Automatic compensation and self-diagnosis system for SOI (Silicon On Insulator) silicon piezoresistive sensing pressure transmitter

Abstract

The invention discloses an SOI (Silicon On Insulator) silicon piezoresistive sensing pressure transmitter automatic compensation and self-diagnosis system, which is characterized by comprising a silicon piezoresistive pressure sensor and a sensor automatic compensation and self-diagnosis system, and the sensor automatic compensation and self-diagnosis system automatically compensates the silicon piezoresistive pressure sensor through software. The high-precision intelligent pressure transmitter is researched and developed by adopting an AI self-learning compensation learning algorithm aiming at the SOI silicon piezoresistive pressure sensor with the independent intellectual property right, and the high-precision intelligent pressure transmitter has the functions of wide range, high precision, high reliability and the like, and is suitable for the SOI silicon piezoresistive pressure sensor with the independent intellectual property right. Intelligent sensing autonomous controllable localization substitution is achieved, and large-scale popularization and application in the industries of petrochemical engineering, new energy, industrial automobiles and the like are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of SOI silicon piezoresistive sensing, and in particular relates to an automatic compensation and self-diagnosis system for a SOI silicon piezoresistive sensing pressure transmitter. Background Art

[0002] Domestic research on intelligent pressure transmitters and their core component, high-temperature pressure sensors, is primarily conducted in Beijing, Shanghai, and Shijiazhuang, primarily within research institutions such as the Chinese Academy of Sciences, the Institute of Electronic Aerospace, and the Institute of Ordnance, as well as key domestic science and engineering universities. Most existing transmitters in China use capacitive sensors as their detection core, which exhibit low accuracy (less than 0.075% FS), poor anti-interference capabilities, and low static pressure tolerance (generally less than 10 MPa). Furthermore, capacitive sensors have a very low overload capacity, resulting in low measurement reliability. Improving transmitter accuracy has long been a technological bottleneck in China. Due to foreign restrictions on advanced sensor technology, the development of sensor and transmitter-related technologies has been significantly restricted, and my country relies heavily on imported finished products. Therefore, high-precision intelligent pressure transmitters have become a hot topic and a key research focus. Summary of the Invention

[0003] To achieve the above object, the technical solution of the present invention is as follows: an automatic compensation and self-diagnosis system for an SOI silicon piezoresistive sensor pressure transmitter, the system comprising a silicon piezoresistive pressure sensor and a sensor automatic compensation and self-diagnosis system; The sensor automatic compensation and self-diagnosis system automatically compensates the silicon piezoresistive pressure sensor through software, and performs self-test on the silicon piezoresistive pressure sensor after power-on to check whether the sensor is normal and make a judgment; The sensor automatic compensation and self-diagnosis system includes: The signal conditioning module is used to perform preliminary amplification, filtering, and common-mode suppression on the original weak differential voltage signal output by the sensor. It uses a common-mode filtering structure, connecting C4, C6, and C7 in parallel through the signal line to form a multi-stage filter, covering broadband noise and ensuring the purity of the input signal. At the same time, the NSA2860 communicates with the CPU through the four-wire SPI protocol and uploads the measurement data to the CPU. The self-diagnosis execution module includes a watchdog circuit, a key signal monitoring comparator, a software fault handling program and an alarm output circuit. Preferably, the system adapts to the nonlinear and time-varying characteristics of the sensor signal by dynamically adjusting the filter parameters through the Kalman filter algorithm, comprising the following steps: Step 1: In the MCU software, treat the output of the pressure sensor as a dynamic system, define the state vector X_k, and define the state transition equation X_k = F * X_{k-1} + W_k, where F is the state transition matrix and W_k is the process noise; Step 2: Define the observation vector Z_k = H * X_k + V_k, where H is the observation matrix and V_k is the observation noise; Step 3: Kalman filter iteration, executed in real time in the MCU: Predicted state: X_k|k-1 = F * X_k-1|k-1; Prediction covariance: P_k|k-1 = F * P_k-1|k-1 * F^T + Q, where Q is the process noise covariance matrix; Calculate the Kalman gain: K_k = P_k|k-1 * H^T * (H * P_k|k-1 * H^T + R)^-1, where R is the observation noise covariance matrix; Update the state estimate: X_k|k = X_k|k-1 + K_k * (Z_k - H * X_k|k-1), where Z_k is the raw pressure observation read by the ADC; Update the covariance estimate: P_k|k = (I - K_k * H) * P_k|k-1; Step 4: The pressure estimate P_k in X_k|k is the output that is closer to the actual pressure after filtering and noise reduction. During operation, the system dynamically adjusts the parameters of the noise covariance matrix Q and R by monitoring the statistical characteristics of the observation residual (Z_k - H * X_k|k-1) or using other sensor information, so that the filter can adapt to changes in the environment and sensor noise characteristics. Step 5: Kalman filtering is performed in the digital domain of the MCU. Its input Z_k is the digital signal converted by the ADC module. To accurately collect the original weak analog signal, it is necessary to rely on the high-gain, low-noise instrument amplifier circuit of the signal conditioning module and the high-resolution, low-noise sampling of the ADC module.

[0004] Preferably, the sensor automatic compensation and self-diagnosis system uses a conditioning circuit to compensate for it. The conditioning circuit is built based on the NSA2860 chip to achieve temperature drift compensation of the sensor output; The conditioning circuit includes connecting the positive and negative output terminals of the silicon piezoresistive sensor to the dedicated sensor input pins of the NSA2860 chip. The NSA2860 provides a programmable excitation current source or voltage source connected to the excitation terminal of the sensor bridge circuit. The chip integrates a high-precision instrumentation amplifier, a digital-to-analog converter for zero-point calibration and temperature compensation, and a temperature sensor.

[0005] Preferably, the sensor automatic compensation and self-diagnosis system, including time drift, comprises the following steps: Step 1: Data collection and preprocessing: During normal operation of the transmitter, the MCU periodically collects and records a set of data, and the output vector is the target; Step 2: Neural Network Architecture and Training: Choose a long short-term memory network with 5 nodes in the input layer, 1-2 hidden layers with 8-16 nodes per layer, using the ReLU activation function, and a linear activation function for the 2 nodes in the output layer. Use historical data sets to train the network outside the MCU to obtain initial weights and biases (W, b). The training goal is to minimize the mean square error between the predicted ΔV_offset and the true value or estimated value. Store the trained network parameters (W, b) in non-volatile memory. Step 3: In the MCU, a simplified recursive least squares and extended Kalman filter algorithm are run simultaneously as a system identifier. The identifier estimates and updates the parameters of the system model online based on the input and output data collected in real time. Step 4: regularly input the collected input vectors into the neural network loaded with initial parameters to predict the current drift ΔV_offset_pred and ΔSensitivity_pred, input the same input vectors into the system identifier, and the identifier outputs its estimate of the current drift ΔV_offset_iden and ΔSensitivity_iden; fuse or compare the neural network prediction value and the identifier estimate value, and combine the confidence assessment to generate the final recognized drift ΔV_offset_final and ΔSensitivity_final; according to ΔV_offset_pred, the final drift value ΔV_offset_final and ΔSensitivity_final are generated. fset_final updates the zero-point compensation parameters and the sensitivity compensation parameters based on ΔSensitivity_final. The final identified drift values ​​ΔV_offset_final and ΔSensitivity_final are used as new training samples. Combined with the corresponding input vectors, the neural network is incrementally learned online to update the network weights (W, b). The updated parameters are stored in non-volatile memory to enable the network to adapt to the sensor's unique long-term drift characteristics. The system identifier parameters are also continuously updated. A permissible drift threshold is set; if the identified drift exceeds the threshold, a maintenance alarm may be triggered.

[0006] Preferably, during the signal acquisition process, from the time the trigger signal occurs until the acquisition is triggered and after the acquisition is completed, the input signal is zero and the output signal is non-zero. The collected output data exists in the form of random noise. The system adopts a zero drift compensation algorithm based on a back-propagation neural network, which can achieve high-precision fitting through a small amount of sample training. It also supports online learning functions and can adapt to slow changes in sensor characteristics, including: Step 1: Establish an accurate mapping relationship between the sensor output and the actual pressure P and temperature T: P = f(V_diff, T). The function f is usually nonlinear. Step 2: The neural network model is implemented in the MCU. The input layer includes two nodes: the raw sensor differential voltage V_diff and the temperature sensor reading T; the hidden layer is one or two layers, the number of nodes is optional, and the activation function uses the Sigmoid function; the output layer includes one node for the compensated pressure estimate P_comp, using a linear activation function; Step 3, training, is done offline. This involves measuring the raw sensor output V_diff_ij at multiple temperatures (T1, T2, ..., Tm) and pressures (P1, P2, ..., Pn) during the sensor calibration phase. This generates a training dataset: input samples X = [V_diff_ij, Ti], target output Y = Pj. The network is trained on a PC using a backpropagation algorithm to minimize the mean squared error between the predicted pressure P_comp and the actual applied pressure Pj. The trained network parameters (weights W, bias b) are stored in the transmitter's non-volatile memory. Step 4: Online compensation, executed by the MCU during runtime, includes the signal conditioning module outputting the original V_diff, the temperature sensor module outputting the current T, the MCU reading V_diff and T as the input vector [V_diff, T], the MCU loading the neural network parameters, and performing forward propagation calculations, including the hidden layer input H_in = W1 * [V_diff; T] + b1, the hidden layer output H_out = Sigmoid(H_in), and the output layer P_comp = W2 * H_out + b2. The calculated P_comp is the final pressure measurement value after comprehensive compensation for zero point, sensitivity, nonlinearity, and temperature, and is used for output.

[0007] Preferably, the online calibration step includes: Step 1, online calibration step, uses historical data and real-time data of the sensor in the working environment to automatically estimate and update the compensation parameters through the algorithm; Step 2: Automatic calibration steps: The zero-point calibration process involves the MCU receiving a "zero-point calibration" command from the host computer via the communication interface. The MCU ensures that the current pressure is a known zero pressure. The MCU controls the ADC to collect the current sensor output (V_diff_zero) and temperature (T). The MCU updates the zero-point offset parameters in the compensation algorithm and stores the new zero-point parameters in non-volatile memory. The full-scale / span calibration process involves the MCU receiving a "full-scale calibration" command and the target full-scale pressure value (P_fs) from the host computer via the communication interface. The MCU ensures that the currently applied pressure is stable at P_fs. The MCU collects the current sensor output (V_diff_fs) and temperature (T), calculates the current sensitivity (Sensitivity_current) = (V_diff_fs - V_diff_zero) / P_fs, compares it with the stored nominal sensitivity (Sensitivity_nominal), calculates the gain compensation coefficient (Gain_comp) = Sensitivity_nominal / Sensitivity_current, and updates the gain parameters in the compensation algorithm and stores the new gain parameters in non-volatile memory. The linear calibration step involves solving the problem in the calibration phase using a high-order polynomial or neural network model, and updating the zero and full-scale points during automatic calibration.

[0008] Preferably, the communication software supports multiple communication protocols used by the pressure sensor, including hardware interfaces and software interfaces. When parsing data, the software has the ability to parse data from the pressure sensor, including analog signals and digital signals. For analog signals, the software can accurately read and convert them into corresponding pressure values; for digital signals, the software can correctly parse and display them. The HART communication module communicates with the microcontroller's UART serial port via the RXD and TXD pins. It uses a 3.6864MHz passive crystal oscillator, capacitors C27 and C28, and resistors R16 and R15 as an external filter to receive the HART signal from the 4-20mA loop and transmit it to the ADC_IP pin. The CD pin is the carrier monitor port. When the CD pin is high, the ADC_IP signal amplitude is greater than 120mA, indicating that the signal is a valid carrier signal. In modulation mode, RTS is low, and the digital signal sent by the microcontroller enters the FSK modulation module. The modulated HART signal is then output to the loop via the HART_OUT pin.

[0009] Preferably, the self-diagnosis function is based on real-time monitoring and analysis of its key parameters. The diagnostic system integrated in the sensor collects this data in real time and compares it with a preset threshold or standard to determine the performance status of the sensor. Specifically, the following steps are included: Step 1: Real-time monitoring of key parameters, including the raw bridge voltage V_diff obtained from the signal conditioning module or ADC module, the compensated pressure value P_comp, the temperature value T, the sensor excitation voltage / current V_bridge / I_bridge, and circuit parameters. The circuit parameters include the power supply voltage V_cc monitored by the MCU internal ADC or monitoring chip, the ADC reference voltage V_ref, the watchdog timer status, and the memory checksum; including operation status monitoring, including software task execution time, stack usage, and communication error counter; Step 2: Execution of diagnostic algorithms, including threshold comparison, zero drift exceeding the limit, full scale exceeding / undershooting, over-range, power supply exceeding the limit, temperature exceeding the limit, communication failure, watchdog reset, pattern recognition / trend analysis, signal noise anomaly, signal glitch detection, temperature-pressure relationship anomaly, and historical trend analysis; Step 3, fault determination and isolation, the diagnostic algorithm compares the monitoring results with the preset multi-level thresholds, and the MCU executes according to the fault level; Step 4: Information reporting, including uploading the diagnostic results to the host computer / control system in real time or periodically through the communication interface module. Locally, an audible / visual indication is provided through an alarm output.

[0010] Compared with the existing technology, the beneficial effects of the present invention are: the present invention targets SOI silicon piezoresistive pressure sensors with independent intellectual property rights, and adopts AI self-learning compensation learning algorithm to develop high-precision intelligent pressure transmitters with functions such as wide range, high precision, and high reliability, realizing independent and controllable domestic substitution of intelligent sensing, and promoting large-scale application in petrochemical, new energy, industrial automobile and other industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a schematic diagram of the development of the SOI silicon piezoresistive high-precision intelligent pressure transmitter described in the present invention; Figure 2 This is a schematic diagram of the adaptive Kalman filter algorithm process described in the present invention. DETAILED DESCRIPTION

[0012] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0013] Example: Figures 1 to 2 As shown, an automatic compensation and self-diagnosis system for a SOI silicon piezoresistive pressure transmitter includes a silicon piezoresistive pressure sensor and a sensor automatic compensation and self-diagnosis system. The sensor automatic compensation and self-diagnosis system automatically compensates the silicon piezoresistive pressure sensor through software, and performs a self-test on the silicon piezoresistive pressure sensor after power is turned on to check whether the sensor is normal and make a judgment. The sensor automatic compensation and self-diagnosis system specifically includes the following functional modules: (1) Microcontroller module (MCU): The MCU module uses GigaDevice GD32L233, which uses GigaDevice's independently developed 40nm ultra-low power (ULP) manufacturing technology. From process technology to chip design, it is domestically produced, and its system power consumption is low (the power consumption in full-power mode is only 66μA / MHz, and the deep sleep current is as low as 2μA), making it suitable for the design of precision instruments that are sensitive to power consumption.

[0014] (2) Analog / Digital and Digital / Analog Conversion Modules (ADC and DAC): The dedicated NSA2860 chip has a built-in ADC that converts the conditioned analog signal into a high-resolution digital signal. It uses the Arnold AD5700-1, a dedicated 4-20mA loop driver, which is HART-compatible and offers low power consumption and high precision.

[0015] (3) Signal conditioning module: This module is primarily based on the dedicated NSA2860 chip, responsible for the initial amplification, filtering, and common-mode suppression of the original weak differential voltage signal output by the sensor. A common-mode filter structure is used, with C4, C6, and C7 connected in parallel through the signal line to form a multi-stage filter that covers broadband noise (100nF filters low frequencies, 10nF filters high frequencies), ensuring the purity of the input signal. At the same time, the NSA2860 communicates with the CPU via the four-wire SPI protocol, uploading the measurement data to the CPU.

[0016] (4) Power Management Module: The power module uses a bridge rectifier structure to effectively prevent reverse connection and has additional TVS static and surge protection. The 24V DC power passes through the D1-D4 voltage regulator diodes and is then input into the TPS709B50 and TPS709B33 voltage dropper chips. The chips have internal overheat protection and undervoltage lockout functions to effectively protect the back-end circuits. The circuit output voltage can output the corresponding voltage according to the setting, providing a stable, low-noise power supply for each module of the system.

[0017] (5) Communication interface module: including HART protocol based on 4-20mA and RS-485 Modbus. The HART communication module uses the AD5700-1 chip from Arnold to implement a two-wire HART modulator and demodulator, and to exchange data with the host computer or control system.

[0018] (6) Self-diagnosis execution module: including watchdog circuit, key signal monitoring comparator, software fault handling program and alarm output circuit (such as driving LED, relay or sending fault code). Inter-module Interaction: After power-up, the MCU first reads the preset parameters from non-volatile memory through the communication interface or the built-in self-test (BIST) routine, and controls the signal conditioning module and ADC module to perform an initial acquisition of the sensor output. The acquired raw data (pressure and temperature signals) is fed into the MCU. The MCU then executes a self-diagnostic algorithm, comparing the current zero-point output with stored historical zero-point data / thresholds, checking the signal's full-scale range, calculating the signal-to-noise ratio, and verifying the plausibility of the temperature sensor reading. It then determines the sensor status (normal, warning, or fault), records the result in memory, and reports it via the communication interface. If the sensor status is normal, the MCU automatically compensates the raw digital pressure signal using a predefined compensation algorithm (such as polynomial fitting, table interpolation, or neural network model) based on the real-time acquired temperature signal and the compensation coefficients (zero-point temperature drift coefficient, sensitivity temperature drift coefficient, and nonlinear coefficient) stored in non-volatile memory, obtaining an accurate pressure measurement. The compensated pressure value is then output through the communication interface module. During operation, the self-diagnostic execution module continuously monitors the status of key hardware (power supply voltage, clock, memory checksum) and software (task execution time, stack overflow), and collaborates with the diagnostic logic in the MCU to achieve real-time fault detection and isolation / alarming. A time drift compensation module (such as a neural network prediction module) runs in the MCU background, periodically or based on trigger conditions (such as accumulated temperature change or runtime), updating compensation coefficients and storing them in non-volatile memory.

[0019] Furthermore, innovative adaptive filters such as the Kalman filter algorithm with self-adjusting function are developed to achieve feature extraction of weak signals, amplification and accurate acquisition of tiny signals, including the use of Kalman filtering to utilize linear system state equations and optimally estimate the system state through system input and output observation data. The Kalman filter algorithm adapts to the nonlinear and time-varying characteristics of sensor signals by dynamically adjusting the filter parameters, significantly improving the signal-to-noise ratio of the signal. The system also integrates a wavelet transform algorithm to further eliminate high-frequency noise and ensure the integrity and accuracy of weak signals. Specific technical means (Kalman filter implementation): (1) State-space modeling: In the MCU software, the pressure sensor output is treated as a dynamic system. Define the state vector X_k (for example, containing the current true pressure value P_k and its rate of change dP / dt_k). Define the state transition equation X_k = F * X_{k-1} + W_k, where F is the state transition matrix (modeling the dynamic characteristics of pressure changes, such as assuming uniform velocity or uniform acceleration) and W_k is the process noise (modeling model uncertainty).

[0020] (2) Observation model: Define the observation vector Z_k = H * X_k + V_k, where H is the observation matrix (representing the directly observed pressure value) and V_k is the observation noise (mainly due to sensor noise and circuit noise).

[0021] (3) Kalman filter iteration (executed in real time in MCU): ① Prediction: Predicted state: X_k|k-1 = F * X_k-1|k-1; Prediction covariance: P_k|k-1 = F * P_k-1|k-1 * F^T + Q (Q is the process noise covariance matrix); ②Update: Calculate the Kalman gain: K_k = P_k|k-1 * H^T * (H * P_k|k-1 * H^T + R)^-1 (R is the observation noise covariance matrix); Update the state estimate: X_k|k = X_k|k-1 + K_k * (Z_k - H * X_k|k-1) (Z_k is the raw pressure observation read by the ADC); Update the covariance estimate: P_k|k = (I - K_k * H) * P_k|k-1.

[0022] (4) Output: The pressure estimate P_k in X_k|k is the output that is closer to the actual pressure after filtering and noise reduction.

[0023] Self-tuning: During operation, the system dynamically adjusts the parameters of the noise covariance matrices Q and R by monitoring the statistical characteristics (such as variance) of the observation residual (Z_k - H * X_k|k-1) or leveraging other sensor information (such as the temperature change rate). (For example, an increase in Q indicates increased model uncertainty, and an increase in R indicates increased observation noise.) This allows the filter to adapt to environmental changes (such as shock, vibration, and temperature transients) and changes in sensor noise characteristics, maintaining optimal estimation performance. This adjustment can be implemented in the MCU using a preset rule table or a simple adaptive algorithm (such as adjustment based on the residual variance). (5) Accurate acquisition of tiny signals: Kalman filtering is primarily performed in the digital domain of the MCU, with its input Z_k being the digital signal converted by the ADC module. Accurate acquisition of the original weak analog signal relies on the high-gain, low-noise instrumentation amplifier circuit (e.g., programmable gain) of the signal conditioning module and the high-resolution, low-noise sampling of the ADC module. Kalman filtering effectively suppresses noise and extracts pressure features by optimally estimating these digitized weak signals, effectively achieving signal "amplification" (improving the signal-to-noise ratio) and precise acquisition in the digital domain.

[0024] Furthermore, the sensor automatic compensation and self-diagnosis system uses a conditioning circuit to compensate for it. The conditioning circuit is built based on the NSA2860 chip and is used to achieve temperature drift compensation of the sensor output.

[0025] (1) Specific circuit structure: The positive output terminal (Vout+) and negative output terminal (Vout-) of the silicon piezoresistive sensor (Wheatstone full bridge) are connected to the dedicated sensor input pins (such as INP and INN) of the NSA2860 chip. The NSA2860 provides a programmable excitation current source or voltage source (V_Bridge) connected to the excitation terminals (In+ and In-) of the sensor bridge circuit to power the sensor. The chip integrates a high-precision instrumentation amplifier (PGA), a digital-to-analog converter (DAC) for zero-point calibration and temperature compensation, and a temperature sensor.

[0026] (2) Temperature drift compensation principle: Before the sensor leaves the factory, its zero-point output V_offset(T) and full-scale output V_fs(T) are measured at multiple temperature points (T1, T2, ..., Tn). The zero-point compensation value (added to the original output) and the full-scale gain compensation coefficient (multiplied by the original output) are calculated for each temperature point. These compensation coefficients (or the fitted compensation polynomial coefficients) are programmed into the registers inside the NSA2860 or the non-volatile memory of the MCU connected to it. When the transmitter is working, the temperature sensor module inside the NSA2860 measures the temperature T in real time. Based on the current temperature T, the NSA2860 retrieves or calculates the zero-point compensation value (DACoffset) and gain compensation coefficient (PGAGain) corresponding to that T from the memory through a look-up table (LUT) or a built-in formula calculator.

[0027] (3) Zero-point compensation: The DAC inside the chip generates a voltage (or current) that is superimposed on the original differential output of the sensor to offset the zero-point offset caused by temperature drift. That is: V_corrected_diff = (Vout+ - Vout-) + DAC_offset(T).

[0028] (4) Sensitivity (Gain) Compensation: The programmable gain amplifier (PGA) inside the chip adjusts the amplification factor according to the calculated gain compensation coefficient Gain(T). That is: V_out = Gain(T) * V_corrected_diff.

[0029] The compensated analog signal V_out is output to the subsequent ADC module for digitization. The MCU can configure the registers of the NSA2860 through the SPI / I2C interface, control the compensation process, and read the status.

[0030] Furthermore, the sensor automatic compensation and self-diagnosis system, including time drift, adopts a dynamic online calibration method based on neural network prediction and system identifier. This method can realize the determination of time drift, select the appropriate neural network architecture, determine the input layer, hidden layer and output layer of the neural network, and the connection method between each layer. By real-time monitoring and analysis of changes in system parameters or data, the neural network is used to predict future trends, and the system model is dynamically updated in combination with the system identifier, thereby realizing accurate determination of time drift. The neural network adopts the LSTM (long short-term memory network) structure, which can effectively handle the long-term dependence of time series data. Combined with the real-time parameter update function of the system identifier, it significantly improves the accuracy and response speed of time drift compensation. Specific technical means (steps and methods): (1) Data collection and preprocessing: During normal operation of the transmitter, the MCU periodically collects and records a set of data (e.g., every hour or based on temperature changes). The input vector includes the current accurate pressure measurement value (from the main compensation output) P_meas, the current temperature T, the sensor's raw bridge voltage / resistance V_bridge / R_bridge, the zero output V_offset, and the operating time t_run. The output vector is the target: the pressure sensor's actual zero drift ΔV_offset and / or sensitivity change ΔSensitivity. (The initial target value must refer to or assume a baseline value under specific stable conditions.)

[0031] (2) Neural network architecture and training (offline / online initialization): Choose a Long Short-Term Memory (LSTM) network, which is better suited for processing time series. Input layer: 5 nodes (P_meas, T, V_bridge / R_bridge, V_offset, t_run). Hidden layers: 1-2 layers, 8-16 nodes per layer, using the ReLU activation function. Output layer: 2 nodes (ΔV_offset, ΔSensitivity), using the linear activation function.

[0032] Initial training: Use a historical dataset (possibly from aging experiments or simulations of similar sensors) to train the network externally (e.g., on a PC) to obtain initial weights and biases (W, b). The training objective is to minimize the mean squared error (MSE) between the predicted (ΔV_offset, ΔSensitivity) and the true value (or estimated value). Store the trained network parameters (W, b) in non-volatile memory.

[0033] (3) System Identifier (Online Modeling): The MCU simultaneously runs a simplified recursive least squares (RLS) and extended Kalman filter (EKF) algorithm as a system identifier. This identifier treats the sensor system as a dynamic model (for example, a first-order or second-order linear time-varying system). Its inputs are pressure, temperature, and other parameters, and its outputs are observed zero points or sensitivity changes. Based on real-time input and output data, the identifier estimates and updates the system model parameters (such as the state transition matrix and the elements of the observation matrix) online.

[0034] (4) Dynamic online calibration (executed in MCU): Periodically (or upon triggering), the collected input vector is fed into a neural network loaded with initial (or last updated) parameters to predict the current drift ΔV_offset_pred and ΔSensitivity_pred. The same input vector is fed into the system identifier, which outputs its estimates of the current drift ΔV_offset_iden and ΔSensitivity_iden.

[0035] Fusion / Decision: Fuse (e.g., weighted averaging, Kalman filter fusion) or compare the neural network prediction value and the identifier estimation value, and combine with the confidence assessment (e.g., the size of the prediction residual) to generate the final recognized drift ΔV_offset_final, ΔSensitivity_final.

[0036] Update compensation coefficients: Update zero compensation parameters (such as the DACoffset register value of the NSA2860 or the zero constant term of the main compensation model) based on ΔV_offset_final. Update sensitivity compensation parameters (such as the PGA gain coefficient of the NSA2860 or the gain term coefficient of the main compensation model) based on ΔSensitivity_final.

[0037] Online Learning: The final drift values ​​ΔV_offset_final and ΔSensitivity_final are used as new training examples. Combined with the corresponding input vectors, the neural network undergoes online incremental learning (using algorithms such as online gradient descent). The network weights (W, b) are updated and the updated parameters are stored in non-volatile memory, enabling the network to adapt to the sensor's unique long-term drift characteristics. The system identifier parameters are also continuously updated.

[0038] Threshold judgment: Set the drift threshold. If the drift exceeds the threshold, a maintenance alarm (part of the self-diagnosis function) may be triggered.

[0039] Furthermore, during the signal acquisition process, from the time when the trigger signal does not occur to the time when the acquisition is triggered and after the acquisition is completed, the input signal is zero and the output signal is not zero. The collected output data exists in the form of random noise. The signal value collected during this period is called zero-point drift. The temperature, pressure and other physical quantities measured by the pressure sensor will not have a strict linear relationship with the output value, so its functional relationship is often in the form of a polynomial. Polynomials can be used to fit nonlinear signals. The key lies in solving its various coefficients. Usually, the formula method in the zero-point temperature compensation algorithm is more complicated, and the fitting accuracy is often limited. The artificial neural network method has the advantages of using a small number of samples, a simple algorithm, and the ability to approximate arbitrary functions. The system adopts a zero-point drift compensation algorithm based on the back propagation (BP) neural network. It can achieve high-precision fitting through a small amount of sample training. It also supports online learning functions and can adapt to slow changes in sensor characteristics. Specific technical means (BP neural network is used for compensation): (1) Compensation goal: Establish an accurate mapping relationship between the sensor output (raw bridge pressure difference V_diff) and the actual pressure P and temperature T: P = f(V_diff, T). This function f is usually nonlinear.

[0040] (2) Neural network model (implemented in MCU): Input layer: 2 nodes: V_diff (raw sensor differential voltage) and T (temperature sensor reading). Hidden layer: 1 (or 2) layers, with an optional number of nodes (e.g., 4-8). The activation function uses the Sigmoid function. Output layer: 1 node: P_comp (compensated pressure estimate), using a linear activation function.

[0041] (3) Training (done offline): During the sensor calibration phase, the corresponding raw sensor output V_diff_ij is measured at multiple temperature points (T1, T2, ..., Tm) and multiple pressure points (P1, P2, ..., Pn).

[0042] Form a training data set: input sample X = [V_diff_ij, Ti], target output Y = Pj.

[0043] The network is trained using the BP algorithm on the PC to minimize the mean square error (MSE) between the predicted pressure P_comp and the actual applied pressure Pj.

[0044] The trained network parameters (weight W, bias b) are stored in the non-volatile memory of the transmitter.

[0045] (4) Online compensation (MCU execution during runtime): The signal conditioning module outputs the original V_diff. The temperature sensor module outputs the current T. The MCU reads V_diff and T as the input vector [V_diff, T]. The MCU loads the neural network parameters and performs the forward propagation calculation: Hidden layer input: H_in = W1 * [V_diff; T] + b1; Hidden layer output: H_out = Sigmoid(H_in); Output layer: P_comp = W2 * H_out + b2; The calculated P_comp is the final pressure measurement value after comprehensive compensation for zero point, sensitivity, nonlinearity and temperature, and is used for output.

[0046] Advantages: Compared to high-order polynomials (which require more coefficients and are prone to numerical instability), neural networks can effectively fit complex nonlinear relationships, including temperature cross-effects, with fewer parameters (weights). Trained models can be executed efficiently in MCUs.

[0047] Zero drift processing: The above compensation model already implicitly includes compensation for the change of zero point (the value of V_diff at zero pressure) with temperature T. Specialized zero point temperature drift compensation is usually part of the entire compensation process or the previous stage.

[0048] Furthermore, online calibration involves performing calibration directly in the pressure sensor's operating environment without removing it from the system. This method is efficient and convenient, reducing errors introduced by disassembling and installing the sensor and ensuring consistency between the calibration environment and the sensor's actual operating environment. Automatic calibration utilizes automated equipment and software to quickly and accurately calibrate the pressure sensor. This method significantly improves calibration efficiency and accuracy while reducing human error. Automatic calibration requires a sophisticated electronic control system and software algorithms. Using pre-set calibration procedures and algorithms, the system automatically adjusts the sensor's internal parameters to minimize the error between its output signal and the actual pressure value. This process may involve multiple steps, including zero-point calibration, full-scale calibration, and linearity calibration. Automatic calibration significantly improves calibration efficiency, reduces errors introduced by human operators, and ensures that the pressure sensor maintains high-precision measurements under various operating conditions. The system utilizes closed-loop automatic calibration technology, utilizing a high-precision reference sensor and PID control algorithm for rapid convergence. Full-scale calibration can be completed in 30 seconds, achieving a calibration accuracy of ±0.05% FS.

[0049] (1) Online calibration (implemented in this system): This is the dynamic online calibration method based on neural network prediction and system identifier described in point 4 above. It does not require removing the sensor. It uses historical and real-time data from the sensor in the working environment to automatically estimate and update compensation parameters (mainly time drift related parameters) through an algorithm.

[0050] (2) Automatic calibration (implemented by this system): Zero-point calibration: The MCU receives a "zero-point calibration" command from the host computer via the communication interface. The MCU ensures that the current pressure is a known zero pressure (for example, after the user confirms that the pressure has been vented to atmosphere). The MCU controls the ADC to acquire the current sensor output (V_diff_zero) and temperature (T). The MCU updates the zero-point offset parameters in the compensation algorithm (this may be done by directly updating the NSA2860's DACoffset register, by updating the constant term in the neural network compensation model or polynomial model, or by updating the term corresponding to the current temperature T in the zero-point compensation LUT). The new zero-point parameters are stored in non-volatile memory.

[0051] Full-scale / Span Calibration: The MCU receives the "Full-Scale Calibration" command and the target full-scale pressure value (P_fs) from the host computer via the communication interface. The MCU ensures that the currently applied pressure is stable at P_fs (user confirmation). The MCU collects the current sensor output (V_diff_fs) and temperature (T). It calculates the current sensitivity (Sensitivity_current) = (V_diff_fs - V_diff_zero) / P_fs. Comparing this with the stored nominal sensitivity (Sensitivity_nominal) (or the expected sensitivity after temperature compensation), the gain compensation coefficient (Gain_comp) = Sensitivity_nominal / Sensitivity_current. The MCU updates the gain parameters in the compensation algorithm (this may involve updating the NSA2860 PGA gain register, the gain term of the neural network / polynomial model, or the term corresponding to the current temperature (T) in the sensitivity compensation LUT).

[0052] Store the new gain parameters into non-volatile memory.

[0053] (3) Linear calibration (optional): This is usually solved during the calibration phase using a high-order polynomial or neural network model. During automatic calibration, the zero and full-scale points are mainly updated. Complex nonlinear deviations may require multi-point calibration (similar to full-scale calibration, which is performed at multiple pressure points and updates the model parameters).

[0054] (4) Automation: The entire calibration process is controlled by the MCU software. The user only needs to send commands through the HART handheld operator, host computer software, or control system and physically apply the corresponding zero pressure and full-scale pressure. The MCU automatically completes data acquisition, calculation, parameter update, and storage.

[0055] Furthermore, the communication software supports multiple communication protocols used by pressure sensors, including both hardware and software interfaces. When analyzing data, the software is capable of interpreting both analog and digital pressure sensor data. For analog signals, the software accurately reads and converts them into corresponding pressure values; for digital signals, the software correctly interprets and displays them. The communication protocols support industrial standards such as Modbus RTU / TCP, HART, and CANopen, and are compatible with mainstream PLC and DCS systems. The software interface provides real-time waveform display, historical data playback, and alarm log query capabilities, facilitating device status monitoring and fault analysis. The AD5700 chip used for the HART communication function offers high integration, low power consumption, and interference resistance. It communicates with the microcontroller's UART serial port via the RXD and TXD pins. A 3.6864MHz passive crystal oscillator is used. Capacitors C27, C28, and resistors R16 and R15 act as an external filter to receive the HART signal from the 4-20mA loop to the ADC_IP pin. The CD pin is the carrier monitor port. When the CD pin is high, the ADC_IP signal amplitude is greater than 120mA, indicating that the signal is a valid carrier signal. In modulation mode, RTS is low, and the digital signal transmitted by the microcontroller enters the FSK modulation module. The modulated HART signal is then output to the loop via the HART_OUT pin.

[0056] Furthermore, the self-diagnosis function is based on real-time monitoring and analysis of its key parameters. Through the diagnostic system integrated in the sensor, this data is collected in real time and compared with the preset threshold or standard to determine the performance status of the sensor. Based on the preset fault detection algorithm, the diagnostic system can identify potential faults or anomalies at an early stage. These algorithms are usually based on historical data of equipment performance and pattern recognition technology, and can predict potential faults or failures. Once a fault is detected, the diagnostic system will activate the corresponding mechanism to isolate the faulty part to prevent it from causing a greater impact on the entire system. This helps to keep the rest of the equipment running and reduce downtime. At the same time, the diagnostic system uploads the fault information to the main control system or monitoring center through the internal communication interface. Through sound, light or other forms of early warning, the operator can quickly know the problem with the sensor and perform maintenance and replacement in time. The diagnostic system adopts a predictive maintenance algorithm based on machine learning. By analyzing the historical data of the sensor to establish a health status model, it can predict potential faults 72 hours in advance with an accuracy rate of more than 90%. Specific technical means: (1) Real-time monitoring of key parameters (data source): Sensor signals: raw bridge pressure V_diff (obtained from the signal conditioning module or ADC module); compensated pressure value P_comp; temperature value T (obtained from the temperature sensor module); sensor excitation voltage / current V_bridge / I_bridge.

[0057] Circuit parameters: power supply voltage V_cc (monitored by the MCU's internal ADC or dedicated monitoring chip); ADC reference voltage V_ref; watchdog timer status; memory checksum.

[0058] Running status: software task execution time (RTOS monitoring); stack usage; communication error counter.

[0059] (2) Diagnostic algorithm execution (in MCU software): Threshold comparison (simple diagnosis): Zero - point drift exceeded: |V_diff (at zero pressure) - Stored_V_offset| > Threshold_V_offset? Full - scale exceeded / insufficient: When a known pressure is applied, |P_comp - Expected_P| > Threshold_P_fs? Over - range: V_diff > Full_Scale_Input_Range? Power supply exceeded: V_cc < Min_Vcc or V_cc > Max_Vcc? Temperature exceeded: T < Min_T or T > Max_T? Communication failure: Continuous communication timeout or CRC error count > Max_Error_Count? Watchdog reset: If watchdog reset event > 1? Pattern recognition / trend analysis (advanced diagnosis): Abnormal signal noise: Calculate the standard deviation σ of P_comp within a certain time window. If σ > Threshold_Noise (which may be adjusted according to the range), it indicates excessive noise, short - circuit or open - circuit.

[0060] Signal glitch detection: Real - time monitor the instantaneous change rate of V_diff or P_comp. If it exceeds the reasonable physical limit, an alarm is issued (such as impact / water hammer).

[0061] Abnormal temperature - pressure relationship: Compare the current combination of P_comp and T with the predicted value of the stored calibration model. Excessive residuals may indicate physical damage or severe drift of the sensor.

[0062] Historical trend analysis: Use the drift rate predicted by the time - drift compensation module. If the predicted drift rate increases sharply, it indicates that the sensor may be about to fail.

[0063] (3) Fault determination and isolation: The diagnostic algorithm compares the monitoring results with preset multi - level thresholds (warning, fault).

[0064] The MCU executes according to the fault level: Warning: Record the log, report the status word and warning code through the communication interface. The output remains normal.

[0065] Fault: Software Isolation: The MCU forces the output to a safe state (e.g., maintaining the last valid value, outputting predefined fault values ​​such as 3.9mA or 22mA, or NaN). Certain non-core functions are disabled. Hardware Isolation (Optional): If a fault occurs in the sensor excitation circuit (e.g., detecting an open or shorted bridge), the MCU may control the power management module through GPIOs to disconnect or limit sensor power. Alternatively, a hardware switch may be used to disconnect the faulty analog output circuit.

[0066] Trigger alarm output: The MCU controls the alarm circuit in the self-diagnosis execution module through GPIO, drives the LED to flash (specific color / frequency), or drives the relay contacts to close, or sends an emergency alarm message through the communication interface.

[0067] (4) Information reporting: Diagnostic results (status words, fault / warning codes, detailed parameters) are uploaded to the host computer / control system in real time or periodically via the communication interface module. Local alarm outputs provide audible / visual indications (such as buzzers and LEDs).

[0068] It should be noted that the above content merely illustrates the technical idea of ​​the present invention and cannot be used to limit the scope of protection of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications all fall within the scope of protection of the claims of the present invention.

Claims

1. An automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter, characterized in that: The system includes a silicon piezoresistive pressure sensor and a sensor automatic compensation and self-diagnosis system; The sensor automatic compensation and self-diagnosis system automatically compensates the silicon piezoresistive pressure sensor through software, and performs self-test on the silicon piezoresistive pressure sensor after power-on to check whether the sensor is normal and make a judgment.

2. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1 is characterized in that: The system adapts to the nonlinear and time-varying characteristics of sensor signals by dynamically adjusting filter parameters using a Kalman filter algorithm, including the following steps: Step 1: In the MCU software, treat the output of the pressure sensor as a dynamic system, define the state vector X_k, and define the state transition equation X_k = F * X_{k-1} + W_k, where F is the state transition matrix and W_k is the process noise; Step 2: Define the observation vector Z_k = H * X_k + V_k, where H is the observation matrix and V_k is the observation noise; Step 3: Kalman filter iteration, executed in real time in the MCU: Predicted state: X_k|k-1 = F * X_k-1|k-1; Prediction covariance: P_k|k-1 = F * P_k-1|k-1 * F^T + Q, where Q is the process noise covariance matrix; Calculate the Kalman gain: K_k = P_k|k-1 * H^T * (H * P_k|k-1 * H^T + R)^-1, where R is the observation noise covariance matrix; Update the state estimate: X_k|k = X_k|k-1 + K_k * (Z_k - H * X_k|k-1), where Z_k is the raw pressure observation read by the ADC; Update the covariance estimate: P_k|k = (I - K_k * H) * P_k|k-1; Step 4: The pressure estimate P_k in X_k|k is the output that is closer to the actual pressure after filtering and noise reduction. During operation, the system dynamically adjusts the parameters of the noise covariance matrix Q and R by monitoring the statistical characteristics of the observation residual (Z_k - H * X_k|k-1) or using other sensor information, so that the filter can adapt to changes in the environment and sensor noise characteristics. Step 5: Kalman filtering is performed in the digital domain of the MCU. Its input Z_k is the digital signal converted by the ADC module. To accurately collect the original weak analog signal, it is necessary to rely on the high-gain, low-noise instrument amplifier circuit of the signal conditioning module and the high-resolution, low-noise sampling of the ADC module.

3. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensing pressure transmitter according to claim 1 is characterized in that: The sensor automatic compensation and self-diagnosis system uses a conditioning circuit to compensate for it. The conditioning circuit is built based on the NSA2860 chip to achieve temperature drift compensation for the sensor output; The conditioning circuit includes connecting the positive and negative output terminals of the silicon piezoresistive sensor to the dedicated sensor input pins of the NSA2860 chip. The NSA2860 provides a programmable excitation current source or voltage source connected to the excitation terminal of the sensor bridge circuit. The chip integrates a high-precision instrumentation amplifier, a digital-to-analog converter for zero-point calibration and temperature compensation, and a temperature sensor.

4. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1 is characterized in that: The sensor automatic compensation and self-diagnosis system, including time drift, comprises the following steps: Step 1: Data collection and preprocessing: During normal operation of the transmitter, the MCU periodically collects and records a set of data, and the output vector is the target; Step 2: Neural Network Architecture and Training: Choose a long short-term memory network with 5 nodes in the input layer, 1-2 hidden layers with 8-16 nodes per layer, using the ReLU activation function, and a linear activation function for the 2 nodes in the output layer. Use historical data sets to train the network outside the MCU to obtain initial weights and biases (W, b). The training goal is to minimize the mean square error between the predicted ΔV_offset and the true value or estimated value. Store the trained network parameters (W, b) in non-volatile memory. Step 3: In the MCU, a simplified recursive least squares and extended Kalman filter algorithm are run simultaneously as a system identifier. The identifier estimates and updates the parameters of the system model online based on the input and output data collected in real time. Step 4: regularly input the collected input vectors into the neural network loaded with initial parameters to predict the current drift ΔV_offset_pred and ΔSensitivity_pred, input the same input vectors into the system identifier, and the identifier outputs its estimate of the current drift ΔV_offset_iden and ΔSensitivity_iden; fuse or compare the neural network prediction value and the identifier estimate value, and combine the confidence assessment to generate the final recognized drift ΔV_offset_final and ΔSensitivity_final; according to ΔV_offset_pred, the final drift value ΔV_offset_final and ΔSensitivity_final are generated. fset_final updates the zero-point compensation parameters and the sensitivity compensation parameters based on ΔSensitivity_final. The final identified drift values ​​ΔV_offset_final and ΔSensitivity_final are used as new training samples. Combined with the corresponding input vectors, the neural network is incrementally learned online to update the network weights (W, b). The updated parameters are stored in non-volatile memory to enable the network to adapt to the sensor's unique long-term drift characteristics. The system identifier parameters are also continuously updated. A permissible drift threshold is set; if the identified drift exceeds the threshold, a maintenance alarm may be triggered.

5. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1, characterized in that: During the signal acquisition process, from the time the trigger signal occurs until the acquisition is triggered, and after the acquisition is completed, the input signal is zero, but the output signal is not zero. The collected output data exists in the form of random noise. The system uses a zero-drift compensation algorithm based on a back-propagation neural network. It can achieve high-precision fitting through a small amount of sample training. It also supports online learning functions and can adapt to slow changes in sensor characteristics. Specifically, the following are some examples: Step 1: Establish an accurate mapping relationship between the sensor output and the actual pressure P and temperature T: P = f(V_diff, T). The function f is usually nonlinear. Step 2: The neural network model is implemented in the MCU. The input layer includes two nodes: the raw sensor differential voltage V_diff and the temperature sensor reading T; the hidden layer is one or two layers, the number of nodes is optional, and the activation function uses the Sigmoid function; the output layer includes one node for the compensated pressure estimate P_comp, using a linear activation function; Step 3, training, is done offline. This involves measuring the raw sensor output V_diff_ij at multiple temperatures (T1, T2, ..., Tm) and pressures (P1, P2, ..., Pn) during the sensor calibration phase. This generates a training dataset: input samples X = [V_diff_ij, Ti], target output Y = Pj. The network is trained on a PC using a backpropagation algorithm to minimize the mean squared error between the predicted pressure P_comp and the actual applied pressure Pj. The trained network parameters (weights W, bias b) are stored in the transmitter's non-volatile memory. Step 4: Online compensation, executed by the MCU during runtime, includes the signal conditioning module outputting the original V_diff, the temperature sensor module outputting the current T, the MCU reading V_diff and T as the input vector [V_diff, T], the MCU loading the neural network parameters, and performing forward propagation calculations, including the hidden layer input H_in = W1 * [V_diff; T] + b1, the hidden layer output H_out = Sigmoid(H_in), and the output layer P_comp = W2 * H_out + b2. The calculated P_comp is the final pressure measurement value after comprehensive compensation for zero point, sensitivity, nonlinearity, and temperature, and is used for output.

6. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1, characterized in that: The online calibration steps include: Step 1, online calibration step, uses historical data and real-time data of the sensor in the working environment to automatically estimate and update the compensation parameters through the algorithm; Step 2: Automatic calibration steps: The zero-point calibration process involves the MCU receiving a "zero-point calibration" command from the host computer via the communication interface. The MCU ensures that the current pressure is a known zero pressure. The MCU controls the ADC to collect the current sensor output (V_diff_zero) and temperature (T). The MCU then updates the zero-point offset parameters in the compensation algorithm and stores the new zero-point parameters in non-volatile memory. The full-scale / span calibration process involves the MCU receiving a "full-scale calibration" command and the target full-scale pressure value (P_fs) from the host computer via a communication interface. The MCU ensures that the currently applied pressure is stable at P_fs. The MCU collects the current sensor output (V_diff_fs) and temperature (T), calculates the current sensitivity (Sensitivity_current) = (V_diff_fs - V_diff_zero) / P_fs, compares it with the stored nominal sensitivity (Sensitivity_nominal), and calculates the gain compensation coefficient (Gain_comp) = Sensitivity_nominal / Sensitivity_current. The MCU then updates the gain parameters in the compensation algorithm and stores the new gain parameters in non-volatile memory. The linear calibration step involves solving the problem in the calibration phase using a high-order polynomial or neural network model, and updating the zero and full-scale points during automatic calibration.

7. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1, characterized in that: The communication software supports multiple communication protocols used by pressure sensors, including hardware interfaces and software interfaces. When parsing data, the software has the ability to parse data from pressure sensors, including analog signals and digital signals. For analog signals, the software can accurately read and convert them into corresponding pressure values; for digital signals, the software can correctly parse and display them. The HART communication module communicates with the microcontroller's UART serial port via the RXD and TXD pins. It uses a 3.6864MHz passive crystal oscillator, capacitors C27 and C28, and resistors R16 and R15 as an external filter to receive the HART signal from the 4-20mA loop and transmit it to the ADC_IP pin. The CD pin is the carrier monitor port. When the CD pin is high, the ADC_IP signal amplitude is greater than 120mA, indicating that the signal is a valid carrier signal. In modulation mode, RTS is low, and the digital signal sent by the microcontroller enters the FSK modulation module. The modulated HART signal is then output to the loop via the HART_OUT pin.

8. The automatic compensation and self-diagnosis system for SOI silicon piezoresistive sensor pressure transmitter according to claim 1, characterized in that: The self-diagnosis function is based on real-time monitoring and analysis of its key parameters. Through the diagnostic system integrated in the sensor, this data is collected in real time and compared with preset thresholds or standards to determine the performance status of the sensor. Specifically, it includes the following steps: Step 1: Real-time monitoring of key parameters, including the raw bridge voltage V_diff obtained from the signal conditioning module or ADC module, the compensated pressure value P_comp, the temperature value T, the sensor excitation voltage / current V_bridge / I_bridge, and circuit parameters. The circuit parameters include the power supply voltage V_cc monitored by the MCU internal ADC or monitoring chip, the ADC reference voltage V_ref, the watchdog timer status, and the memory checksum; including operation status monitoring, including software task execution time, stack usage, and communication error counter; Step 2: Execution of diagnostic algorithms, including threshold comparison, zero drift exceeding the limit, full scale exceeding / undershooting, over-range, power supply exceeding the limit, temperature exceeding the limit, communication failure, watchdog reset, pattern recognition / trend analysis, signal noise anomaly, signal glitch detection, temperature-pressure relationship anomaly, and historical trend analysis; Step 3, fault determination and isolation, the diagnostic algorithm compares the monitoring results with the preset multi-level thresholds, and the MCU executes according to the fault level; Step 4: Information reporting, including uploading the diagnostic results to the host computer / control system in real time or periodically through the communication interface module, and providing sound / light indication locally through alarm output.

Images