Ti3C2Tx-based aerogel sensitive material piezoresistive pressure sensor, preparation method and application of Ti3C2Tx-based aerogel sensitive material piezoresistive pressure sensor in dialect recognition

Abstract

The invention discloses a Ti3C2Tx-based aerogel sensitive material piezoresistive pressure sensor, a preparation method and application of the Ti3C2Tx-based aerogel sensitive material piezoresistive pressure sensor in dialect recognition, and belongs to the technical field of pressure sensors. According to the sensor prepared by the invention, a medical polyurethane film is used as a packaging layer, and Ti3C2Tx / chitosan / PVDF prepared by a freeze drying method is used as a sensitive material layer; the prepared pressure sensor can rapidly obtain rapid and tiny vibration signals of the throat during sound production, dialect recognition is achieved by combining a convolutional neural network (CNN) model after the vibration signals are processed into time-varying images of single current signals, and the CNN model is composed of one input layer, four middle layers and three linear layers. The composite aerogel sensitive material prepared by the invention has the advantages of light weight and recoverable deformation capability under high pressure, and the accuracy rates of the sensor prepared by the invention on six Chinese dialects and seven different languages respectively reach 96.2% and 96.6%.

Description

Technical Field

[0001] The present invention belongs to the technical field of pressure sensors and particularly relates to a Ti3C2T x Piezoresistive pressure sensor based on aerogel sensitive material, preparation method and application in dialect recognition. Background Art

[0002] For people who do not speak Mandarin, it is a challenge to communicate with locals in Mandarin when working or seeking medical treatment outside. For tonal languages, the difference between dialect pronunciations lies in the tone and pitch, which are produced by the laryngeal muscles that control the movement of the hyoid bone and cartilage. The pitch is closely related to the contraction and relaxation of the laryngeal muscles. Therefore, it is effective to obtain interference-free vocalization signals by obtaining the vibration signals of the laryngeal muscles during vocalization.

[0003] Wearable flexible piezoresistive pressure sensors have the advantages of good stability, fast response recovery speed, and strong anti-interference. The specific sensing performance of the sensor is obviously affected by the sensitive material, so it is an efficient way to prepare high-performance pressure sensors by preparing new sensitive materials. Summary of the invention

[0004] The present invention aims to provide a Ti3C2T x The pressure sensor prepared by the present invention uses a medical polyurethane film as the encapsulation layer, and a Ti3C2T4 prepared by a freeze-drying method. x / chitosan / PVDF is used as the sensitive material layer; the prepared pressure sensor can quickly obtain the fast and tiny vibration signals of the throat during vocalization, and combined with the convolutional neural network (CNN) model to realize dialect recognition.

[0005] The Ti3C2T x Based on aerogel sensitive material piezoresistive pressure sensor, such as Figure 1 As shown, the polyurethane upper encapsulation layer, Ti3C2T x The base aerogel sensitive material and the lower polyurethane encapsulation layer are composed of Ti3C2T x The sensitive material of the aerogel is Ti3C2T x / chitosan / PVDF composite aerogel sensitive material, and two copper wires were pasted on the Ti3C2T x / chitosan / PVDF composite aerogel sensitive material, the upper and lower surfaces are used as electrodes; among them, Ti3C2T x The / chitosan / PVDF composite aerogel sensitive material was prepared by freeze drying, and the steps are as follows:

[0006] (1) dissolving PVDF in a mixed solution of DMF and acetone to prepare an electrospinning solution, wherein the concentration of PVDF is 5 to 10 wt %; then preparing a PVDF fiber membrane on a collector by electrospinning, wherein the feed rate of the electrospinning is 0.6 to 1 mL / h, the electrospinning voltage is 15 to 18 kV, and the distance from the nozzle of the electrospinning device to the collector is 16 to 20 cm;

[0007] (2) crushing 0.4-0.6 g of the PVDF fiber membrane obtained in step (1) in 400-600 mL of deionized water using a high-speed mixer for 8-15 minutes to obtain a suspension;

[0008] (3) adding 2 to 3 g of chitosan powder and 3 to 8 mL of acetic acid to the suspension obtained in step (2), and stirring with a magnetic stirrer until a clear solution is obtained;

[0009] (4) 1-3 g of Ti3AlC2 precursor powder is slowly added to an etching solution prepared by mixing 1-3 g of lithium fluoride with 70-90 mL of 35% hydrochloric acid, and the mixture is stirred in a water bath at 40-50° C. for 20-30 hours. The product obtained by the reaction is repeatedly rinsed with deionized water and centrifuged until the pH of the supernatant is 6-7; the rinsed product is then dispersed in 120-140 mL of deionized water and ultrasonically crushed for 50-70 minutes; the upper dispersion is taken after centrifugation to obtain Ti3C2T x The dispersion was dried in a vacuum oven for 20 to 30 hours to obtain Ti3C2T x powder;

[0010] (5) 0.4-0.6 g of the Ti3C2T obtained in step (4) was added x The powder is added to 100 mL of the clear solution obtained in step (3) and ultrasonically dispersed for 20 to 40 minutes. Then, 6 to 10 mL of the obtained mixed solution is poured into a mold and frozen at the bottom of the mold with liquid nitrogen. The frozen mold is freeze-dried at -50 to -40°C for 15 to 20 hours to obtain Ti3C2T x / Chitosan / PVDF composite aerogel sensitive material, with a thickness of 0.4 to 0.6 cm.

[0011] The Ti3C2T x A method for preparing a piezoresistive pressure sensor based on aerogel sensitive material, the steps of which are as follows:

[0012] (1) Preparation of sensitive materials: Ti3C2T x / chitosan / PVDF composite aerogel sensitive material is cut into 1×1 cm size, and copper wires are pasted on the upper and lower surfaces of the sensitive material with conductive silver paste;

[0013] (2) Encapsulation of Ti3C2Tx / Chitosan / PVDF composite aerogel sensitive layer: Cut the medical polyurethane film into 1.5×1.5 cm sizes and paste them on the Ti3C2T x The upper and lower surfaces of the Ti3C2T / chitosan / PVDF composite aerogel sensitive material are encapsulated to obtain the Ti3C2T x Piezoresistive pressure sensor based on aerogel sensitive material.

[0014] Ti3C2T prepared by the present invention x The piezoresistive pressure sensor based on aerogel sensitive material can be used in dialect recognition. The specific operation is as follows:

[0015] (1) A DC power supply is connected in parallel between the two copper wires of the pressure sensor to provide a 0.1V DC voltage to the sensor; a digital multimeter is connected in series between the two copper wires of the pressure sensor to measure the real-time current change signal of the pressure sensor, which is converted into a response current signal image and recorded by a computer;

[0016] (2) The pressure sensor is fixed on the throat of the dialect user. The dialect user repeatedly speaks M words in N dialects. The multimeter will obtain stable N×M groups of response current signals with small differences. Then, the multiple repetitive response current signals of the same word expressed in the same dialect are processed into a single current signal changing over time. The image size is set to 64×64 pixels and the batch size is 512.

[0017] (3) The obtained images of the change of single current signal over time are randomly divided into training data set and test data set after filtering, and combined with convolutional neural network (CNN) model to realize dialect recognition. The convolutional neural network (CNN) model consists of 1 input layer, 4 intermediate layers and 3 linear layers;

[0018] S1: The image of the change of a single current signal over time as a training data set is transmitted to the input layer of the CNN model;

[0019] S2: Each intermediate layer contains at least one convolution layer and one maximum pooling layer. The function of the convolution layer is to extract features from the image. The four maximum pooling layers have 256 filters, 128 filters, 128 filters and 64 filters respectively. The function of the maximum pooling layer is to simplify the complexity, compress the eigenvalues ​​and reduce the amount of calculation. The activation function used after the maximum pooling layer is Relu; a batch normalization layer is set between the convolution layer and the maximum pooling layer to reduce the covariance drift in the model and improve the stability of the model;

[0020] S3: The Dropout layer in the linear layer reduces overfitting by ignoring half of the hidden nodes; the test data set is then used for cross-validation, so that the computer finally outputs the dialect type and the output results of the vocabulary expressed to achieve dialect classification.

[0021] Advantages of the present invention:

[0022] (1) Preparation of Ti3C2T x / Chitosan / PVDF composite aerogel sensitive material has the advantages of light weight and the ability to recover deformation under high pressure;

[0023] (2) The prepared pressure sensor device has the advantages of low hysteresis, wide detection range, low monitoring limit and high stability, and can quickly obtain the weak and rapid vibration signal of the larynx during vocalization;

[0024] (3) The Ti3C2T based x The piezoresistive pressure sensor based on aerogel sensitive material has sensitivity and response value to pressure. Combined with the convolutional neural network (CNN) model, the response current signal images were obtained one by one when volunteers expressed seven common words "stop", "sleep", "turn left", "turn right", "eat", "drink water" and "go home" in six Chinese dialect categories of Nantong, Liaocheng, Jiutai, Shuozhou, Xi'an and Hefei respectively; finally, the response current signal image was used as the input feature value of the computer model, and a model training set was established. By cross-validation, the computer finally output the output results of the dialect type and the expressed vocabulary. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 :The method based on Ti3C2T for dialect recognition described in the present invention x Piezoresistive pressure sensor based on aerogel sensitive materials and Ti3C2T x / Schematic diagram of the structure of chitosan / PVDF sensitive material;

[0026] like Figure 1 As shown in the figure, the pressure sensor consists of a polyurethane upper encapsulation layer, a Ti3C2T x / chitosan / PVDF sensitive material, polyurethane lower encapsulation layer;

[0027] Figure 2 : The Ti3C2T prepared by freeze drying method in the present invention x / SEM image of chitosan / PVDF composite aerogel sensitive material;

[0028] like Figure 2 As shown, it can be seen that the sensitive material is an interconnected layered structure;

[0029] Figure 3: The Ti3C2T prepared by freeze drying method in the present invention x / Infrared spectrum of chitosan / PVDF composite aerogel sensitive material;

[0030] like Figure 3 As shown, with Ti3C2T x (Ti3C2T x Compared with MXene) and chitosan / PVDF (CS / PVDF) aerogels, Ti3C2T x / chitosan / PVDF composite aerogel (MX / CS / PVDF-1) at 2240cm -1 The peak near the α-threshold of chitosan broadens and blue-shifts, which confirms that chitosan and Ti3C2T x The strong hydrogen bonds between them; the vibration peaks of Ti-O, -NH, -CF2 and β phase can be observed in the infrared spectrum, indicating that Ti3C2T x / Chitosan / PVDF composite aerogel sensitive materials were successfully prepared;

[0031] Figure 4 : Using Ti3C2T x When the piezoresistive pressure sensor made of / chitosan / PVDF composite aerogel sensitive material is fixed on the tester's throat, the image of the specific current between the two copper wires of the sensor changes over time when six dialects and seven daily expressions are spoken;

[0032] like Figure 4 As shown in the figure, multiple images of pronunciation tests are taken for each daily language, which are then processed into multiple images of single current signal changes over time, and then randomly divided into training data sets and test data sets, and the convolutional neural network (CNN) model is combined to realize dialect recognition; Figure 4 Only six utterances of each word are listed to show the repeatability and stability of the image. The actual number of images input to the input layer can be more.

[0033] Figure 5 : Schematic diagram of the architecture of the convolutional neural network (CNN) model;

[0034] The model consists of 1 input layer, 4 intermediate layers and 3 linear layers. Each intermediate layer contains at least one convolution layer and one maximum pooling layer. There is also a batch normalization layer between the convolution layer and the maximum pooling layer to reduce the covariance drift in the model and improve the stability of the model. After the 4 intermediate layers, there are three linear layers for final prediction.

[0035] Figure 6:Use the signal image of the current changing with time when the sound is produced measured by the sensor as the training data set and test data set of the convolutional neural network (CNN) model, and the output classification confusion matrix result diagram of the convolutional neural network (CNN) model;

[0036] like Figure 6 As shown, the results show that the pressure sensor of the present invention combined with the convolutional neural network (CNN) model has an accuracy rate of 96.2% and 96.6% for six dialects and seven different dialect words respectively. DETAILED DESCRIPTION

[0037] Example 1

[0038] Ti3C2T was successfully prepared by freeze drying x / chitosan / PVDF composite aerogel is used as a sensitive material to prepare a piezoresistive pressure sensor and test its specific sensing performance. The process is as follows:

[0039] 1. Preparation of PVDF nanofibers: Preparation of PVDF fiber membrane by solution electrospinning. PVDF (Solef6010, Solvay S.A) was dissolved in a mixed solution of DMF and acetone (7 / 3, weight ratio) to prepare an electrospinning solution, with a PVDF concentration of 8wt%; placed in a 10mL syringe equipped with a 19g needle, and then prepared on a collector by electrospinning to obtain a PVDF fiber membrane, with an electrospinning feed rate of 0.8mL / h, an electrospinning voltage of 16kV, and a distance of 18cm from the nozzle of the electrospinning device to the collector; during the electrospinning process, the ambient humidity and temperature were fixed at 58±2% and 20°C;

[0040] 2. Preparation of single-layer Ti3C2T x Powder: Ti3C2T x MXene nanosheets were synthesized by etching Ti3AlC2 (Jilin Yiyi Technology Co., Ltd., China) less vigorously and removing the aluminum layer in the MAX phase using a LiF / HCl mixed solution. First, 2 g of LiF (Aladdin, >99.99%) was slowly dissolved in 80 mL of 35% HCl (Xilong Technology Co., Ltd.) in a Teflon liner and stirred vigorously at room temperature for 15 minutes. Then 2 g of Ti3AlC2 was gradually added to the above mixed solution and reacted in a water bath at 45 °C for 24 hours. After etching, the resulting mixture was repeatedly centrifuged at 8000 rpm for 8 minutes until the pH value of the supernatant reached 6.5; the rinsed product was dispersed in 120 mL of deionized water and ultrasonically crushed for 60 minutes; the upper dispersion was obtained after centrifugation to obtain Ti3C2T x dispersion and dried in a vacuum oven for 24 h to obtain a single layer Ti3C2T xPowder, i.e. Ti3C2T x MXene nanosheets;

[0041] 3. Ti3C2T x Preparation of chitosan / PVDF composite aerogel: 0.5 g of PVDF fiber membrane was placed in 500 mL of deionized water and crushed with a high-speed mixer for 10 minutes to form a uniform suspension. Then 2.5 g of chitosan (Aladdin, deacetylation degree ≥ 75%) and 5 mL of acetic acid (Xilong Technology Co., Ltd., > 99.5%) were dissolved in the above mixture and magnetically stirred for 20 minutes to obtain a chitosan / PVDF clear solution; 100 mL of the clear solution was added with 0.5 g of Ti3C2T x MXene nanosheets were ultrasonically dispersed for 30 min to obtain Ti3C2T x / chitosan / PVDF composite aerogel sensitive material mixture. Take 8mL of the obtained mixture and pour it into a 4×4cm mold, freeze it with liquid nitrogen at the bottom of the mold; freeze-dry the frozen mold at -45℃ for 16 hours to obtain a Ti3C2T3 with a thickness of 0.5cm. x / Chitosan / PVDF composite aerogel sensitive materials;

[0042] 4. Device packaging: Cut the sensitive material obtained in step 3 into 1×1 cm sizes, fix the copper electrodes on the upper and lower surfaces with conductive silver paste; cut the medical polyurethane film into 1.5×1.5 cm sizes and paste them on the Ti3C2T x / chitosan / PVDF composite aerogel sensitive material is encapsulated on the upper and lower surfaces to produce a Ti3C2T x Piezoresistive pressure sensor based on / chitosan / PVDF sensitive materials.

[0043] Example 2

[0044] Sensor signal acquisition: A DC power supply (DPS-305BM) is connected in parallel between the two copper wires of the pressure sensor to provide a 0.1V DC voltage to the sensor; a digital multimeter (Keithley DMM7510) is connected in series between the two copper wires of the pressure sensor to measure the real-time current change signal of the pressure sensor, which is converted into a response current signal image and recorded by a computer. The pressure sensor is fixed on the throat of six dialect users. The dialect users repeat the seven daily command words "stop", "sleep", "turn left", "turn right", "eat", "drink water" and "go home" in six dialects of Nantong, Liaocheng, Jiutai, Shuozhou, Xi'an and Hefei for 6 times. The multimeter will obtain 42 groups of stable and small differences in response current signals; then the multiple repetitive response current signals of the same word expressed in the same dialect are processed into 6 single current signal change images over time. The image size is set to 64×64 pixels and the batch size is 512 for further training and recognition. Finally, the obtained image of the change of single current signal over time is filtered and randomly divided into training data set and test data set as the feature data set of CNN model; the CNN model consists of 1 input layer, 4 middle layers and 3 linear layers, and the image of the change of single current signal over time as the training data set is transmitted to the input layer of CNN model. Each intermediate layer contains at least one convolution layer and one maximum pooling layer. The function of the convolution layer is to extract features from the image (CHEN Y, JIANG H, LIC, et al. Deep Feature Extraction and Classification of Hyperspectral Images Based on Convolutional Neural Networks[J]. ITGRS, 2016, 54(10): 6232-51). The four maximum pooling layers have 256 filters, 128 filters, 128 filters and 64 filters respectively. The function of the maximum pooling layer is to simplify complexity, compress feature values ​​and reduce the amount of calculation (WANG Z, ZHAO H, ZHENG M, et al. A novel time series prediction method based on pooling compressed sensing echo state network and its application in stock market[J]. Neural Netw, 2023, 164: 216-27).In addition, a batch normalization layer is set between the convolution layer and the maximum pooling layer to reduce the covariance drift in the model and improve the stability of the model (CéSPEDES SISNIEGAJ, RODRíGUEZ V, MOLTóG, et al. Efficient and scalable covariate drift detection in machine learning systems with serverless computing [J]. Future Generation Computer Systems, 2024, 161: 174-88). After the four intermediate layers, there are three linear layers. The Dropout layer in the linear layer reduces overfitting by ignoring half of the hidden nodes. The activation function used after the maximum pooling layer is Relu. The test data set is then used for cross-validation, so that the final computer outputs the output results of the dialect type and the vocabulary expressed to achieve dialect classification. The results show that the pressure sensor combined with the convolutional neural network (CNN) model of the present invention has an accuracy of 96.2% and 96.6% for six Chinese dialects and seven different dialect words, respectively.

Claims

1. A Ti3C2T x The piezoresistive pressure sensor based on aerogel sensitive material is characterized by: The polyurethane upper encapsulation layer, Ti3C2T x / chitosan / PVDF composite aerogel sensitive material, a lower polyurethane packaging layer, and two copper wires were pasted on the Ti3C2T x / chitosan / PVDF composite aerogel sensitive material, both sides of the upper and lower surfaces are used as electrodes; Among them, Ti3C2T x The / chitosan / PVDF composite aerogel sensitive material was prepared by freeze drying, and the steps are as follows: (1) dissolving PVDF in a mixed solution of DMF and acetone to prepare an electrospinning solution, and then preparing a PVDF fiber membrane on a collector by electrospinning; (2) crushing 0.4-0.6 g of the PVDF fiber membrane obtained in step (1) in 400-600 mL of deionized water using a high-speed mixer for 8-15 minutes to obtain a suspension; (3) adding 2 to 3 g of chitosan powder and 3 to 8 mL of acetic acid to the suspension obtained in step (2), and stirring with a magnetic stirrer until a clear solution is obtained; (4) 1-3 g of Ti3AlC2 precursor powder is slowly added to an etching solution prepared by mixing 1-3 g of lithium fluoride with 70-90 mL of 35% hydrochloric acid, and the mixture is stirred in a water bath at 40-50° C. for 20-30 hours. The product obtained by the reaction is repeatedly rinsed with deionized water and centrifuged until the pH of the supernatant is 6-7; the rinsed product is then dispersed in 120-140 mL of deionized water and ultrasonically crushed for 50-70 minutes; the upper dispersion is taken after centrifugation to obtain Ti3C2T x The dispersion was dried in a vacuum oven for 20 to 30 hours to obtain Ti3C2T x powder; (5) 0.4-0.6 g of the Ti3C2T obtained in step (4) was added x The powder is added to 100 mL of the clear solution obtained in step (3) and ultrasonically dispersed for 20 to 40 minutes. Then, 6 to 10 mL of the obtained mixed solution is poured into a mold and frozen at the bottom of the mold with liquid nitrogen. The frozen mold is freeze-dried at -50 to -40°C for 15 to 20 hours to obtain Ti3C2T x / Chitosan / PVDF composite aerogel sensitive material, with a thickness of 0.4 to 0.6 cm.

2. A Ti3C2T as claimed in claim 1 x The piezoresistive pressure sensor based on aerogel sensitive material is characterized by: In step (1), the concentration of PVDF is 5-10 wt %, the feed rate of electrospinning is 0.6-1 mL / h, the electrospinning voltage is 15-18 kV, and the distance from the nozzle of the electrospinning device to the collector is 16-20 cm.

3. A Ti3C2T3 according to claim 1 or 2 x A method for preparing a piezoresistive pressure sensor based on aerogel sensitive material, the steps of which are as follows: (1) Preparation of sensitive materials: Ti3C2T x / chitosan / PVDF composite aerogel sensitive material is cut into 1×1 cm size, and copper wires are pasted on the upper and lower surfaces of the sensitive material with conductive silver paste; (2) Encapsulation of Ti3C2T x / Chitosan / PVDF composite aerogel sensitive layer: Cut the medical polyurethane film into 1.5×1.5 cm sizes and paste them on the Ti3C2T x The upper and lower surfaces of the / chitosan / PVDF composite aerogel sensitive material are encapsulated to obtain the Ti3C2T based on x Piezoresistive pressure sensor based on aerogel sensitive material.

4. A Ti3C2T3 according to claim 1 or 2 x The application of aerogel-based sensitive material piezoresistive pressure sensor in dialect recognition is characterized by: (1) A DC power supply is connected in parallel between the two copper wires of the pressure sensor to provide a 0.1V DC voltage to the sensor; a digital multimeter is connected in series between the two copper wires of the pressure sensor to measure the real-time current change signal of the pressure sensor, which is converted into a response current signal image and recorded by a computer; (2) The pressure sensor is fixed on the throat of the dialect user. The dialect user repeatedly speaks M words in N dialects. The multimeter will obtain stable N×M groups of response current signals with small differences. Then, the multiple repetitive response current signals of the same word expressed in the same dialect are processed into a single current signal changing over time. The image size is set to 64×64 pixels and the batch size is 512. (3) The obtained images of the change of single current signal over time are randomly divided into training data set and test data set after filtering, and combined with convolutional neural network (CNN) model to realize dialect recognition. The convolutional neural network (CNN) model consists of 1 input layer, 4 intermediate layers and 3 linear layers; S1: The image of the change of a single current signal over time as a training data set is transmitted to the input layer of the convolutional neural network (CNN) model; S2: Each intermediate layer contains at least one convolution layer and one maximum pooling layer. The function of the convolution layer is to extract features from the image. The four maximum pooling layers have 256 filters, 128 filters, 128 filters and 64 filters respectively. The function of the maximum pooling layer is to simplify the complexity, compress the eigenvalues ​​and reduce the amount of calculation. The activation function used after the maximum pooling layer is Relu; a batch normalization layer is set between the convolution layer and the maximum pooling layer to reduce the covariance drift in the model and improve the stability of the model; S3: The Dropout layer in the linear layer reduces overfitting by ignoring half of the hidden nodes; the test data set is then used for cross-validation, so that the computer finally outputs the dialect type and the output results of the vocabulary expressed to achieve dialect classification.

5. A Ti3C2T as claimed in claim 4 x The application of aerogel-based sensitive material piezoresistive pressure sensor in dialect recognition is characterized by: In step (2), the pressure sensor is fixed on the throat of six dialect users. The dialect users repeat seven daily command words "stop", "sleep", "turn left", "turn right", "eat", "drink water" and "go home" in six dialects of Nantong, Liaocheng, Jiutai, Shuozhou, Xi'an and Hefei for 6 times. The multimeter will obtain 42 groups of stable and small-difference response current signals; the pressure sensor combined with the convolutional neural network (CNN) model has an accuracy rate of 96.2% and 96.6% for six Chinese dialects and seven different dialect words respectively.

Images