Innovations in fiber-based wearable sensors using machine learning

Innovations in fiber-based wearable sensors using machine learning
by Simon Mansfield
Sydney, Australia (SPX) Aug 26, 2024
The last decade's swift advancements in artificial intelligence have significantly bolstered the capabilities of wearable devices in handling intricate data. Machine learning, a key subset of AI algorithms, and specifically deep learning, have been central to this technological surge. Machine learning reduces the need for manual data feature extraction, while deep learning excels at identifying hidden patterns. Both require vast amounts of data, a demand well-suited to today's era of information overload.

This article reviews the machine learning algorithms that have been successfully integrated with fiber sensors, categorizing them into traditional machine learning methods and deep learning techniques. Traditional algorithms include linear regression (LR), k-nearest neighbors (KNN), support vector machine (SVM), random forest, XGBoost, and K-means clustering.

The article also categorizes fiber sensors based on their operational principles and sizes, as depicted in Figure 3. The operational principles fall into two main categories: optical and electrical. Optical sensors include Fiber Bragg Grating (FBG), Fabry-Perot interferometers, Specklegrams, and light intensity sensors, while electrical sensors encompass piezoresistive, triboelectric, electromyography (EMG), and chip-in-fiber technologies.

Fiber sensors present a viable alternative to rigid electronic devices for everyday use, particularly when combined with machine learning, enabling the creation of smart clothing. However, significant challenges remain. Most current fiber sensors utilizing machine learning focus on capturing a single type of signal, typically related to mechanical force and deformation-such as pressure-based gesture recognition in gloves. Other valuable data, like light intensity, color, temperature, humidity, and surface roughness, are often not integrated. Additionally, as machine learning continues to evolve rapidly, newer algorithms like reinforcement learning, generative adversarial networks (GANs), self-supervised learning, and attention mechanisms (e.g., GPT) have seen limited application in this field. As research progresses in these areas, it is anticipated that fiber sensor-based wearable devices, enhanced by artificial intelligence, will become more intelligent, comfortable, and efficient, making their way into everyday life.

Research Report:Advances in Fiber-Based Wearable Sensors with Machine Learning