A multi-modal feature fusion actuator stroke calibration system

Abstract

The application relates to the technical field of actuator sensors, and provides an actuator stroke calibration system based on multi-modal feature fusion, which comprises a sensor module, a signal conditioning module, a data extraction module, a multi-modal feature fusion module and a stroke output and calibration module; the sensor module is used for collecting multi-dimensional information of an actuator stroke and state reflected by multiple sensors; the signal conditioning module is used for amplifying and aligning the multi-dimensional information; the data extraction module is used for performing dimension compression on the aligned multi-dimensional information and outputting a feature vector; the multi-modal feature fusion module is used for fusing the feature vector and outputting a stroke estimation value; and the stroke calibration module is used for correcting the stroke estimation value and outputting a final stroke value. The application solves the problems of limited precision, poor robustness, poor dynamic performance and weak environmental adaptability of a single actuator sensor in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of motion sensor technology, and in particular relates to a multimodal feature fusion actuator stroke calibration system. Background Technology

[0002] Traditional calibration systems typically rely on a single position sensor, such as an LVDT, magnetic scale, or encoder. These systems are susceptible to installation errors, temperature drift, electromagnetic interference, mechanical wear, and nonlinear characteristics, and their accuracy may decrease or they may fail under harsh operating conditions.

[0003] In addition, most calibrations are performed under static or quasi-static conditions, which makes it difficult to reflect the true stroke characteristics of the actuator during dynamic motion. Therefore, a high-precision, high-robust actuator stroke calibration system that can adapt to dynamic processes and complex environments is needed. Summary of the Invention

[0004] In view of the above-mentioned deficiencies of the prior art, this invention proposes a multimodal feature fusion actuator stroke calibration system to solve the problems of limited accuracy, poor robustness, poor dynamic performance, and weak environmental adaptability of single actuator sensors in the prior art. The technical solution of this invention includes the following steps:

[0005] The module includes a sensor module, a signal conditioning module, a data extraction module, a multimodal feature fusion module, and a stroke output and calibration module.

[0006] The sensor module is used to collect multi-dimensional information from multiple sensors reflecting the actuator's stroke and status;

[0007] The signal conditioning module is used to amplify and align multi-dimensional information;

[0008] The data extraction module is used to compress the aligned multi-dimensional information and output a feature vector.

[0009] The multimodal feature fusion module is used to fuse feature vectors and output travel estimates;

[0010] The trip calibration module is used to correct the trip estimate and output the final trip value.

[0011] Preferably, the sensor includes:

[0012] Displacement sensing unit, vibration sensing unit, and environmental sensing unit.

[0013] Preferably, the process of fusing feature vectors and outputting travel estimates includes:

[0014] The feature vectors are passed through the physical model constraint layer and the dynamic weighted fusion layer respectively to output the travel estimate.

[0015] Preferably, the step of correcting the travel estimate and outputting the final travel value includes:

[0016] Identify abnormal sensors, preprocess the abnormal sensor data, calculate the correlation value between the abnormal sensor and the actuator stroke, assign dynamic weights to the abnormal sensors, generate correction values ​​based on the correlation values ​​and dynamic weights, and output the final stroke value based on the correction values ​​and stroke estimates.

[0017] Preferably, the source of the abnormal sensor is identified as including:

[0018] The multi-dimensional information deviations of each sensor are calculated in real time, using the following formula:

[0019]

[0020] In the formula, Let m be the deviation value of sensor. Let m be the data value of sensor m at time t. This is the steady-state reference value for sensor m;

[0021] when The time stamp is marked as an abnormal sensor source. This represents the historical standard deviation.

[0022] Preferably, the preprocessing of the abnormal sensor data is performed using the following formula:

[0023]

[0024]

[0025] In the formula, For the normalized data of sensor m, For the normalized sequence of the travel reference, Let m be the mean value of sensor m within the observation window. Let m be the standard deviation of sensor m within the observation window. As the reference for the trip, for The mean, for The standard deviation.

[0026] Preferably, the correlation value between the calculation of the anomaly sensor and the actuator stroke is calculated using the following formula:

[0027]

[0028] In the formula, This is the correlation value between sensor m and stroke. The total number of samples, To normalize the data mean, This is the normalized benchmark mean.

[0029] Preferably, the dynamic weights assigned to the anomaly sensors are as follows:

[0030]

[0031]

[0032] In the formula, Let m be the fusion coefficient of sensor m. Let be the covariant energy of sensors m and n. For time delay parameters, This represents the maximum time delay.

[0033] Preferably, the formula for generating the correction amount based on the correlation value and dynamic weight is as follows:

[0034]

[0035]

[0036] In the formula, Let m be the local compensation amount for sensor m. For correction amount, This is the adaptive gain factor.

[0037] Beneficial effects:

[0038] This application proposes a multimodal feature fusion actuator stroke calibration system. It uses displacement / vibration / temperature sensors to work together and applies the mutual variable energy weighting to the actuator stroke calibration to generate correction values ​​to provide stroke calibration accuracy, thus solving the problem of actuator stroke calibration in high dynamic and strong interference environments. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of a preferred embodiment of the present invention. Detailed Implementation

[0040] The embodiments of the present invention will be described in detail below. The embodiments described below are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given. However, the protection scope of the present invention is not limited to the embodiments described below.

[0041] This invention designs an actuator stroke calibration system that integrates multimodal feature fusion, such as... Figure 1 As shown, it specifically includes:

[0042] The module includes a sensor module, a signal conditioning module, a data extraction module, a multimodal feature fusion module, and a stroke output and calibration module.

[0043] The sensor module is used to collect multi-dimensional information from multiple sensors reflecting the actuator's stroke and state;

[0044] The signal conditioning module is used to amplify and align multi-dimensional information;

[0045] The data extraction module is used to compress the alignment of multi-dimensional information and output a feature vector;

[0046] The multimodal feature fusion module is used to fuse feature vectors and output travel estimates;

[0047] The travel calibration module is used to correct the travel estimate and output the final travel value.

[0048] Specifically, the signal conditioning module includes a signal amplification and filtering circuit and a synchronous sample-and-hold circuit. The signal amplification and filtering circuit amplifies and impedance-matches weak signals (such as millivolts) or high-impedance signals output by different sensors, and uses an anti-aliasing filter to suppress high-frequency noise. The synchronous sample-and-hold circuit provides a unified sampling clock signal for all sensor channels, ensuring that all modal data are strictly aligned at the time point and eliminating errors introduced by sampling time differences.

[0049] In addition, the data extraction module includes time-domain feature extraction, frequency-domain feature extraction, and feature dimensionality reduction. Among them, time-domain feature extraction extracts effective features from preprocessed sensor data, such as mean, variance, root mean square (RMS), and zero-crossing rate. Frequency-domain feature extraction performs Fast Fourier Transform (FFT) on vibration and other signals to extract features such as dominant frequency, spectral centroid, and frequency band energy. Feature dimensionality reduction uses Principal Component Analysis (PCA) or an autoencoder to reduce the dimensionality of the high-dimensional feature set, remove redundant information, and form a low-dimensional feature vector for use by the subsequent fusion module.

[0050] Preferably, the sensor includes:

[0051] Displacement sensing unit, vibration sensing unit, and environmental sensing unit.

[0052] Specifically, the displacement sensing unit can employ a linear variable differential transformer (LVDT), photoelectric encoder, or laser displacement sensor to directly measure the absolute or relative displacement of the actuator push rod; the vibration sensing unit can employ a MEMS or piezoelectric accelerometer, arranged on the actuator housing or mounting base, to collect vibration acceleration signals generated during actuator operation, with its frequency response characteristics covering the actuator's main operating frequency band; the environmental sensing unit may include temperature and humidity sensors, installed near the actuator, to monitor changes in ambient temperature and humidity, as these parameters may affect the sensor reading materials and the performance of the actuator itself.

[0053] Preferably, the process of fusing feature vectors and outputting travel estimates includes:

[0054] The feature vectors are passed through the physical model constraint layer and the dynamic weighted fusion layer respectively to output the travel estimate.

[0055] Specifically, the physical model constraint layer includes:

[0056] The kinematic model of the actuator is constructed as follows:

[0057]

[0058] In the formula, For thrust coefficient, The control command voltage for the actuator. Let be the Coulomb friction coefficient. The sign function for velocity direction;

[0059] The dynamic weighted fusion layer includes:

[0060] The fusion weight is calculated based on the mutual variable energy, using the following formula:

[0061]

[0062] The estimated travel value is output using the following formula:

[0063]

[0064] In the formula, Let m be the feature vector of sensor m. This is the weighted coefficient matrix of the eigenvectors.

[0065] Preferably, correcting the travel estimate and outputting the final travel value includes:

[0066] Identify abnormal sensors, preprocess the abnormal sensor data, calculate the correlation value between the abnormal sensor and the actuator stroke, assign dynamic weights to the abnormal sensors, generate correction values ​​based on the correlation values ​​and dynamic weights, and output the final stroke value based on the correction values ​​and stroke estimates.

[0067] Preferably, identifying abnormal sensor sources includes:

[0068] The multi-dimensional information deviations of each sensor are calculated in real time, using the following formula:

[0069]

[0070] In the formula, Let m be the deviation value of sensor. Let m be the data value of sensor m at time t. This is the steady-state reference value for sensor m;

[0071] when The time stamp is marked as an abnormal sensor source. This represents the historical standard deviation.

[0072] Preferably, the abnormal sensor data is preprocessed using the following formula:

[0073]

[0074]

[0075] In the formula, For the normalized data of sensor m, For the normalized sequence of the travel reference, Let m be the mean value of sensor m within the observation window. Let m be the standard deviation of sensor m within the observation window. As the reference for the trip, for The mean, for The standard deviation.

[0076] Preferably, the correlation value between the fault sensor and the actuator stroke is calculated using the following formula:

[0077]

[0078] In the formula, This is the correlation value between sensor m and stroke. The total number of samples, To normalize the data mean, This is the normalized benchmark mean.

[0079] Specifically, the formula for normalizing the data mean is as follows:

[0080] .

[0081] The normalized benchmark mean is calculated using the following formula:

[0082] .

[0083] Preferably, dynamic weights are assigned to the anomaly sensors, using the following formula:

[0084]

[0085]

[0086] In the formula, Let m be the fusion coefficient of sensor m. Let be the covariant energy of sensors m and n. For time delay parameters, This represents the maximum time delay.

[0087] Preferably, the correction amount is generated based on the correlation value and dynamic weights, as shown in the following formula:

[0088]

[0089]

[0090] In the formula, Let m be the local compensation amount for sensor m. For correction amount, This is the adaptive gain factor.

[0091] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A multimodal feature fusion actuator stroke calibration system, characterized in that, include: The module includes a sensor module, a signal conditioning module, a data extraction module, a multimodal feature fusion module, and a stroke output and calibration module. The sensor module is used to collect multi-dimensional information from multiple sensors reflecting the actuator's stroke and status; The signal conditioning module is used to amplify and align multi-dimensional information; The data extraction module is used to compress the aligned multi-dimensional information and output a feature vector. The multimodal feature fusion module is used to fuse feature vectors and output travel estimates; The travel output and calibration module is used to correct the travel estimate and output the final travel value; The process of correcting the travel estimate and outputting the final travel value includes: Identify abnormal sensors, preprocess the abnormal sensor data, calculate the correlation value between the abnormal sensor and the actuator stroke, assign dynamic weights to the abnormal sensors, generate correction values ​​based on the correlation values ​​and dynamic weights, and output the final stroke value based on the correction values ​​and stroke estimates. The anomaly detection sensor includes: The multi-dimensional information deviations of each sensor are calculated in real time, using the following formula: In the formula, Let m be the deviation value of sensor. Let m be the data value of sensor m at time t. This is the steady-state reference value for sensor m; when The time stamp is marked as an abnormal sensor source. The historical standard deviation; The preprocessing formula for abnormal sensor data is as follows: In the formula, For the normalized data of sensor m, For the normalized sequence of the travel reference, Let m be the mean value of sensor m within the observation window. Let m be the standard deviation of sensor m within the observation window. As the reference for the trip, for The mean, for The standard deviation.

2. The actuator stroke calibration system based on multimodal feature fusion according to claim 1, characterized in that, The sensor includes: Displacement sensing unit, vibration sensing unit, and environmental sensing unit.

3. The actuator stroke calibration system based on multimodal feature fusion according to claim 1, characterized in that, The process of fusing feature vectors and outputting travel estimates includes: The feature vectors are passed through the physical model constraint layer and the dynamic weighted fusion layer respectively to output the travel estimate.

4. The actuator stroke calibration system based on multimodal feature fusion according to claim 1, characterized in that, The formula for calculating the correlation between the abnormal sensor and the actuator stroke is as follows: In the formula, This is the correlation value between sensor m and stroke. The total number of samples, To normalize the data mean, This is the normalized baseline mean.

5. The actuator stroke calibration system based on multimodal feature fusion according to claim 1, characterized in that, The dynamic weights for assigning abnormal sensors are as follows: In the formula, Let m be the fusion coefficient of sensor m. Let be the covariant energy of sensors m and n. For time delay parameters, This represents the maximum time delay.

6. The actuator stroke calibration system based on multimodal feature fusion according to claim 4 or 5, characterized in that, The formula for generating the correction amount based on the correlation value and dynamic weight is as follows: In the formula, Let m be the local compensation amount for sensor m. For correction amount, This is the adaptive gain factor.

Images