Stability evaluation device and method for solid-state direct-current voltage standard source based on digital twinning

Abstract

The application discloses a solid-state direct-current voltage standard source stability evaluation device and method based on digital twinning, and belongs to the technical field of electrical measurement and testing, and comprises an accelerated aging test module, a temperature and humidity monitoring module, a voltage output monitoring module, a digital twinning simulation module, a digital twinning model domestication and evaluation module, and a real device verification and incremental learning module. By constructing a coupled digital twinning model containing Zener reference aging, resistance network aging, operational amplifier drift and temperature field control, combining accelerated aging test and short-term measured data to identify parameters and adaptively tame the model, and then predicting annual stability indicators through virtual simulation deduction, and using subsequent real device verification data for incremental learning and iterative optimization, the application shortens the traditional 36-month evaluation period to 2-3 months, and the prediction relative error is better than 5%, which significantly improves the efficiency and accuracy of the solid-state direct-current voltage standard source stability evaluation.

Description

Technical Field

[0001] This invention belongs to the field of electrical measurement and testing technology, and in particular relates to a device and method for evaluating the stability of a solid-state DC voltage standard source based on digital twins. Background Technology

[0002] Solid-state DC voltage standard sources are core reference devices for the transmission and traceability of DC voltage values ​​in the field of electrical metrology. The long-term stability of their output voltage is a key performance indicator. High-performance solid-state DC voltage standard sources generally require an annual stability of better than 2 ppm (parts per million). Typical output ranges include standard voltage values ​​such as 10 V, 1 V, and 1.018 V, which can meet the high-precision requirements of DC reference voltage in scenarios such as precision electronic measurement and instrument calibration.

[0003] The core components of existing high-performance solid-state DC voltage standard sources include: a Zener reference device (with an annual stability of up to 3 ppm), a high-precision operational amplifier circuit, a low-drift DC voltage divider resistor network, and a precision temperature control system. Among these, the breakdown characteristics of the Zener reference, the stability of the resistor network's voltage division ratio, the operational amplifier's offset drift, and the temperature fluctuations of the temperature control system are the main factors affecting long-term stability.

[0004] Currently, the traditional method for evaluating the annual stability of solid-state DC voltage standard sources is the "36-month continuous real-time monitoring method," which involves continuously recording the output voltage values ​​over a long period and taking the ratio of the difference between the maximum and minimum values ​​to the average value to obtain the annual stability index. This method has the following insurmountable drawbacks, which have not yet been effectively resolved: The evaluation cycle is too long: it takes 36 months of continuous monitoring to obtain the final indicators, which seriously slows down the progress of product development iteration, finalization and mass production, and fails to expose design defects in a timely manner; Limited prediction accuracy: Based solely on the difference between the maximum and minimum values ​​of long-term measured data, without considering the degradation mechanism of multi-physical field coupling such as electricity, heat, and time, it is impossible to predict long-term drift trends in advance; The testing costs are high: it requires dedicated space, equipment and manpower for long-term monitoring, the data utilization rate is low, and it is not suitable for rapid performance evaluation of batch products. Lack of digital evaluation capabilities: relying solely on actual testing, there is a lack of effective data feedback and model correction mechanisms, making it impossible to continuously improve evaluation accuracy based on actual test data.

[0005] Since its initial proposal by Michael Grieves in 2002, digital twin technology has been widely applied in fields such as power equipment, new energy transformer substations, rail transit power supply and distribution, and water conservancy projects. Its core is to construct a virtual mirror image of a physical entity, enabling status monitoring, trend prediction, and health management through real-time data interaction. For example, invention patent CN121189199B discloses a digital twin assessment method for measuring the wear state of switch contacts. By constructing a digital twin model of contact wear, it achieves accurate prediction of the remaining lifespan of the contacts. Publicly available patents such as CN121431912A, CN121440561A, and CN119377762A also apply digital twins to scenarios such as state estimation and control of new energy transformer substations, operation and maintenance management of railway power supply and distribution, and parameter determination for straight-line traffic conversion. None of these patents address the construction of a dedicated degradation mechanism model and a rapid long-term stability prediction method for solid-state DC voltage standard sources with ppm-level accuracy. Furthermore, none of them disclose a technical solution for applying the closed-loop architecture of digital twins—"model construction—experimental training—accelerated simulation—practical verification"—to this field. In summary, existing technologies lack a rapid evaluation method that can combine digital twin technology, accelerated aging tests, and high-precision parameter identification to accurately predict the annual stability of solid-state DC voltage standard sources within weeks to months, and possess model self-learning capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide a device and method for evaluating the stability of a solid-state DC voltage standard source based on digital twins, which solves the problems of long annual stability evaluation cycle, low efficiency, incomplete mechanism modeling, uncontrollable prediction accuracy, and lack of self-optimization capability in the existing technology.

[0007] To achieve the above objectives, this invention provides a stability evaluation device for a solid-state DC voltage standard source based on digital twins, including an accelerated aging test module, a temperature and humidity monitoring module, a voltage output monitoring module, a digital twin simulation module, a digital twin model training and evaluation module, and an implementation verification and incremental learning module. Each module achieves synchronous communication and data interaction through a data bus. Wherein: The accelerated aging test module is used to apply programmable temperature cycling stress and excess electrical stress to the voltage reference device inside the solid-state DC voltage standard source to accelerate the performance degradation of the device. The temperature control range is -15℃ to 150℃, the temperature control accuracy is not less than ±0.05℃, and the excess electrical stress is 1.2 times the rated operating voltage. The temperature and humidity monitoring module is used to collect and store the temperature and humidity time series data of the environment in which the solid-state DC voltage standard source under test is located in real time. The sampling interval is continuously adjustable in the range of 1 second to 10 minutes, and the sampling accuracy meets the requirements of temperature ±0.01℃ and humidity ±2% RH. The voltage output monitoring module is used to continuously acquire the three output voltage values ​​of a solid-state DC voltage standard source (10 V, 1 V, and 1.018 V) with a measurement accuracy of not less than 0.9 ppm, and synchronously record the original voltage drift data and the acquisition timestamp. The digital twin simulation module is used to construct a fully coupled digital twin model that includes a Zener reference device aging sub-model, a DC resistor voltage divider network aging sub-model, an operational amplifier circuit drift sub-model, and an accelerated aging test constant temperature device temperature field control sub-model, and includes a multi-model coupling relationship matrix between each sub-model. The digital twin model taming and evaluation module is used to identify parameters and adaptively correct the digital twin model based on the measured data collected by the temperature and humidity monitoring module and the voltage output monitoring module, complete the model taming, and output the predicted annual stability index based on the tamed digital twin model through virtual simulation. The implementation verification and incremental learning module is used to compare and verify the predicted annual stability index with the actual annual stability index monitored continuously for the following 12 months, calculate the prediction error, and feed the verification results back to the digital twin model to complete the incremental learning and iterative optimization of the model parameters.

[0008] Preferably, the digital twin simulation module adopts a multi-fidelity modeling strategy, using a high-precision finite element model for the voltage reference device and a reduced-order mechanism model for the DC resistor voltage divider network and operational amplifier circuit.

[0009] Preferably, the aging sub-model of the Zener reference device is constructed based on the Arrhenius model and the Zener reference breakdown voltage drift mechanism, and the Zener reference is a high-stability Zener reference device.

[0010] Preferably, the digital twin model taming and evaluation module adopts a hybrid approach of physical mechanism and deep neural network to identify and correct key parameters of the digital twin model online. The convergence criterion for digital twin model taming is that the mean square error between the model output and the measured data is less than 0.2 ppm.

[0011] Preferably, the digital twin model taming and evaluation module has an active learning function. When the model prediction uncertainty exceeds a set threshold of 0.3 ppm, it automatically recommends to carry out supplementary tests under specific temperature ranges or stress conditions to obtain optimized taming data.

[0012] Preferably, the voltage output monitoring module is based on a high-precision digital multimeter with 8.5 bits or more, combined with a multi-channel programmable scanning switch and signal conditioning circuit.

[0013] This invention also provides a method for evaluating the stability of a solid-state DC voltage standard source based on digital twins, comprising the following steps: Step S1: Based on the physical structure and failure mechanism of the solid-state DC voltage standard source under test, construct a full-state digital twin model containing multiple sub-models and coupling relationship matrices, and initialize the model parameters; Step S2: Place the solid-state DC voltage standard source under test into the accelerated aging test module, apply temperature cycling stress and excess electrical stress, run the accelerated aging test, and simultaneously collect ambient temperature and humidity and output voltage drift data during the test. Step S3: Place the solid-state DC voltage standard source under test in the rated working environment and continuously collect measured data of ambient temperature and humidity and output voltage over a preset time period; Step S4: Input the short-term measured data collected in step S3 into the digital twin model training and evaluation module, and use a hybrid driving method to iteratively correct the parameters of the digital twin model until the model output and the measured data converge, thus completing the model training. Step S5: Based on the tamed digital twin model, input historical annual environmental data, perform a one-year virtual simulation, and obtain the predicted annual stability index by analyzing the voltage time series output by the simulation. Step S6: Collect the actual output voltage and environmental data of the solid-state DC voltage standard source under test for the next 12 months, calculate the true annual stability index, and compare it with the predicted index obtained in step S5 to verify the prediction error. Step S7: Feed the prediction error data obtained in step S6 back to the digital twin model, iteratively update the model parameters, and complete incremental learning; if the prediction error exceeds the preset threshold, return to step S3 to collect additional measured data and retrain the model.

[0014] Preferably, in step S4, model training uses the residual between the digital twin model output and the measured output as the observation, and the key parameters for iterative updating include aging rate, temperature coefficient, and drift coefficient.

[0015] Preferably, the temperature cycling stress in step S2 is a stepped cycle, specifically: -15℃ constant temperature for 24 hours → temperature rise to 25℃ constant temperature for 24 hours → temperature rise to 150℃ constant temperature for 24 hours → temperature drop to 25℃ constant temperature for 24 hours → temperature drop to -15℃ constant temperature for 24 hours, cycled 5 times.

[0016] Preferably, in step S5, the annual stability index is obtained by fitting the voltage output drift time series using linear regression, and the prediction accuracy is greater than [predictive accuracy]. .

[0017] Therefore, the present invention employs the above-described digital twin-based solid-state DC voltage standard source stability evaluation device and method, which has the following beneficial effects: (1) The evaluation cycle is greatly shortened: the traditional annual stability assessment cycle of 36 months is shortened to 2-3 months, which significantly improves the efficiency of R&D, finalization and batch testing of solid DC voltage standard sources and solves the core pain point of slowing down product iteration by traditional methods.

[0018] (2) High accuracy and high reliability: The dual-drive mode of "physical mechanism modeling + measured data training" is adopted. It integrates the multi-physical field coupling degradation mechanism and combines the high-precision parameter identification capability of deep neural network hybrid drive algorithm. The prediction relative error is better than 5%, which is much higher than the traditional statistical extrapolation method. It can accurately reproduce the long-term drift trend of the equipment and meet the ppm-level measurement accuracy requirements.

[0019] (3) The model is tame and can evolve: Through the closed-loop feedback of the actual verification data, the incremental learning of the digital twin model is realized. The model evaluation accuracy continues to improve with the accumulation of data. It can be adapted to different models and batches of solid DC voltage standard sources and has strong versatility.

[0020] (4) Significantly reduced testing costs: Reduced reliance on long-term on-site testing, no need for long-term on-site personnel, reduced equipment occupation and labor costs, and accelerated aging test to quickly stimulate equipment degradation characteristics, further shortening the test cycle, suitable for rapid performance evaluation of batch products.

[0021] (5) Innovative architecture adapted to metrology scenarios: It pioneered a four-in-one evaluation architecture of "accelerated testing + digital twin + practical verification + incremental learning", and applied digital twin technology and model taming concept to the high-precision metrology scenario of solid-state DC voltage standard source for the first time, filling the technological gap in this field.

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the stability evaluation device for a solid-state DC voltage standard source based on digital twins according to the present invention. Figure 2 This is a flowchart illustrating the stability evaluation method for a solid-state DC voltage standard source based on digital twins, as implemented in this invention. Figure 3 The stepped temperature stress loading curve used in the accelerated aging test of this invention is shown in the embodiment of the invention. Figure 4 This is a comparison between the output curves of the digital twin model before and after taming and the measured voltage drift curves in an embodiment of the present invention. Figure 5 This is a comparison chart of the evaluation cycle between the conventional evaluation method and the evaluation method of the present invention in this embodiment of the invention. Detailed Implementation

[0024] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] Please see Figures 1 to 5 A stability evaluation device for a solid-state DC voltage standard source based on digital twins includes an accelerated aging test module, a temperature and humidity monitoring module, a voltage output monitoring module, a digital twin simulation module, a digital twin model training and evaluation module, and an on-site verification and incremental learning module. Each module achieves synchronous communication and data interaction via a data bus. The accelerated aging test module includes a programmable high and low temperature test chamber and a test control unit. The high and low temperature test chamber has a temperature control range of -40℃ to 150℃, a temperature fluctuation of ≤±0.05℃, and a temperature change rate of ±3℃ / min. It can apply cyclic temperature stress through stepped heating, constant temperature holding, and stepped cooling. Under preset temperature shock conditions, a circuit consisting of an aligned reference, a voltage divider network, and a precision operational amplifier is used to apply 1.2 times the rated operating voltage, accelerating the performance degradation of the voltage reference device, the DC resistive voltage divider network, and the operational amplifier circuit, rapidly activating the long-term drift characteristics of the equipment, and shortening the test cycle. In summary, the accelerated aging test module is used to apply programmable temperature cyclic stress and excess electrical stress to the voltage reference device inside a solid-state DC voltage standard source, accelerating device performance degradation. The temperature control range is -15℃ to 150℃, with a temperature control accuracy of no less than ±0.05℃, and the excess electrical stress is 1.2 times the rated operating voltage.

[0026] The temperature and humidity monitoring module consists of a high-precision temperature and humidity sensor, a data acquisition module, and a storage unit. The sensor is deployed in the internal cavity (close to the reference device and the constant temperature cavity) and the outer surface of the solid-state DC voltage standard source under test. The sampling interval is continuously adjustable from 1 second to 10 minutes. It collects and stores temperature and humidity time-series data in real time, with sampling accuracy meeting the requirements of: temperature ±0.01℃ and humidity ±2% RH, providing accurate environmental boundary condition input for the digital twin model; it is used to collect and store the temperature and humidity time-series data of the environment in which the solid-state DC voltage standard source under test is located in real time.

[0027] The voltage output monitoring module, with a high-precision digital multimeter (8.5 digits) as its core and a multi-channel low thermoelectric potential programmable switch, sequentially acquires three typical output voltage values ​​of 10 V, 1 V and 1.018 V from the solid-state DC voltage standard source under test. The measurement accuracy is no less than 0.9 ppm. It also records the original voltage drift data and acquisition time simultaneously to ensure the integrity and traceability of the data.

[0028] The digital twin simulation module is used to construct a fully coupled digital twin model, including an aging sub-model of a Zener reference device, an aging sub-model of a DC resistive voltage divider network, an operational amplifier circuit drift sub-model, and a temperature field control sub-model of an accelerated aging test isothermal device. It also includes a multi-model coupling relationship matrix between the sub-models. The digital twin simulation module employs a multi-fidelity modeling strategy, using a high-precision finite element model for the voltage reference device and a reduced-order mechanism model for the DC resistive voltage divider network and operational amplifier circuit. A detailed analysis of the sub-models and coupling relationships is as follows: (1) Aging sub-model of Zener reference device: Based on the Allennis model and the drift mechanism of Zener reference breakdown voltage, combined with the characteristic parameters of high-stability Zener reference device, an aging equation under temperature-current-time multidimensional stress is established to accurately describe the long-term drift law of reference device. (2) DC resistive voltage divider network aging sub-model: Considering the thermal relaxation effect and time aging characteristics of the resistive film, a mathematical model of the resistance value changing with temperature and time is established. (3) Operational amplifier circuit drift sub-model: Based on the temperature characteristics of the input bias current and offset voltage of the low-drift operational amplifier, a model of the influence of operational amplifier drift on the output voltage is established; (4) Temperature field control sub-model of constant temperature device for accelerated aging test: simulate the temperature field distribution and dynamic response characteristics of constant temperature chamber under PID control, and accurately reproduce the control hysteresis and temperature fluctuation of constant temperature system; (5) Multi-module coupling relationship matrix: Establish the interaction relationship between each sub-model and quantify the comprehensive impact of multi-physical field coupling such as electricity, heat, and time on the long-term stability of the equipment.

[0029] The digital twin model training and evaluation module is used to identify and adaptively correct parameters of the digital twin model based on measured data collected by the temperature and humidity monitoring module and the voltage output monitoring module, thus completing model training. Based on the trained digital twin model, it outputs predicted annual stability indicators through virtual simulation. The module employs a hybrid approach of physical mechanisms and deep neural networks to identify and correct key parameters of the digital twin model online. The convergence criterion for digital twin model training is that the mean square error between the model output and the measured data is less than 0.2 ppm. Furthermore, the module has an active learning function; when the model prediction uncertainty exceeds a set threshold of 0.3 ppm, it automatically recommends supplementary experiments under specific temperature ranges or stress conditions to obtain optimized training data. Specifically, the digital twin model training and evaluation module is mounted on a high-performance computing server, integrating a hybrid driving algorithm of physical mechanisms and deep neural networks for parameter identification and adaptive correction of the digital twin model. Specifically, using 60 days of measured data collected by the temperature and humidity monitoring module and the voltage output monitoring module, the deviation between the digital twin model output and the measured output of the physical entity is used as an observation value. Key parameters in the model, such as aging rate, temperature coefficient, and drift coefficient, are iteratively updated until the mean square error between the model output and the measured data is less than 0.2 ppm, completing model training. Based on the trained digital twin model, one year of operational cycle experience environmental data can be input to accelerate simulation and extrapolate the performance degradation trajectory of the equipment, outputting predicted annual stability indicators.

[0030] The implementation verification and incremental learning module is used to compare and verify the annual stability index predicted by the digital twin model with the actual annual stability index monitored continuously for the next 12 months, calculate the prediction error and generate a verification report; at the same time, the verification results are fed back to the digital twin model to complete the incremental learning and iterative optimization of the model parameters, so that the model evaluation accuracy continues to improve with the accumulation of data.

[0031] Please see Figure 2 A stability evaluation method for a solid-state DC voltage standard source based on digital twins includes the following steps: Step S1, Model Construction: Based on the physical structure, component parameters and failure mechanism of the solid-state DC voltage standard source under test, and combined with the characteristics of the high-stability Zener reference device, a full-state digital twin model is constructed, including the Zener reference device aging sub-model, the DC resistance voltage divider network aging sub-model, the operational amplifier circuit drift sub-model, the accelerated aging test constant temperature device temperature field control sub-model, and the multi-model coupling relationship matrix. The model parameters are then initialized.

[0032] Step S2, Accelerated Aging Test: Place the solid-state DC voltage standard source under test into the accelerated aging test module, set the temperature cycle to -15℃→25℃→150℃→25℃→-15℃, hold each temperature for 24 hours, cycle 5 times, and simultaneously apply 1.2 times the rated operating voltage. Run the accelerated aging test, and synchronously collect temperature, humidity, and output voltage drift data during the test. The duration is 4 weeks. Figure 3 As shown.

[0033] Step S3, Short-term test: Place the solid-state DC voltage standard source under test in the rated working environment (temperature 23±2℃, humidity 45%~65%RH). Through the temperature and humidity monitoring module and the voltage output monitoring module, continuously collect the measured data of ambient temperature and humidity and output voltage for 60 days to ensure that the data covers the typical working conditions of the equipment.

[0034] Step S4, Model Taming: Input the short-term measured data collected in Step S3 into the digital twin model taming and evaluation module. Use an integrated physical mechanism and deep neural network hybrid driving algorithm to construct a state transition matrix with the residual between the digital twin model output and the measured output as the observation. Iteratively update key parameters such as the aging rate, temperature coefficient, and drift coefficient of the model until the model output and the measured data converge (mean square error < 0.2 ppm), thus completing the model taming.

[0035] Step S5, Simulation Evaluation: Based on the tamed digital twin model, input the historical annual measured environmental time series data, simulate the normal working state of the equipment within a 1-year cycle, perform virtual accelerated simulation, collect the voltage output time series during the simulation process, and obtain the predicted annual stability index S through linear regression fitting.

[0036] Step S6, Practical Verification: Put the solid-state DC voltage standard source under test into normal operation, continuously collect actual output voltage and environmental data for 12 months, and calculate the true annual stability index. , with S Compare the results, verify the prediction errors, and generate a verification report. Step S7: Feed the prediction error data obtained in step S6 back to the digital twin model, iteratively update the model parameters, complete incremental learning, and continuously improve the evaluation accuracy of the model; if the relative prediction error exceeds 5%, return to step S3 to collect additional measured data and retrain the model.

[0037] Example 1 This embodiment provides a complete device and method for evaluating the stability of a solid-state DC voltage standard source based on digital twins.

[0038] 1. Device Composition: like Figure 1As shown, the apparatus of this embodiment includes: (1) Temperature and humidity monitoring module: The SHT30 high-precision temperature and humidity sensor is used and is installed in the constant temperature cavity inside the device under test and the external environment. The sampling interval is 30 s, the temperature accuracy is ±0.01℃, the humidity accuracy is ±2%RH, the data is stored in the SD card, and it supports real-time upload to the computing server.

[0039] (2) Voltage output monitoring module: The FLUKE8508A digital multimeter is used in conjunction with the Agilent 34970A multi-channel scanning switch. The measurement accuracy is 0.9 ppm. The three outputs of 10 V, 1 V and 1.018 V are collected in sequence with a 1-minute interval. The data is transmitted to the computing server through the LAN interface.

[0040] (3) Accelerated aging test module: The ESPEC high and low temperature test chamber is used, with a temperature range of -40℃ to 150℃ and a temperature control accuracy of ±0.05℃. The electric stress loading circuit uses a programmable DC power supply with an output voltage range of 0 to 20V, which can be adjusted to 1.2 times the rated working voltage.

[0041] (4) Digital twin simulation module: A full-state coupled model is built using MATLAB, integrating the Arrhenius aging model and the PID temperature control model.

[0042] (5) Digital twin model taming and evaluation module: Based on MATLAB, a deep neural network hybrid driving algorithm program is written to realize online identification and correction of model parameters. The convergence criterion is set as the mean square error between the model output and the measured data <0.2 ppm.

[0043] (6) Implementation verification and incremental learning module: Develop dedicated data comparison software to automatically calculate prediction error, generate verification report, and feed back deviation data to digital twin model to complete parameter update.

[0044] 2. Digital Twin Model Construction: Based on the hardware structure of the solid-state DC voltage standard source under test, the following sub-models are constructed: (1) Aging sub-model of Zener reference device: LTZ1000 is used as the reference device, and the aging equation is established based on the Arrhenius model: ,in This is the reference voltage drift. Thermodynamic temperature (K) For time (h), This is the proportionality coefficient. To activate energy, Boltzmann's constant, The aging index is set based on the LTZ1000 datasheet.

[0045] (2) DC resistive voltage divider network aging sub-model: Considering the thermal relaxation effect of the resistive film layer, a model of the resistance value changing with time is established: ,in This is the initial resistance value. This is the time aging factor. For temperature coefficient, This represents the change in temperature.

[0046] (3) Operational amplifier circuit drift sub-model: Based on the parameters of operational amplifier OP07, an offset voltage drift model is established: ,in The initial offset voltage, For time drift rate, This is the temperature coefficient.

[0047] (4) Temperature field control sub-model of the constant temperature device for accelerated aging test: Simulate PID control algorithm and establish dynamic response model of temperature field in constant temperature chamber: ,in To set the temperature, The initial temperature. This is the response time constant.

[0048] (5) Multi-model coupling relationship matrix: quantifies the interaction between sub-models, accelerates the output temperature of the temperature field control sub-model of the aging test constant temperature device to affect the drift of the reference device, resistor network and operational amplifier, the aging of the resistor network affects the output voltage division ratio, and finally integrates to obtain the overall output voltage drift model of the device.

[0049] 3. The model taming and evaluation process is as follows: (1) Accelerated test: The solid DC voltage standard source was placed in the high and low temperature test chamber and the temperature cycle stress was set (-15℃ constant temperature for 24h → temperature rise to 25℃ constant temperature for 24h → temperature rise to 150℃ constant temperature for 24h → temperature drop to 25℃ constant temperature for 24h → temperature drop to -15℃ constant temperature for 24h), 5 cycles were performed, the electrical stress was set to 1.2 times the rated working voltage, the test duration was 28 days, and environmental and voltage data were collected simultaneously.

[0050] (2) Short-term test: The solid DC voltage standard source was placed in the rated environment of 23℃ and 55% RH, and temperature, humidity and output voltage data were continuously collected for 60 days with a 1-minute interval between collections.

[0051] (3) Model taming: Input 60 days of measured data into the deep neural network hybrid driving algorithm, use the deviation between the model output and the measured output as the observation, construct a state transition matrix (including unknown parameters such as aging rate and temperature coefficient), iteratively update the parameter estimates, and after 1000 iterations, the mean square error between the model output and the measured data drops to 0.15 ppm, completing model taming; if the mean square error between the model output and the measured data is greater than 0.15 ppm, analyze the temperature range involved in the large mean square error, prompt to repeat step (1), supplement the accelerated test data under the temperature range and stress conditions, until the mean square error drops to 0.15 ppm. Figure 4 As shown.

[0052] (4) Simulation evaluation: Input the historical annual environmental temperature spectrum (covering the temperature changes of the four seasons), perform a one-year virtual simulation based on the acclimated model, obtain the voltage output drift time series, and obtain the predicted annual stability index through linear regression fitting. .

[0053] (5) Actual verification: The equipment under test is put into normal operation and continuously monitored for 12 months to calculate the true annual stability index. The prediction relative error is 4.26%, which meets the requirement of being better than 5%.

[0054] (6) Incremental learning: Feed the prediction error data back to the digital twin model to update the aging index in the model. With temperature coefficient The optimized model further improved prediction accuracy, with the mean squared error reduced to 0.12 ppm.

[0055] Example 2 This embodiment is a further optimization of Embodiment 1, mainly focusing on improving the digital twin model and the model taming method. Specifically: 1. The digital twin model adopts a multi-fidelity modeling strategy: a high-precision finite element model is used for the voltage reference device (LTZ1000) to accurately simulate its internal temperature distribution and breakdown characteristics; a reduced-order mechanism model is used for the DC resistor voltage divider network and operational amplifier circuit to reduce the computational complexity of the model while ensuring the evaluation accuracy, thereby improving the simulation efficiency by more than 30%.

[0056] 2. Model training adopts a hybrid approach of physical mechanism and deep neural network: Based on the physical model parameters identified by the hybrid deep neural network driving algorithm, the BP neural network is trained with measured data to compensate for the modeling error of the physical model, so that the mean square error between the model output and the measured data is reduced to less than 0.1 ppm, thereby further improving the prediction accuracy.

[0057] Example 3 This embodiment, as another optimization of Embodiment 1, focuses on adding an active learning function to the digital twin model training and evaluation module. Specifically, this embodiment sets the model prediction uncertainty threshold to 0.3 ppm. When the model prediction uncertainty exceeds this threshold, the system automatically analyzes the source of uncertainty (such as insufficient data in a certain temperature range) and recommends conducting specific supplementary experiments (such as long-term monitoring for 7 days in a temperature range of 40℃~60℃) to obtain the most valuable training data. Practice has proven that after adopting the technical solution of this embodiment, the model training cycle is shortened from 60 days to 30 days, while improving model calibration efficiency and prediction reliability.

[0058] Example 4 This embodiment serves as a verification scheme for Embodiment 3, aiming to systematically verify the actual effect of the active learning function in the digital twin model taming and evaluation module through measured data. Starting with the model from Embodiment 1 that has completed initial taming but has high prediction uncertainty, this embodiment focuses on comparing the differences in the cycle required for model taming, correction efficiency, and final prediction accuracy when the active learning function is enabled versus disabled. Specifically: 1. In this embodiment, two solid-state DC voltage standard sources of the same model (numbered #01 and #02) were used to complete the initial data acquisition under the same accelerated aging test conditions (same as in embodiment 1), and then the model taming stage was entered.

[0059] 2. No active learning function is introduced. The control group (device #01) adopts the conventional taming process in Example 1, that is, 60 consecutive days of measured data in the rated environment are collected at one time, and all data are used for model parameter identification and correction until the model converges.

[0060] 3. Introduction of Active Learning Function. For experimental group #02 equipment, a digital twin model was initially trained based on the measured data from the first 15 days. After training, the mean squared error (MSE) between the model output and the measured data was 0.25 ppm, meeting the convergence criterion (<0.2 ppm). However, the model prediction uncertainty was assessed at 0.35 ppm, exceeding the set threshold of 0.3 ppm. The system automatically triggered the active learning function to analyze the sources of model uncertainty. By analyzing the parameter sensitivity and data coverage range of each sub-module of the model, the system identified the main source of uncertainty as the sparse measured data points used for training within the temperature range of 40℃ to 60℃. This resulted in insufficient accuracy in identifying the temperature coefficient of the Zener reference device aging sub-model within this range. Based on the above analysis, the system automatically generated a recommended solution: "It is suggested to conduct long-term constant temperature monitoring for 7 days at each temperature point within the temperature range of 40℃ to 60℃, with a step size of 10℃, to obtain high-value compliance data for optimizing the model parameters in this temperature range." From day 16 to 30, according to the system recommendation, in the accelerated aging test module, three temperature points were set: 40℃, 50℃, and 60℃, each undergoing a 7-day constant temperature stress test (electrical stress maintained at 1.2 times the rated voltage), while simultaneously collecting output voltage drift data. After fusing the newly acquired data with the previous data, model compliance was re-performed. At this point, the MSE between the model output and the measured data decreased to 0.12 ppm, and the prediction uncertainty assessment value decreased to 0.22 ppm, below the set threshold, and the active learning process ended.

[0061] 4. Comparison of Experimental Results and Data Analysis. Table 1 shows a comparison of the key process indicators and outcome indicators between the control group and the experimental group: Table 1: Comparison of Indicators

[0062] Through the above comparative experiments, after adopting the active learning function, the taming period of the experimental group #02 device was shortened from 60 days in the control group to 30 days, verifying that this function can shorten the model taming period by 50%. Not only was the taming period shorter, but its final MSE (0.12 ppm) and prediction uncertainty (0.22 ppm) were also better than those of the control group. In summary, this embodiment, through rigorous controlled experiments based on measured data, fully demonstrates that the active learning function described in Embodiment 3 can achieve better model taming results in a shorter time, significantly improving the efficiency and reliability of the entire evaluation method.

[0063] Therefore, this invention employs the aforementioned digital twin-based solid-state DC voltage standard source stability assessment device and method. The core of this invention lies in constructing a fully coupled digital twin model and adopting a closed-loop architecture of "accelerated testing—actual taming—simulation deduction—actual verification—incremental learning." First, based on the equipment's physical structure and failure mechanism, a fully coupled digital twin model is constructed, including a Zener reference device aging sub-model, a DC resistive voltage divider network aging sub-model, an operational amplifier circuit drift sub-model, and an accelerated aging test temperature field control sub-model. A multi-model coupling relationship matrix is ​​established between the sub-models. Second, by applying temperature cycling stress and excess electrical stress through accelerated aging testing, the equipment degradation characteristics are rapidly induced. Combined with short-term measured data under rated conditions, a hybrid driving method of physical mechanism and deep neural network is used to identify and adaptively correct the model's parameters. Model taming is achieved with a mean square error less than 0.2 ppm as the convergence criterion. Based on the tamed model, a one-year virtual simulation deduction is performed, which quickly outputs the predicted annual stability index. Finally, the prediction error is verified using subsequent 12 months of actual installation data and fed back to the model, completing incremental learning and iterative optimization. This method shortens the evaluation cycle from 36 months to 2-3 months, with a prediction error of better than 5%, achieving efficient, accurate, and evolvable stability assessment.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A stability evaluation device for a solid-state DC voltage standard source based on digital twins, characterized in that: It includes an accelerated aging test module, a temperature and humidity monitoring module, a voltage output monitoring module, a digital twin simulation module, a digital twin model training and evaluation module, and an on-site verification and incremental learning module. Each module achieves synchronous communication and data interaction via a data bus. The accelerated aging test module is used to apply programmable temperature cycling stress and excess electrical stress to the voltage reference device inside the solid-state DC voltage standard source to accelerate the performance degradation of the device. The temperature control range is -15℃ to 150℃, the temperature control accuracy is not less than ±0.05℃, and the excess electrical stress is 1.2 times the rated operating voltage. The temperature and humidity monitoring module is used to collect and store the temperature and humidity time series data of the environment in which the solid-state DC voltage standard source under test is located in real time. The sampling interval is continuously adjustable in the range of 1 second to 10 minutes, and the sampling accuracy meets the requirements of temperature ±0.01℃ and humidity ±2% RH. The voltage output monitoring module is used to continuously acquire the three output voltage values ​​of a solid-state DC voltage standard source (10 V, 1 V, and 1.018 V) with a measurement accuracy of not less than 0.9 ppm, and synchronously record the original voltage drift data and the acquisition timestamp. The digital twin simulation module is used to construct a fully coupled digital twin model that includes a Zener reference device aging sub-model, a DC resistor voltage divider network aging sub-model, an operational amplifier circuit drift sub-model, and an accelerated aging test constant temperature device temperature field control sub-model, and includes a multi-model coupling relationship matrix between each sub-model. The digital twin model taming and evaluation module is used to identify parameters and adaptively correct the digital twin model based on the measured data collected by the temperature and humidity monitoring module and the voltage output monitoring module, complete the model taming, and output the predicted annual stability index based on the tamed digital twin model through virtual simulation. The implementation verification and incremental learning module is used to compare and verify the predicted annual stability index with the actual annual stability index monitored continuously for the following 12 months, calculate the prediction error, and feed the verification results back to the digital twin model to complete the incremental learning and iterative optimization of the model parameters.

2. The stability evaluation device for a solid-state DC voltage standard source based on digital twin according to claim 1, characterized in that: The digital twin simulation module adopts a multi-fidelity modeling strategy, using a high-precision finite element model for the voltage reference device and a reduced-order mechanism model for the DC resistor voltage divider network and operational amplifier circuit.

3. The stability evaluation device for a solid-state DC voltage standard source based on digital twin according to claim 2, characterized in that: The aging sub-model of the Zener reference device is constructed based on the Arrhenius model and the Zener reference breakdown voltage drift mechanism. The Zener reference is a high-stability Zener reference device.

4. The stability evaluation device for a solid-state DC voltage standard source based on digital twin according to claim 3, characterized in that: The digital twin model taming and evaluation module adopts a hybrid approach of physical mechanism and deep neural network to identify and correct the parameters of the digital twin model online. The convergence criterion for taming the digital twin model is that the mean square error between the model output and the measured data is less than 0.2 ppm.

5. The stability evaluation device for a solid-state DC voltage standard source based on digital twin according to claim 4, characterized in that: The digital twin model taming and evaluation module has an active learning function. When the model prediction uncertainty exceeds the set threshold of 0.3 ppm, it automatically recommends to carry out supplementary tests under specific temperature ranges or stress conditions to obtain optimized taming data.

6. The stability evaluation device for a solid-state DC voltage standard source based on digital twin according to claim 5, characterized in that: The voltage output monitoring module is based on a high-precision digital multimeter with 8.5 bits or more, combined with a multi-channel programmable scanning switch and signal conditioning circuit.

7. A method for evaluating the stability of a solid-state DC voltage standard source based on digital twins, using the stability evaluation device for a solid-state DC voltage standard source based on digital twins as described in any one of claims 1-6, characterized in that, Includes the following steps: Step S1: Based on the physical structure and failure mechanism of the solid-state DC voltage standard source under test, construct a full-state digital twin model containing multiple sub-models and coupling relationship matrices, and initialize the model parameters; Step S2: Place the solid-state DC voltage standard source under test into the accelerated aging test module, apply temperature cycling stress and excess electrical stress, run the accelerated aging test, and simultaneously collect ambient temperature and humidity and output voltage drift data during the test. Step S3: Place the solid-state DC voltage standard source under test in the rated working environment and continuously collect measured data of ambient temperature and humidity and output voltage over a preset time period; Step S4: Input the short-term measured data collected in step S3 into the digital twin model training and evaluation module, and use a hybrid driving method to iteratively correct the parameters of the digital twin model until the model output and the measured data converge, thus completing the model training. Step S5: Based on the tamed digital twin model, input historical annual environmental data, perform a one-year virtual simulation, and obtain the predicted annual stability index by analyzing the voltage time series output by the simulation. Step S6: Collect the actual output voltage and environmental data of the solid-state DC voltage standard source under test for the next 12 months, calculate the true annual stability index, and compare it with the predicted index obtained in step S5 to verify the prediction error. Step S7: Feed the prediction error data obtained in step S6 back to the digital twin model, iteratively update the model parameters, and complete incremental learning; If the prediction error exceeds the preset threshold, return to step S3 to collect additional measured data and re-train the model.

8. The method for evaluating the stability of a solid-state DC voltage standard source based on digital twins according to claim 7, characterized in that: In step S4, model training uses the residual between the digital twin model output and the measured output as the observation, and the parameters that are iteratively updated include aging rate, temperature coefficient and drift coefficient.

9. The method for evaluating the stability of a solid-state DC voltage standard source based on digital twins according to claim 8, characterized in that: In step S2, the temperature cycling stress is a stepped cycle, specifically: -15℃ constant temperature for 24 hours → temperature rise to 25℃ constant temperature for 24 hours → temperature rise to 150℃ constant temperature for 24 hours → temperature drop to 25℃ constant temperature for 24 hours → temperature drop to -15℃ constant temperature for 24 hours, cycled 5 times.

10. The method for evaluating the stability of a solid-state DC voltage standard source based on digital twins according to claim 9, characterized in that: In step S5, the interim stability index is obtained by fitting the voltage output drift time series using linear regression, with a prediction accuracy greater than [missing information]. .

Images