Nonlinear static evaluation method of carbon fiber resistance paste for sensor

Abstract

The invention discloses a non-linear static evaluation method of carbon fiber resistance slurry for a sensor, which comprises the following steps of: S1, loading or unloading a sensor sample in sequence, and collecting a Raman spectrogram of each sensitive wire grid in the same area under different loads; s2, for the same sensitive wire grid area, six parameters of each load point form a data matrix X; s3, according to the data matrix X of each sensitive wire grid area, calculating a nonlinear error evaluation value L and a lag error evaluation value H of each column of parameters, and obtaining six evaluation values L, six 0% load point evaluation values H and six 50% load point evaluation values H; and S4, selecting the maximum value of all the evaluation values L and H as a final prediction index of the performance of the sensor sample. According to the method, various methods are adopted to evaluate the performance of the sensor made of the carbon fiber resistance paste, analysis can be carried out from different ideas, and the accuracy and reliability of sensor performance evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor performance evaluation, and in particular to a nonlinear static evaluation method of carbon fiber resistor slurry for sensors. Background Art

[0002] Currently, various force sensors can be made by utilizing the conductive and resistive strain properties of carbon materials. These materials are first mixed with other solvents to form inks, which are then printed and used in various processes. For example, "CN110105813A - A carbon-based conductive ink, its preparation method and use," "CN 117804326A - Preparation and application of an embedded carbon nanocomposite flexible piezoresistive sensor," and "CN 118516012 A - A sensitive ink for a high-sensitivity coefficient strain gauge, a sensitive ink-printed strain gauge, and its preparation method," all describe the ink manufacturing method and its application in force measurement.

[0003] However, there's still a lack of a predictive method for evaluating the performance of force sensors made using these methods. Currently, performance can only be determined after the sensor is finished, through testing such as calibration, creep, and temperature characteristics. This increases costs and time, and lacks guidance for material or process improvements. However, predicting the performance of these force sensors before they are manufactured would accelerate product development and reduce costs.

[0004] From a microscopic perspective, the performance differences of conductive ink force sensors based on carbon materials, such as carbon nanotubes, graphene and graphite, mainly stem from the physical properties of the carbon material itself and its behavioral changes in the composite material. These changes lead to changes in the resistivity, electron mobility and microstructure of the composite material, thus giving this type of force sensor diverse characteristics. Therefore, analyzing the micromorphology of carbon-sensitive materials is one of the potential implementation paths. At the same time, predicting and optimizing the performance of force sensors through spectral analysis and processing technology can provide an important basis for the development of efficient and accurate sensors. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to propose a nonlinear static evaluation method for carbon fiber resistor slurry for sensors. By adopting multiple evaluation methods and constructing some quantitative indicators to analyze the micromorphology of the sensitive wire grid of the force sensor, the sensor performance can be evaluated and guidance can be provided for subsequent process optimization.

[0006] This is achieved through the following technical solutions: A nonlinear static evaluation method for carbon fiber resistor slurry for sensors, the method comprising the following steps: S1, sequentially loading the sensor sample, collecting Raman spectra of each sensitive wire grid in the same area under 0%, 50% and 100% load; then sequentially unloading the sensor sample until zero load, collecting Raman spectra of each sensitive wire grid in the same area under 50% and 0% load, respectively obtaining spectral data of each sensitive wire grid area under different loading or unloading loads, including 6 parameters: D peak height, D peak half-height width, G peak height, G peak half-height width, D peak frequency shift position, G peak frequency shift Position, or D peak height, D peak half-height width, G peak height, G peak half-height width, 2D peak height, 2D peak half-height width; S2. For the same sensitive wire grid area, the six parameters of each load point are combined into a data matrix X with a dimension of 5×6; S3. Based on the data matrix X of each sensitive wire grid area, the nonlinear error evaluation value L and the hysteresis error evaluation value H of each column parameter are calculated respectively, and 6 evaluation values ​​L, 6 0% load evaluation value H, and 6 50% load evaluation value H are obtained; S4. The maximum value of all evaluation values ​​L and H is selected as the final prediction index of the sensor sample performance. The method of the present invention analyzes the micromorphology of the sensitive wire grid of the force sensor by adopting multiple evaluation methods and constructing some quantitative indicators, avoiding the problems of complex calculations and insufficient precision of traditional evaluation methods, and can improve the accuracy and reliability of sensor performance evaluation.

[0007] Preferably, in step S2, the data matrix X is: ,in : i = 1 for loading 0% FS, i = 2 for loading 50% FS, i = 3 for loading 100% FS, i = 4 for unloading 50% FS, and i = 5 for unloading 0% FS; j = 1 for D peak height, j = 2 for D peak half-height width, j = 3 for G peak height, j = 4 for G peak half-height width, j = 5 for D peak frequency shift position, and j = 6 for G peak frequency shift position; or j = 1 for D peak height, j = 2 for D peak half-height width, j = 3 for G peak height, j = 4 for G peak half-height width, j = 5 for 2D peak height, and j = 6 for 2D peak half-height width. By constructing the spectral data collected under different loads into a data matrix, the response relationship between different loads and spectral data can be established, facilitating subsequent data calculation and processing.

[0008] Preferably, in step S3, the calculation formula of the nonlinear error evaluation value L is: ,in, Indicates the data collected when loading 100% FS. Indicates the data collected when loading 50% FS. Represents data collected when loading or unloading at 0% FS. The nonlinear error evaluation value L for each column parameter of the data matrix X is calculated to provide input data for the subsequent prediction model.

[0009] Preferably, in step S3, the calculation formula of the 0% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 0% FS. The hysteresis error evaluation value H of the 0% load point parameter in each column of the data matrix X is calculated to provide data support for the subsequent prediction model.

[0010] Preferably, in step S3, the calculation formula for the 50% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 50% FS. The hysteresis error evaluation value H of the 50% load point parameter in each column of the data matrix X is calculated to provide data support for the subsequent prediction model.

[0011] Preferably, a neural network is used to construct prediction networks for the nonlinear error evaluation value L and the hysteresis error evaluation value H, respectively, based on all evaluation values ​​L and H obtained in step S3, and the sensor sample performance is analyzed based on the output nonlinear error evaluation values ​​L and hysteresis error evaluation values ​​H. By using a neural network to construct prediction networks for the nonlinear error evaluation value L and the hysteresis error evaluation value H, the prediction networks can be used to independently predict the nonlinear error evaluation value L and the hysteresis error evaluation value H of the sensor and perform separate analyses.

[0012] Preferably, a neural network is used to construct a single prediction network based on all the evaluation values ​​L and H obtained in step S3, and the sensor sample performance is analyzed based on the output nonlinear error evaluation value L and hysteresis error evaluation value H. By using a neural network to construct a single prediction network, the interaction between the nonlinear error evaluation value L and the hysteresis error evaluation value H can be considered, thereby providing a more in-depth analysis of sensor performance.

[0013] Preferably, after obtaining the data matrix X in step S2, the parameters of each column of the data matrix X are normalized. A neural network model is then constructed based on the processed data. The sensor sample performance is analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model. By directly using the collected spectral data to construct the neural network model, the workload of data preprocessing is reduced.

[0014] Preferably, when the sensor sample is loaded and unloaded, spectral data is collected for each sensitive wire grid area under each load, including nine parameters: D peak height, D peak half-width, G peak height, G peak half-width, 2D peak height, 2D peak half-width, D peak frequency shift position, G peak frequency shift position, and 2D peak frequency shift position; each column parameter of the resulting data matrix X is then normalized, and a neural network model is constructed based on the processed data. The sensor sample performance is analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model. By increasing the range of spectral data collection for constructing the neural network model, the dimensionality of the input data can be increased, thereby improving the accuracy of the evaluation.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The technical solution of the present invention analyzes the micromorphology of the sensitive wire grid of the force sensor by constructing some quantitative indicators, avoiding the problems of complex calculations and insufficient precision of traditional evaluation methods, and can improve the accuracy and reliability of sensor performance evaluation and provide guidance for subsequent process optimization; at the same time, by adopting multiple evaluation methods and using neural networks to construct prediction models for nonlinear error evaluation values ​​L, hysteresis error evaluation values ​​H and repeatability error evaluation values ​​R, it is possible to analyze and evaluate sensor performance from different angles, thereby improving the efficiency and accuracy of sensor performance evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flowchart of a nonlinear static evaluation method for carbon fiber resistor paste for sensors; Figure 2 It is the front view of the sensor sample; Figure 3 A top view of the sensor sample. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0018] like Figure 1As shown, it is a flow chart of a nonlinear static evaluation method for carbon fiber resistor slurry for sensors. First, the sensor sample is loaded and unloaded in sequence, and Raman spectra of each sensitive wire grid in the same area under different loads are collected. Secondly, for the same sensitive wire grid area, the six parameters of each load point are combined into a data matrix X. Then, based on the data matrix X of each sensitive wire grid area, the nonlinear error evaluation value L and the hysteresis error evaluation value H of each column parameter are calculated respectively, and six evaluation values ​​L, six 0% load evaluation values ​​H and six 50% load evaluation values ​​H are obtained. Finally, the maximum value of all the evaluation values ​​L and H is selected as the final prediction index of the sensor sample performance. The method of the present invention adopts multiple methods to evaluate the performance of the sensor made of carbon fiber resistor slurry, and can be analyzed from different perspectives, thereby improving the accuracy and reliability of the sensor performance evaluation.

[0019] The method specifically comprises the following steps: S1. Load the sensor sample in sequence, and collect Raman spectra of each sensitive wire grid in the same area under 0%, 50% and 100% loads; then unload the sensor sample in sequence until zero load, collect Raman spectra of each sensitive wire grid in the same area under 50% and 0% loads, and obtain spectral data of each sensitive wire grid area under different loading or unloading loads, including 6 parameters: D peak height, D peak half-height width, G peak height, G peak half-height width, D peak frequency shift position, G peak frequency shift position, or D peak height, D peak half-height width, G peak height, G peak half-height width, 2D peak height, 2D peak half-height width.

[0020] Among them, the D peak is used to indicate the defect density of the carbon material used to make the sensor sample; the G peak is used to indicate the sp² ordering of the carbon material used to make the sensor sample; and the 2D peak is used to identify the electronic structure and number of layers of the carbon material used to make the sensor sample. By collecting spectral data to obtain relevant data such as the D peak and the G peak, important information about the structural characteristics of the carbon material can be obtained.

[0021] Secondly, the sensor sample is fastened to the base by screws, and the left side of the sensor sample is fixed and the right side is not fixed. The loading and unloading methods of the sensor sample are as follows: use a set screw to support the bottom right side of the sensor sample through the threaded hole on the right side of the base to achieve loading of the sensor sample; and unload the sensor sample by pulling the set screw out of the threaded hole on the right side of the base from the bottom right side of the sensor sample.

[0022] like Figure 2 As shown in the figure, it is the front view of the sensor sample. Figure 3 The figure shows the top view of the sensor sample, combined with Figure 2 and Figure 3The sensor sample adopts a parallel beam structure. There are two relatively gentle stress concentration areas on the structural design, on which sensitive wire grids are printed. Each wire grid is 0.2mm wide and 1.5mm long. As the spectral data collection area, there are two positions on the sensor sample: the tensile stress wire grid collection position and the compressive stress wire grid collection position. By designing two spectral data collection areas on the sensor sample, the Raman spectra of the sensitive wire grids in different areas can be collected, and the spectral data in different loading and unloading stages can be obtained.

[0023] S2. For the same sensitive wire grid area, the six parameters of each load point are combined into a data matrix X with a dimension of 5×6. The load points include: 0%, 50% and 100% loading, 50% and 0% unloading. The six parameters include: D peak height, D peak half-height width, G peak height, G peak half-height width, D peak frequency shift position, G peak frequency shift position, or D peak height, D peak half-height width, G peak height, G peak half-height width, 2D peak height, 2D peak half-height width; the data matrix X is: ,in : i=1 is loading 0% FS, i=2 is loading 50% FS, i=3 is loading 100% FS, i=4 is unloading 50% FS, and i=5 is unloading 0% FS; j=1 is the D peak height, j=2 is the D peak half-height width, j=3 is the G peak height, j=4 is the G peak half-height width, j=5 is the D peak frequency shift position, and j=6 is the G peak frequency shift position, or j=1 is the D peak height, j=2 is the D peak half-height width, j=3 is the G peak height, j=4 is the G peak half-height width, j=5 is the 2D peak height, and j=6 is the 2D peak half-height width; among them, FS stands for Full Scale, which means full scale; by constructing the spectral data collected under different loads into a data matrix, the response relationship between different loads and spectral data can be established, which is convenient for subsequent data calculation and processing.

[0024] S3. Based on the data matrix X of each sensitive wire grid area, the nonlinear error evaluation value L and the hysteresis error evaluation value H of each column parameter are calculated respectively, and 6 evaluation values ​​L, 6 0% load evaluation values ​​H and 6 50% load evaluation values ​​H are obtained.

[0025] Specifically, the calculation formula of the nonlinear error evaluation value L is: ,in, Indicates the data collected when loading 100% FS. Indicates the data collected when loading 50% FS. Represents the data collected when loading or unloading 0% FS; the calculation formula for the 0% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 0% FS. represents the data collected when unloading 0% FS; the calculation formula for the 50% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 50% FS. represents the data collected when unloading 50% FS; by calculating the nonlinear error evaluation value L, the 0% load point hysteresis error evaluation value H, and the 50% load point hysteresis error evaluation value H of each column parameter of the data matrix X, it can be used to provide input data for the subsequent prediction model.

[0026] S4. Select the maximum value among all evaluation values ​​L and H as the final prediction indicator of sensor sample performance.

[0027] In this embodiment, based on all the evaluation values ​​L and H obtained in step S3, a neural network is used to construct prediction networks for the nonlinear error evaluation value L and the hysteresis error evaluation value H, respectively. The performance of the sensor sample is analyzed based on the output nonlinear error evaluation value L and the hysteresis error evaluation value H. The specific steps for constructing the prediction networks for L and H to evaluate the sensor performance are as follows: In the first step, a prediction network for the nonlinear error evaluation value L is constructed using a BP network structure, wherein the input layer includes 6 nodes corresponding to the 6 evaluation values ​​L calculated in step S3; the output layer includes 1 node corresponding to the predicted L value output by the sensor; the hidden layer includes 1 layer, and the number of nodes is selected according to the amount of data; Specifically, the mathematical expression of the prediction network of the nonlinear error evaluation value L is: hidden layer: , Represents the result obtained after the input data is transformed and activated. represents the hidden layer weight matrix, Represents the hidden layer bias vector, output layer: , represents the output layer weight matrix of L, represents the output layer bias vector of L, and n is the number of hidden layer nodes; In the second step, a prediction network for the hysteresis error evaluation value H is constructed using a BP network structure. The input layer includes 12 nodes, corresponding to the six 0% load evaluation values ​​H and the six 50% load evaluation values ​​H calculated in step S3; the output layer includes 1 node, corresponding to the predicted H value output by the sensor; and the hidden layer includes 1 layer, with the number of nodes selected based on the amount of data. Specifically, the mathematical expression of the prediction network of the lag error evaluation value H is: Hidden layer: , output layer: , represents the output layer weight matrix of H, represents the output layer bias vector of H, n is the number of hidden layer nodes; and the hidden layer activation function of the prediction network of L and H is the Sigmoid function, and the output layer activation function is the linear activation function.

[0028] Among them, the prediction networks of L and H both analyze the sensor performance through the L value or H value output by the output layer; therefore, by adopting neural networks to construct prediction networks for the nonlinear error evaluation value L and the hysteresis error evaluation value H respectively, the prediction networks can be used to separately predict the nonlinear error evaluation value L and the hysteresis error evaluation value H of the sensor, and the sensor performance can be evaluated separately according to the L value or the H value.

[0029] In this embodiment, a single prediction network is constructed using a neural network based on all evaluation values ​​L and H obtained in step S3, and the performance of the sensor sample is analyzed based on the output nonlinear error evaluation value L and hysteresis error evaluation value H. The specific contents of constructing a single prediction network for sensor performance evaluation include: A single prediction network is constructed using a BP network structure, in which the input layer includes 18 nodes, corresponding to the 6 evaluation values ​​L, 6 0% load evaluation values ​​H, and 6 50% load evaluation values ​​H calculated in step S3; the output layer includes 2 nodes, corresponding to the predicted L value and predicted H value output by the sensor; the hidden layer includes 1 layer, and the number of nodes is selected according to the amount of data.

[0030] Specifically, the mathematical expression of a single prediction network is: Hidden layer: , output layer: and , and denote the output layer weight matrices of L and H respectively, and They represent the output layer bias vectors of L and H respectively, n is the number of hidden layer nodes; and the hidden layer activation function of the single prediction network is the Sigmoid function, and the output layer activation function is the linear activation function.

[0031] Among them, the single prediction network evaluates the sensor performance through the L value or H value output by the output layer; by using a neural network to construct a single prediction network, the interaction between the nonlinear error evaluation value L and the hysteresis error evaluation value H can be considered to conduct a more in-depth analysis of the sensor performance.

[0032] In this embodiment, after obtaining the data matrix X in step S2, the parameters of each column of the data matrix X are first normalized, and then a neural network model is constructed based on the processed data. The performance of the sensor sample is analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model. The specific steps of constructing the neural network model to evaluate sensor performance are as follows: The first step is to normalize the parameters of each column of the data matrix X in step S2. The normalization calculation formula is: ,in, Indicates the spectral data collected when loading 100% FS. Indicates the spectral data collected when loading 0% FS, i is 1, 2...5, and j is 1, 2...6; In the second step, a neural network model was constructed using a multi-task learning framework. The input layer used normalized data, and the feature vector dimension for each sample was 60, i.e., (3 loading + 2 unloading) × 2 regions × 6 parameters = 60. Each sample corresponded to three error metrics, L, H, and R, which were calibrated experimentally or calculated based on historical data. The hidden layer was a fully connected layer. The output layer consisted of three independent branches, each of which was a fully connected layer, outputting the predicted L value, predicted H value, and predicted R value of the corresponding sensor, respectively. The third step is to train the neural network model. The weighted mean square error is used to balance the three independent branches of L, H and R. The training data is then divided into training set, validation set and test set according to the ratio of 8:1:1. Finally, the neural network model is trained. In the fourth step, the feature vector is input into the trained neural network model, which contains a shared feature extraction layer and independent output branches, and outputs the nonlinear error evaluation value L, the hysteresis error evaluation value H and the repeatability error evaluation value R respectively, and judges the sensor performance based on the output error indicators.

[0033] By directly using the collected spectral data to build a neural network model, the workload of data preprocessing can be saved and the efficiency of sensor performance evaluation can be improved.

[0034] In this embodiment, when the sensor sample is loaded and unloaded, spectral data of each sensitive wire grid area under each load is collected, including nine parameters: D peak height, D peak half-height width, G peak height, G peak half-height width, 2D peak height, 2D peak half-height width, D peak frequency shift position, G peak frequency shift position, and 2D peak frequency shift position; each column parameter of the composed data matrix X is then normalized, and a neural network model is constructed based on the processed data. The sensor sample performance is analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model, thereby increasing the dimensionality of the input data and improving the accuracy of the sensor performance evaluation.

[0035] In summary, the present invention analyzes the micromorphology of the sensitive wire grid of the force sensor by constructing some quantitative indicators, thereby avoiding the problems of complex calculations and insufficient precision of traditional evaluation methods, and can improve the accuracy and reliability of sensor performance evaluation, and provide guidance for subsequent process optimization; at the same time, by adopting multiple evaluation methods and using neural networks to construct prediction models for nonlinear error evaluation values ​​L, hysteresis error evaluation values ​​H and repeatability error evaluation values ​​R, the sensor performance can be analyzed and evaluated from different angles, thereby improving the efficiency and accuracy of sensor performance evaluation, which is significantly progressive.

[0036] The above embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A nonlinear static evaluation method for carbon fiber resistor slurry for sensors, characterized in that: The steps include: S1. Load the sensor sample in sequence, and collect Raman spectra of each sensitive wire grid in the same area under 0%, 50% and 100% loads; then unload the sensor sample in sequence until zero load, collect Raman spectra of each sensitive wire grid in the same area under 50% and 0% loads, and obtain spectral data of each sensitive wire grid area under different loading or unloading loads, including 6 parameters: D peak height, D peak half-height width, G peak height, G peak half-height width, D peak frequency shift position, G peak frequency shift position, or D peak height, D peak half-height width, G peak height, G peak half-height width, 2D peak height, 2D peak half-height width; S2. For the same sensitive wire grid area, the six parameters of each load point are combined into a data matrix X with a dimension of 5×6; S3. Calculate the nonlinear error evaluation value L and the hysteresis error evaluation value H of each column parameter based on the data matrix X of each sensitive wire grid area, and obtain 6 evaluation values ​​L, 6 0% load evaluation values ​​H, and 6 50% load evaluation values ​​H; S4. Select the maximum value among all evaluation values ​​L and H as the final prediction indicator of sensor sample performance.

2. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: In step S2, the data matrix X is: ,in : i=1 is loading 0% FS, i=2 is loading 50% FS, i=3 is loading 100% FS, i=4 is unloading 50% FS, i=5 is unloading 0% FS; j=1 is D peak height, j=2 is D peak half-height width, j=3 is G peak height, j=4 is G peak half-height width, j=5 is D peak frequency shift position, j=6 is G peak frequency shift position, or j=1 is D peak height, j=2 is D peak half-height width, j=3 is G peak height, j=4 is G peak half-height width, j=5 is 2D peak height, j=6 is 2D peak half-height width.

3. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: In step S3, the calculation formula of the nonlinear error evaluation value L is: ,in, Indicates the data collected when loading 100% FS. Indicates the data collected when loading 50% FS. Indicates data collected when loading or unloading 0% FS.

4. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: In step S3, the calculation formula of the 0% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 0% FS. Indicates the data collected when 0% FS is unloaded.

5. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: In step S3, the calculation formula for the 50% load point hysteresis error evaluation value H is: ,in, Indicates the data collected when loading 50% FS. Indicates the data collected when 50% FS is unloaded.

6. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: According to all the evaluation values ​​L and H obtained in step S3, a neural network is used to construct prediction networks for the nonlinear error evaluation value L and the hysteresis error evaluation value H, respectively, and the performance of the sensor sample is analyzed based on the output nonlinear error evaluation value L and the hysteresis error evaluation value H.

7. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: According to all the evaluation values ​​L and H obtained in step S3, a single prediction network is constructed using a neural network, and the performance of the sensor sample is analyzed based on the output nonlinear error evaluation value L and hysteresis error evaluation value H.

8. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: After obtaining the data matrix X in step S2, the parameters of each column of the data matrix X are first normalized, and then a neural network model is constructed based on the processed data. The performance of the sensor sample is analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model.

9. The nonlinear static evaluation method of carbon fiber resistor slurry for sensors according to claim 1, characterized in that: When the sensor sample was loaded and unloaded, spectral data of each sensitive wire grid area under each load was collected, including nine parameters: D peak height, D peak half-width, G peak height, G peak half-width, 2D peak height, 2D peak half-width, D peak frequency shift position, G peak frequency shift position, and 2D peak frequency shift position. After normalizing each column parameter of the data matrix X, a neural network model was constructed based on the processed data. The performance of the sensor sample was analyzed based on the nonlinear error evaluation value L, hysteresis error evaluation value H, and repeatability error evaluation value R output by the neural network model.

Images