Pressure sensor calibration system and method based on deep learning

Abstract

The present invention relates to a pressure sensor calibration system and method based on deep learning, including a pressure source, which generates a set pressure required in a calibration environment; a data collector, which obtains the pressure generated by the pressure source as an actual pressure value, and obtains the output value of the pressure sensor to be calibrated corresponding to the actual pressure value; a processing unit, which is configured to: based on the actual pressure value in the calibration environment and the output value of the pressure sensor to be calibrated, use a trained calibration model to obtain a predicted output value, determine the error between the actual pressure value in the calibration environment and the obtained predicted output value, and with the goal of minimizing the error, obtain the partial derivative of each calibration parameter and determine the update amount required for each calibration parameter; a parameter updating unit, which is configured to: obtain updated calibration parameters based on the obtained update amount, input the updated calibration parameters into the pressure sensor to be calibrated, and complete the calibration.

Description

Technical Field

[0001] The present invention relates to the field of sensor calibration technology, and in particular to a pressure sensor calibration system and method based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Sensor performance can fluctuate due to subtle changes during use and the external environment, making testing and calibration difficult. Current sensor calibration methods rely on traditional physical testing and calculation methods, resulting in low test accuracy, lack of traceability, and complex processes. Some physical testing methods rely on expensive equipment, are time-consuming, and lack accuracy. Summary of the Invention

[0004] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a pressure sensor calibration system and method based on deep learning, which uses the actual pressure value obtained in the calibration environment and the predicted value output based on the calibration model as the calibration data source. When the error between the actual pressure value and the predicted value is minimized, the update amount required for each calibration parameter is obtained. The calibration model can be used to correct the output deviation of the sensor itself, reduce the error and drift caused by the traditional calibration process, and reduce time consumption.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A first aspect of the present invention provides a pressure sensor calibration system based on deep learning, comprising:

[0007] A pressure source, generating the set pressure required in the calibration environment;

[0008] The data collector obtains the pressure generated by the pressure source as the actual pressure value, and obtains the output value of the pressure sensor to be calibrated corresponding to the actual pressure value;

[0009] a processing unit configured to: obtain a predicted output value using a trained calibration model based on an actual pressure value in a calibration environment and an output value of the pressure sensor to be calibrated; determine an error between the actual pressure value in the calibration environment and the predicted output value; obtain a partial derivative of each calibration parameter with the goal of minimizing the error; and determine a required update amount for each calibration parameter;

[0010] The parameter updating unit is configured to obtain updated calibration parameters based on the obtained update amount, input the updated calibration parameters into the pressure sensor to be calibrated, and complete the calibration.

[0011] The training data set of the calibration model includes obtaining the set pressure value of the pressure source as actual pressure value data under the calibration environment, and using the corresponding output data obtained by using a pre-calibrated pressure sensor as the training data set.

[0012] During the training of the calibration model, the output data obtained from the pre-calibrated pressure sensor is preprocessed and features are extracted.

[0013] The preprocessing includes at least one of de-averaging, normalization and filtering.

[0014] The extracted features include one or more of the average value, standard deviation, maximum value, minimum value, slope, peak value and energy spectrum density in the pressure sensor output data.

[0015] During the training process of the calibration model, the extracted features are sequentially output through the convolution layer, pooling layer and fully connected layer, and the output is the predicted pressure sensor output value P pred , with P pred The minimum between the target and the actual pressure value is used to complete the training of the calibration model.

[0016] After the calibration model is trained, the actual pressure value and the corresponding output value of the pressure sensor to be calibrated are obtained in the calibration environment, and the calibration model is used to obtain the predicted output value based on the output value of the pressure sensor to be calibrated.

[0017] After the calibration model is trained, the mean square error between the actual pressure value and the predicted output value in the calibration environment is obtained. With the goal of minimizing the mean square error, the partial derivative of each calibration parameter is solved to obtain the required update amount for each calibration parameter.

[0018] The processing unit has a data interface that is communicatively connected to the pressure sensor to be calibrated, and writes the updated calibration parameters into the pressure sensor to be calibrated through the data interface to complete the calibration.

[0019] A second aspect of the present invention provides a pressure sensor calibration method based on deep learning, comprising the following steps:

[0020] Obtain the actual pressure value under the calibration environment and the output value of the pressure sensor to be calibrated, and obtain the predicted output value based on the trained calibration model;

[0021] Determine the error between the actual pressure value under the calibration environment and the predicted output value, and with the goal of minimizing the error, solve the partial derivative of each calibration parameter to obtain the required update amount for each calibration parameter;

[0022] An updated calibration parameter is obtained based on the updated amount and input into the pressure sensor to be calibrated to complete the calibration.

[0023] Compared with the existing technology, one or more of the above technical solutions have the following beneficial effects:

[0024] 1. Using the actual pressure value obtained in the calibration environment and the predicted output value obtained based on the calibration model as the calibration data source, the required update amount for each calibration parameter is obtained when the error between the actual pressure value and the predicted output value is minimized. The calibration model can be used to correct the output deviation of the sensor itself, reducing the error and drift caused by the traditional calibration process and saving time.

[0025] 2. In a calibration environment, the actual pressure values ​​and corresponding sensor output values ​​obtained from a pre-calibrated similar sensor are used as a training data set. After training the calibration model through deep learning, the required calibration parameters are obtained. The obtained calibration parameters are then input into the pressure sensor to be calibrated to complete the update, thereby achieving calibration and effectively improving test accuracy and stability.

[0026] 3. Compared with traditional mechanical or electrical testing equipment, this method can perform testing and calibration more quickly and conveniently, while also reducing costs and human resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0028] Figure 1 is a schematic diagram of the architecture of a pressure sensor calibration system based on deep learning provided by one or more embodiments of the present invention;

[0029] Figure 2 is a schematic diagram of a pressure sensor test and calibration process provided by one or more embodiments of the present invention;

[0030] Figure 3 This is a flowchart of using a CNN model to perform pressure sensor testing and calibration according to one or more embodiments of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0033] As described in the background technology, the current calibration method for pressure sensors has problems such as low test accuracy, non-traceable calibration, and complex process. The following embodiment provides a pressure sensor calibration system and method based on deep learning. By using a sensor testing system and an external pressure calibration instrument, sensor output signal data and actual pressure value data under a series of different pressure values ​​are collected as a training data set. After training the calibration model through deep learning, the required calibration parameters are obtained, and then the obtained calibration parameters are input into the pressure sensor to be calibrated to complete the update, thereby achieving calibration.

[0034] The following embodiments take a MEMS pressure sensor as an example to provide a calibration process, which can also be extended to other types of pressure sensors.

[0035] A MEMS pressure sensor is a thin film element that deforms when subjected to pressure. This deformation can be measured using a strain gauge (piezoresistive sensing) to reflect changes in pressure, or by capacitive sensing of changes in the distance between two surfaces to reflect changes in pressure.

[0036] Example 1:

[0037] like Figure 1 As shown, the pressure sensor calibration system based on deep learning includes:

[0038] A pressure source, generating the set pressure required in the calibration environment;

[0039] The data collector obtains the pressure generated by the pressure source as the actual pressure value, and obtains the output value of the pressure sensor to be calibrated corresponding to the actual pressure value;

[0040] a processing unit configured to: obtain a predicted output value using a trained calibration model based on an actual pressure value in a calibration environment and an output value of the pressure sensor to be calibrated; determine an error between the actual pressure value in the calibration environment and the predicted output value; obtain a partial derivative of each calibration parameter with the goal of minimizing the error; and determine a required update amount for each calibration parameter;

[0041] The parameter updating unit is configured to obtain updated calibration parameters based on the obtained update amount, input the updated calibration parameters into the pressure sensor to be calibrated, and complete the calibration.

[0042] The pressure source is capable of generating a set pressure as the actual pressure value. The pressure source is pre-calibrated and has no specific structural restrictions. For example, a gas pump (typically air or nitrogen) can be used to compress the gas to the desired pressure value, thereby outputting the actual pressure value required for calibration data acquisition. After the calibration model is trained, the pressure source generates the set pressure value, which can be used as the calibration pressure. Alternatively, compressed gas can be supplied from a cylinder or pipeline, and the desired pressure can be output through a pre-calibrated pressure regulating valve. Alternatively, weights and sensors can be used (both weights and sensors are pre-calibrated).

[0043] In this embodiment, a pre-calibrated MEMS pressure sensor is used to obtain the actual pressure value generated by the pressure source. The actual pressure value generated during the process and the corresponding reading of the MEMS pressure sensor are used as the training data set of the calibration model. After the calibration model is trained, the actual pressure value generated by the pressure source and the output value corresponding to the pressure sensor to be calibrated are used as the input of the model, and the output is the predicted value of the pressure sensor to be calibrated. There is an error between the actual pressure value and the obtained predicted output value. The error here includes parameters such as the final reading of the pressure sensor to be calibrated after reaching stability and the time required to reach stability. In addition, when multiple data acquisitions are performed at an actual pressure value, the pressure sensor to be calibrated may output corresponding multiple output values. This part of the output value includes parameters such as maximum value, minimum value and standard deviation. These parameters are all calibration parameters that can affect the performance of the pressure sensor.

[0044] Therefore, this embodiment uses the actual pressure value obtained in the calibration environment and the predicted output value obtained based on the calibration model as the data source for calibration. When the error between the actual pressure value and the predicted output value is minimized, the update amount required for each calibration parameter is obtained. The calibration model can be used to correct the output deviation of the sensor itself, reduce the error and drift caused by the traditional calibration process, and reduce time consumption.

[0045] Example 2:

[0046] like Figure 1-Figure 3 As shown, the pressure sensor calibration method based on deep learning includes the following steps:

[0047] Obtain the actual pressure value under the calibration environment and the output value of the pressure sensor to be calibrated, and obtain the predicted output value based on the trained calibration model;

[0048] Determine the error between the actual pressure value under the calibration environment and the predicted output value, and with the goal of minimizing the error, solve the partial derivative of each calibration parameter to obtain the required update amount for each calibration parameter;

[0049] An updated calibration parameter is obtained based on the updated amount and input into the pressure sensor to be calibrated to complete the calibration.

[0050] Specifically:

[0051] Collect training data: Use a pre-calibrated MEMS sensor in a calibration environment to collect signal data and actual pressure value data output by the MEMS sensor at a range of different pressure values ​​as a training data set.

[0052] Data preprocessing: Filter, denoise and normalize the collected raw data to improve data quality and reduce errors.

[0053] Model training: Select an appropriate deep learning model for training and testing. Convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are commonly used models, but other models can also be used depending on the data characteristics and task requirements. The model is trained using the training set. During training, the model is optimized according to the definition of the loss function to improve its accuracy and generalization capabilities.

[0054] Model testing and tuning: Use the test set to test the trained model and fine-tune the model based on the test results. Models can be evaluated using methods such as cross-validation to determine their performance.

[0055] Calibration parameter adjustment: The output of the model is compared with the actual pressure value to determine the error and adjust the calibration parameters.

[0056] Application: Apply the adjusted calibration parameters to MEMS pressure sensors in real-world scenarios to improve sensor accuracy and reliability. This can be achieved by updating the sensor's internal parameters or by calibrating the sensor's output using an external algorithm.

[0057] The models used for calibration include but are not limited to CNN (Convolutional Neural Network) models. The following is an introduction using the CNN model as an example:

[0058] 1. Sensor Data Acquisition: In a calibration environment, obtain as many pre-calibrated MEMS pressure sensors as possible, including sensor readings at different pressures. A calibration environment refers to an environment where pressure values ​​are controllable and the equipment or device generating the pressure values ​​is pre-calibrated. By collecting MEMS pressure sensor readings and the corresponding actual pressure values ​​(where the actual pressure value is the set test pressure), data from different ranges is collected for training and testing.

[0059] This embodiment uses a pressure source that can generate a set pressure to generate an actual pressure value. The pressure source is pre-calibrated. For example, an air pump can be used to compress the air to the required pressure value, thereby outputting the actual pressure value required for calibration data acquisition. After the model training is completed, the set pressure value is generated as the calibration pressure; or compressed gas is supplied by a gas cylinder or pipeline, and the required pressure is output through a pressure regulating valve (pre-calibrated); or weights and sensors are used for measurement (the weights and sensors here are all pre-calibrated); the specific type of the pressure source used in the calibration environment is not restricted. The actual pressure value generated in the calibration environment corresponds to the reading of the MEMS pressure sensor. The actual pressure value and the corresponding sensor reading are saved as the data set for the training model.

[0060] 2. Data preprocessing: Data is preprocessed, including noise removal, normalization, smoothing, and feature extraction to ensure data quality and consistency. Assume that the preprocessed data is x i , where i represents the number of the data point.

[0061] 3. Divide the data set: Divide the preprocessed data into a training set and a test set. Assume that the training set and the test set are D train and D test .

[0062] 4. Model Design: Design a CNN model, including convolutional layers, pooling layers, and fully connected layers, to extract features from the input data and output corresponding pressure values. The model structure can be adjusted and optimized based on actual conditions to improve model accuracy and robustness.

[0063] In this embodiment, feature extraction refers to extracting features useful for the calibration task from the input sensor reading data so that the model can better perform operations such as classification or regression, which is achieved through the operations of convolutional layers and pooling layers.

[0064] This embodiment can extract one or more of the following features:

[0065] 1. Average: For a set of data, the average is the sum of all data points divided by the number of data points. The average provides a measure of the central tendency of the data. In this embodiment, multiple sensor output values ​​may exist for a given actual pressure value. The average of these multiple sensor output values ​​corresponding to that actual pressure value is calculated as a feature.

[0066] 2. Standard Deviation: The standard deviation provides the degree of dispersion of the data, that is, the range of variation of the data. The larger the standard deviation, the greater the variation of the data.

[0067] 3. Maximum and Minimum Values: Maximum and minimum values ​​can provide the range of the data, i.e., the maximum and minimum values ​​of the data. In this embodiment, there may be multiple sensor output values ​​under an actual pressure value, and there may be a maximum and minimum value among these output values.

[0068] 4. Slope: The slope can provide the trend of the data. In the pressure sensor calibration task, the slope can indicate the speed of change of the data.

[0069] 5. Peak: Peak can provide the local extreme points of the data, that is, the highest and lowest points of the data.

[0070] 6. Energy Spectral Density: Energy spectral density provides the frequency distribution of data. In pressure sensor calibration, energy spectral density can indicate the periodicity and frequency characteristics of the data.

[0071] The features here are extracted from the sensor readings. The sensor reading data can be an analog signal such as voltage or current (depending on the type of pressure sensor), but it needs to be converted into a digital signal with the corresponding ADC (analog-to-digital converter) for subsequent training.

[0072] In this embodiment, the CNN model extracts features useful for the calibration task from the input data through the operations of the convolution layer, pooling layer and fully connected layer, so as to perform subsequent operations. The input data here is the data in the training set. After the training set passes through the convolution layer, pooling layer and fully connected layer in the CNN model, a value of P is obtained. pred .

[0073] 5. Model training: The preprocessed dataset is fed into the CNN model for training. The model parameters are optimized using the backpropagation algorithm to reduce the loss function. Optimization methods, such as stochastic gradient descent and Adam, can be used during training to improve training speed and effectiveness.

[0074] 6. Model testing and tuning: Use the test set to test the trained model and tune the model based on the test results. Cross-validation and other methods can be used to evaluate the model to determine the quality of the model. Common evaluation indicators include root mean square error (RMSE), mean absolute error (MAE), etc. Assume that the prediction value of the model on the i-th test data point is , the true value is y i , the mean squared error (MSE) and mean absolute error (MAE) can be used to evaluate and compare the models.

[0075]

[0076]

[0077] Here, n represents the number of data points in the test set.

[0078] 7. Calibration Parameter Adjustment: Compare the model output with the actual value to determine the error and adjust the calibration parameters. Calibration parameters can be calculated using methods such as linear regression and polynomial regression. Calibration parameter adjustment can be achieved manually or through automated algorithms.

[0079] In CNN model design:

[0080] 1. Convolutional layer: The convolutional layer uses convolution operations to extract features of the input data. The convolution operation can be expressed as:

[0081] h i,j =∑ k,l w k,l x i+k,j+l +b

[0082] Among them, x i,j Represents the (i, j)th element of the input data, w k,l represents the (k, l)th element of the convolution kernel, b represents the bias term, h i,j Represents the (i, j)th element of the output data after the convolution operation.

[0083] 2 Pooling layer: The pooling layer can reduce the dimension and compress the feature map output by the convolution layer to reduce the parameters and computational complexity of the model. Maximum pooling and average pooling are commonly used pooling operations, which can be expressed as:

[0084] y i,j =max k,l x i+k,j+l

[0085]

[0086] Among them, x i,j Represents the (i, j)th element of the input data, y i,j Represents the (i, j)th element of the output data after the pooling operation, and k×l represents the size of the pooling operation.

[0087] 3 Fully connected layer: The fully connected layer can connect and transform the features output by the convolutional layer and the pooling layer to output the final prediction result. The fully connected layer can be expressed as:

[0088] y=Wx+b

[0089] Among them, x represents the characteristics of the input data, W represents the connection weight matrix, b represents the bias term, and y represents the output data.

[0090] In model training, cross entropy can be used as a loss function to evaluate the accuracy of the model's prediction results. The cross entropy loss function can be expressed as:

[0091]

[0092] Among them, n represents the size of the dataset, y i Indicates the actual label value, Represents the label value predicted by the model.

[0093] In model optimization, the stochastic gradient descent algorithm can be used to update the model parameters. The stochastic gradient descent algorithm can be expressed as:

[0094]

[0095] Among them, θ t represents the value of the model parameter, η represents the learning rate, Represents the loss function with respect to the parameter θ t gradient.

[0096] In summary, using a CNN model for pressure sensor testing and calibration can effectively improve the model's accuracy and robustness. When designing and implementing the model, it's important to consider implementation details such as data preprocessing, model structure design, loss function selection, and optimization algorithm selection. Furthermore, it's crucial to select and optimize the model based on the actual application scenario to achieve optimal results.

[0097] In this embodiment, pre-processing includes but is not limited to the following operations:

[0098] 1. Demeaning: By subtracting the mean of the data, the center of the data can be moved to zero, reducing the impact of noise and interference.

[0099]

[0100] Where: x i Represents the i-th data point in the original data, x' i represents the i-th data point after processing, and n represents the total number of data points.

[0101] 2 Normalization: By scaling the data, the value range of the data can be limited to a certain range, so that different data can be compared.

[0102]

[0103] Among them, x i Represents the i-th data point in the original data, x' irepresents the i-th data point after processing, min(x) and max(x) represent the minimum and maximum values ​​in the data, respectively.

[0104] 3. Filtering: Filters can be used to remove high-frequency noise and interference from data.

[0105] The model structure design of this embodiment is performed on the convolutional layer, pooling layer, and fully connected layer in the CNN model. The optimization algorithm is used to adjust the model parameters to minimize the loss function. Common optimization algorithms include stochastic gradient descent (SGD), Adam, and Adagrad. SGD formula: Where θ represents the model parameters, L represents the loss function, and α represents the learning rate.

[0106] For every known P true , using deep learning models to predict sensor output P pred The details are as follows:

[0107] Load test data into the program;

[0108] Use the same preprocessing method as the training set to process the loaded data;

[0109] Use the trained deep learning model and input the preprocessed data into the model;

[0110] The model is used to process the data to obtain the predicted sensor output P pred .

[0111] Get the sensor's predicted value P pred After that, through the known P true The error between them can be obtained, and this error indicator can be minimized by the gradient descent algorithm.

[0112] The pressure sensor calibration operation is performed based on the output results of the CNN model, which is divided into the following steps:

[0113] 1 Collect the actual pressure value P true and the corresponding sensor output value P sensor The actual pressure value can be obtained by measuring other calibrated pressure sensors in a calibration environment.

[0114] 2For every known P true , using deep learning models to predict sensor output P pred .

[0115] If the output value P of the sensor is used sensorThe error and calibration are calculated by comparing the actual pressure value under the calibration environment. Since the output value of the sensor may come from different types of sensors, although these sensors are calibrated, it is difficult to guarantee that the sensors themselves have some uncertainties and limitations. Therefore, using this method for calibration may lead to inaccurate results.

[0116] In this embodiment, the method of calculating the error by comparing the predicted value of the model with the actual pressure value under the calibration environment can correct the output deviation of the sensor through the model, thereby obtaining a more accurate result.

[0117] When the model training is completed, the input values ​​must be consistent with the data used during the original training. The input is the actual pressure value and the output value of the uncalibrated sensor to predict the sensor output value. The uncalibrated sensor output value usually contains errors, and these errors may have a negative impact on the model's prediction.

[0118] 3. Calculate the error between the predicted value and the actual value. For example, you can use the mean square error (MSE) as the error indicator:

[0119]

[0120] Where n represents the number of samples, and They represent the predicted value and actual value of the i-th sample respectively.

[0121] 4. Adjust the calibration parameters of the sensor according to the error index. A gradient descent algorithm or Adam (Adam is a gradient-based optimization algorithm) can be used. This embodiment uses the gradient descent algorithm to minimize the error index, thereby adjusting the calibration parameters of the sensor. Specifically, for the kth calibration parameter w k , can be updated using the following formula:

[0122]

[0123] Where α is the learning rate, represents the error index with respect to the kth calibration parameter w k The partial derivative of .

[0124] The parameters that need to be calibrated are the sensitivity and offset of the sensor. The sensitivity and offset obtained through calibration can be used to correct the output value of the sensor, thereby improving the precision and accuracy of the sensor. Calibration parameters w1, w2, ..., w k It can also be considered as a part of the zero offset and sensitivity. Specifically, w1 can be considered as the zero offset, and w2,…,w k can be considered as a sensitivity coefficient.

[0125] The calibration parameters are adjusted by minimizing the error metric, which is a measure of the difference between the predicted value and the actual value. The mean squared error (MSE) is usually used as the error metric.

[0126] During the calibration process, the error index is minimized, that is, the difference between the predicted value and the actual value is minimized. Therefore, the partial derivative of each calibration parameter can be solved by the error index to obtain the amount that each calibration parameter needs to be updated. Specifically, according to the chain rule, for each calibration parameter w k , its partial derivative can be expressed as:

[0127]

[0128] Where n represents the number of known actual values ​​and predicted values. It can be calculated by the back propagation algorithm Thus we get According to this value, the optimization method of the gradient descent algorithm can be used to update each calibration parameter w k value.

[0129] When calibrating a pressure sensor, the goal is to find a set of calibration parameters that minimizes the error metric. The gradient descent algorithm iteratively updates the calibration parameters, each update following the gradient of the current error function. This helps find the minimum of the error function. Compared to brute-force search or other conventional methods, the gradient descent algorithm is more efficient and can find the minimum of the error function more quickly, enabling rapid calibration of pressure sensors.

[0130] 5 Repeat steps 2-4 until the error indicator converges or the preset stopping condition is reached.

[0131] It should be noted that depending on the sensor, more than one calibration parameter may be required. For example, multiple parameters such as gain and offset may need to be adjusted. Therefore, in actual applications, the calibration parameters need to be selected and adjusted based on the specific sensor type and calibration requirements.

[0132] 6. Write the calibration parameters to the sensor to be calibrated. This allows the sensor to be calibrated during actual measurements. The specific method depends on the sensor model and interface. This is usually accomplished by programming or setting sensor registers. After writing the calibration parameters, the sensor can modify the original measurement value based on the calibration parameters, thereby improving measurement precision and accuracy.

[0133] Errors and drift are common problems in sensor testing. They can be affected by environmental factors such as temperature, humidity, and pressure, as well as by inherent sensor nonlinearity and lifespan loss. Deep learning algorithms, by learning from large amounts of sensor data and building sensor models, can analyze and model sensor operating characteristics and errors, enabling precise calibration and prediction.

[0134] The above method has high precision and stability: the neural network model established using the deep learning algorithm can effectively reduce the error and drift in sensor testing and improve the test accuracy and stability.

[0135] The above method is efficient and convenient: compared with traditional mechanical or electrical test equipment, this method can perform testing and calibration more quickly and conveniently, while also reducing costs and human resource consumption.

[0136] The above method is scalable and applicable: the method can be applied to different types of MEMS pressure sensors and has certain scalability and can be used for testing and calibration of other types of MEMS sensors.

[0137] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A pressure sensor calibration system based on deep learning, characterized in that: include: A pressure source, generating the set pressure required in the calibration environment; The data collector obtains the pressure generated by the pressure source as the actual pressure value, and obtains the output value of the pressure sensor to be calibrated corresponding to the actual pressure value; a processing unit configured to: obtain a predicted output value using a trained calibration model based on an actual pressure value in a calibration environment and an output value of the pressure sensor to be calibrated; determine an error between the actual pressure value in the calibration environment and the predicted output value; obtain a partial derivative of each calibration parameter with the goal of minimizing the error; and determine a required update amount for each calibration parameter; The parameter updating unit is configured to obtain updated calibration parameters based on the obtained update amount, input the updated calibration parameters into the pressure sensor to be calibrated, and complete the calibration.

2. The deep learning-based pressure sensor calibration system according to claim 1, wherein: The training data set of the calibration model includes obtaining the set pressure value of the pressure source as actual pressure value data under the calibration environment, and using the corresponding output data obtained by using a pre-calibrated pressure sensor as the training data set.

3. The deep learning-based pressure sensor calibration system according to claim 2, wherein: During the training of the calibration model, the output data obtained from the pre-calibrated pressure sensor is preprocessed and features are extracted.

4. The pressure sensor calibration method based on deep learning according to claim 3, characterized in that: The preprocessing includes at least one of de-averaging, normalization and filtering.

5. The pressure sensor calibration method based on deep learning according to claim 3, characterized in that: The extracted features include one or more of the average value, standard deviation, maximum value, minimum value, slope, peak value and energy spectral density in the output data.

6. The pressure sensor calibration method based on deep learning according to claim 3, characterized in that: During the training process of the calibration model, the extracted features are sequentially processed through the convolution layer, pooling layer and fully connected layer to obtain the predicted pressure sensor output value P pred , with P pred The goal is to complete the training of the calibration model by minimizing the error between the actual pressure value and the actual pressure value.

7. The pressure sensor calibration method based on deep learning according to claim 1, characterized in that: After the calibration model is trained, the actual pressure value and the corresponding output value of the pressure sensor to be calibrated are obtained in the calibration environment, and the calibration model is used to obtain the predicted output value based on the output value of the pressure sensor to be calibrated.

8. The pressure sensor calibration method based on deep learning according to claim 1, characterized in that: After the calibration model is trained, the mean square error between the actual pressure value and the predicted output value under the calibration environment is obtained. With the goal of minimizing the mean square error, the partial derivative of each calibration parameter is solved to obtain the required update amount for each calibration parameter.

9. The pressure sensor calibration method based on deep learning according to claim 1, characterized in that: The processing unit has a data interface that is communicatively connected to the pressure sensor to be calibrated, and writes the updated calibration parameters into the pressure sensor to be calibrated through the data interface to complete the calibration.

10. A method for calibrating a pressure sensor based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Obtain the actual pressure value under the calibration environment and the output value of the pressure sensor to be calibrated, and obtain the predicted output value based on the trained calibration model; Determine the error between the actual pressure value under the calibration environment and the predicted output value, and with the goal of minimizing the error, solve the partial derivative of each calibration parameter to obtain the required update amount for each calibration parameter; An updated calibration parameter is obtained based on the updated amount and input into the pressure sensor to be calibrated to complete the calibration.

Images