Wide-range magnetostriction crack meter based on machine self-learning

Abstract

The invention relates to the field of structural health monitoring, and discloses a wide-range magnetostriction crack meter based on machine self-learning. Comprising a magnetostriction measuring assembly, and the magnetostriction measuring assembly comprises a measuring rod, a waveguide wire arranged on the measuring rod, a movable permanent magnetic field generating device arranged on the outer side of the waveguide wire in a sleeving mode and a detection coil fixed to one end of the measuring rod; the signal conditioning module is electrically connected with the detection coil and is used for carrying out filtering, amplification and analog-to-digital conversion processing on the induction signal; and the edge calculation unit is electrically connected with the signal conditioning module, and the edge calculation unit constructs a multi-dimensional error model comprising a nonlinear error feature, a temperature drift feature and a hysteresis effect feature by acquiring historical measurement data of the magnetostriction measurement assembly in a preset calibration interval.

Description

Technical Field

[0001] This application relates to the field of structural health monitoring technology, and in particular to a large-range magnetostrictive crack gauge based on machine self-learning. Background Technology

[0002] In the health monitoring of civil engineering, hydraulic engineering, and mechanical structures, accurate measurement of crack displacement is crucial for structural safety assessment. Magnetostrictive crack gauges are widely used due to their advantages such as non-contact measurement, long lifespan, and good stability; however, they face the following technical bottlenecks in practical applications: 1. The contradiction between measurement range and accuracy: Traditional magnetostrictive sensors measure through a single range. In large-range scenarios (such as long-term deformation of bridge structures and displacement monitoring of soil and rock slopes), due to the nonlinear characteristics of waveguide wire materials and uneven magnetic field distribution, it is difficult to balance the measurement range and resolution, resulting in a narrow high-precision measurement range. 2. Insufficient environmental adaptability: Temperature changes can cause the elastic modulus of the waveguide wire to drift and the permanent magnetic field strength to decay. Device aging can lead to an aggravation of the hysteresis effect. Traditional sensors rely on fixed calibration parameters and cannot dynamically compensate for environmental changes and nonlinear errors during long-term operation. 3. Lack of self-learning ability: Existing equipment lacks in-depth utilization of historical data, cannot optimize measurement models through data-driven methods, and accumulates significant errors during long-term monitoring, making it difficult to meet the long-term reliability requirements of structural safety monitoring.

[0003] To address the aforementioned issues, existing technologies employ error correction methods such as segmented calibration or temperature compensation circuits. However, these methods only address a single error factor, fail to construct an error model involving multi-physics coupling, and cannot achieve adaptive updates of model parameters. Summary of the Invention

[0004] This application provides a large-range magnetostrictive crack gauge based on machine self-learning, which aims to solve the problem that existing technologies use segmented calibration or temperature compensation circuits to correct errors. However, such solutions only address a single error factor, do not construct an error model that couples multiple physical fields, and cannot achieve adaptive updates of model parameters.

[0005] In a first aspect, embodiments of this application provide a large-range magnetostrictive crack gauge based on machine self-learning, comprising: A magnetostrictive measurement assembly includes a measuring rod, a waveguide wire disposed on the measuring rod, a movable permanent magnetic field generator sleeved on the outside of the waveguide wire, and a detection coil fixed to one end of the measuring rod. The movable permanent magnetic field generator moves along the axial direction of the waveguide wire as the crack opens and closes. The detection coil is used to collect the induced signal generated by the waveguide wire due to the magnetostrictive effect. The signal conditioning module is electrically connected to the detection coil and is used to filter, amplify, and perform analog-to-digital conversion on the sensed signal. An edge computing unit, electrically connected to the signal conditioning module, acquires historical measurement data of the magnetostrictive measurement component within a preset calibration range and constructs a multidimensional error model including nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics. Based on the multidimensional error model, it preprocesses the real-time acquired induction signal data to generate initial displacement measurement values. It iteratively trains the multidimensional error model using actual crack displacement monitoring data to dynamically optimize model parameters to adapt to environmental changes and device aging characteristics.

[0006] In some embodiments, the device further includes a multi-range fusion module electrically connected to the edge computing unit, the multi-range fusion module being configured to automatically switch the measurement range of the magnetostrictive measurement component based on the dynamic range of the initial displacement measurement value, the measurement range including at least a high-precision short-range mode and a large-range extended mode.

[0007] In some embodiments, in the high-precision short-range mode, the movable permanent magnetic field generator is located in the calibration sensitive range of the waveguide wire, and the detection coil uses a narrowband filtering algorithm to improve resolution; in the large-range extended mode, the waveguide wire is divided into multiple continuous measurement segments by a segmented calibration algorithm, and the cross-segment measurement data is fused and compensated by the self-learning algorithm module.

[0008] In some embodiments, the system further includes: a storage module electrically connected to the edge computing unit for storing the multidimensional error model parameters, historical measurement data, and calibration coefficients; and a communication module electrically connected to the edge computing unit for uploading the processed displacement data and self-learning state parameters to a remote server.

[0009] In some embodiments, constructing a multidimensional error model including nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics includes: collecting magnetostrictive induction signals and standard displacement reference values ​​under different temperature conditions and displacement amounts within a preset calibration interval, and extracting nonlinear characteristic parameters such as signal amplitude, rising edge slope, and zero-crossing time difference; acquiring ambient temperature data in real time through a temperature sensor and establishing a mapping relationship between temperature-zero drift and sensitivity drift; constructing a hysteresis error matrix based on hysteresis loop test data; and using a support vector regression algorithm or artificial neural network to couple and model the nonlinear characteristic parameters, temperature drift data, and hysteresis error matrix to generate a multidimensional mathematical model containing multi-physics error factors.

[0010] In some embodiments, the preprocessing of the real-time acquired sensing signal data based on the multidimensional error model to generate initial displacement measurement values ​​includes: extracting the timestamp, waveform peak value, and zero-crossing time difference of the real-time sensing signal as input features; performing temperature drift compensation on the input features based on real-time ambient temperature data to correct the temperature correlation deviation of the elastic modulus of the waveguide wire material and the magnetic field distribution; mapping the compensated signal features through the nonlinear correction function of the multidimensional error model, and performing hysteresis error compensation on the displacement measurement values ​​in combination with the hysteresis effect correction matrix to output the initial displacement measurement values.

[0011] In some embodiments, the iterative training of the multidimensional error model using actual crack displacement monitoring data to dynamically optimize model parameters to adapt to environmental changes and device aging characteristics includes: establishing a monitoring data quality assessment mechanism to remove abnormal data points and form an effective training dataset; using an incremental learning algorithm to update the multidimensional error model online, calculating the model prediction error by comparing real-time measured values ​​with calibration data from an external reference sensor; and triggering a model parameter optimization process when the prediction error exceeds a preset threshold, adaptively adjusting the nonlinear error coefficient, temperature drift coefficient, and hysteresis effect correction factor to form an iteratively optimized error compensation model.

[0012] In some embodiments, the filtering, amplification, and analog-to-digital conversion processing of the induced signal includes: filtering out power frequency interference and high-frequency noise using a bandpass filter while retaining the characteristic frequency band of the magnetostrictive torsional wave signal; dynamically adjusting the gain of the filtered signal using a programmable gain amplifier to make the signal amplitude match the input range of the analog-to-digital converter; and sampling the amplified analog signal using a high-precision analog-to-digital converter at a sampling frequency not less than twice the highest frequency of the magnetostrictive torsional wave to generate a digital signal sequence.

[0013] In some embodiments, the acquisition of the induced signal generated by the magnetostrictive effect of the waveguide wire includes: the detection coil is wound with a high permeability magnetic core, and the number of coil turns is in a preset ratio to the length of the waveguide wire to match the induction sensitivity of the magnetostrictive torsional wave; when the movable permanent magnetic field generator moves, the magnetostrictive effect is excited on the surface of the waveguide wire to generate a torsional wave, and the detection coil acquires the induced electromotive force signal of the torsional wave propagating to the coil position through electromagnetic induction, and simultaneously records the signal arrival time and waveform characteristics.

[0014] In some embodiments, the edge computing unit is further configured to use a long short-term memory network to model the temporal correlation of historical displacement data sequences and predict the trend of crack displacement; dynamically adjust the regularization parameters of the multidimensional error model by combining a particle swarm optimization algorithm to balance the model fitting accuracy and generalization ability; when the displacement change rate is detected to exceed a preset threshold, automatically trigger the encrypted sampling mode, and use a convolutional neural network to identify the features of abnormal signal waveforms to determine whether it is a sudden crack expansion event.

[0015] This invention constructs a multi-dimensional model incorporating nonlinear errors, temperature drift, and hysteresis effects using historical data from a preset calibration interval. This effectively corrects multi-physics coupling interference, improving measurement accuracy compared to traditional single-parameter compensation methods. It iteratively trains the error model using actual monitoring data, dynamically adapting to time-varying factors such as environmental temperature changes and device aging. This avoids the accuracy degradation issues of traditional sensors due to long-term use, extending the effective monitoring cycle of the equipment. The multi-range fusion module automatically matches high-precision short-range modes with large-range extended modes, extending the measurement range to the meter level while maintaining millimeter-level resolution, meeting the accuracy and range requirements of different monitoring scenarios. A built-in self-learning algorithm module enables local data processing and model updates, reducing reliance on remote servers and improving data processing efficiency and system reliability. This makes it particularly suitable for long-term monitoring in remote areas or environments with limited network access.

[0016] In summary, this invention, through the deep integration of intelligent algorithms and magnetostrictive sensing technology, solves the problems of accuracy decay and poor environmental adaptability of traditional crack gauges in large-range measurements, and significantly improves the automation and intelligence level of structural health monitoring.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of a large-range magnetostrictive crack gauge based on machine self-learning provided in an embodiment of this application; Figure 2 This is a schematic block diagram of a large-range magnetostrictive crack gauge based on machine self-learning provided in an embodiment of this application.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0023] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0024] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0027] In the health monitoring of civil engineering, hydraulic engineering, and mechanical structures, accurate measurement of crack displacement is crucial for structural safety assessment. Magnetostrictive crack gauges are widely used due to their advantages such as non-contact measurement, long lifespan, and good stability; however, they face the following technical bottlenecks in practical applications: 1. The contradiction between measurement range and accuracy: Traditional magnetostrictive sensors measure through a single range. In large-range scenarios (such as long-term deformation of bridge structures and displacement monitoring of soil and rock slopes), due to the nonlinear characteristics of waveguide wire materials and uneven magnetic field distribution, it is difficult to balance the measurement range and resolution, resulting in a narrow high-precision measurement range. 2. Insufficient environmental adaptability: Temperature changes can cause the elastic modulus of the waveguide wire to drift and the permanent magnetic field strength to decay. Device aging can lead to an aggravation of the hysteresis effect. Traditional sensors rely on fixed calibration parameters and cannot dynamically compensate for environmental changes and nonlinear errors during long-term operation. 3. Lack of self-learning ability: Existing equipment lacks in-depth utilization of historical data, cannot optimize measurement models through data-driven methods, and accumulates significant errors during long-term monitoring, making it difficult to meet the long-term reliability requirements of structural safety monitoring.

[0028] To address the aforementioned issues, existing technologies employ error correction methods such as segmented calibration or temperature compensation circuits. However, these methods only address a single error factor, fail to construct an error model involving multi-physics coupling, and cannot achieve adaptive updates of model parameters.

[0029] Please refer to Figures 1-2 This application provides a large-range magnetostrictive crack gauge based on machine self-learning, including a magnetostrictive measurement component. The component includes a measuring rod, a waveguide wire mounted on the measuring rod, a movable permanent magnetic field generator sleeved on the outside of the waveguide wire, and a detection coil fixed to one end of the measuring rod. The movable permanent magnetic field generator moves along the axial direction of the waveguide wire as the crack opens and closes. The detection coil is used to collect the induced signal generated by the magnetostrictive effect of the waveguide wire. A signal conditioning module, electrically connected to the detection coil, is used to condition the induced signal. The signal is filtered, amplified, and converted from analog to digital. An edge computing unit, electrically connected to the signal conditioning module, acquires historical measurement data from the magnetostrictive measurement component within a preset calibration range to construct a multidimensional error model incorporating nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics. Based on this multidimensional error model, the real-time acquired inductive signal data is preprocessed to generate initial displacement measurements. The multidimensional error model is then iteratively trained using actual crack displacement monitoring data to dynamically optimize model parameters to adapt to environmental changes and device aging characteristics.

[0030] Specifically, this invention proposes a magnetostrictive crack gauge that integrates machine self-learning. Through multi-dimensional error modeling and dynamic parameter optimization, it breaks through the bottlenecks of traditional sensors in terms of range, accuracy, environmental adaptability and long-term reliability.

[0031] The magnetostrictive measurement component realizes displacement sensing based on the magnetostrictive effect. It combines a movable permanent magnetic field device and a detection coil to convert the crack opening and closing displacement into an electrical signal. The edge computing-driven intelligent error compensation constructs a multi-dimensional model that includes nonlinear error, temperature drift, and hysteresis effect through the edge computing unit. It uses historical data to learn and dynamically optimize the model parameters, realizing high-precision measurement across the entire range and long-term environmental adaptability.

[0032] The measuring rod serves as a mechanical support structure, fixing the waveguide wire and transmitting the opening and closing displacement of the crack. It is made of high-rigidity, low-expansion-coefficient materials (such as Invar steel) to reduce the interference of ambient temperature on structural deformation.

[0033] The waveguide wire is fixed along the axial direction of the measuring rod and is made of magnetostrictive material (such as Terfenol-D or Galfenol). Its length covers the target measurement range (such as 0-500 mm). The magnetostrictive effect of the waveguide wire causes it to generate torsional stress waves under the action of a magnetic field.

[0034] The movable permanent magnetic field generator is usually a ring-shaped permanent magnet, which is sleeved on the outside of the waveguide wire and connected to the crack displacement transmission mechanism (such as a connecting rod or slider). It moves along the axial direction of the waveguide wire as the crack opens and closes. The uniformity of its magnetic field distribution is improved by optimizing the magnetic circuit design (such as adding a soft magnetic shield).

[0035] The detection coil is fixed to one end of the probe (such as the left end) to sense the change in magnetic field caused by the stress wave generated by the magnetostriction effect of the waveguide wire, and outputs a weak electrical signal (mV level).

[0036] When the permanent magnetic field device is located at a certain position on the waveguide wire, the excitation circuit applies a pulse current to the waveguide wire to generate a ring magnetic field. The interaction between the permanent magnetic field and the ring magnetic field induces a magnetostrictive effect, generating a torsional stress wave that propagates to both ends. The detection coil captures the time signal t of the stress wave arrival. Based on the propagation speed v of the stress wave in the waveguide wire, the position L=vt / 2 of the permanent magnetic field is calculated to achieve displacement measurement.

[0037] The signal processing flow includes: filtering: using a bandpass filter (such as a Butterworth filter) to filter out environmental electromagnetic interference (such as 50Hz power frequency noise); amplification: using a low-noise operational amplifier (such as an AD620 instrumentation amplifier) ​​to amplify the mV level signal to the V level to meet the analog-to-digital conversion requirements; analog-to-digital conversion (ADC): using a high-precision ADC chip (such as a 24-bit Δ-Σ type ADC) to improve signal quantization accuracy and reduce quantization error.

[0038] The auxiliary sensor integration uses a built-in temperature sensor (such as PT100 or digital temperature chip DS18B20) to collect the ambient temperature T of the waveguide wire in real time, which is used as one of the input parameters of the multidimensional error model.

[0039] The construction of the multidimensional error model includes: input features: including original induction signal features (such as pulse peak value, rising edge slope), temperature T, historical displacement data Lhist, and hysteresis loop parameters (fitted by forward and reverse travel data).

[0040] Error factor modeling includes: Nonlinear error: Based on the nonlinear characteristics of the magnetostriction coefficient of the waveguide wire material, the mapping relationship between the input signal and the actual displacement is fitted by piecewise polynomials or neural networks; Temperature drift: Establishing a correction model for the effect of temperature on the elastic modulus E(T) and permanent magnetic field strength B(T) of the waveguide wire (such as the thermocouple effect formula or experimental calibration curve); Hysteresis effect: Constructing a hysteresis loop by collecting forward and reverse travel data, and introducing a Preisach model or empirical hysteresis curve for parameterization. The model structure adopts machine learning algorithms (such as random forest, long short-term memory network LSTM or lightweight neural network), and is initially trained in a preset calibration range (such as key position data such as 0-10% range, 50% range, and 90-100% range collected during the first installation) to form an error correction model f:{S,T,Lhist}→ΔL containing multi-physics coupling relationships, where S is the characteristic of the induced signal and ΔL is the error compensation amount.

[0041] The initial displacement calculation is based on the traditional magnetostrictive time difference algorithm to obtain the preliminary measured value L0=v0*t / 2, where v0 is the initial wave velocity at room temperature (determined by factory calibration).

[0042] Preprocessing and correction: Input L0, real-time temperature T and current signal characteristics S into the multidimensional model, output the error compensation amount ΔL=f(S,T,L0), and obtain the corrected displacement L=L0+ΔL.

[0043] Self-learning iterative optimization includes: Data accumulation: The edge computing unit stores historical monitoring data (including actual displacement true value Ltrue, sensor measurement value L, temperature T, signal characteristics S, etc.). When the amount of data reaches a preset threshold (such as 1000 sets), the model is updated.

[0044] Dynamic training employs online learning algorithms (such as stochastic gradient descent SGD) with the mean square error (MSE) of the actual displacement and the model output as the objective function. It iteratively optimizes the model parameters to adapt to device aging (such as permanent magnetic field decay and waveguide wire performance drift) and long-term environmental changes (such as temperature cycling and humidity effects).

[0045] The calibration mechanism automatically enters calibration mode periodically (e.g., monthly) or when significant error drift is detected (e.g., when 10 consecutive sets of data errors exceed the threshold). It collects calibration data through a built-in calibration device (e.g., a precision displacement stage), resets the model's initial parameters, and avoids error accumulation.

[0046] By breaking through the limitations of traditional single-error compensation (such as temperature compensation only), this model integrates nonlinearity, temperature, and hysteresis effects, capturing complex physical relationships through machine learning and expanding the high-precision measurement range. It eliminates reliance on the cloud, completing data processing and model updates in real-time on the device, reducing latency and communication power consumption, and meeting the needs of long-term offline monitoring. Through historical data-driven model iteration, it automatically adapts to device aging (e.g., when the permanent magnetic field strength decreases by 10% after 3 years of use, the model can autonomously correct the wave velocity parameter), ensuring long-term monitoring accuracy.

[0047] With a 500mm measurement range, a resolution of 0.05%FS (full scale) can be achieved through segmented modeling and error correction, which is more than 5 times higher than that of traditional sensors. Within a temperature range of -40℃ to 80℃, through real-time temperature compensation and model adaptation, the temperature drift error is reduced from ±0.1% / ℃ of traditional methods to ±0.02% / ℃. Through a self-learning mechanism, the cumulative error is controlled within 0.1%FS over long-term monitoring (more than 5 years), meeting the full life-cycle safety monitoring needs of structures such as bridges and slopes. This crack gauge can be integrated into structural health monitoring systems, uploading data via RS485, LoRa, and other interfaces, providing a high-precision, high-reliability displacement measurement solution for real-time safety assessment of civil engineering, hydraulic engineering, and mechanical structures.

[0048] In some embodiments, the device further includes a multi-range fusion module electrically connected to the edge computing unit, the multi-range fusion module being configured to automatically switch the measurement range of the magnetostrictive measurement component based on the dynamic range of the initial displacement measurement value, the measurement range including at least a high-precision short-range mode and a large-range extended mode.

[0049] By using a dynamic range switching mechanism, the contradiction between range and accuracy in traditional sensors is resolved, enabling adaptive switching between high-precision short-range and large-range extended modes.

[0050] The modular architecture integrates a multi-range fusion module into the edge computing unit, which includes range judgment logic circuits and hardware control interfaces to monitor the dynamic range of the initial displacement measurement value L0 in real time (e.g., 0~50mm is the short range, and 50~500mm is the large range).

[0051] Switching strategy: When L0 ≤ 10%FS (FS is full scale), automatically switch to high precision short range mode to activate the calibration sensitive area of ​​the waveguide wire (such as the middle high linearity section). When L0 > 10%FS, switch to large range extension mode and divide the measurement area into segments (e.g., every 50mm is a segment) using a segmented calibration algorithm.

[0052] The hardware coordinates the initial position of the permanent magnetic field device through a servo motor or elastic linkage mechanism, so that it is located in the waveguide wire section corresponding to the target range, and coordinates with the gain parameter adjustment of the detection coil (increasing the amplification factor in short range mode).

[0053] In some embodiments, in the high-precision short-range mode, the movable permanent magnetic field generator is located in the calibration sensitive range of the waveguide wire, and the detection coil uses a narrowband filtering algorithm to improve resolution; in the large-range extended mode, the waveguide wire is divided into multiple continuous measurement segments by a segmented calibration algorithm, and the cross-segment measurement data is fused and compensated by the self-learning algorithm module.

[0054] Design specific signal processing strategies for different measurement ranges to improve full-range measurement performance.

[0055] The high-precision short-range mode uses a permanent magnetic field device to fix the sensitive area in the middle of the waveguide wire (the 10% range section with the best linearity is determined by factory calibration, such as 25~75mm in the 500mm range); the detection coil uses a narrowband filtering algorithm (such as an FIR filter with a bandwidth of 10~50kHz) to filter out wideband noise, improve signal resolution, and achieve 0.01mm-level accuracy in conjunction with a 24-bit ADC. The large range extension mode divides the waveguide wire into N continuous segments (e.g., N=10, each segment is 50mm), and each segment has pre-stored segment calibration coefficients (including wave velocity v and nonlinear correction polynomial coefficients). When measuring across sections, the edge computing unit uses a self-learning algorithm to fuse overlapping data from adjacent sections (such as the overlap between 45-55mm in section 1 and 0-10mm in section 2), compensates for nonlinear errors at the section junctions, and uses a weighted average method to merge the measurement results.

[0056] In some embodiments, the system further includes: a storage module electrically connected to the edge computing unit for storing the multidimensional error model parameters, historical measurement data, and calibration coefficients; and a communication module electrically connected to the edge computing unit for uploading the processed displacement data and self-learning state parameters to a remote server.

[0057] By building a local data storage and remote communication interface, it supports the uploading of model parameters, historical data, and real-time status.

[0058] The storage module uses non-volatile memory (such as EEPROM or Flash) and stores the following: multi-dimensional error model parameters (such as neural network weights and segmented calibration coefficients); historical measurement data (including timestamps, displacement values, temperature, and signal waveform characteristics, with a storage period of the most recent year); and calibration coefficients (factory calibration values ​​and periodic self-calibration correction values).

[0059] The communication module supports multiple communication protocols: RS485 / USB for short distances and LoRa / NB-IoT for long distances, with encrypted transmission (AES-128) during wireless communication; uploaded data includes: corrected displacement values, model version number, self-learning status (such as the last training time and error threshold), and supports remote server wake-up of the device for firmware upgrades.

[0060] In some embodiments, constructing a multidimensional error model including nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics includes: collecting magnetostrictive induction signals and standard displacement reference values ​​under different temperature conditions and displacement amounts within a preset calibration interval, and extracting nonlinear characteristic parameters such as signal amplitude, rising edge slope, and zero-crossing time difference; acquiring ambient temperature data in real time through a temperature sensor and establishing a mapping relationship between temperature-zero drift and sensitivity drift; constructing a hysteresis error matrix based on hysteresis loop test data; and using a support vector regression algorithm or artificial neural network to couple and model the nonlinear characteristic parameters, temperature drift data, and hysteresis error matrix to generate a multidimensional mathematical model containing multi-physics error factors.

[0061] Based on multiphysics data acquisition and machine learning, a composite error model incorporating nonlinearity, temperature, and hysteresis effects is constructed.

[0062] Data acquisition is performed by calibration in a constant temperature chamber, covering a temperature range of -40℃ to 80℃. Magnetostrictive signals of 0 to 100%FS displacement are collected at each temperature point, and reference values ​​of a standard displacement stage (accuracy ±0.01mm) are recorded simultaneously. Extract signal features: amplitude A, rising edge slope k, and zero-crossing time difference Δt (the time difference between the excitation pulse and the zero-crossing point of the induced signal).

[0063] Single-factor modeling includes: Temperature drift: Fitting the temperature-zero drift curve ΔL0(T)=aT 2 +bT+c, Temperature-sensitivity drift coefficient α(T) = dT+e; The hysteresis effect is assessed by collecting hysteresis loop data through forward and reverse stroke scanning (e.g., 0→100%→0%FS), constructing a hysteresis error matrix H(L, direction), and recording the forward and reverse stroke deviations at each displacement point.

[0064] Coupled modeling uses support vector regression (SVR) or BP neural network, with inputs of [A,k,Δt,T, displacement direction] and outputs of total error ΔLtotal. The model hyperparameters are optimized through cross-validation (e.g., the kernel function for SVR is RBF, and the regularization parameter C=10).

[0065] In some embodiments, the preprocessing of the real-time acquired sensing signal data based on the multidimensional error model to generate initial displacement measurement values ​​includes: extracting the timestamp, waveform peak value, and zero-crossing time difference of the real-time sensing signal as input features; performing temperature drift compensation on the input features based on real-time ambient temperature data to correct the temperature correlation deviation of the elastic modulus of the waveguide wire material and the magnetic field distribution; mapping the compensated signal features through the nonlinear correction function of the multidimensional error model, and performing hysteresis error compensation on the displacement measurement values ​​in combination with the hysteresis effect correction matrix to output the initial displacement measurement values.

[0066] Multi-dimensional compensation is performed on real-time signals based on a multi-dimensional model to generate high-precision initial displacement measurements.

[0067] Feature extraction uses the timestamp tarrival, peak waveform Areal, and zero-crossing time difference Δtreal of the acquired inductive signal as input features for the model. Temperature compensation corrects the waveguide wire elastic modulus E=E0*(1+α(Treal)) based on the real-time temperature Treal, thereby adjusting the stress wave propagation velocity v=E / ρ (ρ is the material density) and correcting the initial displacement L0=v*tarrival / 2.

[0068] Nonlinear and hysteresis corrections are applied to a multidimensional model by inputting the compensated signal characteristics [Areal, Δtreal, Treal, direction of movement], and outputting the nonlinear error ΔLnl and hysteresis error ΔLh; the final initial displacement Linit = L0 + ΔLnl + ΔLh.

[0069] In some embodiments, the iterative training of the multidimensional error model using actual crack displacement monitoring data to dynamically optimize model parameters to adapt to environmental changes and device aging characteristics includes: establishing a monitoring data quality assessment mechanism to remove abnormal data points and form an effective training dataset; using an incremental learning algorithm to update the multidimensional error model online, calculating the model prediction error by comparing real-time measured values ​​with calibration data from an external reference sensor; and triggering a model parameter optimization process when the prediction error exceeds a preset threshold, adaptively adjusting the nonlinear error coefficient, temperature drift coefficient, and hysteresis effect correction factor to form an iteratively optimized error compensation model.

[0070] The model parameters are dynamically updated through an online learning mechanism to adapt to device aging and environmental changes.

[0071] Data filtering is performed by establishing a 3σ criterion to remove outlier data (such as points where the signal amplitude changes by more than 20%), and by combining a displacement change rate threshold (such as >5 mm / s being considered vibration interference) to retain valid training data.

[0072] The incremental learning process triggers a model update when 100 sets of valid data are accumulated: The real-time measurement value Linit is compared with the true value Ltrue from an external benchmark sensor (e.g., a laser displacement gauge with an accuracy of ±0.05 mm), and the mean square error MSE is calculated as: MSE = 1 / n * ∑(Linit) Ltrue) 2 If MSE > 0.02%FS, use stochastic gradient descent (SGD) to optimize the model parameters, focusing on adjusting the temperature drift coefficient and hysteresis correction factor, with the number of iterations set to 500 and the learning rate 0.001.

[0073] Aging compensation involves periodically (e.g., annually) collecting data after aging using a built-in calibration device (precision displacement stage), updating waveguide wire material parameters (such as magnetostriction coefficient attenuation rate), and resetting the initial conditions of the model.

[0074] In some embodiments, the filtering, amplification, and analog-to-digital conversion processing of the induced signal includes: filtering out power frequency interference and high-frequency noise using a bandpass filter while retaining the characteristic frequency band of the magnetostrictive torsional wave signal; dynamically adjusting the gain of the filtered signal using a programmable gain amplifier to make the signal amplitude match the input range of the analog-to-digital converter; and sampling the amplified analog signal using a high-precision analog-to-digital converter at a sampling frequency not less than twice the highest frequency of the magnetostrictive torsional wave to generate a digital signal sequence.

[0075] Improve the quality of weak signal acquisition by optimizing the signal preprocessing process.

[0076] The filter design uses an 8th-order Butterworth bandpass filter with a passband frequency of 20~40kHz (matching the characteristic frequency band of magnetostrictive torsional waves) to suppress 50Hz power frequency and high-frequency noise above 100kHz.

[0077] Dynamic gain adjustment automatically adjusts the gain (1 to 100 times) according to the signal amplitude through an integrated programmable gain amplifier (such as AD8253), ensuring that the signal amplitude input to the ADC is within 70% to 90% of the full scale (e.g., the input range of a 3.3V ADC is controlled between 2.3 and 3.0V).

[0078] The analog-to-digital conversion uses a 24-bit Δ-Σ type ADC (such as ADS1256), with a sampling frequency of 200kHz (satisfying the Nyquist sampling theorem of 2.5 times the highest frequency of torsional waves of 80kHz). The output digital signal sequence is transmitted to the edge computing unit after mean filtering (window size 5).

[0079] In some embodiments, the acquisition of the induced signal generated by the magnetostrictive effect of the waveguide wire includes: the detection coil is wound with a high permeability magnetic core, and the number of coil turns is in a preset ratio to the length of the waveguide wire to match the induction sensitivity of the magnetostrictive torsional wave; when the movable permanent magnetic field generator moves, the magnetostrictive effect is excited on the surface of the waveguide wire to generate a torsional wave, and the detection coil acquires the induced electromotive force signal of the torsional wave propagating to the coil position through electromagnetic induction, and simultaneously records the signal arrival time and waveform characteristics.

[0080] By improving the detection coil design and signal acquisition strategy, the sensitivity of magnetostrictive effect sensing is enhanced.

[0081] The coil design employs a high-permeability ferrite core, with the number of coil turns N linearly proportional to the waveguide wire length Ltotal (e.g., N = Ltotal / 10 turns / mm), optimizing the coil inductance L = μ0μrN. 2 A / l (where A is the cross-sectional area of ​​the coil and l is the length of the magnetic core) improves the induction sensitivity.

[0082] When the signal acquisition device moves through the permanent magnetic field, the excitation circuit sends a 10μs current pulse, which triggers the magnetostrictive effect to generate a torsional wave. The detection coil collects the induced electromotive force and synchronously records the signal arrival time (with the rising edge of the excitation pulse as the time zero point) and waveform characteristics (such as the first wave peak value and the second wave attenuation rate), which are used for subsequent time difference calculation and signal quality assessment.

[0083] In some embodiments, the edge computing unit is further configured to use a long short-term memory network to model the temporal correlation of historical displacement data sequences and predict the trend of crack displacement; dynamically adjust the regularization parameters of the multidimensional error model by combining a particle swarm optimization algorithm to balance the model fitting accuracy and generalization ability; when the displacement change rate is detected to exceed a preset threshold, automatically trigger the encrypted sampling mode, and use a convolutional neural network to identify the features of abnormal signal waveforms to determine whether it is a sudden crack expansion event.

[0084] By integrating time series prediction, parameter optimization, and anomaly identification, the level of monitoring intelligence is improved.

[0085] The time series prediction uses a Long Short-Term Memory (LSTM) network to build a displacement prediction model. The input is the displacement sequence of the most recent 24 hours [L1, L2, ..., L1440], and the output is the displacement trend L^t+60 for the next hour, which is used to give early warning of accelerated structural deformation.

[0086] Model optimization uses the particle swarm optimization (PSO) algorithm to periodically (weekly) adjust the regularization parameters of the multidimensional error model (such as the L2 regularization coefficient of a neural network) with the goal of minimizing the validation set MSE. The particle swarm size is set to 30 and the number of iterations is 100.

[0087] Anomaly detection is achieved by triggering encrypted sampling (frequency increased from 1Hz to 10Hz) when the displacement change rate dL / dt > 0.5mm / s, and the collected waveform data is input into a convolutional neural network (CNN) for feature recognition. The CNN model is pre-trained on historical data of sudden crack expansion events (waveforms show an increase in high-frequency components and a sharp drop in the amplitude of the first wave), and outputs the probability of abnormal events. When the probability exceeds 90%, a local audible and visual alarm is triggered and emergency data is uploaded.

[0088] This invention constructs a multi-dimensional model incorporating nonlinear errors, temperature drift, and hysteresis effects using historical data from a preset calibration interval. This effectively corrects multi-physics coupling interference, improving measurement accuracy compared to traditional single-parameter compensation methods. It iteratively trains the error model using actual monitoring data, dynamically adapting to time-varying factors such as environmental temperature changes and device aging. This avoids the accuracy degradation issues of traditional sensors due to long-term use, extending the effective monitoring cycle of the equipment. The multi-range fusion module automatically matches high-precision short-range modes with large-range extended modes, extending the measurement range to the meter level while maintaining millimeter-level resolution, meeting the accuracy and range requirements of different monitoring scenarios. A built-in self-learning algorithm module enables local data processing and model updates, reducing reliance on remote servers and improving data processing efficiency and system reliability. This makes it particularly suitable for long-term monitoring in remote areas or environments with limited network access.

[0089] In summary, this invention, through the deep integration of intelligent algorithms and magnetostrictive sensing technology, solves the problems of accuracy decay and poor environmental adaptability of traditional crack gauges in large-range measurements, and significantly improves the automation and intelligence level of structural health monitoring.

[0090] It should be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. It should be understood that when an element or layer is referred to as “on,” “adjacent to,” “connected to,” or “coupled to” other elements or layers, it may be directly on, adjacent to, connected to, or coupled to other elements or layers, or there may be intervening elements or layers. Conversely, when an element is referred to as “directly on,” “directly adjacent to,” “directly connected to,” or “directly coupled to” other elements or layers, there are no intervening elements or layers. It should be understood that although the terms first, second, third, etc., may be used to describe various elements, components, areas, layers, and / or portions, these elements, components, areas, layers, and / or portions should not be limited by these terms. These terms are merely used to distinguish one element, component, area, layer, or portion from another element, component, area, layer, or portion. Therefore, without departing from the teachings of this application, the first element, component, area, layer, or portion discussed below may be referred to as a second element, component, area, layer, or portion.

[0091] Spatial relation terms such as “below,” “under,” “below,” “under,” “above,” “above,” etc., are used herein for convenience of description to describe the relationship between one element or feature shown in the figure and other elements or features. It should be understood that, in addition to the orientation shown in the figure, spatial relation terms are intended to also include different orientations of the device in use and operation. For example, if the device in the figure is flipped, then the element or feature described as “below,” “under,” or “below” other elements or features will be oriented “above” other elements or features. Therefore, the exemplary terms “below” and “under” can include both above and below orientations. The device may be otherwise oriented (rotated 90 degrees or otherwise) and the spatial descriptive terms used herein will be interpreted accordingly.

[0092] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0093] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A large-range magnetostrictive crack gauge based on machine self-learning, characterized in that, include: A magnetostrictive measurement assembly includes a measuring rod, a waveguide wire disposed on the measuring rod, a movable permanent magnetic field generator sleeved on the outside of the waveguide wire, and a detection coil fixed to one end of the measuring rod. The movable permanent magnetic field generator moves along the axial direction of the waveguide wire as the crack opens and closes. The detection coil is used to collect the induced signal generated by the waveguide wire due to the magnetostrictive effect. The signal conditioning module is electrically connected to the detection coil and is used to filter, amplify, and perform analog-to-digital conversion on the sensed signal. An edge computing unit is electrically connected to the signal conditioning module. The edge computing unit acquires historical measurement data of the magnetostrictive measurement component within a preset calibration range and constructs a multidimensional error model that includes nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics. Based on the multidimensional error model, the real-time acquired induction signal data is preprocessed to generate initial displacement measurement values. The multidimensional error model is iteratively trained using actual crack displacement monitoring data, and the model parameters are dynamically optimized to adapt to environmental changes and device aging characteristics.

2. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, Also includes: A multi-range fusion module is electrically connected to the edge computing unit. The multi-range fusion module is configured to automatically switch the measurement range of the magnetostrictive measurement component according to the dynamic range of the initial displacement measurement value. The measurement range includes at least a high-precision short-range mode and a large-range extended mode.

3. The large-range magnetostrictive crack gauge according to claim 2, characterized in that, In the high-precision short-range mode, the movable permanent magnetic field generator is located in the calibration sensitive range of the waveguide wire, and the detection coil uses a narrowband filtering algorithm to improve resolution; in the large-range extended mode, the waveguide wire is divided into multiple continuous measurement segments by a segmented calibration algorithm, and the cross-segment measurement data is fused and compensated by the self-learning algorithm module.

4. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, Also includes: The storage module, electrically connected to the edge computing unit, is used to store the multidimensional error model parameters, historical measurement data, and calibration coefficients. The communication module is electrically connected to the edge computing unit and is used to upload the processed displacement data and self-learning state parameters to a remote server.

5. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The construction of the multidimensional error model, which includes nonlinear error characteristics, temperature drift characteristics, and hysteresis effect characteristics, includes: Within a preset calibration range, magnetostrictive induction signals and standard displacement reference values ​​under different temperature conditions and displacement amounts are collected, and nonlinear characteristic parameters such as signal amplitude, rising edge slope, and zero-crossing time difference are extracted. Ambient temperature data is acquired in real time by a temperature sensor, and a mapping relationship between temperature, zero-point drift, and sensitivity drift is established. Construct a hysteresis error matrix for the hysteresis effect based on hysteresis loop test data; The nonlinear characteristic parameters, temperature drift data, and hysteresis error matrix are coupled and modeled using support vector regression or artificial neural networks to generate a multidimensional mathematical model that includes multiple physical field error factors.

6. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The preprocessing of the real-time acquired sensing signal data based on the multidimensional error model to generate initial displacement measurement values ​​includes: Extract the timestamp, waveform peak value, and zero-crossing time difference of the real-time sensing signal as input features; Temperature drift compensation is performed on the input features based on real-time ambient temperature data to correct temperature correlation deviations in the elastic modulus of waveguide wire material and magnetic field distribution. The nonlinear correction function of the multidimensional error model is used to map and calculate the characteristics of the compensated signal, and the hysteresis error is compensated for the displacement measurement value by combining the hysteresis effect correction matrix, and the initial displacement measurement value is output.

7. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The step of iteratively training the multidimensional error model using actual crack displacement monitoring data and dynamically optimizing the model parameters to adapt to environmental changes and device aging characteristics includes: Establish a monitoring data quality assessment mechanism to remove outlier data points and form an effective training dataset. An incremental learning algorithm is used to update the multidimensional error model online. The model prediction error is calculated by comparing real-time measured values ​​with calibration data from an external reference sensor. When the prediction error exceeds a preset threshold, the model parameter optimization process is triggered to adaptively adjust the nonlinear error coefficient, temperature drift coefficient, and hysteresis effect correction factor to form an iteratively optimized error compensation model.

8. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The filtering, amplification, and analog-to-digital conversion processing of the sensed signal includes: By using a bandpass filter to remove power frequency interference and high-frequency noise, the characteristic frequency band of the magnetostrictive torsional wave signal is preserved. A programmable gain amplifier is used to dynamically adjust the gain of the filtered signal so that the signal amplitude matches the input range of the analog-to-digital converter. A high-precision analog-to-digital converter is used to sample the amplified analog signal, with a sampling frequency no less than twice the highest frequency of the magnetostrictive torsional wave, to generate a digital signal sequence.

9. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The acquisition of the induced signal generated by the magnetostriction effect of the waveguide wire includes: The detection coil is wound with a high permeability magnetic core, and the number of coil turns is in a preset ratio to the length of the waveguide wire to match the sensing sensitivity of the magnetostrictive torsional wave. When the movable permanent magnetic field generator moves, it excites the magnetostrictive effect on the surface of the waveguide wire to generate a torsional wave. The detection coil collects the induced electromotive force signal of the torsional wave propagating to the coil position through electromagnetic induction, and simultaneously records the signal arrival time and waveform characteristics.

10. The large-range magnetostrictive crack gauge according to claim 1, characterized in that, The edge computing unit is also used to model the temporal correlation of historical displacement data sequences using a long short-term memory network to predict the trend of crack displacement; dynamically adjust the regularization parameters of the multidimensional error model by combining the particle swarm optimization algorithm to balance the model fitting accuracy and generalization ability; when the displacement change rate is detected to exceed the preset threshold, the encrypted sampling mode is automatically triggered, and the abnormal signal waveform is identified by the convolutional neural network to determine whether it is a sudden crack expansion event.

Images