Strain sensor and its application in gesture recognition and silent speech recognition

Abstract

A strain sensor and its application in gesture recognition and silent language recognition, the strain sensor adopts a flexible material as a substrate, adds a thin layer of the same material through an electrospinning process for surface modification, and prints conductive ink on the modified surface, the strain sensor has high sensitivity and good flexibility, is suitable for soft and curved scenes such as human body, and a signal acquisition module and a recognition module are respectively used for measuring resistance change of the strain sensor and recognizing corresponding gestures or languages.

Description

Technical Field

[0001] This invention relates to the field of strain sensors, and in particular to a strain sensor and its application in gesture recognition and silent speech recognition. Background Technology

[0002] Strain sensors convert strain into signals such as resistance and voltage to measure the strain of an object. Traditional strain sensors mostly use materials like polyimide as a substrate and metal wire as the sensing material, which is then attached to the surface of the object being measured. They measure strain by utilizing the change in resistance after the metal is stretched. In industrial production and structural monitoring, traditional strain sensors based on metal strain gauges are widely used in metal structure strain monitoring, building and bridge condition detection, and other fields. However, sensors based on metal strain gauges have the following limitations: existing strain sensors are made of rigid materials such as polyimide and metal, which have poor stretchability and flexibility, making them unsuitable for soft, bending environments such as the human body; existing strain sensors measure strain by utilizing the change in resistance caused by the deformation of a metal conductor under strain, resulting in increased length and decreased width, which leads to low sensitivity.

[0003] To address the aforementioned issues, related research has proposed strain sensors based on electrospinning. Electrospinning is a process that uses high-voltage electrostatics to spin polymer solutions into micron- to nanoscale fibers. Membrane materials manufactured using electrospinning are characterized by good flexibility and inherent surface texture, making them suitable for strain sensor fabrication. Existing research uses electrospinning to manufacture strain sensors, but this process requires dissociating the solution into micron- to nanoscale filaments under high-voltage electrostatics, resulting in a very slow filament generation rate. For example, patent application number CN202411476610.9 discloses a "spider web structure" nanofiber-based composite material, its preparation method, and its applications. Accumulating these filaments into sufficiently thick, mechanically capable membranes requires an extremely long time, leading to low production efficiency and high equipment costs. Other research uses electrospinning to manufacture functional membranes, which are then cut and wrapped around fiber surfaces. For instance, patent application number 202311050580.0 discloses a method for preparing a fiber-based stress-strain sensor based on electrospinning. This method also requires the electrospinned functional membrane to have a high thickness to withstand stress and perform wrapping operations. Furthermore, the packaging process is not a spontaneous self-assembly process and requires active manual operation, resulting in low production efficiency; the fiber core material and the electrospinning material are not the same type of material and have different mechanical properties, making them prone to peeling.

[0004] Given the numerous shortcomings of the existing technologies, there is currently no mature solution. Therefore, there is an urgent need to develop a strain sensor that is highly flexible, highly sensitive, low-cost, and easy to manufacture, and to break through the scenario limitations of existing recognition schemes to achieve stable and efficient gesture recognition and silent speech recognition. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention aims to provide a strain sensor and its application in gesture recognition and silent speech recognition. The conductive ink pattern layer of the strain sensor is printed on the surface of a modified thin layer formed by electrospinning. Under the action of strain, ordered microcracks can be generated under the induction of electrospinning fibers in the modified thin layer, so as to realize a high-sensitivity response of resistance to strain.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A strain sensor includes a flexible substrate 1, a modified thin layer 2, and a conductive ink pattern layer 3 arranged sequentially. The modified thin layer 2 is prepared in situ on the surface of the substrate 1 to form a nanoscale ultrathin coating layer on the surface of the flexible substrate 1 by electrospinning. The conductive ink pattern layer 3 is printed on the surface of the modified thin layer 2. Under strain, the electrospinning fibers of the modified thin layer 2 induce ordered microcracks in the conductive ink pattern layer 3, thereby achieving a highly sensitive response of resistance to strain.

[0007] The modified thin layer 2 and the flexible substrate 1 are homogeneous materials. They spontaneously assemble by the electric field force during the electrospinning process and generate an irreversible bond, ultimately forming an ultrathin capping layer with a thickness of nanometers.

[0008] The specific preparation process of the modified thin layer 2 includes: Step 1: Cut the flexible substrate 1 to the preset size, use anhydrous ethanol to ultrasonically clean and dry its surface to be spun; fix the cleaned flexible substrate 1 flat on the negative electrode receiving plate of the electrospinning equipment to ensure that the flexible substrate 1 and the receiving plate are completely in contact without warping. Step 2: Take polyurethane particles of the same material as the flexible substrate 1, add them to N,N-dimethylformamide solvent, stir to dissolve, and prepare a polyurethane spinning solution with a mass fraction of 8%~12%, and let it stand to remove bubbles; Step 3: Inject the deaerated spinning solution into the syringe of the electrospinning equipment. The syringe needle has an inner diameter of 0.4~0.6mm. Connect the needle to the positive high-voltage power supply of the electrospinning equipment and set the positive voltage to +12kV~+18kV. Connect the negative electrode of the receiving plate with the flexible substrate 1 fixed to it to the negative high-voltage power supply and set the negative voltage to -3kV~-8kV. Control the receiving distance between the needle and the flexible substrate 1 to 10~15cm, set the syringe advance rate to 0.5~1.2mL / h, control the ambient temperature to 22~28℃, and control the relative humidity to 30%~50%. Under the action of the high-voltage electric field, the spinning solution dissociates from the needle tip into nanoscale filaments, which fly to the flexible substrate 1 of the negative electrode under the pull of the electric field and spontaneously assemble and tightly bond with the substrate surface under the action of electrostatic force.

[0009] The conductive ink pattern layer 3 is prepared using flexible conductive inks such as carbon nanotube conductive ink or indium tin oxide conductive ink, and the preparation process is screen printing.

[0010] The conductive ink pattern layer 3 includes one or more independent conductive patterns, which can realize the synchronous measurement of single-point or multi-point strain.

[0011] The single conductive pattern can be a straight line extending in a single direction, used to measure uniaxial strain along the length of the pattern; or an arc-shaped pattern or a serpentine pattern extending along a preset curvature. The arc-shaped pattern is used to measure bending strain along the circumference of the arc, and the serpentine pattern is used to extend the length of the conductive path within a limited area, thereby improving the resolution and sensitivity of strain detection.

[0012] The multiple independent conductive patterns use the same or different line types and are arranged in parallel, cross, or independently in different regions, respectively corresponding to strain detection points in different parts and directions of the object under test; in the case of mutual independence, each independent conductive pattern is led out to the back-end signal acquisition module through an independent metal lead.

[0013] An application of a strain sensor in gesture recognition and silent speech recognition involves attaching the strain sensor to the back of the hand to capture hand movements for gesture recognition, or attaching the strain sensor to the perioral area to capture perioral movements.

[0014] The application is as follows: the strain sensor converts the mechanical strain signal to be measured into a resistance change signal; the strain sensor is electrically connected to the signal acquisition module, the signal acquisition module acquires the resistance change signal of the strain sensor and transmits the signal to the recognition module; the recognition module receives the resistance change signal and processes it to identify the corresponding hand gesture or silent language information.

[0015] The signal acquisition module includes a multiplexer, a resistance measurement module, and a signal transmission module. The multiplexer is used to switch the acquisition channels of multiple strain sensors to achieve synchronous polling acquisition of signals from multiple channels. The resistance measurement module is used to accurately measure the real-time resistance value of the conductive ink pattern layer. The signal transmission module supports wired or wireless transmission (preferably Bluetooth wireless transmission) and can quickly transmit the acquired electrical signals to the identification module.

[0016] The recognition module incorporates a recognition model based on the fusion of a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. The CNN takes a sequence of digital signals showing the resistance of multiple sensors changing over time as input to capture the correlation features between the signals from multiple strain sensors. The LSTM takes a sequence of digital signals showing the resistance of multiple sensors changing over time as input to process the time-series strain signals and obtain the long-range relationships between the channels. The spatial correlation features output by the CNN and the temporal evolution features output by the LSTM are concatenated to complete the feature layer fusion. The fused global features are then input into a fully connected neural network for feature dimensionality reduction and mapping. Finally, a Softmax classifier outputs the classification and recognition results for the corresponding gestures or silent language words, completing the conversion from action to semantics.

[0017] The applications described correspond to gesture recognition scenarios controlled by robots and silent speech recognition scenarios for people with laryngeal dysfunction, respectively.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Flexibility and adaptability: The strain sensor of this invention uses a medical-grade flexible substrate, which has excellent flexibility, stretchability and skin adhesion. It can be stably attached to the skin surface of the human hand, face and other parts of the body, and is suitable for dynamic flexible scenarios such as joint bending and muscle micro-movement. It solves the problem that traditional rigid strain sensors cannot be adapted to human body scenarios.

[0019] 2. High sensitivity: This invention modifies the substrate surface by electrospinning a thin layer of homogeneous material, which can induce the conductive ink to generate ordered microcracks under strain. Based on the opening and closing effect of the cracks, a significant change in resistance is achieved. Compared with the geometric deformation effect of traditional metal strain gauges, the sensitivity is improved by several orders of magnitude. It can accurately capture the tiny strain generated by human micro-movements and meet the high-precision requirements of gesture recognition and facial micro-movement capture.

[0020] 3. Low-cost and high-efficiency preparation process: The electrospinning modified thin film of the present invention only requires a nanometer-level ultrathin thickness to achieve the surface induction effect, without the need for long-term spinning to prepare thick films, which greatly shortens the electrospinning process time, significantly improves production efficiency, reduces equipment costs and preparation threshold, and is easy to scale up production.

[0021] 4. Stability in use: This invention uses homogeneous materials to prepare modified thin layers and flexible substrates, which spontaneously assemble by the electric field force during the electrospinning process and produce an irreversible and tight bond. The bonding force between the two is extremely strong. During dynamic use with repeated bending and stretching, it is not easy to have problems such as delamination, detachment, or performance degradation, which greatly improves the cycle life and long-term working stability of the sensor.

[0022] In summary, this invention, based on homogeneous materials and electrospinning technology, effectively reduces spinning time and manufacturing costs by decreasing the thickness of the electrospinning modified layer. Simultaneously, this modified thin layer achieves in-situ self-assembly on the flexible substrate surface under the action of an electric field, utilizing the physical properties between homogeneous materials to obtain a stable interfacial bond, eliminating the need for additional bonding or hot-pressing processes. This invention simplifies the fabrication process of flexible sensors, reducing production costs while ensuring the structural stability of the device, providing a practical and feasible technical solution for the large-scale fabrication of flexible strain sensors. Attached Figure Description

[0023] Figure 1 is a schematic diagram of the workflow of the sensing identification and application described in this invention.

[0024] Figure 2 is a schematic diagram of the cross-sectional structure of the flexible strain sensor described in this invention.

[0025] Figure 3 is a schematic diagram of the front layout of the strain sensor used for gesture recognition in Embodiment 1 of the present invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings.

[0027] Reference Figure 1 , Figure 2 A strain sensor is disclosed, the core structure of which sequentially comprises a flexible substrate 1, a modified thin layer 2, and a conductive ink pattern layer 3. The modified thin layer 2 is fabricated in situ on the surface of the flexible substrate 1 using an electrospinning process to form a nanoscale ultrathin coating. The conductive ink pattern layer 3 is printed on the surface of the modified thin layer 2. Under strain, the conductive ink pattern layer 3 can generate ordered microcracks induced by the electrospinning fibers of the modified thin layer 2, achieving a highly sensitive response of resistance to strain. When the sensor is subjected to tensile / bending strain, the conductive ink generates ordered microcracks under the induction of the electrospinning fibers. The greater the strain, the more numerous and larger the number of cracks, the more significant the obstruction of the conductive path, and the significantly increased the resistance value. When the strain is removed, the cracks close, the resistance value recovers, thereby achieving a highly sensitive strain-resistance response and completing strain measurement.

[0028] The flexible substrate 1 is made of medical-grade flexible polymer material, preferably medical-grade polyurethane tape, which has excellent skin adhesion, biocompatibility and breathability, and can be safely and stably attached to the surface of human skin, without being toxic or irritating.

[0029] The modified thin layer 2 and the flexible substrate 1 are homogeneous materials. The modified thin layer 2 can achieve surface modification by forming a nanoscale ultrathin capping layer, eliminating the need for long-term spinning to prepare a thick film, thus significantly shortening the process time and reducing the preparation cost. The homogeneous material design relies on the electric field force during the electrospinning process to spontaneously assemble and generate an irreversible bond, which can ensure a very strong bonding force between the modified thin layer 2 and the flexible substrate 1, avoiding the problems of delamination and detachment failure during use, and improving the cycle stability and service life of the sensor.

[0030] The complete electrospinning process for preparing the modified thin layer 2 is as follows: Step 1: Substrate pretreatment. Place the medical-grade polyurethane flexible substrate 1, cut to the preset size, into anhydrous ethanol and ultrasonically clean it for 4 minutes to remove dust, oil and impurities from the substrate surface. After taking it out, blow it dry with clean nitrogen and fix it flat on the negative electrode receiving plate of the electrospinning equipment to ensure that the substrate is free from warping and wrinkles and is completely attached to the receiving plate. Step 2: Preparation of spinning solution. Take medical-grade polyurethane particles of the same composition as the flexible substrate 1, add them to N,N-dimethylformamide solvent, place them in a constant temperature water bath at 60℃, and magnetically stir at 500r / min for 14h until the polyurethane particles are completely dissolved to prepare a polyurethane spinning solution with a mass fraction of 10%. Let the prepared spinning solution stand at room temperature and pressure for 45min to completely remove air bubbles in the solution and avoid defects such as broken fibers and beads during the spinning process. Step 3: High-voltage electrospinning. Inject the deaerated spinning solution into a 10mL medical syringe. The syringe is equipped with a 0.5mm inner diameter flat-tipped stainless steel needle. Connect the needle to the positive high-voltage power supply of the electrospinning equipment and set the positive voltage to +15kV. Connect the negative receiving plate with the flexible substrate 1 fixed to it to the negative high-voltage power supply of the equipment and set the negative voltage to -5kV. Adjust the vertical receiving distance between the needle and the flexible substrate 1 to 12cm. Set the syringe feed rate to 0.8mL / h and control the spinning environment temperature to 25℃ and the relative humidity to 40%. Turn on the high-voltage power supply and the feed pump. Under the action of the high-voltage electric field, the spinning solution forms a Taylor cone at the tip of the needle and dissociates into nanoscale polyurethane filaments. The filaments fly towards the flexible substrate 1 under the pull of the electric field. During the process, the solvent gradually evaporates and finally deposits on the surface of the flexible substrate 1. Under the action of the electric field, the filaments spontaneously assemble with the substrate to form a tightly bonded nanoscale spinning layer.

[0031] Reference Figure 3 The conductive ink pattern layer 3 is prepared from flexible conductive inks such as carbon nanotube conductive ink or indium tin oxide conductive ink, and the preparation process is screen printing, which can accurately prepare the preset conductive pattern. The conductive ink pattern layer 3 includes one or more independent conductive patterns, which can realize the synchronous measurement of single-point or multi-point strain and adapt to multi-dimensional motion capture requirements.

[0032] The single conductive pattern can be a straight line extending in a single direction, used to measure uniaxial strain along the length of the pattern; or an arc-shaped pattern or a serpentine pattern extending along a preset curvature. The arc-shaped pattern is used to measure bending strain along the circumference of the arc, and the serpentine pattern is used to extend the length of the conductive path within a limited area, thereby improving the resolution and sensitivity of strain detection.

[0033] The multiple independent conductive patterns are arranged on the surface of the modified thin layer 2, consisting of two or more mutually insulated and electrically disconnected independent conductive patterns, based on the location, direction, and dimension of the strain to be measured. Each independent conductive pattern can use the same or different line types, arranged in parallel, cross, or independently in different regions, corresponding to strain detection points at different parts and directions of the object under test, realizing synchronous and independent measurement of strain at multiple locations and in multiple dimensions. For example, when used for gesture recognition, independent linear conductive patterns can be arranged on the back of the hand corresponding to the five fingers, the web of the thumb, and the wrist joint; when used for silent speech recognition, independent arc-shaped conductive patterns can be arranged around the mouth corresponding to the upper lip, lower lip, and corner of the mouth. In the case of mutual independence, each independent conductive pattern is led out to the back-end signal acquisition module through an independent metal lead. The signal acquisition module obtains the real-time resistance value of each conductive pattern through polling or synchronous parallel acquisition, realizing synchronous and independent acquisition of strain signals at all independent pattern corresponding points, with no crosstalk between the signals.

[0034] For example, for hand motion capture, seven independent conductive patterns are designed, corresponding to seven strain detection points on the back of the hand: index finger, middle finger, ring finger, little finger, thumb root extension, web of the hand, and wrist joint. Each conductive pattern is a straight line pattern with a length of 15mm and a width of 1mm, and the patterns are kept at an insulation distance of more than 2mm and have no electrical connection. Each pattern has an independent metal lead bonded to both ends with conductive silver glue. All leads are connected to a flexible ribbon cable and then to a multi-channel resistance measurement circuit at the back end. The circuit uses a multiplexer to realize synchronous polling and acquisition of the seven signals, synchronously acquiring the strain signals corresponding to each finger bending, web opening and closing, and wrist joint rotation. Based on the above strain sensor, its application in gesture recognition and silent speech recognition involves measuring the real-time resistance value of the conductive ink pattern in the strain sensor through the signal acquisition module, transmitting the resistance change signal to the recognition module, and finally recognizing the gesture or silent speech information corresponding to the strain signal through the algorithm calculation of the recognition module. The entire process requires no visual intervention and is not limited by the environment or perspective.

[0035] The signal acquisition module includes a multiplexer, a resistance measurement module, and a signal transmission module. The multiplexer is used to switch the acquisition channels of multiple strain sensors to achieve synchronous polling acquisition of signals from multiple channels. The resistance measurement module is used to accurately measure the real-time resistance value of the conductive ink pattern layer. The signal transmission module supports wired or wireless transmission (preferably Bluetooth wireless transmission) and can quickly transmit the acquired electrical signals to the identification module.

[0036] The recognition module incorporates a recognition model based on the fusion of a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. The CNN takes a normalized digital signal sequence of resistance variations from multiple strain sensors over time as input, and performs sliding convolution operations through multiple one-dimensional convolutional kernels to extract spatial correlation features, local abrupt change features, and short-range temporal dependence features of signals from different channels within the same time window, capturing the coupling correlation and action feature differences between signals from multiple strain sensors. The LSTM takes a normalized digital signal sequence from multiple channels (of the same origin as the CNN) as input, and processes long-term strain signals through gating mechanisms such as input gates, forget gates, and output gates, extracting long-range temporal dependence features and dynamic change trend features of the signals within the complete action cycle. This yields the temporal correlation and evolutionary patterns of signals from different channels throughout the entire action process. The spatial correlation features output by the CNN and the temporal evolution features output by the LSTM are concatenated dimensionality-wise to complete feature layer fusion. The fused global features are then input into a fully connected neural network for feature dimensionality reduction and mapping, and finally processed through Softmax. The classifier outputs the classification and recognition results of the corresponding gesture or silent language words, completing the conversion from action to semantics.

[0037] Example 1: Applying the present invention to gesture recognition in robot control In fields such as medical robots and industrial sorting robots, hand gestures are often required to control robot operation. Compared with computer vision-based gesture recognition methods, gesture recognition methods based on strain sensors are not limited by lighting conditions and are more stable. The strain sensor of this invention can be attached to the back of the hand, using the strain sensor to capture hand movements, and transmitting the relevant signals to the recognition module through a signal acquisition module to achieve gesture recognition. Finally, robot control is performed based on the recognized gestures.

[0038] In this embodiment, medical-grade polyurethane tape is used as the base material. Polyurethane tape has the advantages of strong adhesion, non-toxicity, and breathability, allowing it to adhere to the skin on the back of the hand. The same polyurethane material is dissolved in acetone solution and electrospun onto the surface of the polyurethane tape under an electrostatic field. Then, conductive ink made of carbon nanotubes is screen-printed onto the surface of the electrospun polyurethane tape to create conductive patterns. When the conductive ink is stretched, it generates micro-cracks induced by the electrospun fibers, increasing resistance and thus achieving strain measurement. To capture the gestures of each finger, this embodiment includes seven strain sensors, corresponding to the five fingers, the web of the hand, and the wrist joint. Wires are used to lead out the conductive ink for connection to the strain measurement module.

[0039] In this embodiment, the strain measurement module includes a multiplexing module, a resistance measurement module, and a Bluetooth-based wireless transmission module, which can send the acquired resistance signal to the identification module.

[0040] In this embodiment, the recognition module employs a convolutional neural network (CNN) and a long short-term memory network (LSTM) for gesture recognition. The CNN effectively captures the interrelationships between the signals from the seven strain sensors, while the LSTM excels at processing data sequences. The recognition module combines the results from both networks to improve recognition accuracy.

[0041] Example 2: Applying the present invention to silent speech recognition in individuals with laryngeal dysfunction Following laryngectomy and vocal cord removal, individuals with laryngeal dysfunction lose their normal vocal function. In this embodiment, medical-grade polyurethane tape is used as the base material. Polyurethane tape has the advantages of strong adhesion, non-toxicity, and breathability, allowing it to adhere to the skin on the back of the hand. The same polyurethane material is dissolved in acetone solution and electrospun on the surface of the polyurethane tape under an electrostatic field. Then, conductive ink made of indium tin oxide is screen-printed onto the surface of the electrospun polyurethane tape using a screen printing method. Indium tin oxide has transparent properties, ensuring that the sensor does not affect the appearance after being attached to the perioral area. When the conductive ink is stretched, it generates micro-cracks induced by the electrospun fibers, increasing resistance and thus achieving strain measurement. To capture perioral movement, this embodiment includes three sensors corresponding to the upper lip, lower lip, and corners of the mouth. Wires are used to lead out the conductive ink for connection to the strain measurement module.

[0042] In this embodiment, the strain measurement module includes a multiplexing module, a resistance measurement module, and a Bluetooth-based wireless transmission module, which can send the acquired resistance signal to the identification module.

[0043] In this embodiment, the recognition module employs a convolutional neural network (CNN) and a long short-term memory network (LSTM) for speech recognition. The CNN effectively captures the interrelationships between the signals from the three strain sensors, while the LSTM excels at processing data sequences. The recognition module combines the recognition results from both networks to improve accuracy. Finally, the recognition results are used to control the speaker or the screen to display the corresponding text.

Claims

1. A strain sensor, comprising a flexible substrate (1), a modified thin layer (2), and a conductive ink pattern layer (3) sequentially disposed thereon; characterized in that, The modified thin layer (2) is prepared in situ on the surface of the substrate (1) by electrospinning process; the conductive ink pattern layer (3) is printed on the surface of the modified thin layer (2). Under the action of strain, the electrospinning fibers of the modified thin layer (2) induce the conductive ink pattern layer (3) to generate ordered microcracks, thereby realizing a high-sensitivity response of resistance to strain.

2. A strain sensor according to claim 1, characterized in that, The flexible substrate (1) is made of medical-grade flexible polymer material.

3. A strain sensor according to claim 1, characterized in that, The modified thin layer (2) and the flexible substrate (1) are homogeneous materials. They spontaneously assemble by the electric field force during the electrospinning process and generate an irreversible bond, ultimately forming an ultrathin covering layer with a thickness of nanometers. The specific preparation process of the modified thin layer (2) includes: Step 1: Cut the flexible substrate (1) to the preset size, and use anhydrous ethanol to ultrasonically clean and dry the surface to be spun; The cleaned flexible substrate (1) is flattened and fixed on the negative electrode receiving plate of the electrospinning equipment to ensure that the flexible substrate (1) is completely attached to the receiving plate without warping. Step 2: Take polyurethane particles of the same material as the flexible substrate (1), add them to N,N-dimethylformamide solvent, stir to dissolve, prepare a polyurethane spinning solution with a mass fraction of 8%~12%, and let it stand to remove bubbles; Step 3: Inject the defoamed spinning solution into the syringe of the electrospinning equipment. The inner diameter of the syringe needle is 0.4~0.6mm. Connect the needle to the positive high voltage power supply of the electrospinning equipment and set the positive voltage to +12kV~+18kV. Connect the negative electrode of the receiving plate with the flexible substrate (1) fixed to it to the negative high voltage power supply and set the negative voltage to -3kV~-8kV. Control the receiving distance between the needle and the flexible substrate (1) to 10~15cm, set the syringe push rate to 0.5~1.2mL / h, control the ambient temperature to 22~28℃, and control the relative humidity to 30%~50%. Under the action of the high voltage electric field, the spinning solution dissociates from the tip of the needle into nano-scale filaments, which fly to the flexible substrate (1) of the negative electrode under the traction of the electric field and spontaneously assemble and tightly bond with the substrate surface under the action of electrostatic force.

4. A strain sensor according to claim 1, characterized in that, The conductive ink pattern layer (3) is prepared using flexible conductive inks such as carbon nanotube conductive ink or indium tin oxide conductive ink, and the preparation process is screen printing.

5. A strain sensor according to claim 1, characterized in that, The conductive ink pattern layer (3) includes one or more independent conductive patterns, which can realize the synchronous measurement of single-point or multi-point strain; The single conductive pattern is a straight line pattern extending in a single direction, used to measure uniaxial strain along the length of the pattern; or an arc pattern or a serpentine pattern extending along a preset curvature, wherein the arc pattern is used to measure bending strain along the circumference of the arc, and the serpentine pattern is used to extend the length of the conductive path within a limited area, thereby improving the resolution and sensitivity of strain detection. The multiple independent conductive patterns use the same or different line types and are arranged in parallel, cross, or independently in different regions, respectively corresponding to strain detection points in different parts and directions of the object under test; in the case of mutual independence, each independent conductive pattern is led out to the back-end signal acquisition module through an independent metal lead.

6. The application of a strain sensor in gesture recognition and silent speech recognition according to claim 1, characterized in that, Strain sensors are attached to the back of the hand to capture hand movements. This can be used to achieve gesture recognition, or strain sensors can be attached to the perioral face to capture perioral movements.

7. The application of a strain sensor in gesture recognition and silent speech recognition according to claim 6, characterized in that, The application is as follows: the strain sensor converts the mechanical strain signal to be measured into a resistance change signal; the strain sensor is electrically connected to the signal acquisition module, the signal acquisition module acquires the resistance change signal of the strain sensor and transmits the signal to the recognition module; the recognition module receives the resistance change signal and processes it to identify the corresponding hand gesture or silent language information.

8. The application of a strain sensor in gesture recognition and silent speech recognition according to claim 7, characterized in that, The signal acquisition module includes a multiplexer, a resistance measurement module, and a signal transmission module; the multiplexer is used to switch the acquisition channels of multiple strain sensors to achieve synchronous polling acquisition of multi-channel signals; The resistance measurement module is used to accurately measure the real-time resistance value of the conductive ink pattern layer; the signal transmission module supports wired or wireless transmission and transmits the collected electrical signals to the identification module.

9. The application of a strain sensor in gesture recognition and silent speech recognition according to claim 7, characterized in that, The recognition module incorporates a recognition model based on the fusion of a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. The CNN takes a sequence of digital signals showing the resistance of multiple sensors changing over time as input to capture the correlation features between the signals from multiple strain sensors. The LSTM takes a sequence of digital signals showing the resistance of multiple sensors changing over time as input to process the time-series strain signals and obtain the long-range relationships between the channels. The spatial correlation features output by the CNN and the temporal evolution features output by the LSTM are concatenated to complete the feature layer fusion. The fused global features are then input into a fully connected neural network for feature dimensionality reduction and mapping. Finally, a Softmax classifier outputs the classification and recognition results for the corresponding gestures or silent language words, completing the conversion from action to semantics.

10. The application of a strain sensor in gesture recognition and silent speech recognition according to claim 6, characterized in that, The applications described correspond to gesture recognition scenarios controlled by robots and silent speech recognition scenarios for people with laryngeal dysfunction, respectively.

Images