Method for preparing a flexible electronic nose based on laser-induced graphene

Abstract

This invention discloses a method for fabricating a flexible electronic nose based on laser-induced graphene. A pre-defined pattern is fabricated on a selected polyimide substrate using a laser to obtain a laser-induced graphene electrode. Three different metal oxide nanoparticles—tin dioxide, tungsten trioxide, and zinc oxide-indium oxide—are selected as gas-sensitive materials. These materials are deposited on LIG using a simple process to form gas-sensitive units, which are used to construct an orthogonal sensor array. By collecting signals from the gas-sensitive units, specific signal features of the sensing units are selected to build a dataset, which is then used for pattern recognition. The final result shows a high recognition rate for single or mixed gases. The electronic nose of this invention is simple to operate, cost-effective, highly flexible, and environmentally friendly in its gas-sensitive sensing unit design. It can also be easily attached to clothing or other substrates, making it portable and wearable.

Description

Technical Field

[0001] This invention relates to the technical fields of gas detection, intelligent sensing, and pattern recognition. Specifically, it relates to an electronic nose system based on multi-sensor unit fusion and pattern recognition, aiming to effectively solve the technical problems commonly found in traditional single gas sensors when facing complex mixed odor scenarios, such as weak environmental adaptability, insufficient odor recognition generalization ability, and poor classification detection accuracy. This provides a new technical solution for the efficient and accurate detection of complex odors. Background Technology

[0002] Gas detection plays a crucial role in food safety, environmental monitoring, and healthcare. Traditional gas detection methods based on single gas sensors are susceptible to false alarms due to the influence of background gases, as existing gas sensor materials exhibit cross-sensitivity, meaning they cannot be sensitive to only one specific gas. To improve the quantitative identification of specific gas components in complex environments, electronic nose systems based on sensor arrays and machine learning algorithms are widely used. However, traditional electronic noses are limited by sensor performance, signal processing algorithms, and anti-interference capabilities, often resulting in insufficient recognition accuracy in practical applications. Furthermore, the inherent limitations of the devices make them unsuitable for prolonged wear. Flexible electronic nose technology, through flexible substrates and flexible sensors, solves the problem of rigid devices being unable to adapt to wearable scenarios. This provides a foundation for the miniaturization, lightweighting, and body-hugging transformation of electronic noses, driving the development of electronic nose systems from fixed-scene detection to mobile real-time monitoring.

[0003] CN120490228A discloses the design of an electronic nose system for odor classification based on pattern recognition algorithms.

[0004] This electronic nose identifies five scents—mint candy, coffee, chrysanthemum tea, orange peel, and apple—through a hardware platform. By training a machine learning algorithm, it achieves a recognition accuracy of 81.7% under the optimal algorithm model. The electronic nose focuses on the hardware system, using a Raspberry Pi 4B as the MCU. The sensing unit is a BME680. However, due to the rigid structure of the sensing unit, it lacks good mechanical flexibility, cannot conform to flexible carriers, and cannot achieve flexible wearability. Furthermore, the gas recognition accuracy is low even with the optimal algorithm. CN119086658B discloses an oxide semiconductor gas sensor array, its fabrication method, an electronic nose, and its applications. The preparation method involves: preparing a mixture containing 2-methylimidazole, a metal compound, and deionized water; adding a noble metal source and controlling the molar ratio of the metal to the noble metal atoms; immersing a ceramic substrate with a portion of its surface covered by an insulating medium into the mixture at a preset temperature, allowing the noble metal-doped MOF nanosheets to grow only on the surface of the interdigitated test electrode, thus obtaining a precursor; finally, calcining the precursor in air to transform it into noble metal-modified metal oxide nanosheets. This sensor uses a ceramic substrate. The processing and integration methods of the ceramic substrate determine that it is difficult to achieve a thin and miniaturized sensor array, resulting in a large overall volume of the accompanying electronic nose and insufficient portability; furthermore, the immersion growth time of the sensing material is 3-5 hours, and the calcination time is 2-4 hours, leading to a long overall manufacturing cycle for the sensing unit.

[0005] CN119125245A discloses a flexible electronic nose that integrates sensing, storage, and computing based on a fabric-based gas sensing unit. This electronic nose consists of a fabric substrate, a flexible gas molecule information sensing module, a flexible ring oscillator, and a flexible gas molecule information storage and computing module. All three modules are mounted on the fabric substrate, and the flexible gas molecule information sensing module, flexible ring oscillator, and flexible gas molecule information storage and computing module are sequentially electrically connected. This fabric-based electronic nose requires step-by-step fabrication of interdigitated electrodes, a sensitive layer, a ring oscillator, and a memristor array. The material and process compatibility requirements for each layer are high, making it prone to problems such as interlayer delamination and poor contact. The flexible circuitry of the memristor and ring oscillator is also prone to open circuits during wear, exhibiting poor mechanical and environmental tolerance.

[0006] On the one hand, most current electronic nose designs are rigid structures, while flexible electronic nose devices suffer from insufficient portability, lack of green and environmentally friendly concepts, complex sensor unit fabrication processes, and poor stability. On the other hand, traditional machine learning algorithms suffer from low accuracy and long training times in pattern recognition. Therefore, it is necessary to explore and develop better sensor design methods and better algorithm models to improve the performance of gas pattern recognition. Summary of the Invention

[0007] The technical problem this invention aims to solve is to provide a low-cost, simple, and easy-to-implement method for fabricating a flexible electronic nose. The design process eliminates the need for a high-temperature environment, significantly reducing energy consumption while avoiding damage to material properties from high temperatures, and allowing for compatibility with a wider range of low-cost basic materials. By simplifying the fabrication steps and reducing equipment dependence, and primarily using inexpensive and readily available environmentally friendly raw materials, production costs are significantly reduced. Sensor units fabricated using this process exhibit advantages such as high sensitivity, low cost, environmental friendliness, and long-term stability, and can achieve long-term stable operation at room temperature. Furthermore, compared to traditional machine learning algorithms, neural network models, due to their automatic learning of complex features through multi-layer nonlinear transformations, demonstrate stronger fitting and generalization capabilities in large-scale data and complex tasks, providing a practical solution for the intelligent and large-scale application of gas sensing technology.

[0008] This invention is achieved using the following technical solution: The fabrication method of a flexible electronic nose based on laser-induced graphene mainly involves the following steps: 1) Design and fabrication of gas-sensitive unit Step 1: Select PI of a specific thickness as the substrate material; Step 2: Use a carbon dioxide laser to fabricate LIG electrodes on the selected polyimide substrate in one step to obtain the LIG device; Step 3: Prepare SnO2, WO3, and ZnO-In2O3 gas-sensitive materials and deposit them onto the LIG surface to form SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units; 2) Signal Acquisition The response signals of each sensing unit are collected by a signal acquisition device; the gas response characteristics are analyzed, and a gas response signal dataset is constructed. 3) Pattern recognition model establishment The original dataset undergoes data processing to enhance its features. Principal Component Analysis (PCA) is used for data dimensionality reduction and simplification. Model training involves building a classifier model using machine learning algorithms and training the classifier with the dataset to obtain the optimal gas classification model.

[0009] In the above technical solution, preferably, in step 1, the thickness of the polyimide film is selected as 50 μm, 75 μm, 100 μm, 150 μm, 200 μm, etc. In step 2, the laser can be a 10.64 μm CO2 infrared laser, a 355nm ultraviolet laser, etc. The parameters of the CO2 laser are selected as follows: laser power can be 2-5 W (approximately 8.5% of the total power of 30 W), scanning speed is 50-75 mm / s (approximately 5% of the total speed of 1279 mm / s), image density is 5-10, defocus distance is 0 mm, pixel density is 300-500 PPI, and the scanning mode is raster mode. The electrode shape of the prepared LIG is an interdigitated electrode.

[0010] In step 3, the SnO2, WO3, and ZnO-In2O3 gas-sensitive materials are mixed with deionized water and sonicated for 10-20 minutes to form a uniform suspension. After the material is deposited onto the LIG surface, it is dried in a drying oven to form a stable sensing unit device. The heating temperature is 60-80℃, and the drying time is 30-50 minutes. In the above technical solution, preferably, in step 2), the response value of the gas response is selected as the gas feature, and a gas response value dataset is established; In the above technical solution, preferably, in step 3), data cleaning, including filling in or removing missing values ​​and identifying and processing outliers, ensures data quality. PCA reduces the three-dimensional data to two dimensions and displays it as a two-dimensional image to observe data clustering characteristics; the gas classification model uses a backpropagation neural network (BPNN) for classification, with 80% of the dataset used for training and 20% for testing. The BPNN network topology consists of three layers: an input layer, a hidden layer, and an output layer. The input signal contains the response values ​​of three sensing units to different concentrations of gas. Five concentration gradients are set for each gas, and each sensor unit undergoes six cycles of concentration gradient testing. The model is optimized by optimizing the number of neurons in the hidden layer, the activation function of the hidden layer, the activation function of the output layer, the loss function of the entire model, the optimizer, and the training epochs parameters. The classification results are displayed using a confusion matrix.

[0011] Compared with existing technologies, the advantages of the technical solution in this application are as follows: This invention relates to a method for fabricating a laser-induced graphene flexible electronic nose for gas monitoring and identification. Compared to traditional gold or silver electrodes, LIG ​​technology offers advantages in gas-sensing unit design, including simpler fabrication, lower cost, greater customizability, and environmental friendliness, avoiding the difficulties associated with the development and extraction of complex electrode materials. By adjusting laser parameters, the electrode's length, width, and microstructure (porous or dense) can be controlled. This highly flexible process facilitates customized electrode design and improves electrode conductivity. Under specific parameters, the fabrication time for a single LIG interdigitated electrode is less than 3 minutes, significantly improving efficiency. Sensing units are obtained by depositing different gas-sensing materials onto the electrode surface using a simple deposition method. LIG-based sensing units possess excellent gas sensing and mechanical properties, exhibiting high mechanical stability capable of withstanding bending deformation (bending radius > 40 mm). This allows for wearable and integrated applications in various scenarios (chemical plants, mines, etc.) for gas monitoring. Furthermore, compared to traditional machine learning algorithms, BPNN achieves accurate identification of multi-component mixed gases under specific conditions, offering high accuracy and a shorter model training cycle. This flexible electronic nose shows great application potential in the field of flexible electronics.

[0012] Compared to existing electronic nose methods, the method provided by this invention expands the types of preparation and application of flexible electronic noses, improves the performance characteristics of LIG in the field of gas sensing, responds to the concept of green environmental protection, and is conducive to the diversified development of electronic nose technology. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a flexible electronic nose structure. Figure 2 A schematic diagram illustrating the design process of a flexible electronic nose sensing unit; Figure 3 This is a schematic diagram of the structure of the interdigitated electrode; Figure 4 Physical images of SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units; Figure 5 A schematic diagram of the flexible attachment of interdigital electrodes to fabric; Figure 6 The dynamic response diagrams of SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units to 1 ppm H2S are shown. Figure 7 Mechanical stability diagrams of SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units against 500 ppm CH4; Figure 8 This is a confusion matrix diagram of a BPNN neural network for six different types of gases, both single and mixed. Detailed Implementation

[0014] To make the objectives, technical solutions, and beneficial effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Unless otherwise stated, the technical methods used in this invention are all conventional techniques in the art.

[0015] Example 1 like Figure 2 The design flowchart for each sensing unit in the electronic nose design is shown, and the detailed steps are as follows: Step 1: Draw the shape of the interdigitated electrodes using software such as CorelDraw or AutoCAD; Step 2: Fix a 75μm thick PI film, and use a nitrogen dioxide laser to fabricate LIG interdigitated electrodes on the selected PI substrate to obtain a LIG device; Step 3: Prepare gas-sensitive material solutions of different metal oxide nanoparticles at room temperature, then deposit them on the LIG surface. Connect the electrodes to copper sheets using silver paste and lead them out. Place the obtained devices in a drying oven for drying treatment to obtain a sensor device with stable performance. In step 1, the dimensions of the interdigitated electrode are 9.3 mm * 13.26 mm; In step 2, the laser parameters are set as follows: output mode is raster mode, power is 2.58 W, speed is 60 mm / s, PPI is 500, image density is 5, and defocus distance is 0 mm. The metal oxide nanoparticles in step 3 include SnO2, WO3, and ZnO-In2O3. Figure 3 This is a schematic diagram of the interdigitated electrode structure obtained by laser direct writing PI according to the present invention, wherein the dimensions of the interdigitated electrode are 9.3mm * 13.26mm, the finger width is 0.5mm, and the finger spacing is 0.5mm; Figure 4 The image shows a physical diagram of the SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units of the present invention. The sensing unit includes a LIG interdigitated electrode layer and a metal oxide gas-sensitive layer.

[0016] Figure 5 A schematic diagram showing the sensor array attached to the fabric is presented. It demonstrates good adhesion and portability, showcasing the flexible and wearable characteristics of laser-induced graphene-based sensor arrays.

[0017] Figure 6This paper presents the dynamic responses of the SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units of this invention when exposed to hydrogen sulfide gas at room temperature. The hydrogen sulfide gas concentration ranges from 1 to 5 ppm, with a step size of 1. The results show that each sensing unit exhibits good dynamic continuity, and the WO3 / LIG unit has the largest response to hydrogen sulfide, with maximum responses of -30.19%, -33.67%, -38.14%, -42.84%, and -45.81% for the 1-5 ppm ranges, respectively.

[0018] Figure 7 This image shows the performance of the SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units of this invention, tested for their bending stability at room temperature. The response of each sensing unit to 500 ppm methane was tested at different bending radii. The results show that each sensing unit exhibits good mechanical stability.

[0019] Figure 8 This invention utilizes SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units to test the response signals of three single gases (NO2, H2S, and CH4) and three gas mixtures (NO2-CH4, NO2-H2S, and N2S-CH4). A dataset was established, and a BPNN neural network model based on the PyTorch framework was used to train the BP neural network model, obtaining a classification confusion matrix between the actual and predicted categories. The optimal model was obtained by adjusting the network hyperparameters. Specifically, the hidden layer had 6 neurons, the hidden layer activation function was ReLU(), the output layer activation function was SoftMax(), the overall model loss function was cross-entropy loss, and the optimizer was Adam. With 150 training epochs, no overfitting or underfitting was observed. This demonstrates that the BP neural network model exhibits a 100% classification accuracy for gas classification.

[0020] Comparative Example 1: LIG interdigitated electrodes were printed using optimal laser parameters. To illustrate the impact of different model parameters on model performance, with 150 training epochs, changing the number of hidden layer neurons to 4 resulted in underfitting on both the training and test sets. The lack of neurons led to insufficient feature learning from the data, thus affecting the model's recognition accuracy. While changing the number of hidden layer neurons to 10 allowed for accurate gas recognition, the training cycle increased by 12% compared to the 6-neuron configuration. This is not the optimal solution for the BPNN model.

[0021] Comparative Example 2: A 75 μm thick PI film was adhered to a glass slide and placed on a 10.6 μm wavelength CO2 laser stage. The laser output mode was set to grating mode, power was 1W, speed was 100 mm / s, PPI was 500, image density was 5, and defocus distance was 0 mm. These laser process parameters exhibited problems of low power and high scanning speed, resulting in insufficient laser energy input to the polyimide (PI) film. Under these conditions, it was impossible to prepare laser-induced graphene (LIG) with uniform carbonization and excellent performance. Depositing gas-sensitive material onto the LIG electrode surface and measuring the resistance change of the sensing unit using a source meter revealed that the sensing unit resistance was as high as GΩ, equivalent to an open circuit.

[0022] Comparative Example 3: A 75 μm thick PI film was adhered to a glass slide and placed on a 10.6 μm wavelength CO2 laser stage. The laser output mode was set to grating mode, power was 10 W, speed was 20 mm / s, PPI was 500, image density was 5, and defocus distance was 0 mm. These laser process parameters resulted in excessive power and low scanning speed, leading to excessive laser energy input to the polyimide (PI) film. This excessive laser energy caused the PI film to ablate, break, and curl due to thermal accumulation. Simultaneously, the laser-induced microstructure of graphene (LIG) was severely damaged by the high energy, ultimately forming distorted, irregular, and discontinuous patterns on the PI film surface, failing to meet the application requirements of the electrode material.

[0023] It should be understood that the above embodiments are merely illustrative of specific implementations of the present invention and are not intended to limit the scope of the present invention. Those skilled in the art can make improvements or modifications based on the above description, and all such changes and modifications should fall within the protection scope of the appended claims.

[0024] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for fabricating a flexible electronic nose based on laser-induced graphene, characterized in that, Includes the following steps: 1) Design and fabrication of gas-sensitive unit Step 1: Select PI of a specific thickness as the substrate material; Step 2: Use a carbon dioxide laser to fabricate LIG electrodes on the selected polyimide substrate in one step to obtain the LIG device; Step 3: Prepare SnO2, WO3, and ZnO-In2O3 gas-sensitive materials and deposit them onto the LIG surface to form SnO2 / LIG, WO3 / LIG, and ZnO-In2O3 / LIG sensing units; 2) Signal Acquisition The response signals of each sensing unit are collected by a signal acquisition device; the gas response characteristics are analyzed, and a gas response signal dataset is constructed. 3) Pattern recognition model establishment The original dataset is processed to enhance its features. Principal component analysis is used for data dimensionality reduction and simplification to reduce the data dimension. Model training is performed by building a classifier model using machine learning algorithms and training the classifier with the dataset to obtain the optimal gas classification model.

2. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 1, the thickness of the polyimide film is selected as 50 μm, 75 μm, 100 μm, 150 μm or 200 μm.

3. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 2, the laser selected is a 10.64 μm CO2 infrared laser, a 355 nm ultraviolet laser, etc. The parameters of the carbon dioxide laser are selected as follows: laser power of 2-5 W, scanning speed of 50-75 mm / s, image density of 5-10, defocus distance of 0 mm, pixel density of 300-500 PPI, scanning mode of grating mode, and the shape of the prepared LIG electrode is interdigitated electrode.

4. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 3, SnO2, WO3, and ZnO-In2O3 gas-sensitive materials are mixed with deionized water and sonicated for 10-20 minutes to form a uniform suspension. After the material is deposited onto the LIG surface, it is dried in a drying oven to form a stable sensing unit device. The heating temperature is 60-80℃ and the drying time is 30-50 minutes.

5. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 2), the gas response value is selected as the gas feature, and a gas response value dataset is established.

6. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 3), data cleaning, including filling or removing missing values ​​and identifying and processing outliers, ensures data quality. PCA reduces the three-dimensional data to two dimensions and displays two-dimensional images to observe data clustering characteristics.

7. The method for preparing a flexible electronic nose based on laser-induced graphene according to claim 1, characterized in that, In step 3), the gas classification model uses a neural network for classification. 80% of the dataset is used for training and 20% for testing. The BPNN network topology consists of three layers: an input layer, a hidden layer, and an output layer. The input signal contains the response values ​​of three sensing units to different concentrations of gas. Five concentration gradients are set for each gas. Each sensor unit receives six cycles of concentration gradient testing. The model is optimized by optimizing the number of neurons in the hidden layer, the activation function of the hidden layer, the activation function of the output layer, the loss function of the entire model, the optimizer, and the training epoch parameters. The final result is displayed using a confusion matrix.

Images