Sensor zero drift calculation system and method based on small sample experiment data

Abstract

The invention discloses a sensor zero drift calculation system and method based on small sample experiment data, and belongs to the technical field of ship power system monitoring and state prediction. The system for calculating the zero drift of the sensor based on the small sample experimental data comprises an accelerated degradation test module, a physical model parameter identification module, a real-time data acquisition module, an accumulated damage calculation and drift prediction module and a drift compensation and output module. The performance zero drift calculation system comprises a temperature sensor, a pressure sensor and the like. According to the invention, based on small sample accelerated degradation experiment data and in combination with an Arranius physical model, quantitative prediction and real-time compensation are carried out on cumulative performance drift of the sensor caused by long-term working at different operating temperatures. According to the method, the physical model and the data model are subjected to hybrid modeling, and the advantages of real-time adaptive prediction of sensor drift, low modeling cost, high physical fidelity and accurate prediction are realized.

Description

Technical Field

[0001] This invention relates to a sensor performance zero-point drift calculation system and method, specifically to a sensor zero-point drift calculation system and method based on small sample experimental data, belonging to the field of ship power system monitoring and state prediction technology. Background Technology

[0002] Marine diesel engines are the core power source of modern ships, and their operational stability and reliability directly affect navigation safety and economic efficiency. To enable precise control, monitoring, and fault diagnosis of diesel engines, numerous sensors are installed in key components such as the engine block, intake and exhaust systems, cooling systems, and fuel systems. These include K-type or N-type thermocouples for measuring exhaust temperature and pressure sensors for measuring cylinder pressure or intake manifold pressure. However, these sensors, especially temperature and pressure sensors, operate for extended periods in harsh environments near the diesel engine, particularly the exhaust system, characterized by high temperatures, high pressures, vibrations, and corrosive gases. This causes slow changes in the physical and chemical properties of the sensors, resulting in performance drift. For example, thermocouples may experience changes in their thermoelectric potential due to thermal aging, short-term aging, oxidation, or corrosion; pressure sensors may experience zero-point drift or gain drift due to viscoelastic creep of the material or aging of the sensing element.

[0003] This sensor drift is gradual and difficult to detect, but its cumulative effect can severely impact data accuracy. Incorrect sensor data can mislead the engine control unit or fault diagnosis and condition estimation system, causing the engine to deviate from optimal operating conditions, resulting in decreased fuel efficiency, excessive emissions, and even, in extreme cases, incorrect fault diagnosis and safe shutdown.

[0004] Currently, the main methods for addressing sensor drift include: First, the traditional method of periodic calibration or replacement. However, for ocean-going vessels, accurate calibration during navigation is virtually impossible, and port maintenance requires significant downtime and high labor costs, and it cannot resolve drift issues between calibration cycles. Second, hardware redundancy, i.e., installing multiple identical sensors for comparison. However, this method is costly, space-consuming, and cannot address common-mode drift issues where all sensors drift synchronously due to being in the same harsh environment. Third, data-driven models, using historical data to build statistical models of sensor drift. However, this method heavily relies on large-scale, long-term operational data, which is very time-consuming to acquire, and the drift characteristics of sensors from different batches and installation locations vary, resulting in poor model generalization ability. In summary, existing technologies lack a system and method that can actively predict and quantify the cumulative drift of sensors under actual variable-temperature conditions using limited, readily available experimental data combined with physical degradation mechanisms. Summary of the Invention

[0005] To address the problem of unpredictable and uncompensated sensor drift, this invention proposes a calculation system and method for sensor zero-point drift based on small sample experimental data.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: A sensor zero-point drift calculation system based on small sample experimental data, the sensor zero-point drift calculation system based on small sample experimental data includes: The accelerated degradation test module is used to acquire small sample drift data from the sensor to form a small sample accelerated degradation dataset. The physical model parameter identification module identifies and stores one or more key physical parameters of the sensor based on a physical model that reflects the effect of temperature on the degradation rate. This is used to extract key physical degradation parameters from the small sample accelerated degradation dataset formed by the accelerated degradation test module and to establish a data-driven benchmark degradation model. The real-time data acquisition module is used to monitor and record the operating temperature of the target sensor on the ship's diesel engine in real time or near real time during actual operation, and to generate a real-time operating temperature curve that changes over time. The cumulative damage calculation and drift prediction module is used to couple the physical model and data model of the physical model parameter identification module and the real-time data acquisition module, converting the sensor's operating history under variable temperature conditions into cumulative drift and predicting the current drift. The drift compensation and output module is used to correct the original sensor measurement value using the predicted drift amount obtained by the cumulative damage calculation and drift prediction module, and output the compensated measurement value.

[0007] Furthermore, the baseline degradation model of the physical model parameter identification module is a baseline degradation path model between sensor drift and time, specifically represented as follows:

[0008] In the formula, For data-driven models, Select a reference temperature Below, the sensor drift amount.

[0009] Furthermore, the physical model parameter identification module identifies and stores an activation energy of the sensor. The cumulative damage calculation and drift prediction module performs the following calculations: a) Physical model normalization: Define an acceleration factor AF function, which characterizes the multiple of the sensor's drift rate at the real-time operating temperature relative to the drift rate at the reference temperature, as specifically expressed below:

[0010] In the formula, Let be the ideal gas constant. Real-time operating temperature; b) Equivalent time integration: for the real-time operating temperature curve Corresponding acceleration factor Depending on actual running time Perform time integration to calculate the equivalent time. The equivalent time This represents the sensor operating under actual variable temperature conditions. The cumulative damage over time is equivalent to the damage at a constant reference temperature. The runtime is specifically expressed as follows:

[0011] In the formula, The sampling time interval, The current moment; c) Hybrid model prediction: accumulating equivalent time Substitute into the baseline degradation path model Real-time prediction of the current moment Predicted drift Specifically, it is expressed as follows: .

[0012] Furthermore, the physical model parameter identification module identifies and stores at least two different activation energies of the sensor and the temperature thresholds for distinguishing different temperature ranges in different temperature ranges. In the physical model normalization step of the cumulative damage calculation and drift prediction module, the real-time temperature is used as the basis. With temperature threshold By comparing the values, the activation energy of the corresponding temperature range is automatically selected to calculate the acceleration factor AF function.

[0013] Furthermore, the physical model parameter identification module identifies and stores the activation energy of the sensor sensing element. Activation energy of sensor signal processing systems The real-time data acquisition module acquires the actual temperature of the sensor's sensing element. and the ambient temperature of the sensor signal processing system The cumulative damage calculation and drift prediction modules respectively calculate the equivalent time of the sensor sensing element. Then, the baseline degradation path model is input. The predicted sensor element drift is obtained from this. Equivalent time of sensor signal processing system Then, the baseline degradation path model is input. The drift of the predicted sensor signal processing system is obtained. The drift compensation and output module executes a composite compensation algorithm, combining the predicted drift amount of the sensor element. And predict the drift of the sensor signal processing system Correct the original sensor measurement values.

[0014] Furthermore, the physical model parameter identification module identifies the activation energy of the physical model. Subsequently, statistical analysis was performed on the historical operating temperature curves to determine the statistically representative temperature, which was then set as the system reference temperature. The original baseline degradation path model Translate to the new reference temperature Next, a new baseline degradation path model is generated. .

[0015] Furthermore, it also includes a baseline degradation path model adaptive correction module, which is used to implement adaptive correction of the baseline degradation path model based on online sparse data.

[0016] Furthermore, the physical model of the physical model parameter identification module is the Arenius model.

[0017] Furthermore, the data-driven model can be any one of an exponential model, a Weibull model, or a polynomial model.

[0018] A method for calculating sensor zero-point drift based on small sample experimental data, employing the aforementioned sensor zero-point drift calculation system based on small sample experimental data, is implemented through the following steps: S1: Obtain small sample drift data of the sensor. By applying at least two different high-temperature constant accelerated stress temperatures to the sensor, record the drift of the sensor performance parameters over time to form a small sample accelerated degradation dataset. S2: Extract physical degradation parameters and establish a baseline degradation model. Based on a physical model that reflects the effect of temperature on degradation rate, identify and store one or more key physical parameters. At the same time, based on the small sample accelerated degradation dataset mentioned in S1, establish a data-driven baseline degradation path model at a selected reference temperature. S3: Collect the real-time operating temperature of the target sensor of the ship's diesel engine and generate a real-time temperature curve; S4: Calculate the cumulative drift and predict the drift amount; S5: Correct and output the measured value. Correct the original sensor measured value according to the predicted drift amount described in S4, and output the compensated measured value.

[0019] The beneficial effects of this invention are: 1. This invention uses small sample sizes to accelerate experimental data instead of expensive and time-consuming full lifecycle testing, which greatly reduces modeling costs and enables rapid response to the characteristic evaluation of new batches of sensors, resulting in significant economic benefits and timeliness.

[0020] 2. This invention utilizes the Arenius model, a physical model, to handle complex and nonlinear temperature dependence, normalizing the temperature history to equivalent time. It then employs a small-sample data model to describe the time-cumulative degradation path under constant conditions. This hybrid approach of physical normalization and data modeling far surpasses purely data-driven models that require massive amounts of data to learn the combined effects of temperature and time, exhibiting stronger physical fidelity and extrapolation prediction capabilities.

[0021] 3. This invention transforms the complete temperature history experienced by the sensor into cumulative damage through real-time integration of the acceleration factor. This makes drift prediction adaptive and can accurately reflect the different wear and tear on the sensor's lifespan caused by different operating conditions of the diesel engine.

[0022] 4. This invention can correct sensor data in real time, providing more accurate input to the engine control unit and fault diagnosis system, thereby improving the control accuracy, fuel economy, operational safety and condition-based maintenance of the diesel engine. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the main structure and process of one embodiment of the sensor zero-point drift calculation system and method based on small sample experimental data of the present invention. Detailed Implementation

[0024] Specific implementation method one: as follows Figure 1 As shown in this embodiment, a sensor zero-point drift calculation system based on small sample experimental data is provided. The sensor zero-point drift calculation system based on small sample experimental data includes: An accelerated degradation testing module is used to acquire small-sample drift data from the sensor, forming a small-sample accelerated degradation dataset. This module applies at least two different constant accelerated stress temperatures above the normal operating temperature to one or more sensor samples under test, and continuously tests at each stress temperature for a preset period of time, recording the drift of the sensor's performance parameters over time to form the small-sample accelerated degradation dataset. Using small-sample accelerated experimental data replaces expensive and time-consuming full-lifecycle testing, significantly reducing modeling costs and enabling rapid response to the characteristic evaluation of new batches of sensors, demonstrating significant economic efficiency and timeliness.

[0025] The physical model parameter identification module identifies and stores one or more key physical parameters of the sensor based on a physical model that reflects the effect of temperature on the degradation rate. This is used to extract key physical degradation parameters from the small sample accelerated degradation dataset formed by the accelerated degradation test module and to establish a data-driven benchmark degradation model.

[0026] The physical model parameter identification module is based on a physical model reflecting the effect of temperature on the degradation rate, preferably the Arenius model. This model assumes a specific relationship between the sensor's equivalent lifetime (defined as the time it takes for the drift to reach a preset failure threshold) and temperature. Using the small sample accelerated degradation dataset, key physical parameters of the sensor, especially the activation energy, are identified and stored through linear regression or nonlinear fitting. Meanwhile, the physical model parameter identification module also uses small sample data at a selected reference temperature. Next, establish the sensor drift amount. With time Benchmark degradation path model between .this It is a data-driven model, such as an exponential model, a Weibull model, or a multinomial model.

[0027] The baseline degradation model of the physical model parameter identification module is a baseline degradation path model between sensor drift and time, specifically represented as follows:

[0028] In the formula, For data-driven models, Select a reference temperature Below, the sensor drift amount.

[0029] The real-time data acquisition module is used to monitor and record the operating temperature of the target sensor on the ship's diesel engine during actual operation in real time or near real time, forming a real-time operating temperature curve that varies over time. .

[0030] The cumulative damage calculation and drift prediction module is used to couple the physical model and data model of the physical model parameter identification module and the real-time data acquisition module, converting the sensor's operating history under variable temperature conditions into cumulative drift and predicting the current drift.

[0031] The cumulative damage calculation and drift prediction module is the core calculation module of this system. It is used to couple the physical model and the data model, converting the sensor's operating history under varying temperature conditions into its cumulative drift. When the physical model parameter identification module identifies and stores an activation energy of the sensor... The cumulative damage calculation and drift prediction module performs the following calculations: a) Physical model normalization, based on the identified activation energy. and the selected reference temperature An acceleration factor (AF) function is defined. The AF function characterizes the sensor's performance at real-time operating temperature. The drift rate at the reference temperature is relative to the drift rate at the reference temperature. The multiples of the drift rate are expressed as follows: in is the ideal gas constant.

[0032] b) Equivalent time integration, through the real-time operating temperature curve The corresponding acceleration factor Depending on actual running time Perform time integration to calculate an equivalent time. The equivalent time This represents the sensor operating under actual variable temperature conditions. The damage accumulated over time is equivalent to that at a constant reference temperature. Downloaded and running Duration. In discrete digital systems, this integral calculation is achieved through the following summation, specifically represented as follows: in The sampling time interval, This refers to the current moment.

[0033] c) Hybrid model predictions, using the cumulative equivalent time obtained by normalizing the physical model. Substituting into the aforementioned data-driven baseline degradation path model In this way, the sensor can predict the current moment in real time. Predicted drift Specifically, it is expressed as follows: .

[0034] By integrating the acceleration factor in real time, the complete temperature history experienced by the sensor is transformed into cumulative damage. This makes drift prediction adaptive, accurately reflecting the different wear and tear on the sensor lifespan caused by different diesel engine operating conditions.

[0035] The drift compensation and output module is used to correct the original sensor measurement value using the predicted drift amount obtained from the cumulative damage calculation and drift prediction module, to obtain the compensated measurement value, and then output it. The drift compensation and output module obtains the original measurement value from the sensor. (The measured value of the physical quantity output by the sensor at the k-th sampling time without drift compensation) and based on the predicted drift amount Acquiring raw measurement values ​​from sensors Corrections are made to obtain the compensated measurement values. .

[0036] For example: The drift compensation and output module will output the compensated measurement value. Outputs are sent to the engine control unit of the marine diesel engine, the condition monitoring system, or to subsequent fault diagnosis algorithms to provide maintenance decision support.

[0037] The sensor zero-point drift calculation system based on small sample experimental data aims to address how to avoid relying on large-scale, long-term, full-lifetime experimental data and instead use only small sample, short-term accelerated experimental data to characterize the temperature-induced drift characteristics of sensors; and how to establish a computational model (i.e., a physical model) that reflects the physical degradation mechanism of sensors, and combine it with a data model based on experimental data fitting to form a hybrid modeling framework; and finally, using this hybrid model, combined with the actual, variable operating temperature curve of the diesel engine sensor, to calculate its cumulative drift in real time, thereby compensating for the sensor output signal.

[0038] Preferably, to improve the model's fidelity to complex physical mechanisms, based on the small-sample accelerated degradation dataset, the physical model parameter identification module identifies and stores at least two different activation energies and temperature thresholds for different temperature ranges of the sensor in different temperature ranges. In the physical model normalization step of the cumulative damage calculation and drift prediction module, the real-time temperature is used as the basis. With temperature threshold The acceleration factor AF function is calculated by automatically selecting the activation energy for the corresponding temperature range through comparison. For example, based on the small sample accelerated degradation dataset, the activation energy for the low-temperature range is identified in different temperature ranges. and activation energy in high temperature range and the temperature thresholds used to distinguish these intervals Accordingly, the physical model normalization step in the cumulative damage calculation and drift prediction module uses a piecewise acceleration factor (AF) function, which is based on the real-time temperature. With temperature threshold The comparison automatically selects the activation energy in the low-temperature range. Or activation energy in high temperature range To calculate AF.

[0039] Preferably, sensor drift comprises two parts: the sensing element (e.g., a sensing chip) and the signal processing system (e.g., an amplifier). To achieve sensor system-level drift decoupling, the physical model parameter identification module identifies and stores the activation energy of the sensor sensing element. Activation energy of sensor signal processing systems The real-time data acquisition module collects the target sensor's operating temperature, including the actual temperature of the sensing element and the ambient temperature of the signal processing system. The real-time data acquisition module collects the actual temperature of the sensor's sensing element. and the ambient temperature of the sensor signal processing system The cumulative damage calculation and drift prediction modules respectively calculate the equivalent time of the sensor sensing element. Then, the baseline degradation path model is input. The predicted sensor element drift is obtained from this. Equivalent time of sensor signal processing system Then, the baseline degradation path model is input. The drift of the predicted sensor signal processing system is obtained. The cumulative damage calculation and drift prediction modules calculate two independent equivalent times in parallel and substitute them into their respective baseline degradation path models to obtain independent drift amounts. The drift compensation and output module executes a composite compensation algorithm, which combines the prediction of sensor element drift. And predict the drift of the sensor signal processing system Correcting the original sensor measurement value .

[0040] Preferably, in order to improve the numerical stability of integral calculation, the physical model parameter identification module identifies the activation energy of the physical model. Then, statistical analysis is performed on the historical operating temperature curves to determine a statistically representative temperature (e.g., average value). Then, the statistically representative temperature (e.g., average value) is used. Set as the system's official reference temperature. Utilizing known activation energies Using the original accelerated degradation data and the Arenius model, the original baseline degradation path model was transformed. Translate to the new reference temperature Next, generate new , used for subsequent calculations.

[0041] Preferably, the sensor zero-point drift calculation system based on small sample experimental data further includes a baseline degradation path model adaptive correction module, used to implement adaptive correction of the baseline degradation path model based on online sparse data. Specifically, the steps of the baseline degradation path model adaptive correction method based on online sparse data are as follows: a) Whether the real-time monitoring system is in a "quasi-calibrated" state where the "true value" can be obtained, such as when the diesel engine is stable under a certain operating condition, or when a known calibration source is applied during shutdown maintenance; b) In the state of step a), acquire one or more sparse actual drift measurement points. ; c) Using the results obtained in step b) Compared with the system's current prediction The residuals between them are used to optimize the baseline degradation path model using an optimization algorithm, such as Bayesian update, Kalman filtering, or least squares. Parameters (e.g.) or Online corrections are performed to generate a personalized degradation model. ; d) The subsequent cumulative damage calculation and drift prediction module will use the results obtained in step c). replace Perform the calculation.

[0042] The adaptive correction method for the baseline degradation path model based on online sparse data enables the drift model to learn and personalize itself, greatly improving the accuracy of long-term predictions. This addresses the baseline degradation path model established in the physical model parameter identification module. Based on only small sample experimental data, which represents the batch average characteristics, there may be discrepancies with the actual degradation path of a single sensor.

[0043] By enabling real-time correction of sensor data, more accurate inputs are provided to the engine control unit and fault diagnosis system, thereby improving the control precision, fuel economy, operational safety, and condition-based maintenance of the diesel engine.

[0044] A method for calculating sensor zero-point drift based on small sample experimental data, employing the aforementioned sensor zero-point drift calculation system based on small sample experimental data, is implemented through the following steps: S1: Obtain small sample drift data of the sensor. By applying at least two different high-temperature constant accelerated stress temperatures to the sensor, record the drift of the sensor performance parameters over time to form a small sample accelerated degradation dataset. S2: Extract physical degradation parameters and establish a baseline degradation model. Based on a physical model that reflects the effect of temperature on degradation rate, identify and store one or more key physical parameters. At the same time, based on the small sample accelerated degradation dataset mentioned in S1, establish a data-driven baseline degradation path model at a selected reference temperature. S3: Collect the real-time operating temperature of the target sensor of the ship's diesel engine and generate a real-time temperature curve; S4: Calculate the cumulative drift and predict the drift amount; S5: Correct and output the measured value. Correct the original sensor measured value according to the predicted drift amount described in S4, and output the compensated measured value.

[0045] The hybrid modeling framework of this invention is neither a purely data-driven black-box model nor a purely physical model that is difficult to solve. This invention utilizes a physical model (Arrhenius model) to handle complex, nonlinear temperature dependencies, normalizing the temperature history to equivalent time. Then utilize small sample data models This describes the time-cumulative degradation path under constant conditions. This hybrid approach of physical normalization and data modeling makes it far superior to methods requiring massive amounts of data to learn temperature. and time The combined effect of pure data-driven models results in stronger physical fidelity and extrapolation prediction capabilities.

[0046] Example This embodiment discloses a temperature drift calculation system for a type K thermocouple in the exhaust manifold of a marine diesel engine. Diesel engine exhaust temperature is a key parameter for combustion control and thermal load management in the engine control unit, and its drift (usually caused by short-term aging or oxidation) can lead to lower measured values. This system includes an accelerated degradation testing module, a physical model parameter identification module, a real-time data acquisition module, a cumulative damage calculation and drift prediction module, and a drift compensation and output module.

[0047] Module implementation Step 1: Obtain small sample data through the accelerated degradation test module In this embodiment, type K thermocouples from the same batch installed on the ship were selected, for example, N=3 units. Tests were conducted in a laboratory environment at two accelerated stress temperature levels: (1073.15K) and (1173.15K). In The test will continue for 100 hours. The thermocouple was continuously tested for 50 hours. During the test, the output potential of the thermocouple at a standard calibration point (e.g., 1000°C) was measured periodically (e.g., every 5 hours), and the deviation from the initial value, i.e., the drift, was calculated. Record the data to obtain two small sample datasets: and .

[0048] Step 2: Identify physical parameters using the physical model parameter identification module. This module performs the first part of the hybrid modeling of this invention. First, it defines the "lifetime". In this embodiment, it is defined as the thermocouple drift reaching a certain preset failure threshold (e.g., The time at which the result is obtained. Using the data from step 1, we can fit or interpolate to obtain: at... When, reach The drift lifetime is Set to 80 hours; When, reach The drift lifetime is Set to 30 hours.

[0049] Then, the module calls the Arenius physics model. Convert it to linear form: use( )and( By solving this system of two linear equations, we can identify (calculate) the activation energy from these two known data points. :

[0050] Calculated The value is stored as the inherent physical drift characteristic parameter of this batch of K-type thermocouples.

[0051] Meanwhile, the identification module selects As a reference temperature. Utilizing in The dataset obtained below Fit a data-driven baseline degradation path model For example, it can be fitted as an exponential model as shown below: At this point, the two key components of hybrid modeling (physical parameters) have been completed. and data model All calculations have been successfully completed.

[0052] Step 3: Obtain real-time temperature through the real-time data acquisition module. During the operation of the marine diesel engine, the real-time data acquisition module in the system is connected to the target thermocouple installed at the exhaust manifold, in order to The actual operating temperature is continuously acquired at minute sampling intervals. . It is an array that varies with the diesel engine load, for example, at idle speed. (723K), during cruising... (1023K), at high load is (1123K).

[0053] Step 4: Calculate the cumulative drift using the cumulative damage calculation and drift prediction module. The prediction module performs the second part of the hybrid modeling of this invention (coupled computation), and internally maintains an accumulator variable. The initial value is 0. At each sampling time... : a) Physical model normalization: This prediction module obtains the current temperature from the real-time data acquisition module. The prediction module retrieves the stored parameters from the physical model parameter identification module. and Calculate the current sampling interval. Acceleration factor within The calculation formula is as follows:

[0054] b) Equivalent Time Integration: Update the accumulated equivalent time through summation. :

[0055] c) Hybrid model prediction: The calculated cumulative equivalent time... Substitute the data model identified in step 2 In, for example: this This is the predicted drift amount at the current moment (unit: °C).

[0056] Step 5: Output correction values ​​via drift compensation and output module At any moment The engine control unit or monitoring system reads the raw, uncompensated temperature value from this type K thermocouple. (For example, 848.5°C). This compensation module will use the values ​​calculated by the prediction module. (For example This information is provided to the engine control unit. The engine control unit calculates the compensated temperature. :

[0057] It should be noted that the sign of the compensation depends on the physical direction of the drift. This embodiment assumes that the drift "leads to a lower measurement," therefore the compensation is additive. If the drift leads to a higher reading, the compensation should be subtractive. The system of this invention supports compensation in either direction. The engine control unit uses... The actual exhaust temperature is used for subsequent combustion control, heat load calculation, and fault diagnosis.

[0058] This invention is not limited to the embodiments described above. For example, the physical model parameter identification module can also use other physical models, such as models used to describe thermal cycles, or generalized models used to describe the combined effects of multiple mechanisms. (Baseline degradation path model) It is not limited to the exponential model; it can be the cumulative function of the Weibull distribution, the linear model, or the multinomial model, as long as it can fit a small sample of accelerated degradation data.

[0059] This invention utilizes small-sample accelerated degradation data to identify physical model parameters (especially) Establish a data-driven degradation path model And the cumulative damage is calculated by real-time temperature integration. ), ultimately passed Drift prediction through physical-data hybrid modeling falls within the scope of this invention. For pressure sensors, the temperature-dependent creep effect can also be equivalent to a generalized Arenius activation process, and therefore is also applicable to this system.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A calculation system for sensor zero-point drift based on small sample experimental data, characterized in that: The sensor zero-point drift calculation system based on small sample experimental data includes: The accelerated degradation test module is used to acquire small sample drift data from the sensor to form a small sample accelerated degradation dataset. The physical model parameter identification module identifies and stores one or more key physical parameters of the sensor based on a physical model that reflects the effect of temperature on the degradation rate. This is used to extract key physical degradation parameters from the small sample accelerated degradation dataset formed by the accelerated degradation test module and to establish a data-driven benchmark degradation model. The real-time data acquisition module is used to monitor and record the operating temperature of the target sensor on the ship's diesel engine in real time or near real time during actual operation, and to generate a real-time operating temperature curve that changes over time. The cumulative damage calculation and drift prediction module is used to couple the physical model and data model of the physical model parameter identification module and the real-time data acquisition module, converting the sensor's operating history under variable temperature conditions into cumulative drift and predicting the current drift. The drift compensation and output module is used to correct the original sensor measurement value using the predicted drift amount obtained by the cumulative damage calculation and drift prediction module, and output the compensated measurement value.

2. The sensor zero-point drift calculation system based on small sample experimental data according to claim 1, characterized in that: The baseline degradation model of the physical model parameter identification module is a baseline degradation path model between sensor drift and time, specifically represented as follows: In the formula, For data-driven models, Select a reference temperature Below, the sensor drift amount.

3. The sensor zero-point drift calculation system based on small sample experimental data according to claim 2, characterized in that: The physical model parameter identification module identifies and stores an activation energy of the sensor. The cumulative damage calculation and drift prediction module performs the following calculations: a) Physical model normalization: Define an acceleration factor AF function, which characterizes the multiple of the sensor's drift rate at the real-time operating temperature relative to the drift rate at the reference temperature, as specifically expressed below: In the formula, Let be the ideal gas constant. Real-time operating temperature; b) Equivalent time integration: for the real-time operating temperature curve Corresponding acceleration factor Depending on actual running time Perform time integration to calculate the equivalent time. The equivalent time This represents the sensor operating under actual variable temperature conditions. The cumulative damage over time is equivalent to the damage at a constant reference temperature. The runtime is specifically expressed as follows: In the formula, The sampling time interval, The current moment; c) Hybrid model prediction: accumulating equivalent time Substitute into the baseline degradation path model Real-time prediction of the current moment Predicted drift Specifically, it is expressed as follows: .

4. The sensor zero-point drift calculation system based on small sample experimental data according to claim 3, characterized in that: The physical model parameter identification module identifies and stores at least two different activation energies and temperature thresholds for different temperature ranges of the sensor in different temperature ranges. In the physical model normalization step of the cumulative damage calculation and drift prediction module, the real-time temperature is used as the basis. With temperature threshold By comparing the values, the activation energy of the corresponding temperature range is automatically selected to calculate the acceleration factor AF function.

5. The sensor zero-point drift calculation system based on small sample experimental data according to claim 3, characterized in that: The physical model parameter identification module identifies and stores the activation energy of the sensor sensing element. Activation energy of sensor signal processing systems The real-time data acquisition module acquires the actual temperature of the sensor's sensing element. and the ambient temperature of the sensor signal processing system The cumulative damage calculation and drift prediction modules respectively calculate the equivalent time of the sensor sensing element. Then, the baseline degradation path model is input. The predicted sensor element drift is obtained from this. Equivalent time of sensor signal processing system Then, the baseline degradation path model is input. The drift of the predicted sensor signal processing system is obtained. The drift compensation and output module executes a composite compensation algorithm, combining the predicted drift amount of the sensor element. And predict the drift of the sensor signal processing system Correct the original sensor measurement values.

6. The sensor zero-point drift calculation system based on small sample experimental data according to claim 3, characterized in that: The physical model parameter identification module identifies the activation energy of the physical model. Subsequently, statistical analysis was performed on the historical operating temperature curves to determine the statistically representative temperature, which was then set as the system reference temperature. The original baseline degradation path model Translate to the new reference temperature Next, a new baseline degradation path model is generated. .

7. The sensor zero-point drift calculation system based on small sample experimental data according to claim 1, characterized in that: It also includes a baseline degradation path model adaptive correction module, which is used to implement the baseline degradation path model adaptive correction based on online sparse data.

8. The sensor zero-point drift calculation system based on small sample experimental data according to claim 1, characterized in that: The physical model of the physical model parameter identification module is the Arenius model.

9. The sensor zero-point drift calculation system based on small sample experimental data according to claim 2, characterized in that: The data-driven model can be any one of the exponential model, the Weibull model, or the polynomial model.

10. A method for calculating sensor zero-point drift based on small sample experimental data, characterized in that: The sensor zero-point drift calculation system based on small sample experimental data, as described in any one of claims 1-9, is implemented through the following steps: S1: Obtain small sample drift data of the sensor. By applying at least two different high-temperature constant accelerated stress temperatures to the sensor, record the drift of the sensor performance parameters over time to form a small sample accelerated degradation dataset. S2: Extract physical degradation parameters and establish a baseline degradation model. Based on a physical model that reflects the effect of temperature on degradation rate, identify and store one or more key physical parameters. At the same time, based on the small sample accelerated degradation dataset mentioned in S1, establish a data-driven baseline degradation path model at a selected reference temperature. S3: Collect the real-time operating temperature of the target sensor of the ship's diesel engine and generate a real-time temperature curve; S4: Calculate the cumulative drift and predict the drift amount; S5: Correct and output the measured value. Correct the original sensor measured value according to the predicted drift amount described in S4, and output the compensated measured value.

Images