# Real-Time Embedded Smart-Particle Monitoring for Index-Based Evaluation of Asphalt Mixture Compaction Quality

## TL;DR

A smart particle embedded in asphalt mixtures can monitor compaction quality in real time, improving pavement durability and construction efficiency.

## Contribution

A novel embedded smart-particle system with real-time compaction evaluation using nonlinear machine learning models and orientation features.

## Key findings

- The smart particle remains stable at 165 °C and captures densification phases through attitude-angle evolution.
- Nonlinear models using orientation features achieved high accuracy (R2 > 0.960) in reconstructing real-time compaction degree.
- Machine learning analysis revealed mixture-type-dependent features in compaction-state transmission.

## Abstract

What are the main findings?
The embedded smart particle remains stable at 165 °C and effectively quantifies the “rapid-to-slow” densification phases via attitude-angle evolution.Nonlinear machine learning models using orientation features successfully reconstructed the real-time compaction degree with high accuracy (R2 > 0.960).

The embedded smart particle remains stable at 165 °C and effectively quantifies the “rapid-to-slow” densification phases via attitude-angle evolution.

Nonlinear machine learning models using orientation features successfully reconstructed the real-time compaction degree with high accuracy (R2 > 0.960).

What are the implications of the main findings?
This method enables continuous in situ monitoring of pavement quality, solving the time lag problem of traditional post-compaction testing.The proposed sensing framework provides a reliable data foundation for real-time feedback control in intelligent transportation infrastructure construction.

This method enables continuous in situ monitoring of pavement quality, solving the time lag problem of traditional post-compaction testing.

The proposed sensing framework provides a reliable data foundation for real-time feedback control in intelligent transportation infrastructure construction.

Compaction quality governs asphalt pavement durability, but conventional density checks are intermittent. Reliable compaction control of asphalt mixtures requires real-time information on internal responses rather than relying solely on endpoint density measurements. In this study, an embedded smart-particle framework is developed for in situ monitoring and index-based evaluation of vibratory compaction quality, integrating multi-source sensing, feature extraction, and compaction degree mapping. The smart particle integrates inertial/orientation sensing together with thermal–mechanical measurements, and its high-temperature survivability and calibratability are verified through thermal exposure and calibration tests. During laboratory vibratory compaction of representative asphalt mixtures, raw signals are converted into stable attitude responses via attitude estimation and filtering; posture-dominant descriptors are then extracted and used to establish a data-driven mapping from internal responses to compaction degree using regression models. Results show that the device remains stable under typical hot-mix asphalt conditions, with calibration exhibiting high linearity (temperature channel R2 > 0.990; force channel R2 > 0.980 in the relevant range). Filtering markedly enhances inertial-signal usability under strong vibration and improves the interpretability of attitude-response evolution during compaction. The evolution of attitude features is consistent with the “rapid-to-slow densification” process, yielding correlations of |r| ≈ 0.35–0.47 with compaction degree evolution. Nonlinear regressors outperform linear baselines, and the better-performing nonlinear models achieve strong predictive performance across all six specimens, with R2 values reaching 0.740–0.960 and RMSE reaching 0.016–0.043. Moreover, machine-learning-based feature-importance analysis reveals distinct mixture-type-dependent characteristics, indicating that AC and SMA transmit compaction-state information through partly different dominant response features. These findings demonstrate the feasibility of embedded smart particles for online compaction-quality evaluation and provide a basis for real-time feedback in intelligent compaction.