Experimental method and device for optimizing optical fiber hydrogen measurement sensor
By combining the algorithms of fiber optic hydrogen sensors and MOS hydrogen sensors, the response hysteresis and curve consistency issues of fiber optic hydrogen sensors have been resolved, achieving high-precision and long-term stable hydrogen concentration detection, and improving the sensor's dynamic tracking capability and environmental adaptability.
Patent Information
- Application Number
- CN202610060972.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-16
- Publication Date
- 2026-04-07
AI Technical Summary
Existing fiber optic hydrogen sensors suffer from response lag, sensitivity dependence on the performance of the sensitive coating, and poor consistency of response curves when detecting hydrogen concentration, resulting in insufficient detection accuracy and repeatability. Furthermore, existing calibration methods fail to fully utilize the fast response advantage of MOS hydrogen sensors.
A joint algorithm combining extended Kalman filtering and recursive least squares method is used to adaptively correct the dynamic response and sensitivity factor of the fiber optic hydrogen sensor. By placing the fiber optic hydrogen sensor and the MOS hydrogen sensor in the same sealed gas-sensitive test chamber, the fast response information of the MOS hydrogen sensor is used for real-time correction, and an anomaly detection mechanism is combined to handle observed anomalies.
This invention enables fiber optic hydrogen sensors to maintain high accuracy and long-term stability in hydrogen concentration detection in dynamic environments, improving the sensor's response speed and concentration detection accuracy, and ensuring dynamic tracking capability and long-term stability under complex working conditions.
Smart Images
Figure CN121805207A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydrogen measurement technology, and in particular to an experimental method and apparatus for optimizing a fiber optic hydrogen sensor. Background Technology
[0002] Hydrogen, as a clean energy source, boasts advantages such as high calorific value, wide availability, and clean combustion products, leading to its widespread application in fuel cells, electric vehicles, aerospace, and hydrogen storage systems. However, hydrogen's lower explosive limit is only 4 vol.%, and it is colorless and odorless, making leaks extremely easy and difficult to detect. Therefore, developing highly sensitive, fast-response, reliable, and safe hydrogen sensors is crucial for ensuring the safety of hydrogen energy utilization.
[0003] Currently, hydrogen sensors mainly include metal-oxide-semiconductor (MOS) sensors, electrochemical sensors, and fiber optic sensors. Fiber optic hydrogen sensors, as a relatively new type of sensor that has emerged in recent years, rely on optical effects such as fiber Bragg gratings (FBGs) and surface plasmon resonance (SPR) to detect hydrogen. Compared with MOS hydrogen sensors, fiber optic hydrogen sensors have significant advantages such as immunity to electromagnetic interference, resistance to high and low temperatures, long-distance signal transmission, and suitability for high-risk or special environments. However, due to issues such as hysteresis in the response of fiber optic hydrogen sensors to hydrogen concentration, sensitivity dependence on the performance of the sensitive coating, and poor consistency of response curves between different batches of sensors, their accuracy and repeatability in practical applications are insufficient.
[0004] Therefore, how to further improve the accuracy and stability of fiber optic hydrogen sensors while maintaining their inherent advantages is a pressing technical problem that needs to be solved. Existing research employs a reference sensor calibration method, which involves placing the fiber optic hydrogen sensor and another commonly used sensor in the same test environment, comparing their detection results to correct the output value of the fiber optic sensor. Typically, the output signal of a MOS hydrogen sensor is used as a reference for the fiber optic hydrogen sensor to correct its optical response curve. However, this method only reaches the level of static calibration or single-point comparison, failing to fully utilize the advantages of the MOS hydrogen sensor's fast response speed and good real-time performance. Because the differences in the working mechanisms of the two types of sensors are not addressed, existing methods often only yield coarse correction results, making it difficult to ensure the accuracy and long-term consistency of the fiber optic hydrogen sensor's output curve in dynamic environments. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides an experimental method and apparatus for optimizing fiber optic hydrogen measurement sensors, which can ensure that the sensor maintains accurate measurements even in constantly changing environments.
[0006] This invention provides an experimental method for optimizing a fiber optic hydrogen measurement sensor, comprising the following steps: The fiber optic hydrogen sensor and the MOS hydrogen sensor are placed in the same sealed gas-sensitive test chamber. Hydrogen gas is introduced into the gas-sensitive test chamber to form a preset dynamic concentration curve, and the hydrogen concentration data in the chamber measured by the fiber optic hydrogen sensor and the real-time hydrogen concentration data are acquired simultaneously. A joint algorithm based on extended Kalman filtering and recursive least squares is used to process concentration data to adaptively correct the dynamic response and sensitivity factor of the fiber optic hydrogen sensor. Specifically, the extended Kalman filtering uses real-time measured hydrogen concentration data as observations to estimate the instantaneous dynamic bias and slowly varying zero-point drift of the fiber optic hydrogen sensor. The recursive least squares method uses the bias estimation results of the extended Kalman filtering to update the sensitivity scaling factor of the fiber optic hydrogen sensor and feeds the updated sensitivity scaling factor back into the observation model of the extended Kalman filtering. The system outputs the concentration curve of the fiber optic hydrogen sensor after correction by the joint algorithm; at the same time, it handles observation anomalies or parameter anomalies through an anomaly detection mechanism.
[0007] Furthermore, the modeling and execution of the extended Kalman filter includes: Define the state vector: in: : The actual hydrogen concentration inside the cavity; Instantaneous dynamic bias of fiber optic sensors; Slowly varying zero-point drift in fiber optic sensors; Extended Kalman filter modeling: Assuming the MOS hydrogen sensor is an ideal reference with no system bias and only measurement noise; the fiber optic sensor (FOS) includes sensitivity, instantaneous bias, and slowly varying bias: in This refers to the hydrogen concentration actually observed by MOS. ; This refers to the hydrogen concentration actually observed by FOS. ; This is the sensitivity scaling factor for the fiber optic sensor. , For measuring noise; pass Predicting prior estimates before new measurements arrive is used to maintain the continuity of sensor output, where... is the state transition Jacobian matrix, where This represents the attenuation coefficient of the instantaneous bias.
[0008] Furthermore, the adaptive update of recursive least squares (RLS) includes: Target output, bias-free FOS observations: For Kalman filter pairs The posterior estimate; For Kalman filter pairs posterior estimation ; Update sensitivity factor: For the recursive least squares method to update at each time step The estimated value; It is the least squares gain scalar; This represents the prediction error using the least squares method. The updated Let's return to the calculation of the observation function and Jacobian matrix for the extended Kalman filter at the next time step.
[0009] Furthermore, the final corrected concentration curve of the fiber optic hydrogen sensor is obtained through the following compensation formula: .
[0010] Furthermore, dynamic concentration includes at least one of step changes, linear ramp changes, or periodic oscillation changes.
[0011] Furthermore, it also includes anomaly detection, which includes: Innovation detection of Extended Kalman Filter: The innovation amount generated during the recursive calculation of Extended Kalman Filter is normalized to obtain the normalized innovation square. If it exceeds the threshold, the observation is judged to be abnormal. When the EKF innovation detection judges the observation to be abnormal, at least one of the following three measures shall be taken: temporarily increasing the observation noise covariance, suspending RLS update, and alarm recording. RLS anomaly protection using least squares: Calculate the prediction error, and if it exceeds a preset error threshold, skip the current update of the sensitivity scaling factor or update it with limited gain. Sensitivity scaling factor detection of fiber optic sensor: If the sensitivity scaling factor of the fiber optic sensor exceeds the preset reasonable range, the update of the sensitivity scaling factor is abandoned.
[0012] Furthermore, when the extended Kalman filter innovatively detects an anomaly in the observation, the state estimation parameters and covariance matrix of the extended Kalman filter are kept unchanged until the observation value returns to the confidence interval; The condition for restoring the recursive least squares update is that the normalized squared innovation value of the innovation detection by the extended Kalman filter is lower than the preset recovery threshold for N consecutive sampling times.
[0013] An experimental apparatus for optimizing a fiber optic hydrogen sensor includes: a gas-sensitive test chamber, a MOS hydrogen sensor, a fiber optic hydrogen sensor, an optical signal transmission module, an electrical signal transmission module, and a data processing and control unit. The gas-sensitive testing chamber includes an inlet and an outlet. The inlet is connected to an external hydrogen source through an inlet valve, and the outlet is connected to an external exhaust pipe through an outlet valve. The MOS hydrogen sensor and the fiber optic hydrogen sensor are placed inside the gas-sensitive test chamber; One end of the optical signal transmission module is connected to the fiber optic hydrogen sensor, and the other end is used to connect to an external fiber optic demodulator. One end of the electrical signal transmission module is electrically connected to the MOS hydrogen sensor, and the other end is used to connect to an external computer. The data processing and control unit is communicatively connected to the fiber optic demodulator, an external computer, and the inlet and outlet valves, respectively. The data processing and control unit controls the opening and closing states of the inlet and outlet valves to generate a preset dynamic concentration change curve within the gas-sensitive testing chamber; The data processing and control unit receives and processes the signal from the fiber optic hydrogen sensor of the fiber optic demodulator to obtain its measured concentration data, and also receives the signal from the MOS hydrogen sensor connected to an external computer to obtain its measured concentration data; The data processing and control unit performs the experimental method of the optimized fiber optic hydrogen sensor as described in any one of claims 1 to 7.
[0014] The technical solution provided in this invention has the following advantages compared with the prior art: the MOS hydrogen sensor can respond quickly to concentration changes and output an electrical signal through the electrical signal transmission module to calculate the real-time hydrogen concentration; the optical wavelength drift signal output by the fiber optic sensor is processed by a demodulator to obtain the corresponding concentration. Through a joint optimization strategy of extended Kalman filtering (EKF) and recursive least squares (RLS), the dynamic response and sensitivity factor of the fiber optic sensor are adaptively corrected to obtain an optimized concentration curve. This curve improves both response speed and concentration accuracy, achieving high-precision and long-term stable detection of hydrogen concentration. It can evaluate the sensor's dynamic tracking capability and long-term stability under complex operating conditions, ensuring that the sensor maintains correct measurement even in constantly changing environments. Attached Figure Description
[0015] Figure 1 A three-dimensional schematic diagram of an experimental apparatus for optimizing a fiber optic hydrogen sensor provided in an embodiment of the present invention; Figure 2 This is a top view of an experimental apparatus for an optimized fiber optic hydrogen sensor provided in an embodiment of the present invention.
[0016] Explanation of reference numerals in the attached figures: 1. Gas-sensitive test chamber; 11. Air inlet; 111. Air inlet valve; 12. Air outlet; 121. Air outlet valve; 2. Optical signal transmission module; 3. Electrical signal transmission module. Detailed Implementation
[0017] The following detailed description of a specific embodiment of the present invention is provided in conjunction with the accompanying drawings. However, it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.
[0018] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the technical solution of this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.
[0019] The present invention will be described below through several specific embodiments. To keep the following description of the embodiments clear and concise, detailed descriptions of known functions and components may be omitted. When any component of an embodiment of the present invention appears in more than one drawing, the component may be represented by the same reference numerals in each drawing.
[0020] Figure 1 This is a three-dimensional schematic diagram of an experimental apparatus for optimizing a fiber optic hydrogen sensor according to an embodiment of the present invention. Figure 2 This is a top view of an experimental apparatus for an optimized fiber optic hydrogen sensor provided in an embodiment of the present invention.
[0021] like Figure 1 and Figure 2As shown, this embodiment of the invention provides an experimental method for optimizing a fiber optic hydrogen sensor. A fiber optic hydrogen sensor and a MOS hydrogen sensor are placed in the same sealed gas-sensitive test chamber. Control programs are set for the inlet and outlet valves connected to the gas-sensitive test chamber. Hydrogen gas is introduced into the gas-sensitive test chamber according to the control program, forming a preset dynamic concentration curve. Simultaneously, hydrogen concentration data measured by the fiber optic hydrogen sensor and the MOS hydrogen sensor are acquired. The concentration data is processed using a joint algorithm of extended Kalman filtering and recursive least squares to adaptively correct the dynamic response and sensitivity factor of the fiber optic hydrogen sensor. Specifically, the extended Kalman filtering uses the measurement value of the MOS hydrogen sensor as the observation value to estimate the dynamic response bias and slow-varying zero-point drift of the fiber optic hydrogen sensor. The recursive least squares method uses the bias estimation result of the extended Kalman filtering to adaptively update the sensitivity scaling factor of the fiber optic hydrogen sensor and feeds the updated sensitivity scaling factor back to the observation model of the extended Kalman filtering. The concentration curve of the fiber optic hydrogen sensor corrected by the joint algorithm is output. Simultaneously, an anomaly detection mechanism is used to handle observational anomalies or parameter anomalies.
[0022] Specific implementation steps: Initial preparation: The fiber optic hydrogen sensor is fixed at the front end inside the gas-sensitive test chamber 1 and connected to the fiber optic demodulator via the optical signal transmission module 2; the MOS hydrogen sensor is fixed at the rear end inside the gas-sensitive test chamber 1 and connected to the computer via the electrical signal transmission module 3; the control program for the inlet valve 111 and the outlet valve 121 is set to ensure that the hydrogen concentration inside the gas-sensitive test chamber 1 can be controlled.
[0023] Dynamic concentration input: Different concentrations of hydrogen gas are introduced into the gas-sensitive cavity 1 to form a concentration curve that changes in a step, slope, or periodic manner; the MOS hydrogen sensor outputs a voltage signal in real time, which is amplified by the circuit module and transmitted to the computer to obtain the hydrogen concentration in the cavity; the optical response signal of the fiber optic sensor is converted into a wavelength drift value by the demodulator to obtain the hydrogen concentration in the cavity.
[0024] Data comparison and deviation detection: At the same time, the concentration values obtained by the MOS hydrogen sensor and the fiber optic hydrogen sensor are compared, and the difference is calculated. Here, the difference represents the sum of the instantaneous deviation and the slowly varying deviation of the fiber optic sensor relative to the MOS sensor. The instantaneous deviation is mainly caused by dynamic response hysteresis, while the slowly varying deviation is mainly caused by factors such as zero-point drift and the influence of temperature and humidity.
[0025] Optimization methods: Set sampling rate =10Hz Initial EKF value: (Pick (with an initial bias of 0), state covariance (ppm2) (taking the larger diagonal value to represent uncertainty).
[0026] Initial RLS values: (Originally obtained from offline static calibration, wavelength drift data of MOS output and fiber optic sensor were synchronously acquired at several known hydrogen concentration points. The least squares method was used to perform linear fitting between the two to obtain the initial estimate of the fiber optic sensor sensitivity scaling factor.) ), parameter covariance (indicating to) The magnitude of uncertainty (identity matrix), forgetting factor (Determine the relative weights of new and historical data in parameter estimation. In stable environments, a larger weighting is preferable.) It can maintain stable parameters and reduce the impact of noise; under conditions of sudden environmental changes or frequent disturbances, it has a smaller impact. This can accelerate the tracking of sensor sensitivity drift or bias changes, thereby improving the reliability and adaptability of the fiber optic hydrogen sensor optimization algorithm.
[0027] Process noise covariance ; Observation noise covariance Estimated from steady-state data, .
[0028] Furthermore, based on the bias and drift estimates from the extended Kalman filter output and the sensitivity scaling factor updated by recursive least squares, the concentration detection results of the fiber optic hydrogen sensor are compensated to obtain and output the optimized concentration value after dynamic error and drift compensation.
[0029] Furthermore, the extended Kalman filter modeling includes: defining the state vector as the actual hydrogen concentration in the cavity, the instantaneous dynamic bias of the fiber optic hydrogen sensor, and the slowly varying zero-point drift; predicting the state and covariance at the next moment through the state transition model; calculating the Jacobian matrix by combining the observation function, updating the state and covariance using the measurement residuals, and obtaining the estimated values of the dynamic response bias and zero-point drift of the fiber optic hydrogen sensor.
[0030] Extended Kalman Filter (EKF) Modeling Assuming the MOS is an ideal reference with no system bias and only measurement noise; the FOS includes sensitivity, instantaneous bias, and slowly varying bias: in This refers to the hydrogen concentration actually observed by MOS. , This refers to the hydrogen concentration actually observed by FOS. . This is the sensitivity scaling factor for the fiber optic sensor (continuously updated by RLS). , For measuring noise.
[0031] Define the state vector: in: : The actual hydrogen concentration inside the cavity; Instantaneous dynamic bias of fiber optic sensors; Slowly varying zero-point drift of fiber optic sensors.
[0032] EKF Prediction: State transition model: in The attenuation coefficient, representing the instantaneous bias, is a first-order autoregressive process. It is caused by sensor response hysteresis. When the concentration change stops, this bias should gradually decrease (attenuate) until it disappears, rather than accumulating continuously like drift. , , For process noise, the process noise covariance is... Observation noise covariance Estimated from steady-state data, .
[0033] Predicted status: in , is the state transition Jacobian matrix, whose size depends on the dimension of the state vector.
[0034] Predicting covariance: In extended Kalman filtering, the predicted state It provides prior estimates before new measurements arrive to maintain the continuity of sensor output; while predicting covariance... This characterizes the level of uncertainty in the prediction and directly affects the Kalman gain in the update phase, thus determining the trust allocation between prediction and measurement in the system. The combination of these two factors ensures that the optimized estimation process of the fiber optic hydrogen sensor in dynamic environments is both forward-looking and adaptively corrective.
[0035] Furthermore, the adaptive update of the recursive least squares method includes: using the actual intracavity hydrogen concentration estimated by the extended Kalman filter as the regression benchmark; constructing the debiased fiber optic hydrogen sensor observation as the target output; calculating the prediction error and the recursive least squares gain to update the sensitivity scaling factor of the fiber optic hydrogen sensor; feeding the updated sensitivity scaling factor back to the extended Kalman filter observation model; and combining the bias and drift estimates from the extended Kalman filter output with the sensitivity scaling factor updated by the recursive least squares method to correct the concentration detection result of the fiber optic hydrogen sensor.
[0036] Construct the observation function and compute the Jacobian matrix ; Observation function (latest obtained using RLS) ): in, For extended Kalman filter pairs Posterior estimation; For extended Kalman filter pairs Prior prediction; For extended Kalman filter pairs Prior prediction; State The partial derivatives are used to obtain the Jacobian matrix. This describes the local linear relationship between measurement and state, and determines the role of observation information in filter updates: EKF Update Observation vector: Measurement residuals The difference between the actual measurement and the predicted measurement is the correction magnitude for the state update; Innovation Covariance As a weighted benchmark, it measures the reliability of the residuals. A large number indicates significant uncertainty in prediction or observation noise, and excessive correction should be avoided. Conversely, if the observation has strong constraints on the state, substantial correction should be made.
[0037] in, This represents the uncertainty of the predicted state after it is projected onto the measurement space through the observation model. This represents the observation noise covariance.
[0038] Kalman gain matrix Kalman gain is the trade-off coefficient between prediction and observation. It is used to balance the feasibility of the predictive model with that of the observed measurements, indicating how much to trust the measurements when updating the state. A higher value indicates greater trust in the observed measurements.
[0039] Status Update: Covariance update: The results were obtained ; Construct the RLS target and calculate the RLS gain (RLS only estimates...) ) Target output, bias-free FOS observations: For extended Kalman filter pairs The posterior estimate; For extended Kalman filter pairs posterior estimation ; This quantity is approximately equal to [the ideal value]. noise.
[0040] Regression parameters (inputs): The concentration estimate of EKF is used as the regression baseline. in, For extended Kalman filter pairs Prior prediction, RLS (Scalar Case) Formula ( As a scalar, (for scalar covariance) Prediction error : Represents the deviation between the current actual measurement and the model prediction; driving parameter Update. A large prediction error indicates that the current parameter estimates are inaccurate and require a significant update.
[0041] RLS gain vector Determines the current prediction error The degree of influence on parameter updates controls the convergence speed and stability.
[0042] Parameter update: For the recursive least squares method to update at each time step The estimated value; It is the least squares gain scalar; Covariance update: Will Write back to EKF (closed loop): The updated EKF will be written back to EKF. Write back to the observation model of the EKF at the next time step for use in Jacobi calculation and observation function construction: Optimize output After EKF filtering, the instantaneous deviation is obtained. With slow variation deviation The estimated value; After RLS adaptive updating, the sensitivity scaling factor of the fiber optic sensor is obtained. ; The final corrected fiber density curve is as follows: Furthermore, dynamic concentration includes at least one of step changes, linear ramp changes, or periodic oscillation changes.
[0043] Specifically, the step change corresponds to a sudden change in hydrogen concentration in real-world scenarios (the instant of a leak). The test assesses the sensor's dynamic response speed; detects the sensor's rise time and recovery time; and tests whether the sensor will exhibit overshoot, hysteresis, or saturation during sudden changes. This accurately evaluates the sensor's real-time response capability to sudden hydrogen leak events; verifies the dynamic consistency of MOS hydrogen sensors and fiber optic sensors under sudden operating conditions; and improves the accuracy of testing in determining the sensor's rapid response capability.
[0044] Linear increase: This corresponds to a stable increase in hydrogen concentration in reality (chronic leakage). It tests the sensor's output smoothness under slowly changing concentration conditions; whether drift occurs; and its sensitivity to low-speed changes. It evaluates the sensor's linearity, stability, and ability to track low-speed concentration changes; it can verify the consistency and synchronization between fiber optic and MOS sensors under slow changes; and it helps verify the effectiveness of algorithms or compensation models.
[0045] Periodic oscillation: Corresponds to the periodic fluctuations in hydrogen concentration in reality (wind field disturbances, pressure fluctuations). Simulates the fluctuating environment of actual hydrogen concentration; verifies the dynamic stability of the sensor during continuous rises and falls; observes for output drift or dynamic hysteresis. It can evaluate the sensor's dynamic tracking capability and long-term stability under complex operating conditions; ensures the sensor maintains accurate measurements even in constantly changing environments.
[0046] Furthermore, the updated sensitivity scaling factor of the fiber optic sensor is written back into the observation model of the EKF at the next time step for use in Jacobian matrix calculation and observation function construction.
[0047] Furthermore, it also includes anomaly detection, which includes: innovation detection using Extended Kalman Filter (EKF), which measures the innovation amount generated during the recursive calculation of EKF and normalizes the innovation amount to obtain the normalized innovation square. If the innovation exceeds the chi-square threshold, an observation anomaly is determined. When EKF innovation detection determines an observation anomaly, at least one of the following three measures is taken: temporarily increasing the observation noise covariance, suspending RLS updates, and alarm recording. Anomaly protection using least squares method is also included, which calculates the prediction error. If the error exceeds a preset error threshold, the current update of the sensitivity scaling factor is skipped or updated with limited gain. Sensitivity scaling factor detection for fiber optic sensors is also included. If the sensitivity scaling factor of the fiber optic sensor exceeds a preset reasonable range, the update of the sensitivity scaling factor is abandoned.
[0048] EKF Innovation Detection: Calculating the Normalized Square of Innovation ( If this value exceeds the chi-square threshold, an observation anomaly is determined (sensor malfunction or external disturbance), and the following measures are taken: temporarily increase the chi-square threshold. 1. Pause RLS updates and alarm logging. Specifically, V(t) represents the amount of innovation generated during the recursive calculation of the extended Kalman filter.
[0049] RLS anomaly protection: If Exceeding the threshold (e.g.) If so, skip this step. Update or update with limited gain (reduction) (the maximum value).
[0050] Furthermore, when the extended Kalman filter innovation detection determines that the observation is abnormal, the state estimation parameters and covariance matrix of the extended Kalman filter remain unchanged until the observation value returns to the confidence interval; the condition for resuming the recursive least squares update is that the normalized innovation square value of the extended Kalman filter innovation detection is lower than the preset recovery threshold for N consecutive sampling times.
[0051] Furthermore, when an observation is detected as unreliable, the parameters and their covariance are kept constant until the observation returns to the reliable interval. To resume RLS updates, the calculated value or residual index of the formula must be lower than the recovery threshold for N consecutive sampling times.
[0052] Parameter limiting: For Set upper and lower bound constraints (e.g.) ), to prevent numerical divergence.
[0053] To resume RLS updates, the calculated value or residual index of the formula must be below a recovery threshold for *n* consecutive sampling times. For example, for five consecutive iterations, if the calculated value is less than the chi-squared χ² value... 2 (d,0.05), where d is the observation dimension, and the value here is 2.
[0054] If â goes out of bounds, abandon this parameter update, keep â(𝑡) = â(𝑡−1), and set 𝑃 𝛼 Increase the size appropriately (to make the algorithm more "willing" to learn real changes in the future). EKF is responsible for estimating the state and additive bias, and can separate hysteresis and long-term drift, while RLS is only responsible for scalar multiplicative correction. To avoid redundant parameters.
[0055] RLS update Conversely, this improves the observation model of EKF, making the state estimation of EKF more accurate, and the two form a positive convergence closed loop.
[0056] Innovative detection and RLS anomaly protection to avoid contamination of parameters by abnormal samples.
[0057] Through the aforementioned EKF+RLS dual-layer optimization method, this invention can rapidly eliminate the instantaneous difference between the fiber optic hydrogen sensor and the MOS hydrogen sensor during the dynamic response phase, and counteract the slow drift caused by temperature and humidity during long-term operation. Optimized concentration curve. With higher accuracy and stability, it can significantly improve the practical application level of fiber optic hydrogen sensors.
[0058] An experimental apparatus for optimizing a fiber optic hydrogen sensor includes: a gas-sensitive testing chamber 1, a MOS hydrogen sensor, a fiber optic hydrogen sensor, an optical signal transmission module 2, an electrical signal transmission module 3, and a data processing and control unit. The gas-sensitive testing chamber 1 includes an inlet 11 and an outlet 12. The inlet 11 is connected to an external hydrogen source via an inlet valve 111, and the outlet 12 is connected to an external exhaust pipe via an outlet valve 121. The MOS hydrogen sensor and the fiber optic hydrogen sensor are disposed within the gas-sensitive testing chamber 1. One end of the optical signal transmission module 2 is connected to the fiber optic hydrogen sensor, and the other end is used to connect to an external fiber Bragg grating demodulator. One end of the electrical signal transmission module 3 is electrically connected to the MOS hydrogen sensor, and the other end is connected to... The terminal is used to connect to an external computer; the data processing and control unit is communicatively connected to the fiber optic demodulator, the external computer, and the inlet valve 111 and outlet valve 121 respectively; the data processing and control unit controls the opening and closing states of the inlet valve 111 and outlet valve 121 to form a preset dynamic concentration change curve in the gas-sensitive test chamber 1; the data processing and control unit receives and processes the signal from the fiber optic hydrogen sensor of the fiber optic demodulator to obtain its measured concentration data, and receives the signal from the MOS hydrogen sensor connected to the external computer to obtain its measured concentration data; the data processing and control unit executes the experimental method of the optimized fiber optic hydrogen sensor as described in any one of claims 1 to 7.
[0059] Specifically, the optical signal transmission module 2 is fixed to the front of the gas-sensitive test chamber 1 by a rubber sealing ring and a spiral clamping buckle. Both ends of the module have fiber optic interfaces; the end located inside the chamber is connected to the fiber optic hydrogen sensor, and the end located outside the chamber is connected to a fiber optic grating demodulator via a connecting fiber for real-time transmission of optical response signals.
[0060] The electrical signal transmission module 3 is fixed to the rear side of the gas-sensitive test chamber 1 by a rubber sealing ring and a spiral clamping buckle. Both ends of the module are equipped with aviation connectors. The end located inside the chamber is electrically connected to the MOS hydrogen sensor, while the end located outside the chamber is connected to an external computer via a connecting cable. This allows for the stable transmission of the electrical signal generated by the MOS hydrogen sensor to the computer, enabling real-time acquisition and processing of the sensor's output signal.
[0061] This invention arranges a MOS hydrogen sensor and an optical fiber hydrogen sensor simultaneously in a gas-sensitive testing chamber. The MOS sensor serves as a high-sensitivity, fast-response reference, while the optical fiber sensor is the object of optimization.
[0062] During testing, the hydrogen concentration in the gas-sensitive testing chamber 1 is dynamically and controllably adjusted by controlling the inlet valve 111 and the outlet valve 121, causing the concentration within the chamber to exhibit various change modes such as step changes, linear increases, or periodic oscillations. The MOS hydrogen sensor can respond rapidly to concentration changes and output an electrical signal through the electrical signal transmission module 3, serving as a real-time reference for the actual concentration change within the chamber. The fiber optic sensor outputs its optical response to the demodulator through the optical signal transmission module 2.
[0063] The MOS hydrogen sensor can respond quickly to changes in concentration and output an electrical signal through the electrical signal transmission module 3 to calculate the real-time hydrogen concentration. The optical wavelength drift signal output by the fiber optic sensor is processed by a demodulator to obtain the corresponding concentration. By employing a joint optimization strategy combining extended Kalman filtering (EKF) and recursive least squares (RLS), the dynamic response and sensitivity factor of the fiber optic sensor are adaptively corrected, thereby obtaining the optimized concentration curve. The curve shows improvements in both response speed and concentration accuracy, enabling high-precision and long-term stable detection of hydrogen concentration.
[0064] This invention integrates an optical fiber sensor and a MOS hydrogen sensor within a sealed test chamber and establishes a dynamic calibration mechanism based on the reference signal of the MOS hydrogen sensor, thereby achieving complementary advantages between the two types of sensors. The MOS hydrogen sensor is upgraded from a single "static reference" to a "dynamically optimized reference," establishing a collaborative optimization mechanism between the two types of sensors.
[0065] A MOS hydrogen sensor provides sensitive and rapid dynamic response information, while a fiber optic sensor outputs stable and interference-resistant monitoring results. Through data fusion and dynamic calibration, the dynamic reference signal from the MOS hydrogen sensor is input into the fiber optic detection results in real time, correcting its insufficient sensitivity and dynamic response lag. This improves the accuracy of concentration detection by the fiber optic hydrogen sensor, enhancing long-term stability and environmental adaptability while maintaining detection sensitivity. Real-time correction of the fiber optic sensor response curve is achieved through a controlled gas-sensitive cavity environment and dynamic gas input, rather than just single-point calibration. EKF filtering eliminates dynamic lag and noise interference in fiber optic sensor measurements, improving signal real-time performance and stability. RLS adaptive updates of the fiber optic sensor's sensitivity scaling factor and bias coefficient enable dynamic compensation for environmental changes. Ultimately, this significantly improves the accuracy and long-term consistency of the fiber optic sensor in concentration detection.
[0066] The above inventions are merely a few specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. An experimental method for optimizing a fiber optic hydrogen sensor, characterized in that, Includes the following steps: The fiber optic hydrogen sensor and the MOS hydrogen sensor are placed in the same sealed gas-sensitive test chamber. Hydrogen gas is introduced into the gas-sensitive test chamber to form a preset dynamic concentration curve, and the hydrogen concentration data in the chamber measured by the fiber optic hydrogen sensor and the real-time hydrogen concentration data are acquired simultaneously. A joint algorithm based on extended Kalman filtering and recursive least squares is used to process the concentration data to adaptively correct the dynamic response and sensitivity factor of the fiber optic hydrogen sensor. Specifically, the extended Kalman filtering uses the real-time measured hydrogen concentration data as observations to estimate the instantaneous dynamic bias and slowly varying zero-point drift of the fiber optic hydrogen sensor. The recursive least squares method uses the bias reduction estimation result of the extended Kalman filtering to update the sensitivity scaling factor of the fiber optic hydrogen sensor and feeds the updated sensitivity scaling factor back into the observation model of the extended Kalman filtering. An anomaly detection mechanism is used to handle observational or parameter anomalies, and the fiber optic hydrogen sensor concentration curve, corrected by the joint algorithm, is output.
2. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 1, characterized in that, The modeling and execution of the extended Kalman filter includes: Define the state vector: in: The actual hydrogen concentration inside the cavity; Instantaneous dynamic bias of fiber optic sensors; Slowly varying zero-point drift in fiber optic sensors; Extended Kalman filter modeling: Assuming the MOS hydrogen sensor is an ideal reference with no system bias and only measurement noise; the fiber optic sensor (FOS) includes sensitivity, instantaneous bias, and slowly varying bias: in This refers to the hydrogen concentration actually observed by MOS. ; This refers to the hydrogen concentration actually observed by FOS. ; This is the sensitivity scaling factor for the fiber optic sensor. , For measuring noise; pass Predicting prior estimates before new measurements arrive is used to maintain the continuity of sensor output, where... is the state transition Jacobian matrix, where This represents the attenuation coefficient of the instantaneous bias.
3. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 2, characterized in that, The adaptive update of the recursive least squares (RLS) method includes: Target output, bias-free FOS observations: For extended Kalman filter pairs The posterior estimate; For extended Kalman filter pairs posterior estimation ; Update the sensitivity factor: For the recursive least squares method to update at each time step The estimated value; It is the least squares gain scalar; This represents the prediction error using the least squares method. The updated Let's return to the calculation of the observation function and Jacobian matrix of the extended Kalman filter at the next time step.
4. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 3, characterized in that, The final corrected concentration curve of the fiber optic hydrogen sensor was obtained using the following compensation formula: 。 5. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 1, characterized in that, Dynamic concentration includes at least one of step changes, linear ramp changes, or periodic oscillation changes.
6. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 1, characterized in that, The anomaly detection includes: Innovation detection of extended Kalman filter: The innovation amount generated during the recursive calculation of extended Kalman filter is normalized to obtain the normalized innovation square. If the value exceeds the threshold, the observation is judged to be abnormal. When the innovation detection of extended Kalman filter determines that the observation is abnormal, at least one of the following three measures shall be taken: temporarily increasing the observation noise covariance, suspending the least squares method update, and alarm recording. Anomaly protection of least squares method: If the prediction error of the sensitivity scaling factor exceeds the preset error threshold, skip the current update of the sensitivity scaling factor or update it with limited gain. Sensitivity scaling factor detection of fiber optic sensor: If the sensitivity scaling factor of the fiber optic sensor exceeds the preset reasonable range, the update of the sensitivity scaling factor is abandoned.
7. The experimental method for optimizing the fiber optic hydrogen sensor as described in claim 6, characterized in that, When the extended Kalman filter innovation detection determines that the observation is abnormal, the state estimation parameters and covariance matrix of the extended Kalman filter remain unchanged until the observation value returns to the confidence interval. The condition for restoring the recursive least squares update is that the normalized squared innovation value of the extended Kalman filter innovation detection is lower than the preset recovery threshold for N consecutive sampling times.
8. The apparatus used in the experimental method for optimizing the fiber optic hydrogen sensor as described in claim 1, characterized in that, include: Gas-sensitive test chamber (1), MOS hydrogen sensor, fiber optic hydrogen sensor, optical signal transmission module (2), electrical signal transmission module (3), and data processing and control unit; The gas-sensitive test chamber (1) includes an air inlet (11) and an air outlet (12). The air inlet (11) is connected to an external hydrogen source through an air inlet valve (111), and the air outlet (12) is connected to an external exhaust pipe through an air outlet valve (121). The MOS hydrogen sensor and the fiber optic hydrogen sensor are disposed inside the gas-sensitive test chamber (1); One end of the optical signal transmission module (2) is connected to the optical fiber hydrogen sensor, and the other end is used to connect to an external fiber optic demodulator. One end of the electrical signal transmission module (3) is electrically connected to the MOS hydrogen sensor, and the other end is used to connect to an external computer. The data processing and control unit is communicatively connected to the fiber optic demodulator, an external computer, and the inlet valve (111) and outlet valve (121), respectively. The data processing and control unit controls the opening and closing states of the inlet valve (111) and the outlet valve (121) to form a preset dynamic concentration change curve in the gas-sensitive test chamber (1); The data processing and control unit receives and processes the signal from the fiber optic hydrogen sensor of the fiber optic demodulator to obtain its measured concentration data, and also receives the signal from the MOS hydrogen sensor connected to an external computer to obtain its measured concentration data. The data processing and control unit performs the experimental method for optimizing the fiber optic hydrogen sensor as described in any one of claims 1 to 7.