Detailed Description

1. Physics of Traffic-Induced Vibration Propagation Through Pavement Structures

When a vehicle tire contacts an asphalt pavement surface, the dynamic axle load generates stress waves that propagate downward through the pavement structure. The load is not static: vehicle suspension dynamics, tire-road interaction, and road surface irregularities create a dynamic load spectrum spanning 1-80 Hz for body bounce, axle hop, and tire vibration modes (Cebon, Transportation Research Record 1996). A fully loaded Class 8 truck (80,000 lb GVW) produces peak dynamic tire forces of 15,000-25,000 lbs per tire, while a passenger vehicle (4,000 lb) generates 800-1,500 lbs per tire. These dynamic loads create compressive body waves (P-waves) and shear waves (S-waves) that propagate through the layered pavement structure, plus Rayleigh surface waves that dominate at distances greater than approximately one wavelength from the source.

The vibration transfer function H(f) between the pavement surface (load application point) and the buried fiber (measurement point) depends on the mechanical properties of each intervening layer:

The key physical principle is that the vibration transfer function H(f) acts as a fingerprint of the pavement's structural state. A healthy pavement with stiff, well-bonded layers transmits traffic vibrations to the subgrade and fiber with a characteristic frequency-dependent attenuation profile: high-frequency components (above 30 Hz) are rapidly attenuated by viscous damping in the asphalt, while low-frequency components (2-15 Hz) propagate efficiently through the structure. As pavement deteriorates, the transfer function changes in predictable ways:

2. DAS Measurement Configuration and Signal Processing

The system connects a DAS interrogator to one end of an existing dark fiber (unused fiber strand within a telecommunications cable bundle) running along the target road segment. Most telecom cables contain 48-288 fiber strands, and telecommunications carriers typically maintain dark fiber inventory of 20-60% of total strand count for future capacity. The DAS interrogator occupies a single fiber strand, leaving all other strands available for telecommunications traffic. No physical modification to the buried cable is required.

Interrogator configuration: The DAS interrogator operates in phase-sensitive OTDR (Φ-OTDR) mode, injecting narrow-linewidth laser pulses (linewidth less than 3 kHz, pulse width 10-100 ns corresponding to 1-10 meter spatial resolution) at a pulse repetition rate of 2-20 kHz. The Rayleigh backscattered light from each spatial resolution cell interferes coherently at the detector, producing a signal whose phase is proportional to the strain integrated over the gauge length. Changes in strain (from passing traffic vibrations) produce phase changes that are demodulated to recover the dynamic strain waveform at each spatial resolution cell along the fiber. Modern DAS interrogators achieve strain sensitivity of 0.1-1 nanostrain/√Hz per Hartog et al., Optics Express 2018, sufficient to detect traffic-induced vibrations from passenger vehicles at lateral offsets of 5-20 meters from the fiber (the typical distance between a highway travel lane and a buried utility in the right-of-way shoulder).

Traffic event detection and isolation: The system detects individual vehicle passages using a matched filter based on the expected strain response of a moving point load on a layered half-space. The matched filter output provides vehicle detection, speed estimation (from the apparent velocity of the strain response along the fiber), and approximate axle load estimation (from the peak strain amplitude). For multi-axle vehicles, the system resolves individual axles using the known relationship between vehicle speed, axle spacing, and temporal separation of strain peaks. Vehicle detection performance exceeding 99% has been demonstrated for DAS-based traffic monitoring on existing telecom fiber by Martin et al., Scientific Reports 2020.

Transfer function estimation: For each detected vehicle passage at each spatial resolution cell, the system estimates the local vibration transfer function H(f) using the following procedure: (1) The surface load time history is estimated from the vehicle speed, axle configuration, and estimated axle loads using a moving load model on a layered elastic half-space. (2) The DAS-measured strain response at the fiber depth is windowed (Tukey window, α=0.1) to isolate the response to the target vehicle. (3) The transfer function is computed as the ratio of the cross-spectral density of the estimated surface load and the measured fiber response to the power spectral density of the estimated surface load: H(f) = Sxy(f) / Sxx(f). (4) Transfer function estimates from multiple vehicle passages are averaged (Welch's method with 50% overlap) to reduce noise, with separate averages maintained for vehicle weight classes (light: less than 10,000 lb GVW; medium: 10,000-33,000 lb; heavy: greater than 33,000 lb) because the stress-dependent nonlinearity of granular base and subgrade materials causes the transfer function to vary with load magnitude. (5) Daily averaged transfer functions for each weight class at each spatial resolution cell form the input features for the machine learning model.

3. Machine Learning Architecture for Condition Estimation

The system employs a two-stage machine learning architecture: a spatial feature extractor followed by a temporal sequence model.

Stage 1: Spatial feature extraction. The daily averaged transfer function H(f) at each spatial resolution cell is represented as a complex-valued vector of 128 frequency bins spanning 1-80 Hz. The magnitude and phase spectra (256 real-valued features) plus the three weight-class transfer functions (768 features total) feed a 1D convolutional neural network that extracts 64 spatial features per cell. The CNN architecture uses four convolutional blocks (1D convolutions with kernel size 5, batch normalization, GELU activation, max pooling), reducing the 768-dimensional input to a 64-dimensional feature vector that encodes the pavement structural state. Adjacent cells (within a 50-meter sliding window) are processed jointly to capture spatial correlation in pavement condition, as distresses often span multiple meters.

Stage 2: Temporal degradation model. The 64-dimensional feature vector for each cell is tracked over time (daily observations) using a Long Short-Term Memory (LSTM) network with 128 hidden units. The LSTM learns the temporal evolution of pavement condition from the sequence of daily feature vectors, capturing degradation trends, seasonal cycles (freeze-thaw, temperature-dependent asphalt stiffness), and abrupt condition changes (utility cuts, emergency repairs, overlay construction). The LSTM outputs two predictions: (a) current estimated Pavement Condition Index (PCI, 0-100 scale) with uncertainty bounds, and (b) predicted time to reach a threshold PCI (typically PCI=40 for rehabilitation trigger or PCI=25 for reconstruction trigger), constituting the remaining useful life (RUL) estimate.

Training data: The model is trained on paired datasets of DAS measurements and ground-truth pavement condition assessments. Ground truth is obtained from: (1) manual PCI surveys (ASTM D6433) conducted by state DOTs and municipalities on their regular survey cycles, (2) FWD measurements that provide backcalculated layer moduli for validation of the physically-motivated transfer function features, (3) GPR profiles that identify layer thicknesses and subsurface anomalies, and (4) maintenance records (mill-and-overlay dates, crack sealing, patching) that mark condition transitions. The training dataset requires at minimum 500 km of fiber-instrumented road with concurrent ground-truth surveys spanning at least two annual temperature cycles to capture seasonal variability. Transfer learning from synthetic datasets (generated by the pavement response model described in Section 4) accelerates training for new deployments where limited ground truth is available.

4. Physics-Informed Synthetic Data Generation

To bootstrap training before sufficient ground-truth paired data is available, the system generates synthetic training datasets using a forward pavement response model. The model combines:

Domain randomization of model parameters (layer thicknesses ±30%, moduli ±50%, fiber depth ±0.5m, lateral offset ±5m, vehicle parameters per NCHRP Report 1-37A vehicle fleet distributions) ensures that the CNN-LSTM model generalizes across the variability encountered in real deployments.

5. Operational Architecture and Data Flow

The system operates in a continuous monitoring mode with the following data flow:

Edge processing at the DAS interrogator: Raw DAS data rates range from 0.5 to 50 GB/hour depending on fiber length, spatial resolution, and sampling rate. An edge computing unit co-located with the DAS interrogator (GPU-equipped server, e.g., NVIDIA Jetson AGX Orin or equivalent) performs real-time traffic event detection, vehicle classification, and per-event transfer function estimation. Only the extracted transfer function features (approximately 3 KB per vehicle passage per spatial cell) are transmitted to the cloud, reducing bandwidth by 99.9% relative to raw DAS data.

Cloud aggregation and inference: Daily averaged transfer functions for each spatial cell are computed in the cloud, and the CNN-LSTM model produces updated PCI estimates and RUL predictions. Results are served through a GIS-enabled web dashboard and API that integrates with pavement management systems (AgileAssets, Cartegraph, Lucity, Deighton dTIMS) via the AASHTO AASHTOWare Pavement Management data exchange format.

Calibration and validation: The system automatically identifies opportunities for ground-truth calibration by detecting planned pavement surveys (from agency work calendars) and flagging sections where the model's PCI estimate has high uncertainty. When ground-truth data becomes available, the model is retrained with the new data incorporated, continuously improving accuracy. A model performance dashboard tracks estimation accuracy (mean absolute error, root mean square error, and prediction interval coverage probability) against ground-truth surveys, alerting operators if model degradation is detected.

6. Figures Description

Claims

  1. A system for assessing pavement structural condition, comprising: a distributed acoustic sensing interrogator connected to an existing telecommunications fiber optic cable buried adjacent to or beneath a paved roadway; a traffic event detection module that identifies individual vehicle passages from the DAS-measured strain waveform; a transfer function estimation module that computes the frequency-dependent vibration transfer function between estimated surface traffic loads and the DAS-measured fiber response for each vehicle passage at each spatial resolution cell along the fiber; and a machine learning model that estimates the pavement structural condition from the time series of transfer function measurements.
  2. The system of claim 1, wherein the distributed acoustic sensing interrogator operates in phase-sensitive optical time-domain reflectometry mode on a dark fiber strand within an existing telecommunications cable bundle, requiring no physical modification to the fiber or its installation.
  3. The system of claim 1, wherein the transfer function estimation module computes separate averaged transfer functions for multiple vehicle weight classes to account for the stress-dependent nonlinearity of granular base and subgrade materials, using estimated vehicle weight derived from peak DAS strain amplitude and vehicle speed.
  4. The system of claim 1, wherein the machine learning model comprises a convolutional neural network that extracts spatial features from the frequency-domain transfer function representation, followed by a recurrent neural network (LSTM architecture) that tracks the temporal evolution of the spatial features to estimate current Pavement Condition Index and predict remaining useful life.
  5. The system of claim 1, wherein the machine learning model detects interlayer debonding and stripping from characteristic notch features in the transfer function magnitude spectrum caused by acoustic impedance discontinuities at delaminated layer interfaces.
  6. The system of claim 1, wherein the machine learning model detects abrupt pavement condition changes (utility cut settlement, base failure, frost heave damage) as step changes in the transfer function time series, triggering immediate maintenance alerts distinct from the gradual degradation tracking.
  7. The system of claim 1, further comprising a physics-informed synthetic data generation module that produces training datasets by simulating vehicle-pavement interaction using a dynamic load model, propagating loads through a layered viscoelastic half-space model, and converting computed strain fields to synthetic DAS measurements with realistic noise characteristics.
  8. The system of claim 1, wherein an edge computing unit co-located with the DAS interrogator performs real-time traffic event detection and per-event transfer function estimation, transmitting only extracted feature vectors to a cloud aggregation system, reducing data bandwidth by at least 99% relative to raw DAS waveform data.
  9. The system of claim 1, further comprising an automatic calibration module that identifies opportunities for ground-truth validation by detecting planned pavement surveys from agency work calendars and flagging high-uncertainty sections for targeted field verification.
  10. A method for continuously monitoring pavement structural condition using existing telecommunications infrastructure, comprising: connecting a distributed acoustic sensing interrogator to a dark fiber strand in an existing buried telecommunications cable running alongside a paved roadway; continuously measuring vibration waveforms along the fiber caused by passing traffic; detecting individual vehicle passages and estimating vehicle weight class and speed from the measured waveforms; computing a frequency-dependent vibration transfer function between estimated surface loads and measured fiber response for each vehicle passage at each measurement location; aggregating transfer function estimates over time windows to produce daily condition feature vectors; and applying a trained recurrent neural network to the time series of feature vectors to estimate current pavement structural condition index and predicted remaining useful life at each measurement location along the fiber route.
  11. The method of claim 10, wherein the transfer function features capture pavement deterioration mechanisms including fatigue cracking (reduced effective stiffness manifest as increased low-frequency vibration amplitude), rutting (increased surface wave to body wave energy ratio), stripping and debonding (notch features from acoustic impedance discontinuities), base failure (broadband amplitude increase), and moisture infiltration (increased high-frequency attenuation and reduced apparent wave velocity).
  12. The method of claim 10, further comprising seasonal normalization of transfer function features using pavement temperature estimated from a combination of air temperature records and a thermal model of the pavement cross-section, accounting for the strong temperature dependence of asphalt dynamic modulus.

Implementation Notes

The system can be deployed incrementally, starting with Interstate highway segments where dark fiber availability and pavement condition survey frequency are both high. The DAS interrogator represents the primary hardware cost, with commercial units from OptaSense, Silixa, and Fotech priced at $100,000-300,000, capable of monitoring 20-80 km of fiber from a single location. At a deployment density of one interrogator per 40 km of highway, the instrumentation cost is approximately $2,500-7,500 per lane-mile, compared to $2,000-5,000 per lane-mile for a single FWD survey campaign that provides only a snapshot rather than continuous monitoring. Over a 10-year service life, the DAS system's per-measurement cost approaches zero as the continuous data stream replaces periodic surveys.

The primary technical challenge is the lateral offset between the buried telecom fiber and the travel lanes. Telecom fiber is typically installed in highway shoulders, 3-10 meters from the nearest travel lane. Traffic-induced vibrations at these lateral offsets are attenuated by 10-30 dB relative to directly beneath the wheel path, but remain well above the DAS noise floor for heavy trucks (estimated SNR of 20-40 dB for Class 8 trucks at 5-meter lateral offset) and above the noise floor for most passenger vehicles (estimated SNR of 5-15 dB at 5-meter offset). The system preferentially uses heavy truck passages for condition estimation, as these provide both higher SNR and greater sensitivity to structural condition (the nonlinear, stress-dependent response of granular materials is more pronounced under heavy loads).

An alternative deployment configuration uses fiber installed directly within the pavement structure during construction or rehabilitation. Several state DOTs have experimented with embedding fiber optic sensors in pavement (Xiang and Wang, Construction and Building Materials 2020), but this approach requires fiber installation during construction and is not applicable to the existing road network. The telecom fiber approach described in this disclosure works with infrastructure already in the ground, enabling immediate deployment without road construction.

Estimated monitoring coverage based on FCC Broadband Map fiber routes and FHWA highway inventory data: approximately 60-75% of US Interstate highway lane-miles, 30-45% of US route lane-miles, and 15-25% of state route lane-miles have telecommunications fiber within the right-of-way. In total, an estimated 200,000-400,000 lane-miles of US highways could be monitored using existing buried fiber, covering the most heavily trafficked portions of the network where pavement condition has the greatest economic impact.

Prior Art References

  1. FHWA Highway Statistics 2022 – US road network mileage and pavement type statistics
  2. ASCE 2025 Infrastructure Report Card: Roads – Infrastructure condition grades and maintenance backlog estimates
  3. Chatti and Zaabar, Transportation Research Record 2012 – Vehicle operating costs on rough pavements
  4. TRIP National Transportation Research Group – Economic costs of poor road conditions to US drivers
  5. ASTM D6433-20 – Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys
  6. Timm et al., Transportation Research Record 2007 – Inter-rater variability in manual pavement condition surveys
  7. NCHRP Synthesis 573, 2021 – Falling Weight Deflectometer deployment costs and practices
  8. FCC Broadband Map, 2024 – Fiber optic route-mile estimates for US telecommunications infrastructure
  9. Mestayer et al., Proceedings of SPIE 2016 – DAS for pipeline leak detection
  10. Lindsey et al., Nature Communications 2019 – DAS for seismic monitoring using existing telecom fiber
  11. Hartog et al., Optics Express 2018 – DAS interrogator performance specifications and noise characteristics
  12. Martin et al., Scientific Reports 2020 – DAS-based traffic monitoring on existing telecommunications fiber
  13. Cebon, Transportation Research Record 1996 – Dynamic vehicle-road interaction and load spectra
  14. ISO 8608 – Mechanical vibration, road surface profiles, reporting of measured data
  15. Kausel and Roesset, International Journal of Rock Mechanics 1981 – Stiffness matrix method for layered media wave propagation
  16. Pavement ME Design (MEPDG) – Mechanistic-empirical pavement deterioration equations
  17. Xiang and Wang, Construction and Building Materials 2020 – Embedded fiber optic sensors in pavement structures