SmarTEG: An Autonomous Wireless Sensor Node for High Accuracy Accelerometer-Based Monitoring

Michele Magno, Lukas Sigrist, Andres Gomez, Lukas Cavigelli, Antonio Libri, Emanuel M. Popovici, Luca Benini

TL;DR

This paper introduces SmarTEG, an energy-neutral wireless sensor node for monitoring industrial machinery wear. By combining ultra-low-power hardware, thermal energy harvesting, and an efficient on-board algorithm, SmarTEG achieves high accuracy in detecting blade wear on band saws while operating perpetually without external power.

What Problem Does This Paper Address?

Industrial machinery relies heavily on efficient operation and timely maintenance to minimize energy consumption and prevent costly failures. Traditional wired sensor networks are expensive and difficult to deploy in harsh industrial environments. Existing wireless sensor networks (WSNs) often suffer from limited battery life, necessitating frequent replacement or recharging. This paper tackles the challenge of creating a truly autonomous, low-cost WSN capable of continuous, high-accuracy monitoring of machinery health, specifically focusing on tool wear detection in band saws, to improve energy efficiency and reduce operational costs.

Methods

The SmarTEG node is built around a low-power microcontroller (MSP430) and a Zigbee PRO radio (CC2530). Key innovations include:

  • On-board Processing: Instead of transmitting raw sensor data, the node processes accelerometer data locally to detect wear. This significantly reduces the amount of data transmitted, saving considerable energy.
  • Low-Complexity Algorithm: A Mean Absolute Value (MAV) algorithm is employed for wear detection. This is chosen over computationally intensive methods like FFT to be feasible on the low-power microcontroller. The algorithm calculates the MAV of vibration data and compares it to pre-defined thresholds.
  • Energy Harvesting: A thermoelectric generator (TEG) is used to harvest ambient thermal energy from the machinery. This harvested energy, managed by an LTC3107 power management chip, powers the node and charges a supercapacitor, aiming for energy neutrality.
  • Ultra-Low Power Wake-Up Radio (WUR): A nano-power WUR receiver is integrated to wake the node only when necessary, drastically reducing the power consumption associated with constant radio listening. The WUR consumes only 600 nW.
  • Adaptive Duty Cycling & Power Management: The system employs deep-sleep modes and intelligent power management to minimize energy usage, activating components only when needed for sensing, processing, or transmission.

Datasets

The paper does not explicitly mention pre-existing datasets for training or evaluation. Instead, the authors collected their own dataset by performing cutting tests on a band saw using five different materials (aluminum, brass, nylon, steel tube, and steel rod) with three different blades (new, medium, and old). This dataset was used to train the wear detection algorithm off-line on a PC and then to evaluate the performance of the on-board algorithm.

Key Insights and Ideas

  • Processing is Key to Energy Autonomy: The paper strongly emphasizes that performing data processing on the node itself, rather than transmitting raw data, is crucial for achieving energy neutrality in WSNs. This is because radio transmission is orders of magnitude more energy-intensive than on-board processing.
  • MAV Algorithm's Sweet Spot: The MAV algorithm, while simpler than FFT, offers a compelling trade-off between computational cost and accuracy for this specific application, making it suitable for resource-constrained nodes.
  • TEG Viability for Industrial Settings: The study demonstrates that TEGs can be a viable energy harvesting solution for industrial machinery, especially where constant heat sources are present, outperforming solar or vibration harvesting in certain scenarios due to their independence from light and low-frequency vibration limitations.
  • Wake-Up Radio's Critical Role: The nano-power wake-up radio is highlighted as a critical component for enabling energy-neutral operation by allowing the main radio and microcontroller to remain in deep sleep for extended periods.
  • Energy Buffer for Cold Starts: The simulations underscore the importance of an energy buffer (supercapacitor) to manage the initial "cold start" phase when the TEG may not yet be generating sufficient power.

Key Results

  • 100% Classification Accuracy: The proposed MAV algorithm, when using four consecutive evaluations, achieved 100% classification accuracy for distinguishing between new, medium, and old blades across various materials. Even a single MAV evaluation achieved high accuracy (up to 77% in the worst case), offering a trade-off for even lower energy consumption.
  • Energy Neutrality Achieved: Experimental results and simulations demonstrate that the SmarTEG node can achieve energy neutrality, operating perpetually by harvesting enough energy from the TEG to power its sensing, processing, and occasional communication tasks.
  • Low Power Consumption: The node's power consumption in deep sleep mode is as low as 1 µW, with active modes consuming significantly more but still within the capabilities of the energy harvesting system.
  • Efficient Energy Harvesting: The TEG system, despite conversion losses, was shown to harvest sufficient power (around 1-2 mW in real operating conditions) to sustain the node's operation.

Limitations and Future Work

  • TEG Efficiency and Temperature Dependence: The TEG's harvested power is highly dependent on the temperature gradient. While industrial machinery provides a consistent heat source, achieving optimal gradients for maximum power might require careful placement. The paper notes that the harvesting circuit is not able to harvest energy if the TEG input voltage is below 63 mV.
  • Training Phase: The wear detection algorithm requires an off-line training phase to determine optimal thresholds for different materials. This needs to be performed once per material type.
  • Limited Communication: The system prioritizes energy neutrality, meaning communication is infrequent and only transmits the wear status, not raw data. This limits its use for real-time, high-bandwidth data streaming.
  • Future Work: The authors suggest that the SmarTEG node can be adapted for other vibration analysis tasks by tuning the algorithm's thresholds. Further exploration of multi-source energy harvesting and more advanced power management techniques could also enhance performance. The simulations also highlight the potential for more conservative energy accumulation by reducing radio activity further.

Summary updated

Abstract

We report on a self-sustainable, wireless accelerometer-based system for wear detection in a band saw blade. Due to the combination of low power hardware design, thermal energy harvesting with a small thermoelectric generator (TEG), an ultra-low power wake-up radio, power management and the low complexity algorithm implemented, our solution works perpetually while also achieving high accuracy. The onboard algorithm processes sensor data, extracts features, performs the classification needed for the blade’s wear detection, and sends the report wirelessly. Experimental results in a real-world deployment scenario demonstrate that its accuracy is comparable to state-of-the-art algorithms executed on a PC and show the energy-neutrality of the solution using a small thermoelectric generator to harvest energy. The impact of various low-power techniques implemented on the node is analyzed, highlighting the benefits of onboard processing, the nano-power wake-up radio, and the combination of harvesting and low power design. Finally, accurate in-field energy intake measurements, coupled with simulations, demonstrate that the proposed approach is energy autonomous and can work perpetually.
BibTeX
@inproceedings{magno2019smarteg,
  author = {Michele Magno and Lukas Sigrist and Andres Gomez and Lukas Cavigelli and Antonio Libri and Emanuel M. Popovici and Luca Benini},
  title = {{SmarTEG: An Autonomous Wireless Sensor Node for High Accuracy Accelerometer-Based Monitoring}},
  booktitle = {Sensors},
  year = {2019},
  url = {https://openreview.net/forum?id=D6CBpgcMrV},
}

Loading related content...

Figure