A stress sensing material and sensor for motion recognition and preparation method and application thereof

Abstract

The present invention belongs to the technical field of flexible stress sensors, and specifically relates to a stress sensing material and sensor for motion recognition, as well as a preparation method and application thereof. PPy is attached to the surface of MXene nanosheets. The ultra-high conductivity of PPy increases the sensitivity of the composite sensing material. While MXene provides more conductive active sites for PPy, PPy also protects the Ti atoms in MXene from being oxidized. A flexible pressure sensor with a porous structure is prepared by depositing MXene / PPy on the surface of a PDMS sponge skeleton through low-temperature drying. The flexible stress sensor is applied to motion recognition by combining it with a multi-input CNN convolutional neural network. The present invention designs a flexible pressure sensor based on the porous structure of PDMS sponge, and aims at the requirements of large-scale preparation and high performance to realize the accurate detection of human motion signals and the detection and recognition of 16 kinds of motions.

Description

Technical Field

[0001] The present invention belongs to the technical field of flexible stress sensors, and in particular relates to a stress sensing material and sensor for motion recognition, and a preparation method and application thereof. Background Art

[0002] Flexible stress sensors can be attached to various key parts of the body to monitor their various activities. The development of flexible stress sensors has attracted great attention due to the satisfactory detection results and wide application prospects. Due to the demand for high-performance devices, various categories of stress sensors such as capacitive, resistive, piezoelectric, and triboelectric sensing mechanisms have gradually moved from laboratory research to commercialization for the next generation of wearable electronic devices. Among these categories, resistive stress sensors based on the piezoresistive effect have occupied a considerable share in this field due to their advantages of easy signal processing and dynamic sensing. Factors such as poor flexibility, low sensitivity, slow response / recovery, narrow measurement range, and complex preparation process still limit the development of flexible pressure sensors. Therefore, it remains a challenge to design high-performance flexible pressure sensors with simple manufacturing strategies.

[0003] At present, some flexible stress sensors with high sensing performance and good stability have been developed, and have made certain progress in sensitivity, detection limit and other aspects. However, they still face huge challenges in practical application. The reasons are: flexible sensors are difficult to overcome the changing and complex environment. In addition, as a device for detecting human movement, it is necessary to achieve small size, multi-function, low power consumption and other necessary factors. The device also needs to be flexible and fixed on the body without burdening the movement.

[0004] Flexible stress sensors have many practical and potential applications, including health signal monitoring, human motion detection, intelligent perception, and human-computer interaction. They are fixed on the surface of human skin to measure pressure signals due to their good biocompatibility. Through reasonable structural optimization, the sensor can distinguish the activities of different parts of the body. By building a sensor matrix and combining it with neural networks and machine learning, the processing, classification, and identification of large amounts of data can be realized, broadening the use scenarios of wearable flexible pressure sensors. It has a very broad application prospect in the medical field, the field of people's livelihood and health, and the field of sports rehabilitation. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a stress sensing material and sensor for motion recognition, as well as a preparation method and application thereof, to achieve accurate monitoring of motion signals and physiological signals of multiple parts of the human body and detection and recognition of four common motions. The present invention uses a sacrificial sugar template PDMS (polydimethylsiloxane) sponge as a flexible substrate, prepares a MXene / PPy (polypyrrole) composite material as a stress-sensitive material, designs a flexible pressure sensor based on a porous structure, and builds a multi-layer convolutional neural network for the needs of wearable, information processing and classification, to achieve detection and accurate recognition of motion signals.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a stress sensing material for motion recognition, in-situ polymerization of pyrrole in a MXene solution and uniform deposition of pyrrole nanoparticles on the surface of MXene nanosheets to form a coating that protects Ti atoms to prevent MXene from being oxidized, a compound of two-dimensional MXene nanosheets and zero-dimensional PPy nanoparticles increases the conductive pathway within the sensing layer, MXene provides more conductive active sites for PPy, and the ultra-high conductivity of PPy increases the sensitivity of the composite sensing material.

[0007] A method for preparing a stress sensing material for motion recognition, the preparation steps comprising:

[0008] (1) adding lithium fluoride into hydrochloric acid to completely dissolve it, thereby forming a hydrochloric acid / lithium fluoride colorless solution;

[0009] (2) Slowly adding aluminum titanium carbide powder, the hydrochloric acid / lithium fluoride solution reacts with it to etch it, and the solution turns from colorless to black;

[0010] (3) Repeatedly centrifuge and pour off the supernatant to etch the Al atoms in the aluminum titanium carbide and the F atoms in the solution, control the pH of the solution to ensure that the byproducts of the etching reaction are completely discharged, and obtain a black multilayer MXene solution;

[0011] (4) Ultrasonic stripping of the black solution to obtain a few-layer MXene solution;

[0012] (5) Take the DBSA solution and pyrrole monomer and add them to deionized water, stir vigorously in an ice water bath to enhance the conductivity of pyrrole, and then add the few-layer MXene solution to mix the two thoroughly;

[0013] (6) Add FeCl 3 -6H 2 O, continue stirring in an ice-water bath, pyrrole monomers begin to self-polymerize on the surface of MXene nanosheets and form a continuous conductive layer, centrifuge several times to remove DBSA and FeCl 3 -6H 2 O to obtain MXene / PPy composite solution.

[0014] Furthermore, the mass ratio of lithium fluoride to aluminum carbon titanium is 1:1;

[0015] The reaction conditions of (2) are sufficient stirring at 35° C. for 24 hours;

[0016] Centrifuging in (3) until the pH value of the supernatant is greater than 6;

[0017] The volume ratio of DBSA solution to pyrrole monomer in (5) is 4:1; the volume ratio of pyrrole monomer to MXene solution is 1:80, the concentration of MXene solution is 10 mg / mL, and the reaction time is 1 h;

[0018] In (6), FeCl 3 -6H 2 The mass ratio of O to MXene was 5:1, and the reaction time was 6 h.

[0019] A stress sensor for motion recognition is prepared by depositing MXene / PPy on the surface of a PDMS sponge skeleton through low-temperature drying to obtain a flexible pressure sensor with a porous structure.

[0020] Furthermore, the sensing principle of the sensor is:

[0021] The stress sensor belongs to the adsorption type stress sensor. The sensing material is adsorbed on the surface of the PDMS sponge. The change in resistance mainly comes from the conductive path inside the sensor. When no pressure is applied, the conductive path only comes from the mutual contact between the sensitive materials on the same sponge skeleton. When pressure is applied, the sensitive materials on different sponge skeletons are connected to each other to form a new conductive path, which reduces the resistance of the sensor.

[0022] Or, under slight stress, there are still a large number of micropores inside the sponge space. As the pressure increases, the sponge skeleton will collapse rapidly. During the collapse process, sensitive materials are constantly in contact with each other. At this stage, the wrinkled MXene nanosheets are mainly combined with each other, which rapidly increases the conductive pathway inside the sensor and causes the resistance to drop rapidly. As the pressure increases further, most of the air inside the sponge is squeezed out by the pressure, leaving only some pores that have not collapsed. The combination of MXene nanosheets is basically complete, and the increase in the conductive pathway mainly comes from the mutual contact between the PPy nanoparticles, which begins to tend to saturation. When the pressure continues to increase, the increase in stress mainly comes from the resistance of the sponge skeleton. Therefore, the pressure range of this part is larger. At this time, only some PPy nanoparticles are in contact with each other, resulting in a very slow decrease in sensor resistance and a significant decrease in sensitivity.

[0023] A method for preparing a stress sensor for motion recognition, the preparation steps comprising:

[0024] (1) placing sugar granules into a hexahedral container with an open top, and compacting the sugar granules into sugar cubes of regular shape;

[0025] (2) A solution of PDMS and a curing agent mixed in a ratio of 10:1 is injected into a container and vacuum dried, so that the micropores inside the sugar cube are filled with the mixture;

[0026] (3) After PDMS solidifies, the sugar cube / PDMS fusion is immersed in water to remove the white sugar to obtain a PDMS porous sponge of the same size as the original sugar cube;

[0027] (4) Immerse the PDMS porous sponge in the MXene / PPy solution for 1 hour, dry it at a temperature not greater than 40°C for 6 hours, and repeat this process multiple times to obtain a MXene / PPy@PDMS sponge sensor.

[0028] An application of a stress sensor for motion recognition,

[0029] The stress sensor is applied to human motion signal detection or action recognition based on neural network.

[0030] The human motion signal detection includes bite muscle detection, throat movement detection, head tilt movement detection, finger bending detection, elbow bending detection, knee bending detection, walking state detection and radial artery pulse detection.

[0031] Furthermore,

[0032] The neural network includes an input layer, a convolutional layer, a pooling layer, an initial layer and a fully connected layer;

[0033] The convolution layer is used to extract features from the input data, the pooling layer is used to compress the input feature map and simplify the calculation complexity. The initial layer consists of two convolution layers and one pooling layer, and the fully connected layer is used to integrate the extracted features and perform calculation and recognition.

[0034] Furthermore, we first train a multi-input CNN model by using a large amount of measurement data from different actions. Secondly, we iteratively train the model using the model error on the dataset to obtain a model that reasonably fits the dataset. Finally, we apply the trained multi-input CNN model to actual scenarios.

[0035] Furthermore,

[0036] Seven stress sensors were prepared and fixed on the skin of the heel, knee, elbow and back of the neck of the tester to detect the joint activities or contact with the ground in these parts. The basic architecture of the system includes data acquisition, feature extraction, normalization, training and testing procedures. In the training phase, the training data are obtained from the MXene / PPy@PDMS stress sensor and normalized, and a multi-input CNN is trained for all activities. On the other hand, the testing phase includes test data acquisition, normalization and application to the trained multi-input CNN to identify activities or behaviors from the test data.

[0037] The advantages of the stress sensing material and sensor for motion recognition and the preparation method and application thereof of the present invention are as follows:

[0038] (1) The flexible stress sensor prepared by the present invention has a high sensitivity (6.8925 kPa -1 ) and a wide detection range (0.43Pa-275kPa), the porous structure-based MXene / PPy@PDMS sponge stress sensor shows a fast response recovery time (100ms / 110ms) and excellent compression stability of more than 7000s, and can identify strains ranging from small deformations to a wide range, such as pulse signals, joint bending, walking status and vocalization.

[0039] (2) The present invention designs a multi-input CNN neural network algorithm to learn the human fitness exercise data obtained by the MXene / PPy@PDMS sponge stress sensor array, and realizes the recognition and judgment of four actions: squat, lunge, push-up and high leg raise. For each action, the correct action and three corresponding non-standard actions can be judged. The accuracy rate of identifying different actions is 95%. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The preparation process of the MXene / PPy@PDMS stress sensor according to an embodiment of the present invention and the schematic diagram of the microscopic and molecular morphology of MXene and PPy;

[0041] Figure 2 a is a schematic diagram of a piezoresistive response sensing platform according to an embodiment of the present invention;

[0042] b is the compressive stress-strain curves of MXene / PPy@PDMS sponge, MXene@PDMS sponge, and PPy@PDMS sponge according to the embodiment of the present invention;

[0043] c is a cyclic stress-strain curve of the stress sensor according to an embodiment of the present invention;

[0044] d is a stress-strain cycle curve of 1-100 repeated times of the stress sensor according to an embodiment of the present invention;

[0045] e is the maximum pressure and dissipated energy of the stress sensor of the embodiment of the present invention when the stress strain is repeated 1-100 times;

[0046] Figure 3 a is the maximum hysteresis of the stress sensor according to the embodiment of the present invention;

[0047] b is the stress of the MXene / PPy@PDMS stress sensors prepared by the same method in six embodiments of the present invention under 70% strain;

[0048] c is the stress of the MXene / PPy@PDMS stress sensors prepared by the same method in six embodiments of the present invention under 5% strain;

[0049] Figure 4 Schematic diagram of the preparation mechanism of MXene / PPy in the MXene / PPy@PDMS stress sensor according to an embodiment of the present invention;

[0050] Figure 5 Schematic diagram of the piezoresistive sensing mechanism of the stress sensor according to an embodiment of the present invention;

[0051] Figure 6 a and b are photos of the stress sensor according to the embodiment of the present invention connected to an LED bulb in the initial and compressed states;

[0052] Figure 7 a is a comparison of stress sensitivity curves of MXene / PPy@PDMS sponge, MXene@PDMS sponge, and PPy@PDMS sponge in an embodiment of the present invention;

[0053] b is a stress sensitivity curve diagram of the MXene / PPy@PDMS stress sensor according to an embodiment of the present invention;

[0054] c is a strain sensitivity curve of the MXene / PPy@PDMS stress sensor according to an embodiment of the present invention;

[0055] Figure 8 a is the IV curve of the stress sensor of the embodiment of the present invention under strains of 0%, 5%, 10%, 20%, and 30%;

[0056] b is the IV curve of the stress sensor of the embodiment of the present invention at strains of 40%, 50%, 60%, 70%, and 80%;

[0057] c is a comparison of the sensitivity and pressure range performance parameters of the stress sensor of the embodiment of the present invention and other research works;

[0058] d is the response / recovery curve of the stress sensor of the embodiment of the present invention under 40% rapid strain;

[0059] e is a resistance signal diagram of the stress sensor according to an embodiment of the present invention at compression rates of 10 mm / min, 20 mm / min, 30 mm / min, 40 mm / min, and 50 mm / min;

[0060] f is a resistance signal diagram of the stress sensor of the embodiment of the present invention at compression rates of 100 mm / min, 150 mm / min, 200 mm / min, 250 mm / min, and 300 mm / min;

[0061] Fig. 9 a is a 0-80% stress step response curve of the stress sensor according to an embodiment of the present invention;

[0062] The cyclic response curve diagram of the stress sensor of the embodiment of the present invention at the same frequency at 5%, 20%, 40%, 60%, and 80%;

[0063] c is a diagram of a stress sensor according to an embodiment of the present invention subjected to a long-term continuous compression-release test of more than 7000 seconds;

[0064] Fig.10 The stress sensor of the present invention detects human physiological and motion signals;

[0065] a is a resistance signal diagram when the stress sensor of an embodiment of the present invention detects pharyngeal activity (swallowing action, "MXene", "sponge" pronunciation);

[0066] b is a resistance signal diagram of the stress sensor fixed on the occlusal muscle of the embodiment of the present invention when detecting the occlusal action;

[0067] c is a diagram of a stress sensor according to an embodiment of the present invention measuring a human radial artery pulse signal;

[0068] d is a resistance signal diagram of the stress sensor fixed at the back of the neck in the embodiment of the present invention when the head is tilted back;

[0069] e is a resistance signal diagram of the stress sensor fixed at the elbow joint when the elbow is flexed according to an embodiment of the present invention;

[0070] f is a resistance signal diagram of the stress sensor at the finger joint when the finger is bent at different angles according to the embodiment of the present invention;

[0071] g is a resistance signal diagram of the stress sensor at the knee joint when the leg is bent according to an embodiment of the present invention;

[0072] h is a resistance signal diagram of the stress sensor at the heel during walking and running according to an embodiment of the present invention;

[0073] Fig.11 This is a resistance signal diagram of a stress sensor fixed on the joint of the index finger according to an embodiment of the present invention detecting a single or double click of a mouse;

[0074] Fig.12 a is a graph showing the resistance response and stress curve of the stress sensor at 50% strain before cleaning in deionized water according to an embodiment of the present invention;

[0075] b is a graph showing the resistance response and stress curve of the stress sensor at 50% strain after being cleaned in deionized water according to an embodiment of the present invention;

[0076] c is a comparison diagram of resistance signal curves of the stress sensor detecting knee bending before and after cleaning in deionized water according to an embodiment of the present invention;

[0077] d is a comparison diagram of resistance signal curves of the stress sensor detecting finger bending before and after cleaning in deionized water according to an embodiment of the present invention;

[0078] Fig.13 This is a schematic diagram of the minimum detection limit of the stress sensor according to an embodiment of the present invention;

[0079] Fig.14 This is a schematic diagram of a stress sensor according to an embodiment of the present invention being fixed to a joint of a human body;

[0080] Fig.15 It is a curve diagram of resistance signal of stress sensors at different parts during lunge squat, high leg raise, deep squat and push-up movements according to an embodiment of the present invention;

[0081] Fig.16 ad is a schematic diagram of detecting fitness postures in an embodiment of the present invention: correct squatting, insufficient squatting depth, excessive forward leaning of the body during squatting, and heel lifting during squatting;

[0082] eh is a graph showing the resistance response signals of seven stress sensors for detecting fitness postures according to an embodiment of the present invention: correct squatting, insufficient squatting depth, excessive forward leaning of the body during squatting, and heel lifting during squatting;

[0083] Fig.17 ad is a schematic diagram of detecting fitness postures in an embodiment of the present invention: correct push-ups, push-ups with bent knees, push-ups with insufficient elbow flexion, and push-ups with shrugs;

[0084] eh is a graph of seven stress sensor resistance response signals for detecting fitness postures in an embodiment of the present invention: correct push-ups, push-ups with bent knees, push-ups with insufficient elbow flexion, and push-ups with shrugs;

[0085] Fig.18 ad is a schematic diagram of detecting fitness postures according to an embodiment of the present invention: correct high leg raising, high leg raising without arm swinging, high leg raising with insufficient height, and high leg raising with the body leaning forward;

[0086] eh is a graph of seven stress sensor resistance response signals for detecting fitness postures according to an embodiment of the present invention: correct high leg raising, high leg raising without arm swinging, high leg raising with insufficient leg raising height, and high leg raising with body leaning forward;

[0087] Fig.19 AD is a schematic diagram of detecting fitness postures according to an embodiment of the present invention: correct lunge, lunge with knees touching the ground, lunge with body leaning back, and lunge with unstable center of gravity;

[0088] eh is a graph of seven stress sensor resistance response signals for detecting fitness postures according to an embodiment of the present invention: correct lunge, lunge with knees touching the ground, lunge with body leaning back, and lunge with unstable center of gravity;

[0089] Fig. 20 A schematic diagram of data training and recognition based on a multi-input CNN neural network according to an embodiment of the present invention;

[0090] Fig.21 It is a flow chart of a posture recognition system based on a stress sensor proposed in an embodiment of the present invention;

[0091] Fig. 22 A test confusion matrix of a multi-input CNN neural network model according to an embodiment of the present invention;

[0092] Fig.23 The accuracy and loss curves of the test set and training set at different training epochs in the embodiment of the present invention. DETAILED DESCRIPTION

[0093] The specific implementation manner will be further described below in conjunction with the accompanying drawings.

[0094] Embodiment 1:

[0095] A stress sensing material for motion recognition, a MXene nanosheet solution was prepared by chemical etching, and pyrrole nanoparticles were agglomerated on the MXene nanosheets by in-situ polymerization to prepare a MXene / PPy composite nanomaterial. MXene has abundant active sites and a large specific surface area, which can allow PPy to adhere well to its surface, and its ultra-high electrical conductivity can make the stress-sensitive composite material highly sensitive under low stress. At the same time, the uniform deposition of pyrrole nanoparticles on the MXene nanosheets can show the role of a coat to protect Ti atoms and prevent MXene from oxidation.

[0096] A method for preparing a stress sensing material for motion recognition,

[0097] The preparation of MXene / PPy stress-sensitive composite materials mainly includes two steps. The first step is to prepare MXene nanosheet solution. The preparation method is to etch Ti 3 AlC 2The specific process is as follows: First, add 2 grams of lithium fluoride to 20 milliliters of hydrochloric acid (9M) and stir for 5 minutes to completely dissolve it to form a hydrochloric acid / lithium fluoride solution. Then slowly add 2 grams of aluminum titanium carbide and stir at 35°C for 24 hours to ensure that the Al layer is completely separated from the Ti layer. 3 C 2 The interaction between them is weakened, making it easier to be peeled off. The product of the etching reaction is then added to deionized water, centrifuged at 3500rpm for 5 minutes, and the supernatant is poured out. This step is repeated several times until the pH value of the supernatant is greater than 6. The purpose of this step is to ensure that the byproducts of the etching reaction are completely washed away to obtain multilayer MXene (m-MXene). The m-MXene is dispersed in deionized water, ultrasonically peeled for 1 hour, and then centrifuged at 3500rpm for 1 hour. The black-green supernatant is collected to finally obtain a few-layer MXene solution (f-MXene). The second step is the adsorption of MXene nanosheet solution by PPy nanoparticles. 2 ml of DBSA and 0.5 ml of pyrrole monomer were added to 40 ml of deionized water, then vigorously stirred in an ice water bath for 1 hour, and then 40 ml of 10 mg / ml f-MXene solution was added to mix the two thoroughly. The role of adding DBSA is to increase the conductivity of PPy so that it can be more fully deposited on MXene nanosheets through electrostatic adsorption. After 1 hour, 2 grams of FeCl 3 -6H 2 O and stirred in an ice-water bath for 6 h. During this period, the solution gradually turned black, indicating that the pyrrole monomer began to self-polymerize on the surface of the MXene nanosheets and formed a continuous conductive layer. Finally, the reaction product was centrifuged at 8000 rpm for several times to remove DBSA and FeCl 3 -6H 2 O, and a MXene / PPy composite solution was obtained.

[0098] Embodiment 2:

[0099] A stress sensor for motion recognition is prepared by depositing MXene / PPy on the surface of PDMS sponge through an impregnation method and a low-temperature drying method to obtain a flexible stress sensor with a porous structure.

[0100] A method for preparing a stress sensor for action recognition, wherein a PDMS sponge is prepared by a sacrificial sugar template method. White sugar (particle size of 0.8-2.5 mm) is placed in a hexahedral container with an opening at the top and a length and width of about 2 cm. The sugar particles are compacted into a regular-shaped sugar cube (2 cm × 2 cm × 1 cm), and then a solution of PDMS and a curing agent mixed in a ratio of 10:1 is injected into the container. After 2 hours of vacuum drying, the micropores inside the sugar cube are filled with the mixture. After PDMS is cured, the sugar cube / PDMS fusion is immersed in water to remove the white sugar to obtain a PDMS porous sponge of the same size as the original sugar cube. The MXene / PPy@PDMS sponge is prepared by coating the MXene / PPy composite material on the PDMS sponge skeleton by a solution impregnation and low-temperature drying method, that is, the PDMS sponge is immersed in the MXene / PPy solution for 1 hour and then dried at a temperature not greater than 40°C for 6 hours, and this step is repeated multiple times. For comparison, MXene@PDMS sponge and PPy@PDMS sponge were also prepared using the same preparation process. Figure 1 Schematic diagram of the preparation process of MXene / PPy@PDMS stress sensor and the microscopic and molecular morphology of MXene and PPy.

[0101] Figure 2 This is a mechanical property test of the flexible stress sensor. a is a comprehensive performance test and analysis of stress and strain of the device using a ZQ-900A comprehensive tensile tester, an Agilent data acquisition instrument and a host computer. b is the stress-strain curve of MXene / PPy@PDMS sponge, MXene@PDMS sponge and PPy@PDMS sponge. It can be seen from the figure that the composite of MXene and PPy significantly increases the detection range of the sensor, and shows a higher stress response overall under the same strain, especially in the strain stage of 10% to 70%. At the same time, the stress differentiation of the sensor under different strains is improved as a whole. Compared with MXene@PDMS sponge, the stress performance of MXene / PPy@PDMS sponge has increased by 5 times. c is the cyclic compression stress-strain curve of the stress sensor under 10%, 20%, 40%, 60%, and 80% strain. It can be seen that the overall stress response can be divided into two stages. The first stage is 0-40% strain ( Figure 2 c), at this time the increase in stress and strain is almost linear. This is because there are still many pores inside the sponge, and the increase in stress at this stage mainly relies on the support of the sponge skeleton. The second stage is 40%-80% strain. At this stage, as the strain increases, the stress increases rapidly. The reason is that most of the pores inside the sponge disappear at this time, and the sponge skeletons begin to contact each other, causing the resistance to increase rapidly. Therefore, compared with the first stage, the stress increase in the second stage is more significant.

[0102] d is the stress-strain test curve of 100 consecutive compressions at 60% strain. It can be seen that with the increase in the number of cycles, the stress performance of the MXene / PPy@PDMS stress sensor only decreases slightly, and the curves of different stress cycles almost overlap, which proves that the porous structure of the PDMS sponge prepared by the sacrificial sugar template method has good stability. e is a further analysis of the 100 compression stress-strain of the MXene / PPy@PDMS stress sensor. With the increase in the number of cycles, the maximum pressure required for the sensor to reach 60% strain (the stress was 61.23 kPa at the first time and 58.74 kPa at the 100th time) did not change significantly. The reason for the stress drop is the stress relaxation of the sponge skeleton. In addition, after 100 compression stress-strain tests, the MXene / PPy@PDMS stress sensor still showed 0.02174 kJ m -3 The higher dissipated energy further proves that the structure has excellent fatigue resistance and repeated stability.

[0103] Figure 3 a is the hysteresis curve of the MXene / PPy@PDMS stress sensor. Because the stress-strain curve of the device during compression and the curve during release basically coincide, the hysteresis shows a small value of 10%. b and c are six MXene / PPy@PDMS stress sensors made using the same preparation method and the stress-strain tests of the six devices under the same conditions. Under large strain (70%) and small strain (5%), the six sensors all show almost the same stress response (the maximum stress and the minimum stress differ by only 4%).

[0104] The sensing principle of the porous structure MXene / PPy@PDMS stress sensor is:

[0105] MXene / PPy@PDMS stress sensor is an adsorption-type stress sensor, and the sensing material is adsorbed on the surface of the PDMS sponge skeleton. Figure 4 Schematic diagram of the preparation process of MXene / PPy in MXene / PPy@PDMS stress sensor; Figure 5 The schematic diagram of the piezoresistive behavior mechanism of MXene / PPy stress-sensitive materials; MXene / PPy@PDMS stress sensors are prepared by electrostatically adsorbing PPy nanoparticles on MXene nanosheets using in-situ polymerization, and silver electrodes are encapsulated on the upper and lower sides of the sponge to complete the device preparation. The composite structure of zero-dimensional and two-dimensional materials can increase the conductive sites and further increase the conductive pathways inside the sensor. Figure 5As shown in the figure, the MXene nanosheets attached to the sponge skeleton play a good linking role. When no pressure is applied, the conductive path only comes from the mutual contact of the sensitive materials on the same sponge skeleton. When pressure is applied, the MXene nanosheets on different sponge skeletons connect to form a new conductive path, which reduces the resistance of the sensor.

[0106] like Figure 6 As shown in Figure 1, a simple circuit was constructed by connecting the MXene / PPy@PDMS stress sensor with a light-emitting diode (LED). The LED bulb can be ignited by the negative piezoresistance effect of the MXene / PPy@PDMS stress sensor, which is caused by the increase in internal conductive paths during compression in the sensor, resulting in a decrease in resistance.

[0107] Comparison of sensitivity of porous structure MXene / PPy@PDMS stress sensor:

[0108] Sensitivity is defined as the change in relative resistance per unit pressure change and is calculated as follows:

[0109] S = ΔR / (R 0 ×P)

[0110] Among them, ΔR represents the difference between the current resistance and the initial resistance; R 0 It is represented by the initial resistance; P is represented by the current pressure value. S determines the sensitivity performance of the pressure sensor. The larger S is, the better the sensing performance of the sensor.

[0111] In order to prove that composite materials have advantages over single materials, PDMS sponge sensors based on MXene, PPy and MXene / PPy were developed. Figure 7 As shown in a, the most sensitive pressure areas of MXene / PPy@PDMS, MXene@PDMS and PPy@PDMS stress sensors are 0-15kPa, 0-25kPa and 25-50kPa, respectively. MXene / PPy@PDMS stress sensors and MXene@PDMS stress sensors show obvious resistance response at the beginning of pressure loading. This change is attributed to the following reasons. Under the action of pressure, the MXene nanosheets that are wrinkled and stacked on each other will contact and separate. This contact separation and tunneling effect at nanometer distances form local conductive paths, which makes the electrical properties of the two sponges and PPy@PDMS stress sensors show obvious changes. The stress response curve quickly reaches the maximum sensitivity in the initial stage, and then maintains linear growth until the turning point, while the PPy@PDMS stress sensor will have a resistance retention space in the initial stage. The maximum sensitivity of MXene / PPy@PDMS, MXene@PDMS and PPy@PDMS stress sensors is 6.8925kPa, respectively.-1 , 1.3888kPa -1 , 1.2601kPa -1 This is because the combination of two-dimensional nanomaterials MXene and PPy allows the sensor to produce more conductive pathways when pressure is applied, so the MXene / PPy@PDMS stress sensor has a higher sensitivity than the MXene@PDMS stress sensor and PPy@PDMS stress sensor.

[0112] Figure 7 Curve b shows the relative resistance of the MXene / PPy@PDMS stress sensor as the stress changes. The curve is mainly divided into three parts. The first part is the stress change of 0-15kPa, and the sensor sensitivity is 6.8925kPa -1 There are still a lot of micropores inside this part of the sponge space. As the pressure increases, the sponge skeleton will collapse rapidly. During the collapse process, the sensitive materials are constantly in contact with each other. In this stage, the wrinkled MXene nanosheets are combined with each other, which rapidly increases the conductive path inside the sensor. Therefore, the resistance value drops rapidly and the sensitivity is significantly higher than other stages. The second part is the stress change of 15kPa-75kPa, and the sensor sensitivity is 1.0136kPa -1 At this stage, most of the air inside the sponge is squeezed out by pressure, and only some pores remain uncollapsed. The bonding of MXene nanosheets is basically complete. The increase in conductive pathways mainly comes from the mutual contact between PPy nanoparticles, which begins to approach saturation. The third part is 75kPa-275kPa, and the sensor sensitivity is 0.0236kPa -1 The increase in stress in this part mainly comes from the resistance of the sponge skeleton, so the pressure range in this part is larger. At this time, only some PPy nanoparticles are in contact with each other, resulting in a very slow decrease in sensor resistance and a significant decrease in sensitivity. The quality factor (GF) is defined as:

[0113]

[0114] Where R 0 is the resistance in the initial state, R is the resistance in the stretched state, and ε is the mechanical strain applied to the sensor. Figure 7 c is the GF test curve of the MXene / PPy@PDMS stress sensor. It can be seen that the sensor has a high quality factor of 1.0474. In addition, with the change of strain, the resistance response value also shows a linear change, which proves the good linearity of the device. Figure 8 a and Figure 8b shows the change of current of MXene / PPy@PDMS stress sensor under 0, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, and 80% strain as the voltage changes from -1V to 1V. It can be seen that no matter what the strain state is, the curve appears as a straight line, which indicates that when the pressure is fixed, the resistance of the sensor is relatively stable. Moreover, as the strain increases, the slope of the IV curve also increases, reflecting the good linearity of the sensor. Comparison with other work also proves that MXene / PPy@PDMS stress sensors have higher sensitivity and wider measurement range in the industry, making up for the regret that flexible pressure sensors have low sensitivity under a wide measurement range ( Figure 8 c).

[0115] Dynamic performance test of porous structure MXene / PPy@PDMS stress sensor:

[0116] In order to verify whether the porous structure MXene / PPy@PDMS stress sensor can be stably applied in the field of human motion detection in the long term, it is necessary to test the dynamic performance of the sensor device, such as response time and repeatability. By setting the loading program of the test platform, the response and recovery time, compression at different frequencies, step response, compression at different speeds, and long-term stability of the MXene / PPy@PDMS stress sensor were tested.

[0117] Figure 8 d represents a 30% instantaneous strain applied to the MXene / PPy@PDMS stress sensor. The sensor is able to respond to the strain in a timely manner, with a response time and recovery time of 100ms and 110ms, respectively. In order to study the cyclic response behavior of the MXene / PPy@PDMS stress sensor at different frequencies. Figure 8 e and Figure 8 As shown in Figure f, under a fixed strain of 40%, in order to adjust the frequency, the compression speed of the tensile machine was set to 10mm / min, 20mm / min, 30mm / min, 40mm / min, 50mm / min, 100mm / min, 150mm / min, 200mm / min, and 300mm / min. Under the same strain and different compression frequencies, the resistance response remained stable and the curve shape remained consistent, proving that the MXene / PPy@PDMS stress sensor has good frequency stability.

[0118] The MXene / PPy@PDMS stress sensor was subjected to 5%, 20%, 40%, 60%, and 80% stage strain step tests ( Fig. 9a) For different strain states, the MXene / PPy@PDMS stress sensor shows good discrimination. In addition, the resistance response under the state of releasing pressure is basically the same as that under the state of applying pressure, which broadens the application of the sensor in the field of strain. The MXene / PPy@PDMS stress sensor was tested under pressure cycles of 5%, 20%, 40%, 60%, and 80%. The results are shown in the figure. Fig. 9 As shown in Fig. 2a, the signal shape and peak value of the MXene / PPy@PDMS stress sensor under different strains remain stable, and the resistance change has good repeatability. In addition, the MXene / PPy@PDMS stress sensor was subjected to long-term continuous compression-release test for more than 7000s. Fig. 9 c), the sensor still has a very stable signal output, which effectively proves the excellent long-term stability of the MXene / PPy@PDMS stress sensor and provides stability guarantee for its feasibility in the field of human motion.

[0119] Embodiment 3:

[0120] An application of a stress sensor for motion recognition, wherein the flexible stress sensor can be applied to monitoring human motion signals.

[0121] Due to the high sensitivity and stability of MXene / PPy@PDMS stress sensors, they can be used as pressure sensors to detect human body movements. Fixing the sponge at different joints of the human body can detect the signal performance of different parts of the body. Record the resistance changes of the pressure sensor with different movements, such as Fig.10 shown.

[0122] Fig.10 a is to fix the MXene / PPy@PDMS stress sensor on the tester's throat to detect the vibration of the throat. When used for swallowing detection, the sensor can sensitively produce corresponding signal changes every time a swallowing action is performed. When used for vocalization detection, there is a characteristic peak when the tester reads the word "swallow". When reading the word "MXene", two characteristic peaks appear, corresponding to "MX-" and "-Xene" respectively, indicating that the MXene / PPy@PDMS stress sensor can distinguish different words. When the same word is repeated, the resistance response changes are almost the same, which means that it has good reproducibility. Therefore, the excellent sound recognition ability of stress enables this sensor to have great potential in vocal rehabilitation exercises and human-computer interaction.

[0123] Fig.10b is to fix the MXene / PPy@PDMS stress sensor on the tester's cheek occlusal muscle to detect the vibration force of the occlusal muscle. When the tester bites once, the sensor will show obvious signal changes. The analysis of bite force can help determine the chewing and eating ability of the elderly, so as to provide better care and nursing for the elderly.

[0124] Fig.10 c shows the situation of fixing the MXene / PPy@PDMS stress sensor on the human radial artery to detect the pulse. The small picture is an enlarged picture of a pulse signal. The high sensitivity of the sensor can detect the typical pulse frequency, which is expected to help users adjust their exercise intensity. The MXene / PPy@PDMS stress sensor can detect the three peaks of P (percussion wave), T (tidal wave), and D (diastolic wave) in the pulse signal. By analyzing the ratio of the peak width and peak height between the PT peak and the TD peak, the cardiovascular condition of the tester can be determined.

[0125] Fig.10 d shows the MXene / PPy@PDMS stress sensor fixed at the back of the neck to detect the head tilt. The sensitive detection of this part lays an important foundation for the testing and identification of subsequent actions. Fig.10 e shows the MXene / PPy@PDMS stress sensor fixed on the elbow to detect elbow bending. The bending and recovery states of the joint can be clearly captured. Fig.10 f shows the situation where the MXene / PPy@PDMS stress sensor is fixed on the index finger joint to detect the bending of the finger. The response of the sensor gradually increases with the bending angle of the finger. It can be seen that the sensor can determine the degree of finger bending (0, 20°, 40°, 60°, 80°, 90°). Fig.10 g shows the situation where the MXene / PPy@PDMS stress sensor is fixed at the knee to detect knee bending.

[0126] Fig.10 h is the MXene / PPy@PDMS stress sensor fixed at the heel to detect the state of walking and running. The walking state of the test person can be judged by the frequency of the signal (the signal frequency is lower when walking and higher when running). This reflects the application prospects of MXene / PPy@PDMS stress sensors in the field of human body detection.

[0127] Fig.11The MXene / PPy@PDMS stress sensor is fixed on the index finger joint to detect the mouse single and double click activities. The sensor can clearly determine the single peak (the tester clicks the mouse) and the double peak (the tester double clicks the mouse), so the MXene / PPy@PDMS stress sensor can be used to detect the tester's clicks when using the mouse single and double click keys.

[0128] In practical applications, whether used in wearable devices or electronic skin, sensors must exhibit good environmental adaptability. Fig.12 a, b The pressure and response performance of the MXene / PPy@PDMS stress sensor under 50% strain after washing in deionized water for 1 hour was compared with that before washing. It can be seen that there is no obvious change in the pressure and resistance response waveforms and peak values. Fig.12 c and d tested the resistance change of the sponge fixed at the knee and finger joints before and after washing with deionized water to detect the knee bending and finger bending. Except that the deionized water increased the resistance inside the sponge, the waveform output by the sponge after washing was basically the same as that before washing. This excellent environmental adaptability is attributed to the stable combination of MXene and PPy and the hydrophobic nature of PDMS itself.

[0129] In order to explore the minimum detection pressure of the MXene / PPy@PDMS stress sensor, a polyurethane sponge block was selected for testing. Its weight was 0.01g and the stress was only 0.43Pa. When it was placed on the MXene / PPy@PDMS stress sensor, the resistance change curve was as follows: Fig.13 When the polyurethane sponge was placed and removed at 2s and 8s, the sensor produced obvious responses, indicating that the minimum detection limit of the MXene / PPy@PDMS stress sensor can reach 0.43Pa.

[0130] Embodiment 4:

[0131] An application of stress sensor for action recognition, human action recognition based on multi-input CNN neural network.

[0132] Deep learning can effectively extract features from sample data and predict results by learning the inherent laws and connections of sample data. It is one of the effective ways to realize artificial intelligence. In recent years, deep learning has been increasingly widely studied in combination with other fields because of its advantages such as high learning ability, wide coverage, strong adaptability, and good anti-interference. We used the same process to prepare seven MXene / PPy@PDMS stress sensors, such as Fig.14 As shown, they are fixed on the skin of the heels (two), knees (two), elbows (two) and the back of the neck of the tester (a healthy adult male) to detect the joint activities or contact with the ground in these parts.

[0133] Fig.15 When the test subjects performed four movements, namely squats, push-ups, high leg raises, and lunges, the prepared MXene / PPy@PDMS high-sensitivity stress sensors can generate specific signals for activities in different parts. For example, when performing squats, except for the knee joints, the rest of the body, such as the elbows, will not produce large-scale movements, so only the two MXene / PPy@PDMS stress sensors at the knees produce more obvious signals. When performing other movements, the devices at the corresponding parts generate corresponding specific signals. The above results show that the prepared high-sensitivity sensors can distinguish the activities of different parts.

[0134] Fig.16 The schematic diagram of the tester performing a squat and the MXene / PPy@PDMS stress sensor signal curve. The squatting movements include: correct squatting, insufficient squatting depth, leaning forward during squatting, and lifting the heels during squatting. When the squatting depth is insufficient, the stress sensor response at the knee will decrease. When the body leans forward during squatting, the stress sensor at the back of the neck will respond to varying degrees. When the heels are lifted during squatting, the stress sensor at the heel will produce a positive response value due to the release of pressure.

[0135] Fig.17 The following is a schematic diagram of the tester performing push-ups and the MXene / PPy@PDMS stress sensor signal curve. The push-ups include: correct push-ups, knee bending during push-ups, insufficient elbow bending during push-ups, and shrugging during push-ups. When the knees are bent during push-ups, in addition to the response signal of the elbow stress sensor, the stress sensor at the knee will also respond. When the elbow is insufficiently bent, the response value of the elbow stress sensor is small. When shrugging during push-ups, the stress sensor at the back of the neck will respond to a certain extent.

[0136] Fig.18 The schematic diagram of the tester performing a high leg lift and the MXene / PPy@PDMS stress sensor signal curve diagram. The actions of high leg lift include: raising the legs correctly, not swinging the arms when raising the legs, raising the legs insufficiently, and leaning forward when raising the legs. When the arms are not swung, the elbow stress sensor that should have responded did not respond. When the leg lift height is insufficient, the stress sensor at the knee will also have a corresponding decrease in response value. When the elbow is not bent enough, the response value of the elbow stress sensor is small. When doing push-ups and shrugging, the stress sensor at the back of the neck will respond to a certain extent.

[0137] Fig.19The following is a schematic diagram of the tester performing a lunge and a signal curve of the MXene / PPy@PDMS stress sensor. The lunge movements include: correct lunge, knee touching the ground during lunge, body leaning back during lunge, and unstable body center of gravity during lunge. When the knee touches the ground during lunge, the stress sensor at the knee will produce a messy signal response based on the original response due to irregular contact with the ground. When the body leans back, the stress sensor at the back of the neck will produce a certain degree of response due to the contraction of the neck. When the center of gravity is unstable, the stress sensor at the heel will produce some stray peaks due to irregular contact with the ground.

[0138] Fig. 20 The basic structure of the multi-input deep CNN used in this work is described. The sampled data is normalized and then applied to the multi-input deep convolutional neural network (CNN) algorithm to model different motion patterns. According to the ratio of training set to test set 4:1, 640 resistance response data are randomly selected from the database as the training set to train the multi-input CNN model. The remaining 160 data are used as the test set to verify the final recognition ability of the model. The structure of the multi-input CNN consists of input layer, convolution layer, pooling layer, initial layer, fully connected layer and output layer. Among them, the convolution layer is used to extract features from the input data, and the pooling layer can compress the input feature map and simplify the computational complexity. The initial layer consists of two convolution layers and one pooling layer. Adding a convolution layer not only increases the network width, but also increases the adaptability of the network to scale. The fully connected layer is used to integrate the extracted features and perform computational recognition. Deep learning models need to continuously learn and adjust to obtain optimal performance. First, the multi-input CNN model is trained by using a large amount of measurement data from different actions. Secondly, the model is iteratively trained using the model error on the data set to obtain a model that reasonably fits the data set.

[0139] Fig.21 The basic architecture of the proposed system is described, including data acquisition, feature extraction, normalization, training, and testing procedures. In the training phase, training data is obtained from the MXene / PPy@PDMS stress sensor and normalized, and a multi-input CNN is trained for all activities. On the other hand, the testing phase includes test data acquisition, normalization, and application to the trained multi-input CNN to identify activities or behaviors from the test data. Finally, the trained multi-input CNN model is applied to the actual scenario.

[0140] Fig. 22 The confusion matrix of the multi-input CNN neural network model training action, with the increase of the training data set training iteration number, the recognition accuracy of the training set gradually increased to 95%. This shows that the multi-input CNN network has been well trained and achieved a great performance improvement.

[0141] Fig.23 The loss function and accuracy curve of the multi-input CNN neural network model training action as the number of epochs increases. The loss function describes the difference between the model prediction value and the true value. In general, the smaller the loss function value, the stronger the model robustness and the better the model fits the data. It can be observed that as the number of epochs increases, the loss function value decreases, indicating that the model converges. This confirms that the best training effect has been achieved.

[0142] For the porous MXene / PPy@PDMS stress sensor, the sacrificial sugar template method ensures the reliability of mass preparation of the device. PPy, due to its ultra-high conductivity, can make the strain sensor have ultra-high sensitivity under low strain, while MXene can extend the sensing range of the strain sensor and improve the overall performance of the sensor.

[0143] A unique sensor structure and a new type of nanosensing material were developed. PPy was in situ polymerized on MXene nanosheets by electrostatic adsorption, which improved the stress sensitivity of the sensor. The PDMS sponge structure can improve the stability of the stress sensor. The conductive network of the composite material is introduced into the flexible sensor, and the synergistic effect between the conductive materials will make up for the shortcomings of a single material while maintaining the advantages of each material.

[0144] The wearable flexible stress sensor is integrated with the artificial intelligence neural network, and the characteristics of the multi-site stress sensors required for judging fitness postures are combined to design a multi-input CNN convolutional neural network. The judgment and recognition of 16 fitness postures are achieved, with a recognition accuracy of about 95%.

[0145] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for preparing a stress sensing material for action recognition, wherein the stress sensing material is an in-situ polymerization of pyrrole in a MXene solution and uniform deposition of pyrrole nanoparticles on the surface of MXene nanosheets to form a coating that protects Ti atoms to prevent MXene from being oxidized, a compound of two-dimensional MXene nanosheets and zero-dimensional PPy nanoparticles increases the conductive path in the sensing layer, MXene provides more conductive active sites for PPy, and the ultra-high conductivity of PPy increases the sensitivity of the composite sensing material, wherein the stress sensing material is characterized in that: The preparation steps include: (1) adding lithium fluoride into hydrochloric acid to completely dissolve it, thereby forming a hydrochloric acid / lithium fluoride colorless solution; (2) Slowly adding aluminum titanium carbide powder, the hydrochloric acid / lithium fluoride solution reacts with it to etch it, and the solution turns from colorless to black; (3) Repeatedly centrifuge and pour off the supernatant to etch the Al atoms in the aluminum titanium carbide and the F atoms in the solution, control the pH of the solution to ensure that the byproducts of the etching reaction are completely discharged, and obtain a black multilayer MXene solution; (4) Ultrasonic stripping of the black solution to obtain a few-layer MXene solution; (5) Take the DBSA solution and pyrrole monomer and add them to deionized water, stir vigorously in an ice water bath to enhance the conductivity of pyrrole, and then add the few-layer MXene solution to mix the two thoroughly; (6) FeCl3-6H2O was added and stirred in an ice-water bath. The pyrrole monomer began to self-polymerize on the surface of the MXene nanosheets and formed a continuous conductive layer. The DBSA and FeCl3-6H2O were removed by centrifugation several times to obtain a MXene / PPy composite solution.

2. The method for preparing a stress sensing material for motion recognition according to claim 1, characterized in that: The mass ratio of lithium fluoride to aluminum carbon titanium is 1:1; The reaction conditions of (2) are sufficient stirring at 35° C. for 24 hours; Centrifuging in (3) until the pH value of the supernatant is greater than 6; The volume ratio of DBSA solution to pyrrole monomer in (5) is 4:1; the volume ratio of pyrrole monomer to MXene solution is 1:80, the concentration of MXene solution is 10 mg / mL, and the reaction time is 1 h; The mass ratio of FeCl3-6H2O to MXene in (6) is 5:1, and the reaction time is 6h.

3. A stress sensor for motion recognition, the sensor comprising a stress sensing material, the stress sensing material being prepared by the method according to any one of claims 1 to 2, characterized in that: A flexible pressure sensor with a porous structure was produced by depositing MXene / PPy on the surface of the PDMS sponge skeleton through low-temperature drying.

4. The stress sensor for motion recognition according to claim 3, characterized in that: The sensing principle of the sensor is: The stress sensor belongs to the adsorption type stress sensor. The sensing material is adsorbed on the surface of the PDMS sponge. The change in resistance mainly comes from the conductive path inside the sensor. When no pressure is applied, the conductive path only comes from the mutual contact between the sensitive materials on the same sponge skeleton. When pressure is applied, the sensitive materials on different sponge skeletons are connected to each other to form a new conductive path, which reduces the resistance of the sensor. Under slight stress, there are still a large number of micropores inside the sponge space. As the pressure increases, the sponge skeleton will collapse rapidly. During the collapse process, sensitive materials are constantly in contact with each other. At this stage, the wrinkled MXene nanosheets are mainly combined with each other, which rapidly increases the conductive pathway inside the sensor and causes the resistance to drop rapidly. As the pressure increases further, most of the air inside the sponge is squeezed out by the pressure, leaving only some pores that have not collapsed. The combination of MXene nanosheets is basically complete, and the increase in the conductive pathway mainly comes from the mutual contact between the PPy nanoparticles, which begins to tend to saturation. When the pressure continues to increase, the increase in stress mainly comes from the resistance of the sponge skeleton. Therefore, the pressure range of this part is larger. At this time, only some PPy nanoparticles are in contact with each other, resulting in a very slow decrease in sensor resistance and a significant decrease in sensitivity.

5. A method for preparing a stress sensor for motion recognition according to any one of claims 3 to 4, characterized in that: The preparation steps include: (1) placing sugar granules into a hexahedral container with an open top, and compacting the sugar granules into sugar cubes of regular shape; (2) A solution of PDMS and a curing agent mixed in a ratio of 10:1 is injected into a container and vacuum dried, so that the micropores inside the sugar cube are filled with the mixture; (3) After PDMS solidifies, the sugar cube / PDMS fusion is immersed in water to remove the white sugar to obtain a PDMS porous sponge of the same size as the original sugar cube; (4) Immerse the PDMS porous sponge in the MXene / PPy solution for 1 hour, dry it at a temperature not greater than 40°C for 6 hours, and repeat this process multiple times to obtain a MXene / PPy@PDMS sponge sensor.

6. Application of a stress sensor for action recognition, the sensor having the structure according to any one of claims 3 to 4, or prepared according to the method according to claim 5, characterized in that: The stress sensor is applied to human motion signal detection or action recognition based on neural network. The human motion signal detection includes bite muscle detection, throat movement detection, head tilt movement detection, finger bending detection, elbow bending detection, knee bending detection, walking state detection and radial artery pulse detection.

7. The application of the stress sensor for motion recognition according to claim 6, characterized in that: The neural network includes an input layer, a convolutional layer, a pooling layer, an initial layer and a fully connected layer; The convolution layer is used to extract features from the input data, the pooling layer is used to compress the input feature map and simplify the calculation complexity. The initial layer consists of two convolution layers and one pooling layer, and the fully connected layer is used to integrate the extracted features and perform calculation and recognition.

8. The application of the stress sensor for action recognition according to any one of claims 6 or 7, characterized in that: First, a multi-input CNN model is trained by using a large amount of measurement data from different actions. Secondly, the model error on the dataset is used to iteratively train the model to obtain a model that reasonably fits the dataset. Finally, the trained multi-input CNN model is applied to actual scenarios.

9. The application of the stress sensor for action recognition according to claim 7, characterized in that: Seven stress sensors were prepared and fixed on the skin of the heel, knee, elbow and back of the neck of the tester to detect the joint activities or contact with the ground in these parts. The basic architecture of the system includes data acquisition, feature extraction, normalization, training and testing procedures. In the training phase, the training data are obtained from the MXene / PPy@PDMS stress sensor and normalized, and a multi-input CNN is trained for all activities. On the other hand, the testing phase includes test data acquisition, normalization and application to the trained multi-input CNN to identify activities or behaviors from the test data.

Images