72 lines
44 KiB
JSON
72 lines
44 KiB
JSON
[
|
||
{
|
||
"id": 1,
|
||
"chunk": "# Self-Powered Machine-Learning-Assisted Material Identification Enabled by a Thermogalvanic Dual-Network Hydrogel with a High Thermopower \n\nYunsheng Li, Wenxu Wang, Xiaojing Cui, Ning Li, Xueliang Ma, Zhaosu Wang, Yuyou Nie, Zhiquan Huang,\\* and Hulin Zhang\\* \n\nWearable devices equipped with high-performance flexible sensors that can identify diverse physical information free from batteries are playing an indispensable role in various fields. However, previous studies on flexible sensors have primarily focused on their elasticity and temperature-sensing capability, with few reports on material identification. In this paper, a thermogalvanic dual-network hydrogel is fabricated with $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-/4-}$ as a redox couple and lithium magnesium silicate, $\\mathbf{C}\\mathbf{d}\\mathbf{m}^{+}$ and lithium bromide as key electrolytes to optimize the interconnected porous structure of the gel, which shows excellent mechanical and thermoelectric properties with a thermopower as high as $4.01\\mathrm{\\mV}\\mathsf{K}^{-1}$ . A self-powered material identification ring is developed based on the temperature-triggered thermoelectric response of the gel in conjunction with machine learning, which can actively infer materials without an external power connection by analyzing the voltage signals correlated with interfacial heat transfer produced upon contact with different materials. The proposed gel ring has important applications for future areas such as human–computer interaction and haptic-associated artificial intelligence.",
|
||
"category": " Introduction"
|
||
},
|
||
{
|
||
"id": 2,
|
||
"chunk": "# 1. Introduction \n\nThe sense of touch is one of the most important ways humans understand the world around them. When humans touch different materials, they can often make general inferences about the type of materials by sensing the temperature.[1] With the advent of the information age, people’s living standards improved with advanced electronics. Expectations for tactile sensors, such as electronic skin, continue to grow with micro and nanofabrication technology advances. \n\nSuccessful identification of a wide range of physical information through sensing elements, using the principles of piezoresistance,[2] capacitance,[3] piezoelectricity,[4–6] and triboelectricity,[7,8] has been reported and applied in healthcare or artificial intelligence.[9–11] Piezoresistive and capacitive sensors depend on external power supports, which makes them problematic in terms of portability and then limits their application scenarios.[12,13] Additionally, piezoelectric and triboelectric devices lack sufficient applicability due to the indispensableness of contact deformations or interfacial displacements and are complicated to fabricate.[14–16] Thermogalvanic hydrogels have excellent application prospects in physical information recognition by converting arbitrary heat, which is imperceptible to human beings, into electricity, and have the advantages of favorable low cost, biocompatibility, mechanical adaptability, and ease of preparation.[17–19] \n\nHere, we report a thermogalvanic hydrogel for self-powered material identification with acrylamide as the matrix, gelatin as the reinforcing agent, and $\\mathrm{[Fe(CN)6]^{3-/4-}}$ as the redox couple, respectively. In addition, lithium magnesium silicate (LMS) in the thermogalvanic dual-network hydrogel (TG-DNH) enlarges its pore size, ${\\mathrm{Li^{+}}}$ and $\\mathrm{Br^{-}}$ extend its effective working time and ${\\mathrm{Gdm}}^{+}$ significantly increases its thermopower $(4.01\\mathrm{~mV~K^{-1}};$ ). Based on the temperature-triggered thermogalvanic effect of the TG-DNH in conjunction with machine learning, we have developed a self-powered material identification ring (MIR) that can actively infer the type of materials without an external power connection by analyzing the voltage signals correlated with interfacial heat transfer produced upon contact with different materials. This study shows a great application potential of thermogalvanic hydrogels for human–computer interaction and artificial tactile sensation. \n\n \nFigure 1. Design of the MIR. a) Schematic representation of the TG-DNH and the MIR. b) Photos show that the TG-DNH can be deformed from external forces and withstand a weight of $\\boldsymbol{\\mathsf{100g}}$ . Scale bar (1 cm). c) Photos of a circular DNH gel, the QR code, and the results obtained after successfully scanning the code using a smartphone. Scale bar (1 cm) d) Comparison of the TG-DNH with existing quasi-solid-state thermogalvanic gel in terms of stretchability, current density, conductivity, thermopower, and maximum power.[30–33]",
|
||
"category": " Introduction"
|
||
},
|
||
{
|
||
"id": 3,
|
||
"chunk": "# 2. Results and Discussion",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 4,
|
||
"chunk": "# 2.1. Design and Fabrication of the TG-DNH \n\nThermogalvanic hydrogels have demonstrated outstanding potential in converting temperature stimuli into electrical signals for wearable sensing.[20–23] A general illustration of the TGDNH-based self-powered material identification is sketched in Figure 1a. When one user touches an unknown material with the MIR made of the TG-DNH, a temperature gradient is established across the TG-DNH. At this time, the electrical signal correlated with interfacial heat transfer upon contact with different materials is actively generated by the thermoelectric conversion of the redox couple in the gel, which is then transmitted out for analyzing the interfacial heat transfer characteristics with the aid of machine learning, precisely inferring the type of materials. Polyacrylamide (PAAm) and gelatin are among the best candidates for the preparation of thermogalvanic hydrogels due to their non-toxicity, environmental friendliness, and low price. In the preparation of PAAm/gelatin dual-network hydrogels via an aqueous solution polymerization route, gelatin forms the first network by physical cross-linking, while PAAm forms the second network by chemical cross-linking. It is worth noting that the gelatin chains undergo the coil-helix transition and then aggregate into triple helices when the temperature decreases. However, the unfolding and polymerization of gelatin chains is reversible. Once the temperature increases, the triple helix structure of gelatin chains will unfold, consequently improving its fluidity and hydrogel adhesion.[24] Figure S1 (Supporting Information) demonstrates the excellent adhesion of the TG-DNH to different materials. The FTIR spectra of PAAm, gelatin, and the TG-DNH and the detailed descriptions are shown in Figure S2 and Text S1 (Supporting Information). The absorption associated with the characteristic groups of PAAm and gelatin can be seen, and no significant positional changes or new peaks occur, indicating that PAAm and gelatin are connected by hydrogen bonding. \n\nConsidering that conventional hydrogels lose water when exposed to the atmosphere and thus affect their performance, we tried to add ${\\mathrm{Li}^{+}}$ and $\\mathrm{Br^{-}}$ to hydrogels. It is evident from Figures S3 and S4 (Supporting Information) that the addition of these two ions at the concentration of $\\mathbf{0.1\\mu_{M}}$ significantly enhanced water retention of the gel. The reason for this change is that the existence of $\\mathrm{Li^{+}}$ and $\\mathrm{Br^{-}}$ diminishes the saturation vapor pressure of water in the hydrogel to a level similar to the partial pressure of atmospheric vapor in the environment, slowing the rate of water loss from the hydrogel.[25] The DTG plots of Figure S5 (Supporting Information) and the explanation in Text S2 (Supporting Information) show that the degradation peak of the TG-DNH weakens after the addition of LiBr, which suggests amelioration in its thermal stability.[26] \n\nFigure 1b shows that the TG-DNH can withstand various deformations from external forces, including bending, stretching, twisting, cutting, and withstanding a weight of $100\\ \\mathrm{g}$ which reveals that the TG-DNH has favorable mechanical strength and flexibility. Figure 1c exhibits a photograph of a circular DNH gel with a thickness of $5\\mathrm{mm}$ . Due to the excellent transparency, the \n\n \nFigure 2. Impact of LMS and mechanical property of the TG-DNH. a) SEM image and pore size distribution of the TG-DNH (Hitachi SU8010). Scale bar $(70\\upmu\\mathrm{m})$ . b) Stretching of the TG-DNH with different LMS contents and immersion time (inset). c) The thermopower and current, d) Frequencydependent conductivity, and e) Cyclic voltammetry curves of the TG-DNH with different LMS content (Keithley 2400 and CHI660E). f) The tensile stress– strain curves and g) toughness of the TG-DNH with different gelatin content. h) Cyclic stretching-releasing curves (UTM, Qualitest). The illustration is stretching-recovery images (Scale bar: $\\mathsf{1c m}$ ). i) Cyclic compression-releasing curves with deformation of $50\\%$ . The illustration is compression-recovery images (Scale bar: $\\mathsf{1c m}_{}^{\\mathsf{^{\\prime}}}$ ). The results in (b–i) are represented as the mean (sample size $n=5$ ). \n\nQR code behind the gel can be clearly seen. We also compared the mechanical and thermoelectric properties of the TG-DNH with those of reported quasi-solid thermogalvanic gels in terms of stretchability, current density, conductivity, thermopower, and maximum power (Figure 1d), and the results show that the TGDNH has an exceptional overall performance, especially in terms of thermoelectricity.",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 5,
|
||
"chunk": "# 2.2. Impact of LMS and Mechanical Property of the TG-DNH \n\nAs shown in Figure 2a, a scanning electron microscope is utilized to examine the microstructural morphologies of the TG-DNH. The hydrogel has a porous mesh structure that facilitates unimpeded ion transport and exhibits a narrow pore size distribution when it does not contain LMS. However, when the gel includes a certain amount of LMS, its pore size notably enlarges because when LMS mixes with water, it holds up the LMS along its lamellar crystal structure; at this point, the LMS swells rapidly until the lamellae separate.[27] Therefore, the strain of the hydrogel shows a decrease with the increasing LMS content (Figure 2b). In the absence of LMS, the tensile strain of the TG-DNH is as high as $370\\%$ , whereas the tensile length is only $106\\%$ after the addition of $4\\%$ LMS. In the subplot of Figure 2b, we compare the tensile lengths for different times of immersion in electrolyte solutions containing solutes of $\\mathrm{K_{4}[F e(C N)_{6}]/K_{3}[F e(C N)_{6}]}.$ . It can be observed that the TG-DNH immersed for ${\\approx}1\\mathrm{~h~}$ has the best strain property, and the TG-DNH that we have used since then have been obtained by immersing them in electrolytes for $^{1\\mathrm{h}}$ . Not only for the mechanical property, the content of LMS in the gel also affects thermoelectricity. The comparison of the thermopower and current of the TG-DNH containing different concentrations of \n\nLMS is shown in Figure 2c. Compared to that without LMS, both the thermopower and short-circuit current of the TG-DNH with $1\\%$ LMS show a substantial increase that is up to $4.01\\mathrm{~mV~K^{-1}}$ and $1.18\\mathrm{mA}$ , respectively. With the continuous addition of LMS to the gel, the thermopower of the TG-DNH begins to decrease, and its current also declines accordingly. The conductivity variation with the frequency of the TG-DNH at different LMS contents is depicted in Figure 2d. The trend observed in the EIS data is consistent with that shown in Figure 2c. While LMS is introduced to the gel, the resistance of the TG-DNH decreases and the conductivity increases until the turning point of $1\\%$ LMS content. The inner diagram in Figure 2d plots the frequency-dependent conductivity of the hydrogel at $1\\%$ LMS content. Figure 2e describes the highly symmetric CV curves at different LMS contents. The figure shows the presence of symmetrical peaks, i.e., a reversible reaction exists, where the peak of oxidation occurs upward, while the one of reduction is observed downward. As the LMS content increases from $1\\%$ to $4\\%$ , the peak intensity is gradually weakening, indicating that the conductivity is decreasing, which is in accordance with the conclusion of Figure 2d. The variance in the horizontal positioning of the oxidation and reduction peaks is negligible, suggesting that the polarization phenomenon is basically absent. \n\nExcellent mechanical performance is crucial for flexible wearable bioelectronics. Hydrogels with good mechanical properties can return to their original shapes when deformed by external forces, significantly improving the applicability of gel-based bioelectronics. We have demonstrated above the remarkable flexibility of the TG-DNH in bending, stretching, twisting, and cutting, as well as the ability to hold a weight of $100\\mathrm{g}$ . Herein, we further investigated the tensile property of the TG-DNH with different gelatin contents. As the gelatin concentration rises from 0 to $10\\mathrm{\\mt.}\\%$ , the tensile strain of the gel increases from $169\\%$ to a maximum of $325\\%$ , and the maximal fracture stress is $0.11\\mathrm{MPa}$ . However, there is a decrease in tensile strain and fracture stress when gelatin concentration is increased to $15\\mathrm{\\wt.\\%}$ (Figure 2f). A proximately positive correlation exists between toughness and gelatin concentration in the gel, as depicted in Figure 2g, where toughness is calculated from the stress–strain curves. Notably, the toughness of PAAm- $10\\%$ Gel attains 164.15 KJ $\\mathbf{m}^{-3}$ , about two times higher than that of PAAm $(74.07\\ \\mathrm{KJ\\m^{-3}},$ ). This trend in Figure $^{2\\mathrm{f},\\mathrm{g}}$ occurs because of the gelatin’s unique triple helix structure, which provides energy dissipation, thus increasing the toughness of the hydrogel. However, if the gelatin concentration is too high, the gelatin chains can stack to form a crystalline structure, leading to a reduction in mechanical performance. Unless otherwise indicated, the TG-DNH with the gelatin concentration of $10\\mathrm{wt.\\%}$ is used in the following experiments. In addition, we further explored the elasticity of the TG-DNH by tensile cycling test. The results of cyclic stretching-recovery processes at different degrees of deformations are depicted in Figures 2h and S6 (Supporting Information), and the built-in inset shows the state of the gel during stretching and recovery. As can be seen from the figure, there is almost no displacement between the curves during the stretching and releasing, indicating that the TG-DNH has excellent fatigue resistance. As summarized in Figure 2i, the TG-DNH is compressed and released for 50 cycles at $50\\%$ deformation, and the insets show a full recovery from $50\\%$ compression. The above results indicate that an appropriate proportion of gelatin in the PAAm network leads to an improved mechanical property of the gel. The optimization of the hydrogel can make it ideally applicable for human wearable applications.",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 6,
|
||
"chunk": "# 2.3. Thermoelectric Property of the TG-DNH \n\nThermopower is one of the most important parameters for assessing thermoelectric conversion, primarily determined by the type of redox pairs, differences in solvation structure entropy, and ion concentration gradients. In order to make the TG-DNH have a desirable thermoelectric conversion efficiency and stable electrical output, we chose $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ among p-type redox ions and used ${\\mathrm{Gdm}}^{+}$ to improve its thermopower. We investigated the thermoelectric property of the TG-DNH with a length of ${\\approx}20\\mathrm{mm}$ , a width of ${\\approx}10\\mathrm{mm}$ , and a thickness of ${\\approx}4$ mm using a homemade temperature-controlled platform (Figure S7, Supporting Information). Figure 3a illustrates the operating principle of the TG-DNH concerning the $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ redox reaction. At the hot end, the thermodynamic oxidation reaction Equation 1 occurs, with electrons move concurrently toward the anode electrode, accompanied by an electrochemical potential increase and a decrease in electrode potential. The product $\\mathsf{\\Gamma}[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-}$ obtained from oxidation will return to the cathode by diffusion, convection, and migration. On the contrary, at the cold end, the cathode undergoes a reduction reaction Equation 2. The attraction of electrons from the cold end leads to a reduction in electrochemical potential and a rise in electrode potential. The product $[\\mathrm{Fe}(\\mathrm{CN})_{6}]^{4-}$ obtained from the reduction process is then returned to the anode to continue the redox reaction. Therefore, based on the thermogalvanic effect of the TG-DNH, the temperature stimuli can be transformed into forms that are more intuitive, such as voltage and current signals. \n\n$$\n\\begin{array}{r l}&{\\mathrm{Hotend:~}\\left[\\mathrm{Fe}(\\mathrm{CN})6\\right]4\\rightarrow\\mathrm{~e-}+[\\mathrm{Fe}(\\mathrm{CN})6]3-}\\\\ &{\\mathrm{Cold}\\mathrm{end:~}\\left[\\mathrm{Fe}(\\mathrm{CN})6\\right]^{3-}+\\mathrm{e^{-}\\rightarrow~}\\left[\\mathrm{Fe}(\\mathrm{CN})6\\right]^{4-}}\\end{array}\n$$ \n\nThe solvation structures of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ and ${\\mathrm{Gdm}}^{+}$ in solution are shown schematically in Figure 3b, where $\\mathrm{[Fe(CN)_{6}]~^{4-}}$ is on the left and $\\mathsf{\\Gamma}[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-}$ is on the right. It is seen that there is a notable difference between the solvation shells of $\\mathrm{[Fe(CN)_{6}]~^{4-}}$ and $\\mathsf{[F e(C N)_{6}]^{3-}}$ following the introduction of ${\\mathrm{Gdm}}^{+}$ . Compared to slightly affecting the solvation shell of $\\mathrm{[Fe(CN)_{6}]^{3-}}$ , $\\mathrm{Gdm^{+}}$ cations tend to distribute around $\\mathrm{[Fe(CN)_{6}]^{4-}}$ and cause its solvation shell to rearrange, which in turn widens the entropy difference of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ .[28] The thermopower, on the other hand, is mainly affected by the redox reaction entropy difference $\\Delta{\\sf S}$ (Equation 3), where $n$ represents the mole count of electrons transferred, $F$ denotes Faraday’s constant, Vh and Vc denote the open circuit voltages at the hot and cold ends of the sample, respectively, and Th and Tc are the temperatures at the hot and cold ends of the sample, respectively.[29] Thus, while the entropy difference of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ is widened by adding ${\\mathrm{Gdm}}^{+}$ , the thermopower also increases. \n\n$$\n\\mathrm{S_{e}}=\\Delta\\mathrm{S}/\\mathrm{nF}=\\left(\\mathrm{Vh}-\\mathrm{Vc}\\right)/\\left(\\mathrm{Th}-\\mathrm{Tc}\\right)\n$$ \n\nTo explore the influence of electrolyte ions on the thermoelectric property of the TG-DNH, we varied the concentrations of $[\\mathrm{Fe(CN)}_{6}]^{3-/4-}$ and ${\\mathrm{Gdm}}^{+}$ , respectively (Figures 3c and S8, Supporting Information). The $S_{\\mathrm{e}}$ of the TG-DNH increases with the enhancing concentration of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ and gets maximum at a concentration of $0.4\\textbf{M}$ . Further, when ${\\mathrm{Gdm^{+}}}$ is added to the TG-DNH, there is a significant elevation in the $S_{\\mathrm{e}}$ that increases with the enhancement of the ${\\mathrm{Gdm^{+}}}$ concentration and reaches the maximal value at a concentration of $0.6\\textbf{M}$ . It is evident that ${\\mathrm{Gdm}}^{+}$ plays a vital role in improving the thermopower of electrolyte $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ . The symmetrical peaks of CV curves at different $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ and ${\\mathrm{Gdm}}^{+}$ concentrations are presented in Figures 3d and S9 (Supporting Information), respectively, showing that $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ undergoes a reversible reaction. While the concentration of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ rises from 0.1 to $0.4\\textbf{M}$ , the intensity of the redox peak gradually strengthens, suggesting that the conductivity is elevating. The concentration of ${\\mathrm{Gdm^{+}}}$ has a similar effect on the trend observed from the CV curves. We fixed the temperatures at both the hot and cold ends and thus obtained the current/power-voltage curves at various concentrations of $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ .When the $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ concentration is \n\n \nFigure 3. Thermoelectric property of the TG-DNH. a) The operating principle of the TG-DNH concerning the $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-/4-}$ redox reaction. b) The schematic solvation formations of $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-}$ and $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{4-}$ in $\\mathsf{C d m^{+}}$ solution. c) Thermal voltage at varied $\\Delta\\mathsf{T}$ d) Cyclic voltammetry curves, and e) current/power-voltage curves of the TG-DNH with different concentrations of $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-/4-}.\\mathsf{f})$ Frequency-dependent conductivity and g) PF and ZT of the TG-DNH with different concentrations of $\\mathsf{G}\\mathsf{d}\\mathsf{m}^{+}$ . h) The maximum specific output power density. i) Short-circuit current curve of the thermogalvanic gel at different temperatures. The results in (c–i) are represented as the mean (sample size ${\\mathsf n}=5$ ). \n\n$0.1\\textbf{\\ m}$ , the short-circuit current of the TG-DNH is ${\\approx}0.5~\\mathrm{mA}$ , the open-circuit voltage is $25.0~\\mathrm{mV},$ and the maximum power is $3.12\\upmu\\mathrm{W},$ approximately. However, at a concentration of $0.4\\mathrm{~m~}$ $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ , the maximal power of the TG-DNH approximately quadruples to $12.98\\upmu\\mathrm{W}$ compared to the concentration of $0.1\\textbf{M}$ (Figure 3e). Next, by varying the temperature of the hot end at the constant cold end temperature of $293\\mathrm{~K~}$ the current/powervoltage curves under different temperature gradients are plotted in Figure S10 (Supporting Information). When the temperature difference $(\\Delta\\mathrm{T})$ increases from 5 to 25 K, the short-circuit current and open-circuit voltage progressively rise from $0.4\\mathrm{mA}$ and $20.0\\mathrm{mV}$ to $2.06\\mathrm{mA}$ and $100.0\\mathrm{mV},$ respectively, with the maximal power increasing from $2.0\\upmu\\mathrm{W}$ to as high as $51.5\\upmu\\mathrm{W}.$ \n\nThe porous structure of the TG-DNH facilitates the free migration of ions, making it suitable as a solid-state electrolyte. The variation of conductivity with frequency at different concentrations of ${\\mathrm{Gdm}}^{+}$ and $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ is shown in Figures 3f and S11 (Supporting Information), respectively, indicating the conductivity of the TG-DNH shows a pattern of increasing and then decreasing as the concentrations of $\\mathrm{Gdm^{+}}$ and $[\\mathrm{Fe}(\\mathrm{CN})_{6}]^{3-/4-}$ grow, identical with the trend in Figures 3c and S8 (Supporting Information). It may be because when the ion concentration is too high, the migration of ions in the hydrogel becomes hindered, leading to a decline in conductivity. Both power factor (PF) and figure of merit (ZT), which are essential indicators of the hydrogel thermoelectricity, reach their maximal values at $0.6\\mathrm{~M~Gdm^{+}}$ (Figure 3g), coherent with the trends of ${\\sf S}_{\\mathrm{e}}$ and ionic conductivity in Figure S12 (Supporting Information). Figure $3\\mathrm{h}$ depicts the maximal power density $(\\mathrm{P}_{\\mathrm{max}})$ at different temperature differences, which is approximately proportional to the square of temperature difference and increases substantially from 0.09 to $0.99\\mathrm{~W~m~}^{-2}$ , in agreement with the theoretical results of Equation 4. The specific output power density $(\\mathrm{P_{max}}/\\Delta\\mathrm{T}^{2})$ is also an essential parameter for evaluating the output performance, which is independent of the $\\Delta\\mathrm{T}$ and approximately constant, consistent with Equation 5. \n\n$$\n\\begin{array}{r}{\\mathrm{P}_{\\mathrm{max}}=\\left(\\sigma\\mathrm{Se}^{2}/4\\mathrm{d}\\right)\\Delta\\mathrm{T}^{2}}\\\\ {\\mathrm{P}_{\\mathrm{max}}/\\Delta\\mathrm{T}^{2}=\\sigma\\mathrm{Se}^{2}/4\\mathrm{d}}\\end{array}\n$$ \n\nwhere $\\sigma$ and d are the conductivity and effective length of the TG-DNH, respectively. \n\nAs shown in Figure 3i, the output current of the TG-DNH demonstrates a remarkable linear correlation with the temperature difference, indicating high sensitivity and relevance to temperature stimuli. In contrast, the conductivity of the TG-DNH fluctuated slightly between 1.05 and $1.24~\\mathrm{S~m^{-1}}$ as the temperature difference raised from 20 to $60~\\mathrm{K}$ . This suggests that the gel provides relatively stable conductivity even at large temperature differences (Figure S13, Supporting Information). Besides, the TG-DNH maintains a high output current after bending over 800 times, which proves its reliable durability (Figure S14, Supporting Information). Even when a pressure of ${\\approx}5\\mathrm{\\kPa}$ and tensile strain of $\\approx50\\%$ were applied on the TG-DNH, respectively, both the output voltage and current nearly remained constant (Figure S15a,b, Supporting Information), indicating that the TGDNH has a highly excellent thermoelectric reliability and antiinterference performance. To summarize, these results fully illustrate that the TG-DNH developed in this study exhibits an outstanding comprehensive performance based on the thermogalvanic effect.",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 7,
|
||
"chunk": "# 2.4. Machine-Learning-Assisted Self-Powered Material Identification by the MIR \n\nIdentifying the material by differing between interfacial heat transfer upon contact has the advantages of high accuracy and low cost. We selected seven typical materials and then ranked them according to their capability of interior thermotransport (Figure 4a). To visualize the interfacial heat transfer process of different materials, we preheated the bottoms of seven materials by a heater at the same temperature of $313\\mathrm{~K~}$ (Figure S16, Supporting Information). In the case of the same heating time, the surfaces of different materials have different temperature distributions. The interior thermotransport simulation of three materials, aluminum (i), ceramics (ii), and wood (iii), under this heating condition is shown in Figure 4b. It is observable that accelerated heat transfer and more heat flow within the materials correlate with the superior thermal conductivity. This phenomenon underpins the theoretical basis for employing interfacial heat transfer to distinguish between various materials. Figure 4c,d shows the photograph and the infrared scanning image while wearing the MIR in contact with the top surface of the aluminum. With the interfacial heat transfer triggered during this process, due to the temperature difference between the two ends of the TG-DNH, the MIR generates distinguishable voltage signals when in contact with different materials based on its inherent thermogalvanic effect without an external power supply (Figure 4e,f). In Figure ${4\\mathrm{g}}$ , a detailed analysis of the voltage signal features is provided, where $\\Delta\\mathrm{V_{A}}$ represents the difference between the voltage at the moment of separation and the initial voltage at the beginning of contact, and $\\Delta\\mathrm{V}_{\\mathrm{B}}$ signifies the discrepancy between the peak voltage $V_{\\mathrm{m}}$ and the voltage at the end of measurement. These four characteristics, $\\Delta\\mathrm{V_{A}}$ , $\\Delta\\mathrm{V}_{\\mathrm{B}}$ , $\\Delta\\mathrm{V_{A}}/\\Delta\\mathrm{V_{B}}$ and $\\mathrm{V_{m}}$ , are designated as eigenvalues in machine learning. \n\nMachine learning has been shown as an efficient tool for achieving material recognition precisely through multiple algorithms. The random forest algorithm integrates multiple decision trees and achieves high identification accuracy through integrated learning, whose core process is described in Figure 4h. In a 2D framework, the distribution of data is characterized by the axes of principal component 1 (PC1) and principal component 2 (PC2), with Principal Component Analysis (PCA). The different colors indexed to different materials indicate most dataset boundaries are clear with favorable intra-class compactness and significant inter-class separability. Figure 4i demonstrates the confusion matrix of the verifying results, where each row is the predicted material and each column is the actual material, with an average recognition accuracy of $97.2\\%$ for seven types of materials. The flowchart of self-powered material identification using the machine-learning-assisted MIR is described in Figure S17 (Supporting Information). Figure 4j presents some of the signal data used for machine learning in the sampling process. During operation with the MIR (Figure 4k), the acquired signal waveforms and final recognition results are visualized (Figures 4l and S18, Supporting Information). The enlarged terminal interface of machine-learning-assisted material identification is exhibited in Figures 4m and S19 (Supporting Information), where the left of the interface shows the real-time voltage signals during measurement and the right depicts material identification results. The whole material identification process is recorded in Movies S1– S3 (Supporting Information). Herein, by discerning differences in the interfacial heat transfer generated upon contact with different materials, the MIR can utilize the intrinsic thermogalvanic effect of the TG-DNH to convert temperature stimulus into a voltage signal and then actively identify the type of materials in conjunction with machine learning, which highlights a great application potential of thermogalvanic hydrogels for human–computer interaction and artificial tactile sensation.",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 8,
|
||
"chunk": "# 3. Conclusion \n\nIn summary, we reported a thermogalvanic hydrogel for selfpowered material identification with AAm as the matrix, gelatin as the reinforcing agent, and $\\mathrm{[Fe(CN)_{6}]^{3-/4-}}$ as the redox pair, respectively. The dual-network gel has a favorable thermoelectric output performance due to LMS enlarging its pore size and ${\\mathrm{Gdm}}^{+}$ widening the entropy difference of the redox couple, achieving a thermopower of $4.01\\ \\mathrm{mV}\\ \\mathrm{K}^{-1}$ . Meanwhile, ${\\mathrm{Li}^{+}}$ and $\\mathrm{Br^{-}}$ slow down the rate of water loss by adjusting the saturation vapor pressure of water in the hydrogel, which extends its effective working time. Supported by excellent mechanical and thermoelectric properties, we have developed a self-powered MIR based on the temperature-triggered thermoelectric response of the TG-DNH in conjunction with machine learning to accomplish the identification of seven materials with an impressive accuracy of $97.2\\%$ by analyzing the voltage signals correlated with interfacial heat transfer produced upon contact with different materials. This research can be further integrated with human– computer interaction and extended to haptic-associated artificial intelligence in the future. \n\n \nFigure 4. Machine-learning-assisted material identification with the MIR. a) Ranking of interior thermotransport capability of typical materials. The capability of interior thermotransport enhances from the bottom to the top. b) Temperature distribution of aluminum (i), ceramics (ii), and wood (iii) for the same time of heating by COMSOL software. c) A photo of the tester wearing the MIR and d) the infrared scanning image during the test (FLIR E6-XT) (Scale bar: $10\\:\\mathsf{m m}$ e,f ) Voltag of differ materials mea red by the TG-DNH. g) Analysis of voltage profiles and selection of features. h) Simple flowchart of the RF orith and the distributio 2D pac nd PC2 axes. i) Confusion map of the machine-learnin sult of the arning during the pling process and photos of several test materials $(\\mathsf{n}=270)$ . k) T f the MIR (Scale bar: $\\mathsf{10}\\mathsf{m m}$ ). l) With th of m achine le rning, the aluminum is successfully identified using the MIR, and the results pla ed on the omputer. m) Computer interface for the material identification system when identifying aluminum.",
|
||
"category": " Conclusions"
|
||
},
|
||
{
|
||
"id": 9,
|
||
"chunk": "# 4. Experimental Section \n\nMaterials: Acrylamide (AAm, $\\begin{array}{c c l}{\\mathsf{M}\\mathsf{w}}&{=}&{71.08}\\end{array}$ , $99.0\\%$ , Gelatin $(\\mathsf{M w}=279.66)$ ), Silicic acid, lithium magnesium salt (LMS, $\\mathsf{M w}=$ 290.74, $99.0\\%$ , N, $N^{\\prime}$ -methylene bisacrylamide (MBA, $\\mathsf{M w}=754.77$ , $\\geq98\\%)$ ), Ammonium persulfate (APS, $\\mathsf{M w}=228.2,\\geq98\\%$ ), $\\mathsf{N},\\mathsf{N},\\mathsf{N}^{\\prime},\\mathsf{N}^{\\prime}.$ - Tetramethylethylenediamine (TEMED, $\\mathsf{M w}=776.27,\\geq98\\%)$ , ${\\sf K}_{3}\\sf r e(C N)_{6}.3$ $H_{2}O$ ( $\\mathsf{M w}=422.4$ , $299\\%$ ), ${\\sf K}_{4}\\sf F e(C N)_{6}$ $(\\mathsf{M w}=329.2,\\ge99\\%$ ), Guanidine hydrochloride $(\\mathsf{M w}=95.53$ , $99.0\\%$ , Lithium bromide $(\\mathsf{M w}=86.85$ , $99.0\\%$ were bought from Aladdin Industrial Corporation. If not specified differently, the substances used in this study are employed without additional purification. \n\nPreparation of the TG-DNH: The TG-DNH is prepared primarily based on an aqueous solution polymerization method. First, acrylamide, LMS, and gelatin were added to $70~\\mathsf{m L}$ of deionized water and stirred at $70~^{\\circ}\\mathsf C$ until wholly dissolved to obtain a mixed solution. After the mixed solution was cooled down, N, $\\mathsf{N}^{\\prime}$ -methylene bisacrylamide, and initiator ammonium persulfate were added sequentially and stirred at room temperature until completely dissolved. Then, catalyst $\\mathsf{N},\\mathsf{N},\\mathsf{N}^{\\prime},\\mathsf{N}^{\\prime}.$ - Tetramethylethylenediamine was added, gently mixed, and quickly poured into the prepared mold for ${\\approx}2\\mathsf{h}$ to obtain DNH. Finally, the prepared DNH was immersed in a mixture of $[\\mathsf{F e}(\\mathsf{C N})_{6}]^{3-/4-}$ , guanidine hydrochloride, and lithium bromide for $2h$ to obtain the TG-DNH. \n\nThermoelectric Measurement: Each sample used for thermoelectric analysis measured dimensions of $20\\times20\\times4m{\\times}^{3}$ . The samples were placed on a homemade Seebeck coefficient measurement setup, where a DC power source regulated the temperature of both Peltier chips, which was then measured using a thermocouple (NAPUI TR230X). Measurements of the output voltage and current were conducted using a Keithley 2400 source. The temperature difference between the two ends of the sample was controlled by connecting the homemade Seebeck coefficient measurement setup to a controlled DC power supply, and a series of current, voltage and temperature variation data were obtained. A workstation for electrochemistry (CHI660e, Shanghai Chenhua Instrument Co. Limited.) was utilized for assessing the hydrogels’ electrochemical impedance spectroscopy (EIS) and cyclic voltammetry (CV). The ionic conductivity of the TG-DNH was calculated according to $\\sigma=\\mathsf{d}/\\mathsf{R A}$ , where d, R, and A represent the thickness, impendence value, and cross-sectional area of the TG-DNH, respectively.[26] \n\nComposition of the MIR: The TG-DNH, PDMS, PU membrane, graphite electrodes and copper leads. The TG-DNH and the graphite electrodes and copper wires attached to its surface are encapsulated by a PU membrane and use the thermogalvanic effect to generate electrical signals. The PDMS acts as a scaffold in the MIR (Figure 4k). \n\nMachine Learning-Assisted Data Measurement: To begin with, wires were used to connect the MIR to a data acquisition system (Keithley DAQ6510) and left it in place for $20~\\mathsf{s}$ to acclimatize to the environment and stabilize the voltage signal. Then, the MIR is used to contact the material to be tested with a force of 7N until the signal peaks. Finally, the MIR is allowed to separate from the material until the end of the test to obtain the voltage signal. The obtained signal is modeled to get the prediction result after the process of acquisition, preprocessing and signal extraction. \n\nMachine Learning-Assisted Data Analysis: First, the tester needs to wear the MIR and expose it to known materials for data acquisition and preprocessing with data acquisition system and prepared Python script (Figure S20, Supporting Information). With data features extracted and selected by the statistical feature extraction, $75\\%$ of the materials was chosen at random from the database for training purposes. The remaining $25\\%$ of the materials are considered test samples. The RF algorithm uses the training samples to build and train a multi-iteration machine learning model. With the expansion of training data, the accuracy of the machine learning model steadily neared the actual truth. In addition, test samples are used to assess whether the model is valid. At the same time, cross-validation and hyperparameter optimization will make the machine learning model more accurate. Finally, the machine learning model predicts the data in real time and identifies the material when the tester wears the MIR in contact with an unknown material, such as aluminum, brass, stainless steel, ceramics, acrylic, wood, and foam. \n\nCharacterization: At room temperature, Universal Tensile Testing (UTM, Qualitest, USA) was employed to conduct tests for both tensile and compression. Use Keithley 2400 instrument for voltage and current measurements. The morphology of the TG-DNH was characterized using the SEM method (Hitachi SU8010). The electrochemical AC impedance spectra, resistance, and cyclic voltammetry curves of the TG-DNH were measured using an electrochemical workstation (CHI660E, Shanghai Chenhua Instrument Co. Ltd.). The temperature was regulated through a homemade temperature platform, with real-time monitoring of temperature variations using thermocouple (NAPUI TR230X). Thermogravimetric analysis is performed by using TGA-601 thermal analysis instrument with a scan rate of $20^{\\circ}C\\mathsf{m i n}^{-1}$ . Use Keithley DAQ6510 for data acquisition. The temperature distribution of aluminum, ceramics, and wood for the same time of heating is simulated by COMSOL. A FLIR E6-XT infrared camera was utilized to capture infrared photographs. \n\nStatistical Analysis: The experimental result was subjected to preprocessing procedures such as de-baselining and expressed as mean values. The samples used for the determination of toughness, $S_{\\mathrm{e}}$ , and conductivity at each variation of the TG-DNH were ${\\mathfrak{n}}=5$ . All the statistical analysis was carried out using Microsoft Excel, Origin and COMSOL software. Each data point represents the mean. \n\nStatements: Ethics approval and informed consent were obtained from all participants, and the protocol was approved by the Ethics Review Committee of Taiyuan University of Technology (Approval No. TYUT2024092301). The study was carried out under the Helsinki Declaration.",
|
||
"category": " Materials and methods"
|
||
},
|
||
{
|
||
"id": 10,
|
||
"chunk": "# Supporting Information \n\nSupporting Information is available from the Wiley Online Library or from the author.",
|
||
"category": " References"
|
||
},
|
||
{
|
||
"id": 11,
|
||
"chunk": "# Acknowledgements \n\nThis work was supported by the National Natural Science Foundation of China (52475600), Special Project of Science and Technology Cooperation and Exchange of Shanxi Province (202304041101021) and Shanxi Province Graduate Science and Technology Innovation Project (2023XSY066).",
|
||
"category": " References"
|
||
},
|
||
{
|
||
"id": 12,
|
||
"chunk": "# Conflict of Interest \n\nThe authors declare no conflict of interest.",
|
||
"category": " Results and discussion"
|
||
},
|
||
{
|
||
"id": 13,
|
||
"chunk": "# Data Availability Statement \n\nThe data that support the findings of this study are available from the corresponding author upon reasonable request.",
|
||
"category": " Conclusions"
|
||
},
|
||
{
|
||
"id": 14,
|
||
"chunk": "# Keywords \n\nmachine learning, material identification, self-powered, thermogalvanic gel, thermopower \n\nReceived: July 15, 2024 \nRevised: October 18, 2024 \nPublished online: October 31, 2024 \n\n[1] R. S. Johansson, J. R. Flanagan, Nat. Rev. Neurosci. 2009, 10, 345. \n[2] J. Park, M. Kim, Y. Lee, H. S. Lee, H. Ko, Sci. Adv. 2015, 1, 1500661. \n[3] L. Beker, N. Matsuhisa, I. You, S. R. A. Ruth, S. Niu, A. Foudeh, J. B.-H. Tok, X. Chen, Z. Bao, Proc. Natl. Acad. Sci. 2020, 117, 11314. \n[4] Z. Liu, Q. Zheng, Y. Shi, L. Xu, Y. Zou, D. Jiang, B. Shi, X. Qu, H. Li, H. Ouyang, R. Liu, Y. Wu, Y. Fan, Z. Li, J. Mater. Chem. B 2020, 8, 3647. \n[5] B.-Y. Lee, D. H. Kim, J. Park, K.-I. Park, K. J. Lee, C. K. Jeong, Sci. Technol. Adv. Mater. 2019, 20, 758. \n[6] H. S. Wang, S. K. Hong, J. H. Han, Y. H. Jung, H. K. Jeong, T. H. Im, C. K. Jeong, B.-Y. Lee, G. Kim, C. D. Yoo, K. J. Lee, Sci. Adv. 2021, 7, eabe5683. \n[7] T. Chen, Q. Shi, M. Zhu, T. He, L. Sun, L. Yang, C. Lee, ACS Nano 2018, 12, 11561. \n[8] T. Jin, Z. Sun, L. Li, Q. Zhang, M. Zhu, Z. Zhang, G. Yuan, T. Chen, Y. Tian, X. Hou, C. Lee, Nat. Commun. 2020, 11, 5381. \n[9] Z. Sun, M. Zhu, Z. Zhang, Z. Chen, Q. Shi, X. Shan, R. C. H. Yeow, C. Lee, Adv. Sci. 2021, 8, 2100230. \n[10] J. Kim, M. Lee, H. J. Shim, R. Ghaffari, H. R. Cho, D. Son, Y. H. Jung, M. Soh, C. Choi, S. Jung, K. Chu, D. Jeon, S.-T. Lee, J. H. Kim, S. H. Choi, T. Hyeon, D.-H. Kim, Nat. Commun. 2014, 5, 5747. \n[11] G. Lee, J. H. Son, S. Lee, S. W. Kim, D. Kim, N. N. Nguyen, S. G. Lee, K. Cho, Adv. Sci. 2021, 8, 2002606. \n[12] S. Wang, J. Xu, W. Wang, G.-J. N. Wang, R. Rastak, F. Molina-Lopez, J. W. Chung, S. Niu, V. R. Feig, J. Lopez, T. Lei, S.-K. Kwon, Y. Kim, A. M. Foudeh, A. Ehrlich, A. Gasperini, Y. Yun, B. Murmann, J. B.-H. Tok, Z. Bao, Nature 2018, 555, 83. \n[13] S. K. Ravi, N. Paul, L. Suresh, A. T. Salim, T. Wu, Z. Wu, M. R. Jones, S. C. Tan, Mater. Horiz. 2020, 7, 866. \n[14] L. Zhao, H. Li, J. Meng, A. C. Wang, P. Tan, Y. Zou, Z. Yuan, J. Lu, C. Pan, Y. Fan, Y. Zhang, Y. Zhang, Z. L. Wang, Z. Li, Adv. Funct. Mater. 2020, 30, 1907999. \n[15] Y. Pang, Y. Cao, M. Derakhshani, Y. Fang, Z. L. Wang, C. Cao, Matter 2021, 4, 116. \n[16] X. Chen, L. Gao, J. Chen, S. Lu, H. Zhou, T. Wang, A. Wang, Z. Zhang, S. Guo, X. Mu, Z. L. Wang, Y. Yang, Nano Energy 2020, 69, 104440. \n[17] C. Bai, X. Li, X. Cui, X. Yang, X. Zhang, K. Yang, T. Wang, H. Zhang, Nano Energy 2022, 100, 107449. \n[18] J. Li, Z. Wang, S. A. Khan, N. Li, Z. Huang, H. Zhang, Nano Energy 2023, 113, 108612. \n[19] R. Fang, X. Li, S. A. Khan, Z. Wang, X. Cui, H. Zhang, Z.-H. Lin, ACS Appl. Polym. Mater. 2023, 5, 4628. \n[20] X. Yang, Z. Zhang, S. A. Khan, L. Sun, Z. Wang, X. Cui, Z. Huang, H. Zhang, J. Mater. Chem. C 2024, 12, 3298. \n[21] C. Tian, C. Bai, T. Wang, Z. Yan, Z. Zhang, K. Zhuo, H. Zhang, Nano Energy 2023, 106, 108077. \n[22] S. Sang, C. Bai, W. Wang, S. A. Khan, Z. Wang, X. Yang, Z. Zhang, H. Zhang, Nano Energy 2024, 123, 109366. \n[23] H. Yang, S. A. Khan, N. Li, R. Fang, Z. Huang, H. Zhang, Chem. Eng. J. 2023, 473, 145247. \n[24] Y. Zhang, Y. Dai, F. Xia, X. Zhang, Nano Energy 2022, 104, 107977. \n[25] S. Pu, Y. Liao, K. Chen, J. Fu, S. Zhang, L. Ge, G. Conta, S. Bouzarif, T. Cheng, X. Hu, K. Liu, J. Chen, Nano Lett. 2020, 20, 3791. \n[26] N. Li, S. A. Khan, K. Yang, K. Zhuo, Y. Zhang, H. Zhang, Compos. Sci. Technol. 2023, 239, 110077. \n[27] Y. Yu, Y. Feng, F. Liu, H. Wang, H. Yu, K. Dai, G. Zheng, W. Feng, Small 2023, 19, 2204365. \n[28] J. Duan, G. Feng, B. Yu, J. Li, M. Chen, P. Yang, J. Feng, K. Liu, J. Zhou, Nat. Commun. 2018, 9, 5146. \n[29] L. Liang, H. Lv, X.-L. Shi, Z. Liu, G. Chen, Z.-G. Chen, G. Sun, Mater. Horiz. 2021, 8, 2750. \n[30] P. Yang, K. Liu, Q. Chen, X. Mo, Y. Zhou, S. Li, G. Feng, J. Zhou, Angew. Chem. Int. Ed. 2016, 55, 12050. \n[31] Y. Zhou, S. Zhang, M. A. Buckingham, L. Aldous, S. Beirne, C. Wu, Y. Liu, G. Wallace, J. Chen, Chem. Eng. J. 2022, 449, 137775. \n[32] W. Gao, Z. Lei, C. Zhang, X. Liu, Y. Chen, Adv. Funct. Mater. 2021, 3, 2104071. \n[33] X. Li, X. Xiao, C. Bai, M. Mayer, X. Cui, K. Lin, Y. Li, H. Zhang, J. Chen, J. Mater. Chem. C 2022, 10, 13789.",
|
||
"category": " References"
|
||
}
|
||
] |