关键词: 电子纺织品, 机器学习, 人机交互, 运动分析, 健康监测, 柔性传感器, 智能纺织品

Abstract:

Significance E-textiles, which integrate sensing, communication, and interactive capabilities into fabrics through embedded flexible electronics, are heralding a transformative shift in the textile industry. These smart textiles have the potential to revolutionize various domains, including human-computer interaction, motion analysis, and health monitoring, by offering versatile applications that go beyond conventional garment functionality. However, the widespread adoption of e-textiles is currently hindered by several challenges. Key barriers include the complexity of adapting machine learning (ML) models to meet the demanding computational constraints of embedded devices, the difficulty of generalizing models across a diverse range of users, and the high costs associated with optimizing both materials and production processes. Despite the significant potential of e-textiles, existing research has yet to systematically explore the inherent relationships between ML models and the various e-textile structures and applications. The matching of specific ML models to the unique characteristics of e-textile structures and their intended use cases remains an unresolved challenge. By addressing this gap, the review seeks to offer valuable insights that will inform the development of intelligent, efficient, and deployable textile systems, facilitating their transition from lab prototypes to practical, real-world applications.

Progress The review examines the evolving role of machine learning in e-textiles along three key dimensions: past developments, current trends, and future directions. Historically, the application of ML models in e-textiles has been shaped by the interplay between model characteristics, the structural properties of e-textiles, and the requirements of different application scenarios. The review discusses how specific ML models have been adapted to address various needs, including human-computer interaction, motion analysis, and health monitoring. For each of these applications, the strengths and weaknesses of the algorithms are critically assessed, providing a comprehensive understanding of their effectiveness in real-world contexts. Additionally, the review highlights a series of documented cases and statistically analyzes the correlations between e-textile structures, the best-performing ML models, and application-specific requirements. These relationships are visually represented through Sankey diagrams, illustrating how the structural characteristics of e-textiles influence the selection of appropriate ML models for different application contexts. The review also identifies key trends currently shaping the field, including the integration of ML in signal processing, process optimization, and scenario-based implementations. By analyzing the temporal evolution of relevant literature, the review offers a snapshot of ongoing advancements and the trajectory of these developments.

Conclusion and Prospect The review concludes that the successful integration of ML models in e-textiles is largely dependent on the specific use case, data characteristics, and the inherent structure of the textile system. ML models like CNNs, LSTMs, and SVMs have demonstrated significant progress in enhancing performance across key application areas, including human-computer interaction, motion analysis, and health monitoring. CNNs, for example, excel in handling spatially rich data, such as pressure distribution images, making them ideal for applications like gesture recognition and handwriting analysis. LSTMs are particularly effective at modeling temporal dependencies, which is crucial for applications involving continuous motion signals. SVMs, on the other hand, offer efficiency and robustness, particularly in scenarios with limited data and well-defined features, making them a popular choice for health monitoring applications. The field is experiencing three major shifts: the movement from simple classification tasks to the decoupling of multi-modal signals in signal processing; the transition from experience-based trial-and-error approaches to model-driven optimization in manufacturing; and the evolution from generalized monitoring to personalized, precision services in application contexts. Despite these advancements, challenges remain, such as the need for lightweight models that can operate within the computational constraints of embedded devices, the difficulty of ensuring model generalization across diverse user populations, and the high cost of optimizing performance. Future research should focus on developing more efficient, lightweight models, enhancing model generalization with diverse and large datasets, and incorporating cost considerations into the design and optimization of ML-driven e-textiles. By addressing these challenges, the review lays a foundational framework for matching ML models with e-textile applications, which will be crucial in driving the intelligent development and industrial scalability of this emerging technology.

Key words: electronic textiles, machine learning, human-computer interaction, motion analysis, health monitoring, flexible sensor, smart textiles