An adaptive front-end sensor data collection method based on AI detection

Abstract

The present application belongs to the field of electronic information, and relates to a data acquisition technology of a sensor, and specifically provides a self-adaptive front-end sensor data acquisition method based on AI detection, which is used to solve the problem of poor universality of an existing data acquisition system for low-resistance value sensors and high-resistance value sensors. In the self-adaptive front-end sensor data acquisition method based on AI detection provided by the present application, a random forest classifier with the advantage of light weight is creatively used as an AI model, and feature parameters are respectively proposed in combination with the working characteristics of low-resistance value sensors and high-resistance value sensors, and the prediction accuracy of the AI model is further improved through multi-feature extraction, so as to realize accurate prediction of the sensor type, and realize adaptive switching of the front end according to the prediction result, so as to complete universal and high-precision data acquisition of low-resistance value sensors and high-resistance value sensors based on a universal data acquisition board.

Description

Technical Field

[0001] This invention belongs to the field of electronic information and relates to sensor data acquisition technology, specifically providing an adaptive front-end sensor data acquisition method based on AI detection. Background Technology

[0002] Currently, sensor data acquisition systems mostly adopt an integrated "acquisition circuit-sensor" design. However, due to the significant differences in base resistance values ​​between different sensors—for example, low-resistance sensors are typically below 1kΩ, while high-resistance sensors can reach several gigaΩ—the current output range drops significantly from the mA level to the nA level. Therefore, dedicated acquisition boards need to be designed separately for sensors with different resistance specifications. Acquisition boards adapted for low-resistance sensors (below 1kΩ) cannot be directly used for high-resistance sensors (several gigaΩ) because the bias voltage and signal amplification factor are fixed. For high-resistance sensors, the extremely weak current requires necessary amplification and noise isolation measures. Therefore, to improve the versatility and deployment efficiency of data acquisition systems, this invention proposes an adaptive front-end sensor data acquisition method based on AI detection. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based adaptive front-end sensor data acquisition method to solve the problem of poor versatility of existing data acquisition systems for both low-resistance and high-resistance sensors.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] An adaptive front-end sensor data acquisition method based on AI detection, characterized by the following steps:

[0006] Step 1: Initialize data acquisition using the data acquisition module. The data acquisition module acquires the sensor's static voltage signal U1 with low-impedance adapter parameters, and simultaneously acquires the bias voltage U. bias ;

[0007] Step 2: Based on the static voltage signal U1 and the bias voltage U bias Calculate the initial resistance value and use the initial resistance value as a criterion to predict the sensor type, including low-resistance sensors and suspected high-resistance sensors;

[0008] Step 3: Extract multiple features based on the predicted sensor category;

[0009] If it is initially determined to be a low-resistance sensor, then the fluctuation variance σ of the static voltage signal U1 is extracted as the characteristic parameter φ.

[0010] If it is initially determined to be a suspected high-impedance sensor, switch the data acquisition module to the high-impedance adaptation parameter, acquire the static voltage signal U2 of the sensor again, and extract the rise time τ of the static voltage signal U2 as the characteristic parameter φ.

[0011] Step 4: Input the initial resistance value R0 and the characteristic parameter φ as input data pairs into the AI ​​resistance recognition model. The AI ​​resistance recognition model will output a sensor type identification label, where label 0 indicates a low-resistance sensor and label 1 indicates a high-resistance sensor.

[0012] Step 5: Identify the sensor type label based on the output of the AI ​​resistance identification model, and the front-end switching module automatically switches to the appropriate front-end. For low-resistance sensors, a low-resistance adaptive front-end is used, and for high-resistance sensors, a high-resistance adaptive front-end is used, thereby completing the adaptive acquisition of sensor data.

[0013] Furthermore, in step 2, the initial resistance value is calculated as follows: R0 = U1 / I bias I bias = U bias / R C R0 is the initial resistance value.

[0014] Furthermore, in step 2, the process of predicting the sensor category based on the initial resistance value is as follows: set the resistance threshold R. th If the initial resistance R0 < 10kΩ, it is initially determined to be a low-resistance sensor; if the initial resistance R0 ≥ 10kΩ, it is initially determined to be a suspected high-resistance sensor.

[0015] Furthermore, in step 4, the AI ​​resistance recognition model uses a random forest classifier, whose input features are the initial resistance value R0 and the feature parameter φ, and whose output is a binary classification label.

[0016] Furthermore, in step 4, the training process of the random forest classifier is as follows:

[0017] A training sample set was constructed, including a low-resistivity sample set and a high-resistivity sample set. For the low-resistivity sample set, 100 sets of sample data of low-resistivity sensors with base resistance values ​​in the range of 100Ω to 10kΩ were collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%~70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=σ) and the label being 0. For the high-resistivity sample set, 100 sets of sample data of high-resistivity sensors with base resistance values ​​in the range of 100MΩ to 10GΩ were collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%~70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=τ) and the label being 1.

[0018] Training was completed using the Python sklearn library based on the above training sample set.

[0019] Based on the above technical solution, the beneficial effects of the present invention are as follows:

[0020] This invention provides an adaptive front-end sensor data acquisition method based on AI detection. It creatively proposes to use a lightweight random forest classifier as the AI ​​model, and proposes feature parameters based on the working characteristics of low and high resistance sensors. Through multi-feature extraction, the prediction accuracy of the AI ​​model is further improved, thereby achieving accurate prediction of sensor type. Based on the prediction results, the method can adaptively switch the front end, thus completing universal and high-precision data acquisition of low and high resistance sensors based on a general data acquisition board.

[0021] More specifically, the present invention has the following advantages:

[0022] 1) Fully automated: No manual intervention is required. The data acquisition system can complete resistance identification and front-end switching within 100ms after power-on, which is suitable for unattended scenarios, such as automatic monitoring stations and batch sensor detection.

[0023] 2) High compatibility: It can be adapted to low-resistance sensors of 100Ω~10kΩ and high-resistance sensors of 100MΩ~10GΩ. When adding a new sensor type, only the sample library needs to be supplemented, and no hardware modification is required.

[0024] 3) High reliability: The integration of initial resistance and characteristic parameters (variance / rise time) for multi-feature recognition improves the AI ​​model's resistance to temperature and humidity interference by 40%, and the recognition accuracy is ≥98%;

[0025] 4) Lightweight and easy to deploy: The AI ​​model size is ≤100KB and the inference time is ≤50ms. It can be directly integrated into the embedded MCU without the need for additional computing chips, which greatly reduces hardware costs. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the AI-based adaptive front-end sensor data acquisition method of the present invention.

[0027] Figure 2 This is a schematic diagram of the front-end switching module in this invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0029] This embodiment provides an AI-based adaptive front-end sensor data acquisition method. It utilizes a universal data acquisition board to achieve universal, high-precision data acquisition from sensors with different resistance values, such as low and high resistance. The process is as follows: Figure 1 As shown, the specific steps include:

[0030] Step 1: Initialization and Data Acquisition;

[0031] The data acquisition module, by default, uses low-impedance adapter parameters (sampling resistor R1, bias current I1) to acquire the sensor's static voltage signal U1 (50 samples per second, averaged, noise suppressed), and simultaneously acquires the bias voltage U. bias ;

[0032] Step 2: Initial resistance calculation and preliminary judgment;

[0033] The AI ​​unit calculates the initial resistance value according to the following formula: R0 = U1 / I bias I bias = U bias / R C R C The value of the sampling resistor is given; the sampling resistor is connected in series with the sensor.

[0034] If R0 < 10kΩ, it is preliminarily determined to be a low-resistance sensor;

[0035] If R0≥10kΩ, it is initially determined to be a suspected high-resistance sensor, triggering the acquisition of high-resistance adaptation parameters;

[0036] Step 3: Multi-feature extraction;

[0037] If it is initially determined to be a low-impedance sensor, the fluctuation variance σ (unit: mV) of the static voltage signal U1 is extracted as the characteristic parameter φ. Under normal circumstances, the fluctuation variance σ1 is used to reflect the signal stability of the low-impedance sensor. When σ≤5mV, it is usually determined to be an effective low-impedance feature.

[0038] If it is initially determined to be a suspected high-impedance sensor, switch the data acquisition unit to the high-impedance adaptation parameters (sampling resistor R2, bias current I2), acquire the sensor's static voltage signal U2, and extract the rise time τ (unit: μs) of the static voltage signal U2 as the characteristic parameter φ; under normal circumstances, the voltage of a high-impedance sensor is slow to build up, and when τ≥100μs, it is usually determined to be a valid high-impedance characteristic;

[0039] Step 4: AI resistance recognition model inference;

[0040] The initial resistance value R0 and the characteristic parameter φ are used as input data pairs and input to the AI ​​resistance recognition model. The AI ​​resistance recognition model outputs a sensor type identification label, with label 0 indicating a low-resistance sensor and label 1 indicating a high-resistance sensor.

[0041] Specifically, the AI ​​resistance recognition model uses a random forest classifier, which has the advantage of being lightweight and suitable for embedded MCU deployment; its input features are the initial resistance value R0 and the feature parameter φ, and the output is a binary classification label;

[0042] During the training process, the AI ​​resistance recognition model first constructs a training sample set, including a low-resistance sample set and a high-resistance sample set. For the low-resistance sample set, 100 sets of sample data of low-resistance sensors with base resistance in the range of 100Ω to 10kΩ are collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%-70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=σ) and the label being 0. For the high-resistance sample set, 100 sets of sample data of high-resistance sensors with base resistance in the range of 100MΩ to 10GΩ are collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%-70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=τ) and the label being 1.

[0043] Based on the above training sample set, the AI ​​resistance recognition model is trained using the Python sklearn library. After training, it is converted to .tflite format (size ≤100KB) to adapt to embedded storage. Furthermore, in the actual model deployment process, it is parsed and run by an embedded MCU (such as STM32F103 series), with a single inference time ≤50ms, which does not affect the real-time performance of data acquisition.

[0044] Step 5: The front-end switching module adapts to the switching process;

[0045] The front-end switching module automatically switches to adapt to the front-end based on the sensor type identification label output by the AI ​​resistance recognition model, such as... Figure 2 As shown, this completes the adaptive acquisition of sensor data.

[0046] The beneficial effects of the present invention will be described in detail below with reference to specific embodiments.

[0047] In this embodiment, the hardware foundation is as follows: an STM32 series chip is used as the main control MCU, integrating a data acquisition module, an AI resistance recognition model running module, a front-end switching module, and a storage module; AI resistance recognition model loading: the trained .tflite model is stored on an SD card, and the MCU reads the model into Flash after power-on and initializes the model inference environment; parameter configuration: preset low resistance adaptation parameters (R1=100Ω, I1=50μA), high resistance adaptation parameters (R2=1MΩ, I2=0.1μA) and front-end matching rules (label 0: low resistance adaptation front-end, label 1: high resistance adaptation front-end).

[0048] Based on the aforementioned AI-based adaptive front-end sensor data acquisition method, a universal and high-precision data acquisition system is implemented for sensors with different resistance values, such as low and high resistance, using a general-purpose data acquisition board. Through AI multi-feature recognition logic, it can cover two core ranges: 100Ω~10kΩ low resistance and 100MΩ~10GΩ high resistance. The acquisition parameters (sampling resistance, bias current) can be switched via software configuration without modifying the motherboard hardware. Furthermore, when adding a new sensor resistance range (such as medium resistance 10kΩ~100MΩ), only one sample set needs to be added and the model trained, demonstrating strong scalability.

[0049] More importantly, this invention proposes specific feature parameters φ for low-resistance and high-resistance sensors, and dual feature cross-validation avoids misjudgment. In addition, random noise can be suppressed by averaging during the sample acquisition stage. The model training covers temperature interference scenarios of 20~40℃ and humidity interference scenarios of 30%~70%RH, with strong environmental adaptability. Finally, the recognition accuracy can be improved to ≥98%, the resistance to temperature and humidity interference can be improved by 40%, and the front-end selection error rate can be reduced by more than 90%.

[0050] In addition, this invention uses a random forest classifier, which has a smaller parameter size (model size ≤100KB after training) compared to deep learning models such as CNN and LSTM, and does not require a lot of computing power. After the model is converted to .tflite format, it can be directly integrated into mid-to-low-end MCUs of the STM32 series, with a single inference time ≤50ms. No additional GPU or NPU computing power chips are required, only an additional 4GB SD card is needed to store the model.

[0051] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. An adaptive front-end sensor data acquisition method based on AI detection, characterized in that, Includes the following steps: Step 1: Initialize data acquisition using the data acquisition module. The data acquisition module acquires the sensor's static voltage signal U1 with low-impedance adapter parameters, and simultaneously acquires the bias voltage U. bias ; Step 2: Based on the static voltage signal U1 and the bias voltage U bias Calculate the initial resistance value and use the initial resistance value as a criterion to predict the sensor type, including low-resistance sensors and suspected high-resistance sensors; Step 3: Extract multiple features based on the predicted sensor category; If it is initially determined to be a low-resistance sensor, then the fluctuation variance σ of the static voltage signal U1 is extracted as the characteristic parameter φ. If it is initially determined to be a suspected high-impedance sensor, switch the data acquisition module to the high-impedance adaptation parameter, acquire the static voltage signal U2 of the sensor again, and extract the rise time τ of the static voltage signal U2 as the characteristic parameter φ. Step 4: Input the initial resistance value R0 and the characteristic parameter φ as input data pairs into the AI ​​resistance recognition model. The AI ​​resistance recognition model will output a sensor type identification label. Label 0 indicates a low resistance sensor and label 1 indicates a high resistance sensor. Step 5: Identify the sensor type label based on the output of the AI ​​resistance recognition model, and the front-end switching module will automatically switch to adapt to the front end to complete the adaptive acquisition of sensor data.

2. The adaptive front-end sensor data acquisition method based on AI detection according to claim 1, characterized in that, In step 2, the initial resistance is calculated as follows: R0 = U1 / I bias I bias = U bias / R C R0 is the initial resistance value.

3. The adaptive front-end sensor data acquisition method based on AI detection according to claim 1, characterized in that, In step 2, the process of predicting the sensor type based on the initial resistance value is as follows: Set the resistance threshold R. th If the initial resistance R0 < 10kΩ, it is initially determined to be a low-resistance sensor; if the initial resistance R0 ≥ 10kΩ, it is initially determined to be a suspected high-resistance sensor.

4. The adaptive front-end sensor data acquisition method based on AI detection according to claim 1, characterized in that, In step 4, the AI ​​resistance recognition model uses a random forest classifier, whose input features are the initial resistance value R0 and the feature parameter φ, and whose output is a binary classification label.

5. The adaptive front-end sensor data acquisition method based on AI detection according to claim 4, characterized in that, In step 4, the training process of the random forest classifier is as follows: A training sample set was constructed, including a low-resistivity sample set and a high-resistivity sample set. For the low-resistivity sample set, 100 sets of sample data of low-resistivity sensors with base resistance values ​​in the range of 100Ω to 10kΩ were collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%~70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=σ) and the label being 0. For the high-resistivity sample set, 100 sets of sample data of high-resistivity sensors with base resistance values ​​in the range of 100MΩ to 10GΩ were collected under an environment with a temperature of 20~40℃ and an air relative humidity (RH) of 30%~70%. The sample data includes input data pairs and labels, with the input data pair being (R0, φ=τ) and the label being 1. Training was completed using the Python sklearn library based on the above training sample set.

Images