A portable potentiometric electronic tongue leveraging smartphone and cloud platforms

Patrick W. Ruch, Rui Hu, Luca Capua, Yuksel Temiz, Stephan Paredes, Antonio Lopez Marin, Jorge Barroso Carmona, Aaron Cox, Eiji Nakamura, Keiji Matsumoto

Introduction

Traditional chemical analysis, while precise, often relies on expensive, stationary laboratory equipment like mass spectrometers. There is a growing demand for portable "electronic tongues" (e-tongues) that can perform rapid, on-site food authentication and environmental monitoring. Researchers from IBM have developed a miniaturized, cloud-connected platform that simplifies sensor hardware by offloading complex data processing to cloud-based machine learning models.

System Architecture and Methods

The platform integrates a potentiometric sensor array with a smartphone interface and a cloud backend. The "tongue" utilizes four 2 mm circular electrode pads coated with polypyrrole (PPy), a conducting polymer. By adjusting electropolymerization conditions and dopants, the sensors are tuned for specific sensitivities to ions such as K⁺, Na⁺, and Cl⁻.

The end-to-end workflow from training on sample classes to cloud-based inference and classification results on a smartphone.

A key innovation is the reference-less design, which records differential voltages between sensors, eliminating the need for a bulky reference electrode. The hardware includes a custom 3D-printed package housing a 16-bit ADC and an Arduino-based microcontroller. Data is transmitted via USB On-The-Go (OTG) to an Android app, which relays time-series data to IBM Watson Machine Learning for classification using a Random Forest model.

Hardware components (c), sensor sensitivity plots (a, b), and the device in use (d).

Key Insights: The Role of Transients

The system leverages non-equilibrium potentiometry. The researchers discovered that the transient data captured within the first 20 seconds of immersion is more informative for classification than the final stabilized voltage. This insight allows for rapid testing cycles, with the entire process from immersion to result taking less than 60 seconds. By shifting the "intelligence" from the physical sensor to the cloud model, the hardware remains inexpensive and easily deployable.

Experimental Results

The system was evaluated using two primary datasets:

  • Beverages: The model achieved 95.3% accuracy in identifying three different soft drinks using leave-one-out cross-validation.
  • Mineral Waters: In a more challenging test involving four brands of still mineral water, the system achieved 61.7% accuracy. While lower than the beverage results due to subtle ppm-level ion differences, this significantly outperformed human testers, who averaged only 30.6% accuracy on the same samples.

Conclusion and Future Outlook

This hardware-to-cloud platform demonstrates a functional prototype for field-ready chemical fingerprinting. While the current four-electrode array has limitations in selectivity for low-concentration mineral waters, future iterations will scale the number of sensing elements to improve resolution and test environmental robustness over hundreds of uses.

Summary updated

BibTeX
@inproceedings{ruch2019portable,
  author = {Patrick W. Ruch and Rui Hu and Luca Capua and Yuksel Temiz and Stephan Paredes and Antonio Lopez Marin and Jorge Barroso Carmona and Aaron Cox and Eiji Nakamura and Keiji Matsumoto},
  title = {{A portable potentiometric electronic tongue leveraging smartphone and cloud platforms}},
  booktitle = {ISOEN},
  year = {2019},
  doi = {10.1109/isoen.2019.8823244},
}
Figure