Soft Polymer Optical Fiber Sensors for Intelligent Recognition of Elastomer Deformations and Wearable Applications

Nicheng Wang, Yuan Yao, Pengao Wu, Lei Zhao, Jin-Hui ChenSensors 2024 · DOI

TL;DR

This paper introduces a novel soft polymer optical fiber (POF) sensor system capable of accurately detecting multi-dimensional deformations in elastomers. By combining POF sensors with machine learning, the system achieves high-precision recognition of pressing, bending, and twisting, paving the way for more sophisticated soft robots and wearable devices.

What Problem Does This Paper Address?

Soft robotics is a rapidly growing field, with soft robots offering unique advantages in applications requiring compliance and adaptability, such as in confined spaces or for delicate manipulation. A critical challenge in soft robotics is endowing these robots with effective perception capabilities to understand their own deformations. Traditional electronic sensors, while mature, often struggle in harsh environments (humidity, chemicals, electromagnetic interference) and can be bulky. While optical fiber sensors offer advantages like immunity to electromagnetic interference and high bandwidth, existing solutions often focus on single-axis deformations or require complex, expensive sensing arrays. This paper addresses the need for a robust, scalable, and cost-effective soft sensing solution that can simultaneously detect multiple types of deformation in elastomers, crucial for advanced proprioception in soft robots and for applications like wearable sensing and prosthetics.

Methods

The core of this work is a soft polymer optical fiber (POF) sensor system. POF sensors are fabricated by cleaving and polishing commercial POFs, then connecting them with a small air gap and encapsulating them in silicone rubber. This design leverages the principle that deformation of the elastomer alters the light coupling between the POF endfaces, modulating the transmitted light intensity.

The key innovation lies in how these POF sensors are utilized:

  • Sensing Array: Multiple POF sensors are arranged in an array within a silicone elastomer.
  • Image-Based Detection: Instead of individual photodetectors for each fiber, a single commercial image sensor (camera) captures the light output from the entire array. This significantly reduces cost and complexity, enabling scalability.
  • Machine Learning for Recognition: The system employs machine learning algorithms, primarily Random Forest, to interpret the complex patterns of light intensity changes from the sensor array. This data-driven approach allows for the recognition of multi-dimensional deformations without relying on explicit physical models.
  • Multi-Modal Deformation: The system is designed to detect strain, bending, and twisting. For pressing, a 2D indentation position is mapped. For combined deformations, the machine learning model learns the correlations between different types of strain.

Datasets

The paper describes experiments that generate datasets for training and evaluating the machine learning models.

  • Pressing Recognition: A 13x13 grid of indentation points was used to generate data. The system was evaluated using a 3x3 sensor array, with 50 output fiber images collected per indentation position.
  • Multi-Dimensional Deformation: Data was collected for combined bending/stretching and twisting. The elastomer was subjected to manual displacements and rotations, with sensor responses recorded.
  • Hand Gesture Recognition: A dataset of 16 distinct hand gestures was collected. For each gesture, 100 images were captured, resulting in a total of 1600 images. 80% of this dataset was used for training, and 20% for validation.

No specific benchmark datasets were introduced; rather, the authors generated their own datasets based on the experimental setups described.

Key Insights and Ideas

  • Scalability through Imaging: Using a single camera to read out multiple POF sensors is a significant step towards cost-effective and scalable soft sensing systems, avoiding the need for numerous individual sensors or complex optical components.
  • Data-Driven Deformation Recognition: The success of the Random Forest algorithm in accurately recognizing complex, multi-dimensional deformations highlights the power of machine learning in interpreting optical sensor data, overcoming the limitations of traditional physics-based models for soft materials.
  • Anisotropic Pressing Response: The observation that pressing along the axial direction of the POF shows higher sensitivity than the orthogonal direction provides valuable insight for sensor placement and interpretation in tactile sensing applications.
  • Robustness of POF Sensors: The demonstrated stability and repeatability of the POF sensors over numerous cycles for strain and bending indicate their suitability for long-term, dynamic applications.

Key Results

  • 2D Indentation Position Recognition: The intelligent sensing system achieved a prediction accuracy of approximately 99.17% for recognizing 2D indentation positions.
  • Combined Strain and Twist Recognition: The average prediction accuracy for combined strain and twist deformations was 98.4% using the Random Forest algorithm.
  • Hand Gesture Recognition: An integrated intelligent glove achieved a high recognition accuracy of 99.38% for 16 different hand gestures.
  • Strain Sensitivity: Approximately -0.2 (normalized intensity change per unit strain).
  • Bending Sensitivity: Approximately -2.1%/° in the linear response range (30–60°).
  • Durability: The sensors demonstrated stable responses over 677 cycles for 100% strain and 1000 cycles for bending.

Limitations and Future Work

  • Curved UV Glue Breakage: The authors note that the repeatability of the sensors is spoiled when strain reaches 200% due to the breakage of the UV glue. This suggests a need for more robust encapsulation methods for extreme strain applications.
  • Bending Sensitivity at Small Angles: The sensor is less sensitive to small bending angles (< 20°), which is attributed to the POF's numerical aperture and gap distance. While tuning the air gap can improve sensitivity, it may compromise the effective angle range.
  • Footprint of Detection System: While the paper demonstrates an integrated, lightweight detection system for the glove, the initial experimental setups used a bulky microscope imaging system. Future work could focus on further miniaturizing the entire sensing and detection unit.
  • Dataset Expansion for Gestures: The authors envision that with more datasets of hand gestures, the intelligent glove could recognize and model a wider range of hand motions with even greater precision.
  • Broader Applications: The work holds promise for applications beyond soft robots and gloves, including human prosthetics and advanced human-machine interaction interfaces, which could be explored further.

Summary updated

BibTeX
@article{wang2024soft,
  author = {Nicheng Wang and Yuan Yao and Pengao Wu and Lei Zhao and Jin-Hui Chen},
  title = {{Soft Polymer Optical Fiber Sensors for Intelligent Recognition of Elastomer Deformations and Wearable Applications}},
  journal = {Sensors},
  year = {2024},
  doi = {10.3390/s24072253},
}
Figure