Design and development framework
To develop the framework for a smart glove system, we closely worked with researchers in bio-robotics labs to understand the user needs and required performances. One major requirement was customization in the development of embedded sensor units and the glove shapes. To meet the user specifications, customization properties were added into the framework and achieved through digital fabrication technology.
Analysis of performance requirements and properties
Figure 2 presents the design and development framework of a seamless knitted smart glove using a digital knitting system modified from the wearable motherboard (Park and Sundaresan 2001; Rajamanickam et al. 1998). The user performance task for the wearable glove sensor system was to detect finger gestures and wrist motion of the human hand using embedded textile-based strain sensor units. The detailed and specific set of performance requirements are as follows: functionality, wearability, durability and stability, manufacturability, customized design, and connectivity. The functionality of the smart glove sensor system required that it be able to detect finger and wrist motion with the embedded sensor unit. Wearability required that it be lightweight, breathable, conformable to the skin, comfortable to wear, and easy to assemble with peripheral devices such as a circuit board. Another critical requirement was the durability to endure repeated deformation with a stable electrical signal. A wearable glove system will eventually be manufactured in large quantities over a range of sizes and designs. In particular, glove design elements should be programmable in the manufacturing process, such as the three-dimensional shapes for the hand, design variation for peripheral units, and the location, profile, and dimension of sensor units. Ultimately, the developed glove sensor should be able to connect to microcontroller units to compute the acquired data.
The identified essential performance requirements for the wearable glove sensor systems were then translated and categorized into appropriate sensing and comfort properties. Based on these properties, we determined specific design components for the wearable glove system (electrical conducting, comfort, form-fitting, electrical sensing, and customized design). These design components can be achieved through appropriate choice of materials and fabrication technologies.
Materials selection and specification
The selected materials and specifications for the glove sensor system are described in “Materials” section. For the electrical conducting component, silver-plated yarn with non-conductive yarn was chosen due to the excellent balance between conductivity and resistance. This yarn formed the piezoresistive sensor to detect the angle of the finger joint and wrist as a strain sensor. Not only the resistance of the materials, but also the knitted sensor geometry and knit structure affect bulk electrical properties. The comfort component is related to the total hand quality, such as softness, smoothness or roughness, air permeability, moisture absorption and stretchability. The texturized nylon selected for the base materials had irregular crimps on the filament and provided warmth and absorbency with trapped air, as well as being soft, flexible, and having high extensibility and excellent comfort properties. For form-fitting, elastomeric fiber was inserted both in the sensor unit and the base glove area, leading to improved form-fit for the user, and more importantly, placement of the sensor unit at the designated area during dynamic movement. While a knitted piece can be somewhat elastic due to the interlocking loop structure, the fabric does not return to its original shape quickly after stretching. To address this issue, we used spandex yarn to create an elastic fabric with low hysteresis, a high degree of stretchability, dimensional stability, durability, and shape retention ability.
Computerized knitting system for fabrication
Digital knitting technology was utilized for the development of the glove sensory system through a programmable and automated process for creating a seamless three-dimensional (3D) glove. The digital knitting machine consists of a two-bed system with multiple yarn carriers, which enable knitting of a two-layered tubular shape with various yarns. The knitting machine is combined with the CAD software, which allows a user to observe, select, and modify the dimensional structural design parameters with visualized results.
Figure 3 illustrates the automatic software, Apex3 (Shima Seiki), including the pre-set glove shapes, the detailed design options, and the manipulation of measurements. Detailed design selection like 3D shape or thumb type and size measurement was closely related to the form-fitting component. The embedded sensor component referred to the type of knit operations, including knit, tuck, transfer, or miss, as well as the sensor placement and configuration. Flat drawing tools for assigning knit structure and color were utilized for determining the embedded sensor component, such as sensor location, conductive path configuration, and knit structure. The knit structure applied for both the base glove and the sensor unit was single jersey, employing the plating technique, where the conductive yarn appears on the backside of the surface. For the sensor configuration, two different conductive profiles, a rectangular shape and a rectangular horseshoe shape, were developed and placed on the figure joint areas and wrist joint area. The customized design component included hand size adjustment for different users, and an opening on the hand back for easy assembly with a circuit board. Advanced design manipulation was limited to the pre-set shape offered by the automation software. However, the development of the machine programming language allowed for extended freedom in design.
Characterization of a knit textile sensor
First, we evaluated the electrical resistance on single-wale and two-wale types of embedded knitted sensor units, as well as the relative change in resistance with respect to the applied strain. Table 2 shows the average initial resistances of the fabricated glove sensors with conductive areas with one- and two-wale rectangular shapes. Since their initial resistances were different, further experiments were conducted to determine which one was more suitable for the glove sensors: producibility and sensitivity.
We connected electrical wires to the fabricated, seamless glove sensors along the wale direction, as shown in Fig. 4a. Since they were wired along the wale direction, each of the wales were electrically connected in parallel; the single-wale sensor showed a higher initial resistance R0 than that of the two-wale (Li et al. 2009; Xie et al. 2019).
Due to the specifications of the digital knitting machine, a knitted sensor with conductive yarns in two or more wales yielded a more stable output than knitting in a single wale, causing a smaller standard deviation of the average initial resistance. Although their initial resistances are different, their relative resistance changes from strain, i.e. sensitivities, showed similar characteristics, as seen in Fig. 4b. The relative resistance changes of the samples show a linear characteristic with a constant gauge factor. However, as the extended loops start to show more contact areas the relative resistance change undergoes gradual flattening, decreasing the gauge factor. Since the glove sensors with conductive areas in single- and two-wale rectangular shapes show similar sensitivities, all the further characterizations were conducted with the single-wale sample.
To investigate the electrical performance of the embedded knit sensor, the relative changes in resistance with respect to the applied strain and bending angle were measured. When the glove sensor was released from the exerted load, it exhibited nearly the same characteristics with the stretch case, as shown in the hysteresis loop in Fig. 5a under stretching and (b) under bending. These linear characteristics of resistance change were also observed in experiments with gradually increasing step loads, with the results presented in Fig. 5c for stretching and Fig. 5d for bending. By comparing the measured response data with the input load, the response time can be calculated, as shown in Fig. 5e and f for stretching and bending, respectively. The average response time for the developed glove sensor was found to be 130 ms.
Fabrication of seamless glove sensor prototypes
Based on the results from extensive analysis of the framework, we explored two seamless glove sensor systems for the human hand and robotic hand and evaluated the sensor characteristics. During the initial procedure to customize glove design, accurate and efficient hand dimensions from a user were extracted from 3D scanned images, which were captured from a 3D scanner (DRAKE, THOR3D, Russia). Similarly, robotic hand measurements can be obtained from the 3D drawing in the 3D CAD system shown in Fig. 6. A previous study proved that there is no significant difference in hand measurements between the 3D scanning method and direct measurement systems (Yu et al. 2013).
Table 3 shows the total measurements of five fingers for both human and robotic hands, including the vertical and horizontal dimensions of each finger and the additional width of three fingers, specified as the size input parameters in the digital knitting software. The size of the circuit box on the robotic hand back was determined.
On the 15-gauge knitting machine used, there were 15 needles/inch on the front and back beds, and the number of stitches was pre-set at 7.8 × 8.7/cm2 (wale × course) using the software. However, the actual number of stitches for the base glove was observed to be 9 × 14/cm2 (wale × course). The elastane yarns in the knit structure caused deformation of the loop shape due to their elasticity; therefore, the loop length became shorter, and the density, particularly in the course direction, increased dramatically (Sadek et al. 2012; Sitotaw 2018). The glove knitted with spandex provided a comfortable shape, fit, smoothness, elasticity, and excellent elastic recovery.
Figure 7 illustrates the sensor units in the automatic software, the virtually knitted simulations, and the physical prototype photographs for the human hand and robotic hand. In the CAD software, the sensor units were placed on the front side of the hand for detection of bending and motion in the five fingers and wrist. There are various conductive profiles possible for strain sensors, such as rectangular, oval, diamond or horseshoe shape. Among these cases, two of the most popular choices are introduced in Fig. 7: conductive profiles of a rectangular horseshoe shape for a human hand glove and a rectangular shape for a robotic hand. Extra conductive rectangles were also added on the palm as textile electrodes for attaching external pressure sensor units. By using the intarsia knitting technique, the various sensor profiles could be knitted with conductive yarns in the designated areas. It should be noted that automatic software processing was limited to the standard pre-set shapes; therefore, the opening for the circuit box on the back of the robotic hand with two-line electrodes was manipulated during the development stage (Fig. 3b, right image) with the machine programming language. The successfully developed opening with the additional textile electrodes are shown in the virtual simulation. Finally, seamless glove sensors for a human hand and robotic hand were fabricated with the digital knitting machine mentioned in “Fabrication of a smart glove using CAD/CAM” section.
Figure 8a and b show two prototype seamless glove sensor systems worn on the human hand and robotic hand. The intended application of the seamless glove sensor for the human hand is for the glove to be worn on one healthy hand in order to control a prosthetic hand on the opposite side (Sebelius et al. 2006). The seamless glove sensor for the robotic hand detects the robotic finger motion and acquires the applied force by moving the robotic finger. We observed that both gloves fit well in size and assembly with the circuit board for both human hand and robotic hands.
To evaluate the electrical signal of actual finger motion, we attached the wires to the finger sensor, which was connected to the circuit board with an Arduino microcontroller for detecting resistance change during finger motion. The electrical resistance of the change in angle related to the finger bending motion is presented in Fig. 8c and d. Although we used bulky crocodile clips for performance evaluation, the wires and the circuit board can be embedded within the glove when it is ready for out-of-the-lab deployment or mass production.