High-precision temperature sensor calibration method based on cryogenic environment

Abstract

This invention discloses a high-precision temperature sensor calibration method based on cryogenic environments, comprising the following steps: S1, acquiring the original detection data and environmental sensing data of the temperature sensor under cryogenic conditions. The original detection data includes sensor electrical output data and noise signal data, while the environmental sensing data includes sensor micro-deformation data and encapsulation cavity characteristic signal data; S2, performing zero-point drift self-reference calibration on the sensor based on quantum tunneling noise spectrum feature inversion to eliminate zero-point offset errors caused by quantum tunneling effects under cryogenic environments. This invention relates to the field of temperature sensor calibration technology. This high-precision temperature sensor calibration method based on cryogenic environments, through the zero-point drift self-reference calibration method based on quantum tunneling noise spectrum feature inversion, can effectively eliminate zero-point drift errors under cryogenic environments, accurately correct the sensor's detection values, and thus improve the accuracy of temperature measurement.

Description

Technical Field

[0001] This invention relates to the field of temperature sensor calibration, and more specifically, to a high-precision temperature sensor calibration method based on cryogenic environments. Background Technology

[0002] Accurate temperature sensor measurements face numerous challenges in cryogenic environments. Firstly, the quantum tunneling effect in cryogenic environments induces zero-point offset errors in temperature sensors, which traditional calibration methods struggle to eliminate, leading to inaccurate measurements. Secondly, the thermal shrinkage of temperature sensor encapsulation materials in cryogenic environments generates mechanical stress that interferes with the sensor output, preventing temperature readings from accurately reflecting the actual temperature. Existing calibration techniques cannot effectively separate the contributions of temperature and mechanical stress to the sensor output, making it difficult to eliminate stress interference. Furthermore, as sensors age and their performance changes with environmental variations, traditional calibration methods lack dynamic optimization mechanisms, failing to update calibration parameters promptly and thus hindering long-term high-precision measurements. Therefore, a temperature sensor calibration method capable of high-precision calibration in cryogenic environments with dynamic optimization capabilities is needed. Summary of the Invention

[0003] The purpose of this invention is to provide a high-precision temperature sensor calibration method based on cryogenic environment. It solves the problem that existing temperature sensor calibration methods often cannot effectively solve the error problems caused by factors such as zero drift and stress interference in cryogenic environment, and lack dynamic optimization capabilities, thus failing to meet the usage requirements.

[0004] This invention achieves the above objective through the following technical solution: a high-precision temperature sensor calibration method based on a cryogenic environment, comprising the following steps: S1. Acquire the raw detection data and environmental perception data of the temperature sensor in a cryogenic environment. The raw detection data includes the sensor's electrical output data and noise signal data, while the environmental perception data includes the sensor's micro-deformation data and the packaging cavity's characteristic signal data. S2. The zero-drift self-reference calibration method based on the inversion of quantum tunneling noise spectrum characteristics is used to correct the zero-point drift of the sensor and eliminate the zero-point offset error caused by the quantum tunneling effect in the cryogenic environment. S3. By using the cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model, the contributions of temperature and mechanical stress to the sensor output are separated and the temperature reading is dynamically corrected, eliminating stress interference caused by thermal shrinkage of the packaging material. S4. A self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points is adopted to establish a self-evolving calibration model and update calibration parameters, thereby completing high-precision calibration of the sensor and realizing dynamic optimization of the calibration model.

[0005] Furthermore, the implementation of the zero-drift self-reference calibration method based on quantum tunneling noise spectrum feature inversion in step S2 includes the following steps: The temperature sensor was placed in a cryogenic steady-state environment with a temperature of ≤-150℃. The noise signal at the metal-semiconductor interface of the sensor was collected and the noise power spectral density function was obtained by power spectral density analysis. Core feature parameters are extracted from the low-frequency range of the function, and a nonlinear mapping model of tunneling noise spectrum-absolute temperature offset is trained based on the corresponding samples of noise spectrum feature parameters and temperature offset. The core feature parameters are input into the model to obtain the absolute zero drift, and the original temperature value detected by the sensor is corrected based on the drift.

[0006] Furthermore, the core feature parameters include: Rate of change of slope in the low-frequency region Inflection point frequency and peak noise of energy level transition Low-frequency region slope change rate The logarithm of the noise power spectral density function and the logarithm of the frequency are in The first derivative of an interval; The training of the nonlinear mapping model of tunneling noise spectrum-absolute temperature offset involves normalizing the core feature parameters, selecting a neural network to build the basic model and setting appropriate training parameters to complete the training. After verification and optimization, the model parameters are solidified and embedded into the sensor calibration algorithm to support online inference calculation.

[0007] Furthermore, the criteria for determining the cryogenic steady-state environment are as follows: Set temperature fluctuation thresholds and noise spectrum abrupt change judgment thresholds; When the fluctuation range of the original temperature detection value of the sensor is less than or equal to the temperature fluctuation threshold within the preset continuous detection period, and the rate of change of the power spectral density function of the acquired noise signal in the whole frequency band is less than or equal to the noise spectrum change judgment threshold, it is determined to be a cryogenic steady-state environment. The temperature fluctuation threshold is determined based on the sensor's nominal accuracy class, which is [value missing]. Level when taking The nominal accuracy is lower than Level according to nominal accuracy Values.

[0008] Furthermore, the noise signal acquisition frequency band covers the low-frequency to mid-frequency range, and a threshold for effective extraction of noise spectrum features is set. Core feature parameters are extracted only when the signal-to-noise ratio of the noise signal in the target frequency band is ≥ the threshold; otherwise, the noise signal is reacquired. The signal-to-noise ratio threshold is dynamically adjusted based on the noise figure NF of the acquisition device. The adjustment rule is to add the basic signal-to-noise ratio threshold to the noise figure NF of the acquisition device.

[0009] Furthermore, the implementation of the cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model described in step S3 includes the following steps: Based on the material physical properties of the temperature sensor packaging material and the sensing element, a thermal shrinkage coefficient difference matrix model is constructed to characterize the difference in thermal shrinkage between the two in different directions; The minute deformation of the sensitive element is obtained by detecting the element and converted into a thermal stress value; A three-dimensional coupled equation system of temperature-stress-electrical output is established. The equation system is solved iteratively using a numerical inversion algorithm to separate the pure temperature contribution and the resistance value of the sensitive element under no stress interference, and the theoretical value corresponding to pure temperature is obtained by back-reasoning. Based on this theoretical value, the temperature value after zero drift correction is dynamically stress decoupling corrected.

[0010] Furthermore, the numerical inversion algorithm is the Newton-Raphson inversion algorithm. An iterative convergence threshold is set for this algorithm, which is adaptively determined based on the sensor's calibration accuracy requirements. When the deviation between the pure temperature theoretical values ​​obtained from two consecutive iterations is less than or equal to the iteration convergence threshold, the iteration is considered to have converged and the calculation is stopped. In the dynamic stress decoupling correction process, the stress decoupling correction weight coefficient is adaptively determined based on the stress sensitivity coefficient of the sensor's sensitive element. The larger the absolute value of the stress sensitivity coefficient, the closer the correction weight coefficient is to 1.

[0011] Furthermore, the implementation of the self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points described in step S4 includes the following steps: A micro-medium storage cavity is pre-set inside the temperature sensor packaging cavity and a high-purity reference medium is introduced. The reference medium has at least three known and stable cryogenic phase transition temperatures, and each phase transition temperature uniformly covers the target cryogenic calibration range of the sensor. The characteristic signals inside the encapsulation cavity are collected in real time by the deployed detection elements. A threshold for judging abrupt changes in the characteristic signals is set. Based on the threshold, it is determined whether the reference medium has undergone a cryogenic phase change. The real-time detection value of the sensor corresponding to the transient characteristic point of the phase change is captured and paired with the standard phase change temperature of the reference medium to obtain a calibration sample pair. The temperature value after decoupling and correction of the three-field coupling and the historical calibration data of the sensor are used as input. A self-evolutionary calibration model is trained by machine learning algorithm, and the detection value is accurately calibrated by nonlinear interpolation algorithm. By continuously capturing new phase transition feature points during the daily detection process of sensors, supplementing the calibration sample set, and updating the model parameters through incremental learning, the calibration model can achieve self-evolutionary optimization.

[0012] Furthermore, the reference medium is one or more combinations of high-purity rare gas, organic microcrystalline material or inorganic low-melting-point microcrystalline material, and its cryogenic phase transition type includes adsorption-desorption phase transition, glass transition or solid phase crystal transformation. Set an effective threshold for the phase transition signal of the reference medium and dynamically adjust the threshold according to the sensor usage time to ensure that the phase transition feature points can still be accurately captured after long-term use of the sensor. The machine learning algorithm is a gradient boosting regression tree algorithm. Before model training, outlier removal and standardization are performed on calibration sample pairs. Model initialization parameters and training process are set, and regularization constraints are added to prevent overfitting. When the mean absolute error of the test set reaches the preset standard, the model is deemed to be qualified for training. When the proportion of new samples is greater than or equal to the preset proportion, the model is triggered to retrain. The parameters of the new model and the original model are fused through a weight fusion formula.

[0013] Furthermore, the cryogenic environment refers to a temperature... In low-temperature environments, the target cryogenic calibration range of the sensor is consistent with the temperature range of the cryogenic environment.

[0014] The beneficial effects of this invention are as follows: 1. The zero-drift self-reference calibration method based on quantum tunneling noise spectrum inversion can effectively eliminate zero-point drift error in cryogenic environments, accurately correct sensor detection values, and thus improve the accuracy of temperature measurement.

[0015] 2. A three-field coupling decoupling model of cryogenic heat-stress-resistance is adopted to dynamically correct the sensor temperature reading, eliminating stress interference caused by factors such as thermal shrinkage of packaging materials, and ensuring that the temperature measurement results are more accurate and reliable.

[0016] 3. The self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points uses machine learning algorithms to self-evolve and optimize the sensor's calibration model. This method can continuously improve calibration accuracy as the sensor is used, adapt to performance changes after long-term use, and ensure that the calibration effect remains reliable.

[0017] 4. Through self-evolving calibration technology, the traditional temperature sensor calibration process can be simplified, reducing manual intervention and calibration time, improving work efficiency, and reducing labor costs. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the overall calibration process of the present invention; Figure 2 This is a flowchart of the quantum tunneling calibration process of the present invention; Figure 3 This is a flowchart of the three-field coupling decoupling process of the present invention; Figure 4 This is a flowchart of the self-evolutionary calibration process of the present invention. Detailed Implementation

[0019] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content.

[0020] Example 1: Please see Figure 1-4 This invention provides a technical solution: a high-precision temperature sensor calibration method based on a cryogenic environment, the method comprising: S1. Acquire the raw detection data of the temperature sensor and the environmental perception data in the cryogenic environment. The raw detection data includes the sensor's electrical output data and noise signal data, while the environmental perception data includes the sensor's micro-deformation data and the packaging cavity's characteristic signal data. Cryogenic environments typically refer to environments with extremely low temperatures, generally far below room temperature, potentially reaching liquid nitrogen temperatures or even lower. Under these extreme low-temperature conditions, the physical and chemical properties of substances undergo significant changes, exerting complex influences on the performance and measurement accuracy of temperature sensors. The raw detection data includes: sensor electrical output data (when operating in a cryogenic environment, the internal electrical components of the temperature sensor, such as thermistors and thermocouples, will generate corresponding electrical signal outputs due to temperature changes, such as resistance and voltage values), which directly reflect the sensor's initial temperature perception; and noise signal data (in cryogenic environments, sensors are affected by various interference factors, generating noise signals, which may originate from the sensor's own electrical components). Noise signal data, including electromagnetic interference from components and the surrounding environment, contains this interference information and can affect the accuracy of sensor measurements. Environmental sensing data includes sensor micro-deformation data, which is obtained by measuring the minute deformations of the sensor and its packaging structure under extremely low temperatures in cryogenic environments. Measuring these micro-deformations allows us to understand the impact of environmental temperature changes on the sensor's physical structure, providing a reference for subsequent calibration. Packaging cavity characteristic signal data is also important. Sensors are typically packaged in specific cavities, and the material properties of these cavities, such as their coefficient of thermal expansion, change under cryogenic conditions. This alters characteristic signals within the cavity, such as pressure and optical properties. This characteristic signal data helps analyze the impact of the packaging environment on sensor measurements. S2. A zero-drift self-reference calibration method based on quantum tunneling noise spectrum characteristics is used to correct the zero-point drift of the sensor and eliminate the zero-point offset error caused by the quantum tunneling effect in a cryogenic environment. Among them, the quantum tunneling noise spectrum feature inversion method addresses the phenomenon that microscopic particles can pass through potential barriers with higher energies than themselves under certain conditions. In cryogenic environments, the quantum tunneling behavior of microscopic particles in sensors generates specific noise spectra. By analyzing and inverting these noise spectrum features, information related to the quantum tunneling effect can be obtained. The zero-drift self-reference calibration method addresses the deviation of the sensor's output value from its input when it is zero, caused by changes in time or environment. In cryogenic environments, the quantum tunneling effect may cause zero-point offset errors in the sensor. The zero-drift self-reference calibration method uses the results of quantum tunneling noise spectrum feature inversion to automatically determine the sensor's zero-point reference and correct the zero-point drift, thereby eliminating the zero-point offset error caused by the quantum tunneling effect and improving the sensor's measurement accuracy near the zero point. S3. By using the cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model, the contributions of temperature and mechanical stress to the sensor output are separated and the temperature reading is dynamically corrected, eliminating stress interference caused by thermal shrinkage of the packaging material. The cryogenic thermal-stress-resistance three-field coupling involves the interaction between three physical fields: temperature change, mechanical stress change, and sensor resistance change. Temperature changes cause thermal contraction of the encapsulation material, resulting in mechanical stress. Mechanical stress, in turn, affects the resistance of the internal electrical components of the sensor. Changes in resistance reflect temperature changes, and this complex coupling relationship has a comprehensive impact on the sensor output. The decoupling inversion calibration model is used to accurately obtain temperature information. It separates the contributions of temperature and mechanical stress to the sensor output. By analyzing and calculating the interrelationships between the thermal, stress, and resistance fields, and utilizing known physical laws and experimental data, the model isolates the portion of the sensor output caused by mechanical stress and dynamically corrects the temperature reading. This eliminates stress interference caused by thermal contraction of the encapsulation material and improves the accuracy of temperature measurement. S4. A self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points is adopted to establish a self-evolving calibration model and update calibration parameters, thereby completing high-precision calibration of the sensor and realizing dynamic optimization of the calibration model. Among these methods, the following are key features: Dynamic capture of cryogenic phase transition reference points: In cryogenic environments, certain substances undergo phase transitions, such as the vaporization of liquid nitrogen and the solidification of certain materials. These phase transitions are accompanied by energy absorption or release, and the phase transition temperature is relatively stable. By dynamically capturing these cryogenic phase transition reference points, accurate temperature reference values ​​can be obtained. Self-evolving calibration methods: Self-evolving calibration methods are a type of method that can automatically adjust and optimize calibration parameters based on actual measurement conditions and environmental changes. It utilizes the results of dynamically captured cryogenic phase transition reference points, combined with historical and real-time measurement data from the sensor, to continuously update calibration parameters through certain algorithms and models. This allows the calibration model to adapt to the dynamic changes in the cryogenic environment, achieving dynamic optimization of the calibration model and thus improving the high-precision calibration effect of the sensor. Self-evolving calibration models: Models established based on self-evolving calibration methods can automatically adjust their parameters and structure according to new measurement data and environmental information to continuously improve the accuracy and adaptability of calibration, ensuring that the sensor maintains high-precision measurement performance in cryogenic environments.

[0021] It should be noted that during use, S1 comprehensively acquires raw and environmental sensing data, providing a rich information foundation for subsequent accurate calibration. S2 uses quantum tunneling noise spectrum characteristics inversion to perform zero-drift self-reference correction, which can specifically eliminate zero-point offset errors caused by quantum tunneling effects under cryogenic conditions and improve measurement accuracy near the zero point. S3 uses a cryogenic thermal-stress-resistance three-field coupling decoupling inversion model to effectively separate the influence of temperature and mechanical stress on the output, dynamically correct temperature readings, eliminate interference from thermal shrinkage of packaging materials, and ensure the accuracy of temperature measurement. S4 adopts a self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points, which can establish and dynamically optimize the self-evolving calibration model and update parameters, enabling calibration to adapt to dynamic changes in the cryogenic environment, continuously maintain high-precision sensor measurements, and meet the stringent requirements for accurate temperature measurement in the cryogenic field.

[0022] In one embodiment, a zero-drift self-reference calibration method based on quantum tunneling noise spectrum characteristic inversion performs zero-point drift self-reference correction on the sensor to eliminate zero-point offset errors caused by quantum tunneling effects in cryogenic environments, including: The temperature sensor was placed in a cryogenic steady-state environment with a temperature ≤ -150℃. A high-precision noise acquisition module was used to collect the noise signal at the metal-semiconductor interface of the sensor in real time. Power spectral density analysis was performed on the collected noise signal to obtain the noise power spectral density function across the entire frequency band. ,in Noise frequency, unit: ; Feature extraction is performed on the low-frequency region of the noise power spectral density function to obtain the rate of change of the slope in the low-frequency region. Inflection point frequency and peak noise of energy level transition Three core feature parameters, among which the slope change rate in the low-frequency region reflect The formula for calculating the change in the ratio of noise to white noise is:

[0023] In the formula, For the corresponding frequency The noise power spectral density is given below, in units of , The inflection point frequency, i.e., low frequency. The critical frequency at which noise transitions to white noise; Based on a massive number of noise spectrum feature parameters and temperature offset corresponding samples in cryogenic environments, training, validation, and test sets are constructed, with samples divided in a 7:2:1 ratio. A nonlinear mapping model between tunneling noise spectrum and absolute temperature offset is trained using a nonlinear fitting algorithm.

[0024] This model can accurately characterize the intrinsic correlation between noise spectrum characteristics and sensor zero-point drift under the quantum tunneling effect; Extract the slope change rate in the low-frequency region Inflection point frequency and peak noise of energy level transition Input a nonlinear mapping model, and calculate the current absolute zero drift of the sensor by inverse modeling. The sensor's original detected temperature value is then corrected based on this zero-point drift, using the following formula:

[0025] In the formula, This is the temperature value after zero drift correction. These are the original, uncorrected temperature values ​​detected by the sensor; all units are... .

[0026] This design, by collecting noise signals from the sensor's metal-semiconductor interface under cryogenic conditions, analyzing the power spectral density and extracting characteristic parameters, constructs a nonlinear mapping model to inversely calculate the zero-point drift, corrects the sensor's original detection value, and accurately eliminates the zero-point offset error caused by the quantum tunneling effect in cryogenic environments. By utilizing the inherent correlation between noise spectrum characteristics and zero-point drift, self-reference correction is achieved without the need for additional complex equipment, improving the sensor's measurement accuracy in cryogenic environments, reducing errors caused by zero-point drift, and ensuring the reliability of measurement results.

[0027] In one embodiment, the training steps and parameter settings for the tunneling noise spectrum-absolute temperature offset nonlinear mapping model are as follows: Step 1: Sample preprocessing, processing the collected noise spectrum characteristic parameters Normalization is performed using the following formula:

[0028] in, These are the original eigenvalues. The minimum value of the characteristic. The maximum eigenvalue is the eigenvalue after normalization, which is then mapped to... interval; Step 2: Model structure selection. A 3-layer fully connected neural network is used as the basic model, with 3 neurons in the input layer. Three features: the first hidden layer has 64 layers and uses ReLU activation function; the second hidden layer has 32 layers and uses ReLU activation function; the output layer has 1 neuron, corresponding to the temperature offset. The activation function is a linear function; Step 3: Model training parameter settings. The optimizer is Adam, the initial learning rate is set to 0.001, the learning rate decay strategy is cosine annealing, and the learning rate decays to 1 / 10 of the initial value every 100 iterations. The batch size is set to 32, the number of training epochs is set to 500, and the loss function is the root mean square error (RMSE). The loss calculation formula is:

[0029] In the formula, This is the actual temperature offset. To predict the temperature shift for the model, This refers to the batch sample size. Step 4: Model Validation and Tuning. Every 10 iterations, the model accuracy is evaluated using the validation set. Training stops when the validation set loss shows no decrease for 20 consecutive iterations, employing an early stopping strategy to prevent overfitting. The model's goodness of fit on the test set is then evaluated. The model is deemed to have passed training at that time. Step 5: Model deployment. The trained model parameters are solidified and embedded into the sensor calibration algorithm to support online inference calculation.

[0030] This design specifies the steps for model training, including sample preprocessing, structure selection, parameter setting, validation and optimization, and deployment. Sample preprocessing makes the data more standardized, which is beneficial for model training; reasonable selection of model structure and parameters can better fit the data relationship; validation and optimization prevent overfitting and ensure the model's generalization ability; after the model is deployed, it supports online inference, which can accurately calculate the temperature offset in real time, improve the efficiency and accuracy of sensor calibration, and enhance the practicality of the system.

[0031] In one embodiment, a cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model is used to separate the contributions of temperature and mechanical stress to the sensor output and dynamically correct the temperature reading, eliminating stress interference caused by thermal shrinkage of the encapsulation material, including: Based on the material physical properties of the temperature sensor packaging material and the sensing element, a three-dimensional thermal shrinkage coefficient difference matrix model is constructed. This matrix is ​​used to characterize the differences in thermal shrinkage between the packaging material and the sensitive element in different directions. The matrix expression is as follows:

[0032] In the formula, The coefficient of thermal expansion of the sensor's sensitive element, in units of... , The encapsulation materials are respectively in The coefficients of thermal expansion in three spatial directions, in units of ; Miniature strain gauges or piezoelectric thin film sensing elements are placed at the junction of the sensor's sensing element and the package to measure in real time the micro-deformation of the sensing element caused by thermal shrinkage of the package material under cryogenic conditions. ,in They are respectively The deformation in three directions, combined with the material's equivalent elastic modulus matrix, converts the micro-deformation into the thermal stress value experienced by the sensitive element. The stress calculation formula is:

[0033] In the formula, This is the equivalent elastic modulus matrix of the packaging material and the sensitive element, in units of... , This is the three-dimensional thermal stress tensor, with units of . ; Combining the sensor's resistance-temperature characteristics, stress-resistance characteristics, and temperature-stress coupling characteristics, a three-dimensional coupling equation set of temperature-stress-electrical output is established to accurately characterize the coupling relationship between temperature, thermal stress, and the sensor's electrical output. The equation set is as follows:

[0034] In the formula, For the sensor's sensitive element at temperature ,stress The actual resistance value below, in units of ; This is the reference resistance value of the sensitive element under normal temperature and pressure, in units of... ; The temperature coefficient of resistance of the sensitive element, in units of . ; The stress resistivity of the sensitive element is expressed in units of . ; The temperature-stress coupling coefficient of the sensitive element, in units of ; The reference temperature is the room temperature. ; The actual electrical output voltage of the sensor, in units of... ; The constant supply current to the sensor, in units of ; The Newton-Raphson inversion algorithm is used to iteratively solve the three-dimensional coupled equations. Through multiple rounds of iterative calculations, the resistance values ​​of the sensitive element under pure temperature contribution and stress-free conditions are separated. And based on this pure temperature resistance value, the theoretical value corresponding to the pure temperature in the sensor detection value is derived. ; The theoretical value of pure temperature obtained by reverse calculation Temperature value after zero drift correction Dynamic stress decoupling correction is performed, and the correction formula is as follows:

[0035] In the formula, The temperature value is the result of decoupling correction for the three-field coupling, in units of... ; The stress decoupling correction weighting coefficient is based on the stress sensitivity coefficient of the sensor's sensing element. Adaptive determination, the determination rule is: , Stress sensitivity coefficient The larger the absolute value, the more significant the influence of stress on the component, and the more important the adjustment weighting coefficient. The closer it gets to 1.

[0036] This design constructs a thermal shrinkage coefficient difference matrix model, arranges detection elements to measure micro-deformation and convert it into thermal stress, establishes a set of coupled equations, uses the Newton-Raphson algorithm to solve for the separation of pure temperature contribution, corrects the temperature value, separates the influence of temperature and mechanical stress on the sensor output, eliminates the interference of thermal shrinkage stress of the packaging material, and improves the accuracy of temperature measurement by the sensor in complex environments through precise calculation and correction, so that the measurement results more realistically reflect the actual temperature.

[0037] In one embodiment, a self-evolving calibration method based on dynamic capture of a cryogenic phase transition reference point is employed to establish a self-evolving calibration model and update calibration parameters, thereby achieving high-precision calibration of the sensor and realizing dynamic optimization of the calibration model, including: A miniature dielectric storage cavity is pre-set inside the packaging cavity of the temperature sensor, and a controllable trace amount of high-purity reference medium is introduced into it. The reference medium has at least three known and stable cryogenic phase transition temperatures. Furthermore, the phase transition temperatures uniformly cover the target cryogenic calibration range of the sensor. ; Miniature acoustic emission sensors and dielectric sensors are deployed inside the encapsulation cavity to collect characteristic signals inside the cavity in real time, and a threshold for detecting sudden changes in characteristic signals is set. The rule for determining this threshold is as follows: ,in The steady-state mean of the characteristic signal. The standard deviation of the characteristic signal under steady state is given when an instantaneous change in the characteristic signal is detected. Furthermore, if the duration is ≥5ms, the reference medium is determined to have undergone a cryogenic phase transition; Accurately capture the real-time sensor detection values ​​corresponding to transient feature points of phase transition. The standard phase transition temperature is known from the reference medium. As a dynamic internal temperature standard point, it is paired one-to-one with the sensor detection values ​​corresponding to the captured phase transition transient characteristic points to obtain multiple sets of calibration sample pairs. ; Temperature value after decoupling and correction of three-field coupling Using historical calibration data from the entire sensor lifecycle as input, a training set and a test set are divided in an 8:2 ratio. A gradient boosting regression tree machine learning algorithm is used to train a multi-phase transition node-nonlinear interpolation self-evolutionary calibration model. The core parameters and iteration steps for model training are set. For detection values ​​located between any two adjacent phase transition standard points, accurate calibration is achieved through a nonlinear interpolation algorithm. The nonlinear interpolation calculation formula is as follows:

[0038] In the formula, These are high-precision temperature values ​​after interpolation calibration, in units of... ; , The phase transition standard temperature for the corresponding range. These are the sensor detection values ​​for the phase transition characteristic points in the corresponding interval; As the sensor is used over time, new phase transition feature points in the reference medium are continuously captured during the sensor's daily detection process and added to the calibration sample set. Sample update weighting coefficients are then set. The rules for determining it are as follows:

[0039] in, The detection time for newly captured phase transition feature points. For the detection time of historical samples, Using the historical sample size, the parameters of the self-evolving calibration model are automatically updated through incremental learning, achieving self-evolving optimization and accuracy iteration of the calibration model, ultimately outputting a result refined through multiple rounds of correction. As a high-precision temperature value after sensor calibration.

[0040] This design incorporates a miniature medium storage cavity within the sensor's encapsulation chamber, introduces a reference medium, and deploys sensors to collect characteristic signals to determine phase transitions. It captures the detection values ​​corresponding to the phase transition points, trains a self-evolving calibration model using machine learning algorithms, and updates the parameters. By utilizing the phase transition points of the reference medium as dynamic internal standard points, high-precision calibration is achieved. By continuously capturing new phase transition feature points and incrementally learning and updating the model, the calibration model can adapt to changes throughout the sensor's entire lifecycle, maintaining high-precision calibration capabilities over the long term.

[0041] In one embodiment, the training steps and parameter settings for the gradient boosting regression tree self-evolutionary calibration model are as follows: Step 1: Sample preprocessing, processing the sensor detection values ​​in the calibration sample pairs Outlier removal is performed using... Criteria for identifying outliers, i.e. Values ​​identified as outliers are removed at certain times. The sample mean. The standard deviation is the sample standard deviation. The samples after removing outliers are standardized using the following formula:

[0042] The standardized sample has a mean of 0 and a standard deviation of 1. Step 2: The model initialization parameters are set as follows: the base learner is a regression decision tree, the maximum depth of the decision tree is set to 8, the minimum number of sample splits is set to 10, the minimum number of leaf node samples is set to 5, the feature sampling ratio is set to 0.8, and the sample sampling ratio is set to 0.9; the learning rate of gradient boosting is set to 0.05, the number of iterations (number of weak learners) is set to 500, and the loss function is the mean squared error loss. Step 3: Model training process, Initialize the weak learner using the sample mean as the initial prediction value:

[0043] Iteratively train the weak learner, for the th Round iteration ( ): ① Calculate the negative gradient as the residual:

[0044] ② Train a regression decision tree to fit the residuals , obtained the Leaf node region of a decision tree , , The number of leaf nodes; ③ Calculate the optimal output value for each leaf node:

[0045] ④ Update the model:

[0046] in, For indicator functions; Regularization constraints are applied using L2 regularization, and the regularization terms are:

[0047] Prevent model overfitting; Step 4: Model evaluation and update, mean absolute error on the test set The model is deemed to have passed training at any time; during incremental updates, the proportion of new samples is... The model is retrained, and the weight fusion formula between the new and original models is as follows:

[0048] This design involves sample preprocessing, setting model initialization parameters, defining the training process, and performing model evaluation and updates. Sample preprocessing removes outliers and standardizes the data to improve data quality. Reasonable setting of initialization parameters and training process enables the model to effectively learn data features. The evaluation and update mechanism ensures that the model training is qualified, and during incremental updates, the old and new models are balanced through weight fusion to improve the performance and stability of the calibrated model.

[0049] In one embodiment, the criterion for determining a cryogenic steady-state environment is to set a temperature fluctuation threshold. Threshold for detecting sudden changes in noise spectrum The temperature fluctuation threshold is determined based on the sensor's nominal accuracy class, which is [value missing]. Level when taking The nominal accuracy is lower than For the grade, take 1 / 10 of the nominal accuracy; When the sensor operates within a continuous 30-second detection cycle, the fluctuation range of the original temperature detection value is ≤ Furthermore, the rate of change of the power spectral density function of the acquired noise signal across the entire frequency band is ≤ At that time, it was determined to be a cryogenic steady-state environment to ensure the accuracy of noise spectrum feature extraction.

[0050] This design sets temperature fluctuation thresholds and noise spectrum mutation judgment thresholds. When the sensor meets the corresponding conditions within a continuous 30-second detection cycle, it is determined to be in a cryogenic steady-state environment. The criteria for determining a cryogenic steady-state environment are clearly defined, and the temperature fluctuation threshold is determined based on the sensor's nominal accuracy level to ensure the rationality of the judgment. Combined with the noise spectrum change rate, it ensures that noise signals are collected in a stable environment, improves the accuracy of noise spectrum feature extraction, and provides a reliable foundation for subsequent accurate sensor calibration.

[0051] In one embodiment, the reference medium is one or more combinations of high-purity rare gas, organic microcrystalline material, or inorganic low-melting-point microcrystalline material, and its cryogenic phase transition type includes adsorption-desorption phase transition, glass transition, or solid-state crystal transformation. The effective threshold of the phase transition signal of the reference medium is set to a sudden change amplitude ≥ 20%, and the correction rule for this threshold is as follows: Based on sensor usage time Dynamic adjustments are made, and the adjustment formula is as follows:

[0052] Ensure that phase transition feature points can still be accurately captured after long-term use.

[0053] This design selects one or more combinations of high-purity rare gases as reference media, specifies their cryogenic phase transition type, sets and dynamically adjusts the effective threshold of the phase transition signal, and multiple reference media combinations can provide multiple stable phase transition temperature points, uniformly covering the target calibration range; clearly defining the phase transition type facilitates accurate capture of the phase transition signal; dynamically adjusting the effective threshold ensures that the sensor can still accurately capture phase transition feature points after long-term use, guaranteeing the accuracy and reliability of calibration.

[0054] In one embodiment, the noise signal acquisition frequency band covers In the low-to-mid frequency range, a threshold for effective extraction of noise spectrum features is set, namely the inflection point frequency. The recognition threshold is the signal-to-noise ratio. When the signal-to-noise ratio of the noise signal in the target frequency band If the signal-to-noise ratio (SNR) is not specified, feature parameters are extracted; otherwise, the noise signal is re-acquired. The SNR threshold is determined based on the noise figure of the acquisition device. Adjustments have been made, and the adjustment rules are as follows:

[0055] in, The noise figure of the data acquisition device, in units of 1. This ensures the completeness and accuracy of feature parameter extraction.

[0056] This design specifies the noise signal acquisition frequency band, sets an effective threshold for noise spectrum feature extraction, and adjusts it according to the noise figure of the acquisition device to cover the low-frequency to mid-frequency range and comprehensively acquire noise signals. Setting a signal-to-noise ratio threshold ensures that feature parameters are extracted when the effective signal is strong, avoiding noise interference. Adjusting the threshold according to the noise figure of the acquisition device makes the standard adaptable to different devices, ensuring complete and accurate feature parameter extraction and improving sensor calibration accuracy.

[0057] In one embodiment, the iterative convergence threshold of the Newton-Raphson inversion algorithm is set to... This threshold is adaptively determined based on the sensor's calibration accuracy requirements, and the determination rule is as follows: Calibration accuracy requirements When the pure temperature theoretical value obtained from two consecutive iterations Deviation ≤ When the iteration converges, the calculation is stopped to ensure that the calculation accuracy of stress decoupling matches the calibration requirements of the sensor.

[0058] This design adaptively determines the iterative convergence threshold based on the sensor calibration accuracy requirements. Convergence is determined when the deviation between two adjacent iterations of the pure temperature theoretical value meets the condition, thus matching the iterative convergence criterion with the sensor calibration accuracy requirements. This avoids excessive iteration that wastes resources or insufficient iteration that leads to inaccurate calculations. It also ensures that the stress decoupling calculation accuracy meets the sensor calibration requirements, improving the accuracy of temperature measurement by the sensor in complex environments.

[0059] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A high-precision temperature sensor calibration method based on cryogenic environment, characterized in that, Includes the following steps: S1. Acquire the raw detection data and environmental perception data of the temperature sensor in a cryogenic environment. The raw detection data includes the sensor's electrical output data and noise signal data, while the environmental perception data includes the sensor's micro-deformation data and the packaging cavity's characteristic signal data. S2. The zero-drift self-reference calibration method based on the inversion of quantum tunneling noise spectrum characteristics is used to correct the zero-point drift of the sensor and eliminate the zero-point offset error caused by the quantum tunneling effect in the cryogenic environment. S3. By using the cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model, the contributions of temperature and mechanical stress to the sensor output are separated and the temperature reading is dynamically corrected, eliminating stress interference caused by thermal shrinkage of the packaging material. S4. A self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points is adopted to establish a self-evolving calibration model and update calibration parameters, thereby completing high-precision calibration of the sensor and realizing dynamic optimization of the calibration model.

2. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 1, characterized in that, The implementation of the zero-drift self-reference calibration method based on quantum tunneling noise spectrum feature inversion in step S2 includes the following steps: The temperature sensor was placed in a cryogenic steady-state environment with a temperature of ≤-150℃. The noise signal at the metal-semiconductor interface of the sensor was collected and the noise power spectral density function was obtained by power spectral density analysis. Core feature parameters are extracted from the low-frequency range of the function, and a nonlinear mapping model of tunneling noise spectrum-absolute temperature offset is trained based on the corresponding samples of noise spectrum feature parameters and temperature offset. The core feature parameters are input into the model to obtain the absolute zero drift, and the original temperature value detected by the sensor is corrected based on the drift.

3. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 2, characterized in that, The core feature parameters include: Rate of change of slope in the low-frequency region Inflection point frequency and peak noise of energy level transition Low-frequency region slope change rate The logarithm of the noise power spectral density function and the logarithm of the frequency are in The first derivative of an interval; The training of the nonlinear mapping model of tunneling noise spectrum-absolute temperature offset involves normalizing the core feature parameters, selecting a neural network to build the basic model and setting appropriate training parameters to complete the training. After verification and optimization, the model parameters are solidified and embedded into the sensor calibration algorithm to support online inference calculation.

4. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 3, characterized in that, The criteria for determining the cryogenic steady-state environment are as follows: Set temperature fluctuation thresholds and noise spectrum abrupt change judgment thresholds; When the fluctuation range of the original temperature detection value of the sensor is less than or equal to the temperature fluctuation threshold within the preset continuous detection period, and the rate of change of the power spectral density function of the acquired noise signal in the whole frequency band is less than or equal to the noise spectrum change judgment threshold, it is determined to be a cryogenic steady-state environment. The temperature fluctuation threshold is determined based on the sensor's nominal accuracy class, which is [value missing]. Level when taking The nominal accuracy is lower than Level according to nominal accuracy Values.

5. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 2, characterized in that: The noise signal acquisition frequency band covers the low-frequency to mid-frequency range. A threshold for effective extraction of noise spectrum features is set. Core feature parameters are extracted only when the signal-to-noise ratio of the noise signal in the target frequency band is ≥ the threshold; otherwise, the noise signal is reacquired. The signal-to-noise ratio threshold is dynamically adjusted based on the noise figure NF of the acquisition device. The adjustment rule is to add the basic signal-to-noise ratio threshold to the noise figure NF of the acquisition device.

6. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 1, characterized in that, The implementation of the cryogenic thermal-stress-resistance three-field coupling decoupling inversion calibration model in step S3 includes the following steps: Based on the material physical properties of the temperature sensor packaging material and the sensing element, a thermal shrinkage coefficient difference matrix model is constructed to characterize the difference in thermal shrinkage between the two in different directions; The minute deformation of the sensitive element is obtained by detecting the element and converted into a thermal stress value; A three-dimensional coupled equation system of temperature-stress-electrical output is established. The equation system is solved iteratively using a numerical inversion algorithm to separate the pure temperature contribution and the resistance value of the sensitive element under no stress interference, and the theoretical value corresponding to pure temperature is obtained by back-reasoning. Based on this theoretical value, the temperature value after zero drift correction is dynamically stress decoupling corrected.

7. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 6, characterized in that: The numerical inversion algorithm is the Newton-Raphson inversion algorithm. An iterative convergence threshold is set for this algorithm, which is adaptively determined based on the sensor's calibration accuracy requirements. When the deviation between the pure temperature theoretical values ​​obtained from two consecutive iterations is less than or equal to the iteration convergence threshold, the iteration is considered to have converged and the calculation is stopped. In the dynamic stress decoupling correction process, the stress decoupling correction weight coefficient is adaptively determined based on the stress sensitivity coefficient of the sensor's sensitive element. The larger the absolute value of the stress sensitivity coefficient, the closer the correction weight coefficient is to 1.

8. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 1, characterized in that, The implementation of the self-evolving calibration method based on dynamic capture of cryogenic phase transition reference points described in step S4 includes the following steps: A micro-medium storage cavity is pre-set inside the temperature sensor packaging cavity and a high-purity reference medium is introduced. The reference medium has at least three known and stable cryogenic phase transition temperatures, and each phase transition temperature uniformly covers the target cryogenic calibration range of the sensor. The characteristic signals inside the encapsulation cavity are collected in real time by the deployed detection elements. A threshold for judging abrupt changes in the characteristic signals is set. Based on the threshold, it is determined whether the reference medium has undergone a cryogenic phase change. The real-time detection value of the sensor corresponding to the transient characteristic point of the phase change is captured and paired with the standard phase change temperature of the reference medium to obtain a calibration sample pair. The temperature value after decoupling and correction of the three-field coupling and the historical calibration data of the sensor are used as input. A self-evolutionary calibration model is trained by machine learning algorithm, and the detection value is accurately calibrated by nonlinear interpolation algorithm. By continuously capturing new phase transition feature points during the daily detection process of sensors, supplementing the calibration sample set, and updating the model parameters through incremental learning, the calibration model can achieve self-evolutionary optimization.

9. The high-precision temperature sensor calibration method based on cryogenic environment according to claim 8, characterized in that: The reference medium is one or more combinations of high-purity rare gas, organic microcrystalline material or inorganic low-melting-point microcrystalline material, and its cryogenic phase transition type includes adsorption-desorption phase transition, glass transition or solid phase crystal form transition. Set an effective threshold for the phase transition signal of the reference medium and dynamically adjust the threshold according to the sensor usage time to ensure that the phase transition feature points can still be accurately captured after long-term use of the sensor. The machine learning algorithm is a gradient boosting regression tree algorithm. Before model training, outlier removal and standardization are performed on calibration sample pairs. Model initialization parameters and training process are set, and regularization constraints are added to prevent overfitting. When the mean absolute error of the test set reaches the preset standard, the model is deemed to be qualified for training. When the proportion of new samples is greater than or equal to the preset proportion, the model is triggered to retrain. The parameters of the new model and the original model are fused through a weight fusion formula.

10. The high-precision temperature sensor calibration method based on a cryogenic environment according to any one of claims 1 to 9, characterized in that: The cryogenic environment is a temperature In low-temperature environments, the target cryogenic calibration range of the sensor is consistent with the temperature range of the cryogenic environment.

Images