基于石墨烯和聚氨酯海绵柔性触觉手套，可穿戴设备和电子皮肤领域

摘要：研究提出人工智能技术的飞速发展推动柔性触觉传感器在多个领域获得了广泛的应用前景。柔性触觉传感器可将有源动态触觉传感信号转换为数字信号，通过使用机器学习方法分析数字信号，提供实时洞察和预测能力。本文，安徽建筑大学王菲露教授在《ACS Appl. Electr

1成果简介

研究提出人工智能技术的飞速发展推动柔性触觉传感器在多个领域获得了广泛的应用前景。柔性触觉传感器可将有源动态触觉传感信号转换为数字信号，通过使用机器学习方法分析数字信号，提供实时洞察和预测能力。本文，安徽建筑大学王菲露教授在《ACS Appl. Electron. Mater》期刊发表名为“A Haptic Glove with Flexible Piezoresistive Sensors Made by Graphene and Polyurethane Sponge for Object Recognition Based on Machine Learning Methods”的论文，研究介绍了一种利用多孔海绵结构制造柔性压阻传感器的低成本高效策略。

所制备的基于聚氨酯（PU）海绵和石墨烯的柔性压阻传感器具有优异的性能，例如灵敏度高（0-55 kPa 压力下为 1.7356 kPa-1）、响应/恢复时间快（147 ms/59 ms）、滞后误差小（6.51%）以及重复性稳定（可进行 2000 次循环压力测试）。由于该传感器的灵敏度范围广，设计过程快速、成本效益高，因此非常适合用于可穿戴设备。因此，利用柔性压阻传感器设计了一种用于物体识别的触觉手套。通过佩戴触觉手套，1500 组时间序列信号在 15 个不同物体的抓取过程中被精确检测和收集。然后，通过触觉手套检测到的触觉时间序列信号，构建了具有强大特征提取和泛化能力的残差网络（ResNet）来识别这 15 个物体，相应的识别准确率达到 95.67%。这项工作将柔性触觉传感器与机器学习方法相结合，为柔性触觉传感器在更多创新应用中提供了一种有效方法。

2图文导读

图1、Sensor and array preparation process. (a) Preparation process of PU sponge with graphene conductive nanoparticles and sensor structure. (b) Preparation process of sensor array with PDMS as substrate.

图2. Comparison of the properties of three conductive materials and electron micrographs. (a) Variation of relative resistance with increasing pressure for sensors prepared from three conductive materials. (b) Variation of sensitivity with increasing pressure for sensors prepared from three conductive materials. (c) Scanning electron microscopy of the pressure-sensitive layer at different magnifications.

图3. Sensing mechanism and bending test. (a) Sensing mechanism of the pressure sensor. (b) Sensing mechanism of strain sensor. (c) Change in relative resistance at different bending angles.

图4. Piezoresistive sensor performance test. (a) Optical photograph and schematic of the flexible PU/Graphene sensor. (b) Resistive response of the sensor to pressure in the range of 0–100 kPa. (c) Response and recovery time of the sensor to a 100 g weight pressure. (d)Loading/unloading curve of the transducer in the range 0–100 kPa. (e) Repeatability of sensor resistance response to pressures of 10, 20, and 30 kPa. (f) Repeatability of sensor resistance in response to frequencies of 0.5, 1, and 2 Hz. (g) The variation of relative resistance values of sensors in different temperature and humidity environments. (h) Stability of the pressure transducer over 2000 loading/unloading cycles, the inset shows the current response curves of the transducer during the initial (left) and later (right) phases of the test.

图5. Various applications of piezoresistive sensors. (a) Voltage response of a multiple-press sensor. (b) Pressing the sensor indicates the Morse code for the word “SENSOR”. (c) The fingertips of the attached sensor pinch a glass and slowly add water to the glass. (d) Voltage response when bending the sensor. (e) Voltage response when twisting the sensor. (f) Response of the sensor fixed at the finger joint when the finger is bent. (g) Response of the sensor fixed at the inner finger joint when the finger is bent. (h) Response of the sensor fixed at the inside of the wrist when the wrist is bent.

图6. piezoresistive sensing array design and application. (a) Optical photograph of the flexible PU/Graphene sensor array. (b) Schematic and structural view of the flexible PU/Graphene sensor array. (c-e) Placing a square, semicylinder, and triprism on the pressure sensor array. (f-h) Voltage response of each sensing unit.

图7. Haptic gloves based on piezoresistive sensors. (a) Schematic of the distribution of sensors in the haptic glove. (b) (I–III) Schematic diagram of the action decomposition of the three stages of contact-grip-grip during grasping an object, (IV) Voltage response of ten sensors during grasping an object. (c) Schematic diagram of the ResNet-18 network structure. (d) Confusion matrix for 15 object classifications. (e)Radar chart comparing the accuracy of multiple algorithms. (f) Confusion matrix for categorization by object shape. (g) Confusion matrix for classifying objects according to their softness.

3小结

在本研究中，我们比较了 MWCNTs、PSS: PEDOT 和石墨烯三种导电材料制备的压敏层的特性，最终选择石墨烯作为制备柔性压阻传感器的导电材料。制备的传感器灵敏度为 1.7356 kPa-1（0-55 kPa），响应时间为147毫秒，恢复时间为 59 毫秒，最大滞后误差为 6.51%，重现性良好（2000 次循环压力测试）。在不同的频率、压力、温度和湿度条件下，它都能表现出相对稳定的输出。基于出色的检测性能，该传感器被用于检测人体关节弯曲运动和外部压力变化，而基于该传感器设计的压力传感器阵列可以区分物体的形状并检测压力的分布。此外，还开发了一种基于柔性压阻传感器的触觉手套，通过记录抓取物体过程中的触觉时间序列信号，并利用 ResNet-18 模型提取和识别每个物体的独特特征，最终实现了对 15 种不同形状和软硬物体的有效识别，识别准确率达到 95.67%，单一属性的识别准确率分别达到 97.33% 和 97.99%。总之，这项研究为运动检测、可穿戴电子设备和人机交互系统的未来发展提供了一条简单而经济的途径。

文献：