A Proximity-Based Approach for Dynamically Matching Industrial Assets and Their Operators Using Low-Power IoT Devices

Silvano Cortesi, Michele Crabolu, Prodromos-Vasileios Mekikis, Giovanni Bellusci, Christian Vogt, Michele MagnoIEEE Internet of Things Journal 2025 · DOI

TL;DR

This paper introduces a novel, low-power IoT system for dynamically matching industrial assets (like power tools) with their operators. By leveraging Bluetooth Low Energy (BLE) and an Extended Kalman Filter (EKF) for distance estimation, the system achieves high accuracy in identifying who is using which tool, even in dynamic environments, with minimal power consumption.

What Problem Does This Paper Address?

While asset tracking systems are becoming common in industries like construction, a significant gap remains in precisely identifying which operator is using which asset at any given time. This lack of granular information hinders organizations from gaining valuable insights into asset utilization, operator safety, and maintenance needs. Existing solutions often rely on complex and power-hungry real-time location systems (RTLS) or biometric readers, which are not always feasible or cost-effective in dynamic industrial settings. The paper aims to fill this gap with a more practical and energy-efficient approach.

Methods

The proposed system comprises three main components:

  1. Sensor Tags: Low-power BLE devices attached to or embedded in assets (e.g., power tools). These tags are equipped with accelerometers and use a TinyML algorithm to detect the asset's activity status (e.g., if a power tool's motor is running). They periodically broadcast this status.
  2. Wearable Devices: BLE-enabled devices worn by operators. These devices scan for BLE advertisements from sensor tags. Upon detecting an active asset, they run an Extended Kalman Filter (EKF) algorithm to estimate the distance between the operator and the asset using Received Signal Strength Indicator (RSSI) measurements.
  3. Cloud-Based Algorithm: The wearable devices transmit estimated distances to a central server. Here, a cloud-based algorithm processes these distances to perform asset-user matching. This algorithm formulates the problem as an optimization task to assign the nearest operator to each active asset, considering constraints like an operator not being assigned to multiple assets simultaneously.

The core innovation lies in using an EKF for distance estimation on the low-power wearable device, balancing accuracy with computational and energy constraints. The asset-user matching is then handled by an optimization algorithm on the server.

Datasets

The paper does not explicitly mention training datasets. However, it describes extensive real-world validation experiments conducted in both indoor and outdoor construction settings. These experiments involved:

  • Ground Truth Data: Collected using an OPTITRACK system for precise distance measurements.
  • Video Recording: A NIKON Z6 II camera was used to record tests for sanity checks.
  • RSSI Measurements: Collected from BLE advertisements broadcast by sensor tags.

The experiments were designed to simulate realistic industrial scenarios, including various numbers of workers, tools, and changing distances.

Key Insights and Ideas

  • Proximity as the Key: The fundamental insight is that the operator of a handheld asset is most likely the one closest to it when it's active. This simplifies the problem from precise localization to relative distance estimation.
  • EKF for Edge Computing: Employing an Extended Kalman Filter on the wearable device is crucial. It allows for real-time distance estimation with improved accuracy over raw RSSI, while remaining computationally feasible for low-power devices.
  • Energy Efficiency is Paramount: The system prioritizes low power consumption. BLE tags are designed for multi-year battery life, and the wearable device's data transmission is minimized (daily updates) to extend its battery life significantly.
  • Dynamic Matching is Achievable: The system demonstrates effectiveness in dynamic scenarios where workers might change tools, proving its adaptability to real-world industrial workflows.
  • Trust Level for Confidence: The asset-user matching algorithm provides a "trust level" (SURE/UNSURE), indicating the confidence in the assigned match. This is vital for understanding the reliability of the system's outputs.

Key Results

  • Distance Estimation Accuracy: The system achieved a median distance estimation error of 0.49 m.
  • Asset-User Matching Accuracy:
    • Indoor environments: 89.7% overall accuracy.
  • Outdoor environments: 98.6% overall accuracy.
  • High Confidence Matches: In the indoor setup, 70.8% of correct matches were classified with high confidence ("SURE"). In the outdoor setup, this rose to 95.8% recall and 100% precision for "SURE" matches.
  • Power Consumption: The wearable device (ECOTRACK) has an average power consumption of 16.06 mW, allowing for a runtime of approximately 28 days on a 10800 mWh battery. BLE tags are estimated to last over three years.

Limitations and Future Work

  • Indoor vs. Outdoor Performance: While the system performs exceptionally well outdoors, the accuracy in indoor environments is slightly lower. The authors suggest that addressing uncertainty, especially in indoor scenarios, is a focus for improvement.
  • Environmental Sensitivity: The paper acknowledges that the path-loss model's accuracy can be influenced by environmental factors, and the current model is optimized for the tested construction settings.
  • Future Work: The authors suggest extending testing to different environments and conducting a sensitivity analysis on the algorithm's tuning parameters to validate its robustness and generalization capabilities.

Summary updated

BibTeX
@article{cortesi2025proximity,
  author = {Silvano Cortesi and Michele Crabolu and Prodromos-Vasileios Mekikis and Giovanni Bellusci and Christian Vogt and Michele Magno},
  title = {{A Proximity-Based Approach for Dynamically Matching Industrial Assets and Their Operators Using Low-Power IoT Devices}},
  journal = {IEEE Internet of Things Journal},
  year = {2025},
  doi = {10.1109/jiot.2024.3479458},
}
Figure