Cable force testing device and testing method thereof

Abstract

The invention discloses a cable force testing device and a testing method, belongs to the technical field of engineering structure health monitoring, and solves the problems that real-time and high-precision measurement is difficult in a complex environment and the existing cable force testing method is influenced by temperature drift and strain transmission loss. According to the scheme, a laser Doppler vibration measurement module integrated on a movable bearing platform collects vibration signals V (t) in a non-contact mode, an infrared thermal imaging module obtains a temperature field T (x, t), and a flexible strain gauge array is attached to the surface of a cable body to collect strain data epsilon (x, t); a signal processing unit (extracting f1 and f2), a temperature processing unit (calculating Tavg and delta T), a strain processing unit (calculating epsilon max) and an LSTM neural network model are arranged in the edge calculation terminal, and a tension value Tpred is predicted through feature vectors [f1, f2, epsilon max, Tavg, delta T, L and D]. The method is used for real-time cable force monitoring and health assessment of large-scale structures such as bridge stay cables and suspension cables.

Description

Technical Field

[0001] The present invention belongs to the technical field of engineering structure health monitoring, and particularly relates to a cable force testing device and a testing method thereof. Background Art

[0002] Cable force measurement is a key link in bridge structure health monitoring, but the existing technologies have significant limitations. Contact mechanical measurement devices such as lug pressure sensors and SL series tensiometers need to be directly installed on the surface of the cable body, and the installation process often requires traffic interruption or erection of an aerial work platform. When such devices are exposed for a long time, they are easily affected by temperature changes and produce zero drift. The typical temperature drift coefficient reaches ±0.05%FS / ℃, and the device weight exceeds 8 kg, making it difficult to deploy in high-altitude scenarios such as the top of a bridge cable tower or a narrow anchorage chamber. The vibration frequency method for inverse cable force based on vibration signals has a precision decline of more than 15% for short cables less than 20 m due to modal sparsity. This method requires manual pasting of reflective marks or relies on specific devices such as millimeter-wave radars, with complex operations and being interfered by environmental light. More critically, this method does not consider the drift effect of temperature on material parameters. Engineering field measurements show that a 20℃ temperature difference can cause changes in linear density and elastic modulus, resulting in a tension calculation deviation of more than 12%.

[0003] Although structural improvement technologies such as automatic positioning devices and coupling devices have optimized the sensor installation process, they are still restricted by the measurement principle of a single frequency parameter and cannot solve the problem of strain transmission distortion caused by the medium layer on the cable body surface. Research shows that the strain signal attenuation in the 70 - 100 Hz frequency band can reach 25% due to grease adhesion, and the attenuation in the 150 - 200 Hz frequency band reaches 18% due to water film coverage. Indirect monitoring means such as wire displacement sensors and magnetic flux methods need to be fixedly deployed for a long time. Restricted by the nonlinearity of material magnetic permeability and the creep characteristics of the cable body, they cannot meet the requirements of temporary detection. Such methods only rely on single parameters such as displacement or magnetic flux and are difficult to jointly process the coupling error of temperature and strain.

[0004] The existing technologies generally have common defects: the vibration frequency method ignores the influence of temperature drift on material parameters, the strain measurement method does not compensate for the signal attenuation caused by the surface medium layer, and there is a lack of a multi-source data collaborative processing mechanism for vibration, temperature, and strain. At the same time, traditional measurement devices are bulky and difficult to adapt to the requirements of rapid high-altitude deployment. These defects lead to insufficient measurement accuracy under complex working conditions such as a solar radiation temperature gradient exceeding 3℃ / m and cable body surface medium pollution. Summary of the Invention

[0005] The purpose of the present invention is to provide a cable force testing device and a testing method thereof. Through multi-source sensing fusion (vibration, temperature, strain) and LSTM neural network modeling, high-precision real-time prediction of cable force is achieved, significantly improving the measurement robustness under complex working conditions; the design of a movable carrying platform supports flexible deployment and is suitable for online monitoring of high-altitude and long-span cable structures.

[0006] Solve the problem of large cable force measurement error caused by complex environmental temperature interference and strain transfer distortion.

[0007] Solve the problem that it is difficult to quickly install and the fitting is unstable for the flexible strain gauge array on the curved cable body.

[0008] Solve the problem that the linear density of the material μ and elastic modulus E drift lead to the inaccuracy of the theoretical model.

[0009] Solve the problems of insufficient visualization of multi-source data and delayed early warning of abnormal cable force.

[0010] Solve the problem that the feature fusion fails due to the asynchronous acquisition time of the vibration, temperature, and strain three channels.

[0011] Solve the problem of strain transfer attenuation caused by grease / water film / oxide layer on the cable body surface.

[0012] Solve the problem that the temperature field measurement is distorted due to the interference of environmental radiation on infrared thermal imaging.

[0013] Solve the problem that it is difficult to adjust the fixture gap when the UAV platform adapts to cables with different diameters.

[0014] Solve the problem of insufficient generalization of the LSTM model when predicting cable force using the features of a single sensor.

[0015] Solve the problem that it cannot be dynamically corrected when there is a conflict between the traditional physical model and the prediction result.

[0016] To solve the above problems and achieve the objectives and other advantages of the present invention, a cable force testing device is provided, including: A laser Doppler vibrometry module, which is integrated on a movable bearing platform and is used for non-contact acquisition of the vibration signal V (t); An infrared thermal imaging module, which is integrated on a movable bearing platform and is used for non-contact acquisition of the temperature field T (x, t); A flexible strain gauge array, which is attached to the surface of the cable body and is used for acquiring the strain data on the surface of the cable body ε (x, t); An edge computing terminal, which receives the vibration signal V (t), the temperature field T (x, t) and the strain data ε (x, t); The edge computing terminal is built-in with: A signal processing unit, which is configured to perform variational mode decomposition on V (t) and extract the first-order modal frequencyf 1st and second order modal frequencies f 2; A temperature processing unit configured to calculate the average temperature based on T (x, t) T avg and the temperature gradient ▽ T ; A strain processing unit configured to calculate the maximum absolute strain value at all positions on the surface of the cable body during the measurement period based on ε (x, t) ε max ; An LSTM neural network model, whose input feature vector f 1, f 2, ε max , T avg , ▽ T , L , D , and outputs a predicted tension value T pred ; Wherein, L is the length of the cable body; D is the diameter of the cable body.

[0017] Preferably, in the cable force testing device, it further includes: A self-locking fixture module for fixing the flexible strain gauge array on the surface of the cable body, which includes: A split fixture body, which includes: A mounting seat, which is circular and coaxially sleeved outside the cable body. The inner diameter of the mounting seat is larger than the diameter of the cable body, and a gap is formed between the two; At least three rigid flaps are evenly distributed on a circular plane of the mounting seat along the circumferential direction of the mounting seat. One end of each flap is hinged to the circular plane through a rotating shaft, so that each flap can open and close radially around the cable body; a pressing curved surface is provided on the inner side of each flap, and when closed, it fits the surface of the cable body; the other end of each flap is a free end and has an outwardly protruding arc-shaped curved surface. The arc-shaped curved surfaces at the free ends of each flap together form a horn-shaped structure, and the opening diameter of the horn mouth of the horn-shaped structure is larger than the diameter of the cable body, facilitating the cable body to be inserted from the horn mouth; the curvature of the arc-shaped curved surface matches the surface of the cable body, guiding the cable body to enter the inside of each flap along its axial direction, ensuring that each flap is coaxial with the cable body when closed; a rectangular groove array is provided on the pressing curved surface; A strain gauge carrier plate array corresponding to the rectangular grooves one by one. Flexible strain gauges are pasted on the surface of each strain gauge carrier plate, and each strain gauge carrier plate is respectively embedded in the corresponding rectangular groove; Multiple groups of shape memory alloy wires or springs, with both ends of each shape memory alloy wire or spring connected to adjacent flaps respectively; An axial guide rail, which is fixed to the movable bearing platform and is parallel to the axis of the cable body; A slider, which is arranged on the circumferential surface of the mounting seat, matches the axial guide rail, and is provided with a locking device, and the locking device can fix the slider to the axial guide rail; Multiple anti-rotation pins, which penetrate the radial screw holes of the mounting seat and are pressed against the surface of the cable body to circumferentially fix the mounting seat and the cable body to prevent relative rotation between the two; An electrothermal control unit, which outputs pulsed current to activate the shape memory alloy wires or springs.

[0018] Preferably, in the cable force testing device, the edge computing terminal also internally has: A dynamic compensation unit, which is configured to according to T avg Revised linear density μ And elastic modulus E , Revised linear density μ eff ( T ), Elastic modulus E eff ( T )The formula is as follows: μ eff ( T )= μ 0 / [1+ α L ×( T avg - T ref )] E eff ( T )= E 0×[1+ α E ×( T avg - T ref )] Wherein, μ 0 is the reference linear density, α L Is the coefficient of thermal expansion, E 0 is the reference elastic modulus, α E Is the thermosensitive coefficient, T ref Is the reference temperature; A cable force prediction error calculation unit, which is based on μ eff( T ) Calculate the theoretical tension value E eff ( T ) as follows: T phys , the formula is as follows: Wherein, n is the harmonic order, n = 1 or 2; f n represents the n -order vibration frequency; Compare T pred with T phys . When | T pred - T phys | > δ , calculate the loss function based on T pred and T phys : Wherein, δ is the preset threshold; Update the weight parameters of the LSTM neural network by backpropagation: Wherein, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the gradient of the loss function with respect to the parameter θ ; η is the learning rate; Output the corrected predicted cable force value , , the expression is: When | T pred - T phys | ≤ δ , output T pred as the final cable force value; When | T pred - T phys | > δ , output As the final cable force value.

[0019] Preferably, the cable force testing device further includes: A visualization terminal, which is connected to the edge computing terminal through a communication module and is used to receive T pred 、 V (t), T (x, t), ε (x, t), and based on T (x, t) to generate a heat distribution cloud map, and based on V (t) to generate a vibration mode curve. When | T pred - T phys | > δ, a deviation alarm signal is triggered.

[0020] Preferably, the cable force testing device further includes: A time synchronization control module, which is configured to: Provide a time reference signal with microsecond-level accuracy through a GPS timing unit or an oven-controlled crystal oscillator unit integrated on the movable bearing platform; Send hardware trigger pulses to the laser Doppler vibrometer module, the infrared thermal imaging module, and the flexible strain gauge array synchronously, so that the three start data acquisition at the same moment; Embed the time stamp of the trigger moment into the collected vibration signal V (t), temperature field T (x, t), and strain data ε (x, t) t k ; The edge computing terminal also has built-in: A time domain alignment unit, which is configured to, based on the time stamp t k Align V ( t k ), T (x, t k ), ε (x, t k ) to a unified time axis by resampling, and output a time-aligned feature vector f 1( t k [[ID=7l]]) f 2( t k ) ε max ( t k ), Tavg ( t k ), ▽ T ( t k )];in: The signal processing unit is configured to process the resampled V ( t k ) Perform variational mode decomposition and extract the first-order mode frequency f 1( t k ) and the second-order modal frequency f 2( t k ); The temperature processing unit is configured based on the resampled T (x, t k )Calculate the average temperature T avg ( t k ) and temperature gradient ▽ T ( t k ); The strain processing unit is configured based on the resampled ε (x, t k ) Calculate the maximum absolute strain value at all locations on the cable surface during the measurement period ε max ( t k ); The LSTM neural network model aligns the time-aligned feature vectors [ f 1( t k ), f 2( t k ), ε max ( t k ), T avg ( t k ), ▽ T ( t k ), L , D ] as input and output the predicted tension value T pred ( t k ).

[0021] Preferably, in the cable force testing device, the edge computing terminal further has a built-in: The strain transfer compensation unit is connected to the flexible strain gauge array signal and is configured as follows: Obtain the original cable surface strain data collected by the flexible strain gauge array ε (x, t); Extracting strain data ε High-frequency harmonic components in the 50-200 Hz frequency band in (x, t); Calculate the energy amplitude of high-frequency harmonic components A h ; when A h Below the preset threshold A th When the media type identification is triggered: Grease layer identification: The attenuation of the 70-100Hz frequency band is greater than that of other frequency bands; Water film layer identification: The attenuation amplitude of the 150-200Hz frequency band is greater than that of other frequency bands; Oxide layer identification: uniform attenuation across the entire frequency band; Select the compensation factor based on the identified media type k : Grease layer: k =1.15; Water film layer: k =1.08; Oxide layer: k =1.05; Output compensated strain data: ε comp (x, t) = k × ε (x, t); Based on compensated strain data ε comp (x, t), calculate the maximum absolute strain value of all positions on the cable surface during the measurement period ε max .

[0022] Preferably, in the cable force testing device, the infrared thermal imaging module includes: A reference blackbody plate is fixed on a movable carrying platform, is located within the field of view of the infrared thermal imaging module, and is parallel to the surface of the cable body. Its emissivity is fixed at 0.97. The environmental radiation compensation unit is configured as follows: Use the infrared thermal imaging module to shoot the reference black body plate to obtain the ambient temperature T amb ; Obtain the temperature value of the original measurement of the cable body surface by the infrared thermal imaging module T meas ; Obtain the surface type of the cable body judged manually and determine the emissivity ε obj , the specific method is as follows: If the surface of the cable body is the surface of a newly galvanized steel cable, and the surface feature is a silver-white metallic luster, then ε obj = 0.85; If the surface of the cable body is the surface of an ordinary steel cable, and the surface feature is a gray-black carbon steel substrate without coating and without rust, then ε obj = 0.88; If the surface of the cable body is the surface of a rusty steel cable, and the surface feature is reddish-brown oxidation rust spots, then ε obj = 0.92; If the surface of the cable body is a PE sheath, then ε obj = 0.94; If the surface of the cable body is an anti-corrosion coating, then ε obj = 0.91; Correct the true temperature of the cable body surface through the following formula: The temperature processing unit is configured to calculate the average temperature T corr based on the corrected temperature field T avg and the temperature gradient ▽ T .

[0023] Preferably, in the cable force testing device, the movable carrying platform is a drone; when the cable body diameter is 50 - 100 mm, the gap between the mounting seat and the cable body is 1.0 - 1.5 mm; when the cable body diameter is 100 - 200 mm, the gap between the mounting seat and the cable body is 1.5 - 2.0 mm; when the cable body diameter is > 200 mm, the gap between the mounting seat and the cable body is 2.0 - 3.0 mm.

[0024] Preferably, in the cable force testing device, the LSTM neural network model includes: Two LSTM layers, each LSTM layer contains 128 LSTM units; A fully connected output layer, and its activation function is a linear function; The LSTM neural network model is configured to: Receive the input feature vector f 1, f 2,ε max , T avg , ▽ T , L , D ; The input feature vector is processed through two LSTM layers to generate the final hidden state; The hidden state is input into the fully connected output layer to output the predicted tension value T pred .

[0025] The test method using the cable force test device includes the following steps: The vibration signal of the cable body is collected non - contact through the laser Doppler vibrometer module V (t); The surface temperature field of the cable body is collected non - contact through the infrared thermal imaging module T (x, t); The surface strain data of the cable body is collected through the flexible strain gauge array ε (x, t); The collected V (t), T (x, t), ε (x, t) are transmitted to the edge computing terminal; The signal processing unit performs variational mode decomposition on V (t) and extracts the first - order modal frequency f 1 and the second - order modal frequency f 2; The temperature processing unit calculates the average temperature based on T (x, t) and the temperature gradient ▽ T avg ; T ; The strain processing unit calculates the maximum absolute strain value at all positions on the surface of the cable body during the measurement period based on ε (x, t) ε max ; The feature vector f 1, f 2, ε max , T avg , ▽ T , L , D is input into the LSTM neural network model to output the predicted tension value T pred ; where, L is the length of the cable body; D is the diameter of the cable body; Based on T avg , the line density is corrected by the dynamic compensation unit μ and the elastic modulus E , the corrected line density μ eff ( T ) and the elastic modulus E eff ( T ) are given by the following formulas: μ eff ( T ) = μ 0 / [1 + α L × ( T avg - T ref )] E eff ( T ) = E 0 × [1 + α E × ( T avg - T ref )] Wherein, μ 0 is the reference line density, α L is the coefficient of thermal expansion, E 0 is the reference elastic modulus, α E is the thermosensitive coefficient, T ref is the reference temperature; The theoretical tension value μ eff ( T ) and E eff ( T ) are calculated by the cable force prediction error calculation unit based on T phys , and the formula is as follows: Wherein, n is the harmonic order, n = 1 or 2; f n represents the n th vibration frequency; Compare T pred with T phys , when | Tpred - T phys |> δ When, based on T pred and T phys calculate the loss function: where, δ is a preset threshold; Update the weight parameters of the LSTM neural network through backpropagation: where, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the loss function with respect to the parameter θ gradient; η is the learning rate; Output the corrected predicted cable force value , The expression is: When | T pred - T phys | ≤ δ output T pred as the final cable force value; When | T pred - T phys | > δ output as the final cable force value.

[0026] The present invention at least includes the following beneficial effects: By integrating multi-source data such as vibration frequencies (fundamental frequency and second-order frequency), maximum strain value, average temperature, temperature gradient, cable length and diameter, and using a deep learning model (LSTM) for comprehensive prediction, the present invention significantly improves the cable force measurement accuracy under complex working conditions such as temperature changes, uneven temperature distribution, and surface contamination, and overcomes the limitations of traditional single methods (such as relying only on vibration frequency).

[0027] The present invention dynamically and online corrects the linear density and elastic modulus parameters of the cable material based on the real-time average temperature measurement. The corrected parameters are used to calculate the theoretical cable force and cross-validate it with the model prediction. If the two deviate significantly, the model parameters are updated online and the corrected prediction is output. This mechanism effectively eliminates systematic errors caused by material property drift due to temperature changes, ensuring the accuracy of measurement results in temperature-fluctuating environments.

[0028] This invention can automatically detect strain signal attenuation caused by surface deposits (such as grease, water films, and oxide layers) on the cable body. It identifies the type of deposit by analyzing the attenuation pattern of specific high-frequency components in the strain signal and intelligently applies the corresponding amplification factor to compensate for the raw strain data. This effectively eliminates the problem of strain measurement distortion caused by surface contaminants and improves the reliability of strain data.

[0029] This method integrates a reference blackbody plate and manually sets an appropriate emissivity based on the cable surface condition (e.g., galvanized, ordinary steel, rusted, PE-sheathed, or anti-corrosion coated) to correct for ambient radiation in the raw temperature data measured by infrared thermal imaging. This design significantly reduces interference from ambient thermal radiation (especially sunlight reflection) on the surface temperature measurement of the metal cable, improving the accuracy of average temperature and temperature gradient data.

[0030] This method uses a high-precision time reference (GPS or oven-controlled crystal oscillator) and hardware trigger pulses to force the vibration, temperature, and strain sensors to initiate data acquisition at the same microsecond instant, embedding precise timestamps in all data. Time-aligned resampling is performed during the data processing phase. This fundamentally resolves the problem of data time misalignment caused by response delays between sensors, ensuring strict consistency and effective fusion of vibration, temperature, and strain characteristics in physical time. This is particularly advantageous in dynamic environments.

[0031] The flared guide design of the clamp of the present invention is combined with a curved surface that matches the curvature of the cable body, allowing the cable body to slide easily, quickly and centrally into the installation position. The shape memory alloy is activated by electrothermal force to generate a contraction force, driving the petal structure to close synchronously and evenly compress the surface of the cable body to ensure that the flexible strain gauge array fits tightly. The stepped installation gap design (different gap values are set according to the cable body diameter range) ensures that the clamp can adapt to various specifications of cables. The anti-rotation pin and axial slider locking device effectively prevent the clamp from rotating and sliding axially relative to the cable body. After power failure, the shape memory alloy cools and recovers, and the clamp automatically pops open, allowing for quick disassembly without tools, greatly improving the efficiency of high-altitude operations and shortening single-point operation time.

[0032] The deep learning model (double-layer LSTM structure) adopted by the present invention has stronger learning ability, can better capture the complex non-linear relationship between cable forces and multiple physical quantities, and improve the adaptability of the model to different working conditions (model generalization). More importantly, by comparing the theoretical cable force value with the model prediction value in real time, when significant deviations are found, the model parameters can be automatically updated online, and the corrected prediction value can be output. This "physical model + data model" cross-validation and collaborative correction mechanism greatly improves the fault tolerance and reliability of the entire system.

[0033] The present invention provides a visual interface to display key information such as predicted cable force values, temperature distribution cloud maps, and vibration mode curves in real time. When the deviation between the model prediction value and the theoretical value exceeds the safety threshold, an audible and visual alarm is immediately triggered. This provides intuitive diagnostic basis for engineers (such as locating high-temperature areas and observing abnormal vibrations), and greatly shortens the delay in discovering potential problems and responses.

[0034] Through innovative multi-sensor fusion, intelligent data processing algorithms (dynamic temperature compensation, strain distortion identification and compensation, infrared radiation correction, high-precision time synchronization, deep learning modeling), ingenious mechanical structure design (self-locking fixture), and the dynamic collaborative mechanism of theoretical models and data models, the present invention systematically solves the core pain points faced by existing cable force testing technologies in complex environments, such as insufficient accuracy, large interference from temperature / surface contamination, difficult and time-consuming installation, data asynchronization, and poor model adaptability. It realizes real-time, high-precision, and high-robustness measurement of cable forces of large-scale engineering cables (such as stay cables and suspension cables), providing more reliable technical support for structural health monitoring.

[0035] Other advantages, objectives, and features of the present invention will be partially reflected by the following description, and partially will be understood by those skilled in the art through the research and practice of the present invention. Detailed implementation mode

[0036] The following provides a further detailed description of the present invention so that those skilled in the art can implement it with reference to the text of the specification.

[0037] The cable force testing device includes a laser Doppler vibrometry module, an infrared thermal imaging module, and a flexible strain gauge array integrated on a movable bearing platform. The laser Doppler vibrometry module collects the cable vibration signal V (t) in a non-contact manner, and the infrared thermal imaging module synchronously obtains the surface temperature field T (x, t) of the cable, and the flexible strain gauge array adheres to the cable surface to collect strain data ε (x, t). The edge computing terminal receives the above data and internally includes a signal processing unit, a temperature processing unit, a strain processing unit, and an LSTM neural network model. The signal processing unit processes the vibration signal V(t) Perform variational mode decomposition to extract the first-order modal frequency f 1 and the second-order modal frequency f 2; The temperature processing unit calculates the average temperature based on the temperature field T (x, t) T avg and the temperature gradient ▽ T ; The strain processing unit calculates the maximum absolute strain value at all positions on the surface of the cable during the measurement period ε max , specifically, during the measurement period, all position coordinates x and all time points t are traversed to determine the compensated strain data ε (x, t) ε max . The LSTM neural network model takes the feature vector f 1, f 2, ε max , T avg , ▽ T , L , D as the input and outputs the predicted tension value T pred , where L is the length of the cable, D is the diameter of the cable.

[0038] Specifically, the cable force test device includes: A laser Doppler vibrometry module, which is integrated on a movable carrying platform and is used to non-contact collect the vibration signal of the cable V (t); An infrared thermal imaging module, which is integrated on a movable carrying platform and is used to non-contact collect the surface temperature field of the cable T (x, t); A flexible strain gauge array, which is attached to the surface of the cable and is used to collect the surface strain data of the cable ε (x, t); An edge computing terminal, which receives the vibration signal V (t), the temperature field T (x, t) and the strain data ε (x, t); The edge computing terminal is built-in with: A signal processing unit, which is configured to perform variational mode decomposition on V (t) and extract the first-order modal frequency f 1 and the second-order modal frequency f 2; A temperature processing unit, which is configured to calculate the average temperature based on T (x, t)T avg and temperature gradient T ; The strain processing unit is configured based on ε (x, t) calculates the maximum absolute strain value of all positions on the cable surface during the measurement period ε max ; LSTM neural network model, whose input feature vector [ f 1, f 2, ε max , T avg ,▽ T , L , D ] and output the predicted tension value T pred ; in, L is the cable length; D is the cable body diameter.

[0039] The closest existing technology is the single vibration frequency method, which relies only on the vibration signal to extract the fundamental frequency. f 1 and through physical formulas T =4 μL 2 f 1 2 Calculate tension. This method has significant defects: it does not integrate infrared thermal imaging data and cannot perceive the dynamic changes of the temperature field; it does not integrate strain data and lacks direct monitoring of the mechanical state of the cable surface; it does not use multi-modal frequency joint analysis and is difficult to suppress the spectrum aliasing caused by environmental noise. T avg 、▽ T and strain extremes ε max , forming a multidimensional input vector, enabling the LSTM model to mine correlation patterns beyond the vibration signal from complex physical quantities. This implementation constructs a vibration-temperature-strain coupling relationship model through the collaborative collection of three-source data and deep learning integration.

[0040] Example 1 The main cable of a suspension bridge with a diameter of 80mm and a length of 150m was tested. The ambient temperature was 26℃ and the cable surface was clean. The laser vibration measurement module collected the vibration signal and obtained the f 1=0.85Hz, f 2=1.72Hz; average temperature measured by infrared thermal imaging T avg =28.1℃, temperature gradient ▽ T= 1.2℃ / m; Strain gauge array output ε max = 218 με Input the feature vector [0.85, 1.72, 218, 28.1, 1.2, 150, 80] into the LSTM model, and the output T pred = 3250 kN. After calibration by the hydraulic sensor, the actual cable force is 3320 kN, with a relative error of 2.1%.

[0041] Comparative Example 1 Under the same working conditions, the traditional vibration frequency method is adopted: only extract based on the vibration signal f f1 = 0.83 Hz, through the formula T = 4 μ L 2 f f1 2 Calculate the tension of 3060 kN ( μ Take the fixed calibration value of 39.1 kg / m. Based on the steel density of 7800 kg / m³, the theoretical linear density of the steel cable with a diameter of 80 mm is about 39.3 kg / m), with a relative error of 7.8%. Since the temperature and strain data are not fused, the frequency identification deviation caused by environmental interference cannot be corrected.

[0042] Effect: In Example 1, through multi-source data fusion modeling, the error is controlled within 2.1%; in Comparative Example 1, due to the limitations of single parameters, the error increases to 7.8%. The measurement accuracy of this embodiment in a complex environment is significantly better than the traditional method.

[0043] The self-locking fixture module is used to fix the flexible strain gauge array on the surface of the cable body. The split fixture body includes an annular mounting seat, the inner diameter of which is larger than the diameter of the cable body to form a mounting gap. At least three rigid petals are circumferentially distributed on a circumferential plane of the mounting seat. One end of the petal is hinged to the circumferential plane through a rotating shaft, so that the petal can open and close radially around the cable body. A pressing curved surface is arranged inside the petal, which is completely attached to the surface of the cable body when closed; the free end of the petal is designed as a convex circular arc curved surface, and the circular arc curved surfaces of multiple petals together form a horn-shaped structure, the diameter of the horn mouth of which is larger than the diameter of the cable body for the cable body to be inserted. The curvature of the circular arc curved surface matches the surface of the cable body, guiding the cable body to slide into the inside of the petal along the axis to ensure coaxial alignment when closed. A rectangular groove array is arranged on the pressing curved surface, and a strain gauge carrier plate with a flexible strain gauge pasted on the surface is installed in each groove. The adjacent petals are connected by shape memory alloy wires or springs, and the electrothermal control unit outputs a pulsed current to activate the contraction of the memory alloy, driving the petals to close and lock synchronously. A slider is arranged on the circumferential surface of the mounting seat, which is fixedly matched with the axial guide rail of the movable bearing platform, and the position of the slider is fixed through a locking device. The anti-rotation pin penetrates the radial screw hole of the mounting seat and then presses against the surface of the cable body to prevent circumferential rotation.

[0044] Specifically, the cable tension testing device further includes: A self-locking fixture module for fixing the flexible strain gauge array on the surface of the cable, which includes: A split fixture body, which includes: A mounting seat, which is circular and coaxially sleeved outside the cable. The inner diameter of the mounting seat is larger than the diameter of the cable, and a gap is formed between them; the mounting seat has two annular planes, and the annular planes are perpendicular to the axis of the mounting seat; At least three rigid flaps, which are evenly distributed on one annular plane of the mounting seat along the circumferential direction of the mounting seat. One end of each flap is hinged to the annular plane through a rotating shaft, so that each flap can open and close radially around the cable; a pressing curved surface is provided on the inner side of each flap, and it fits with the surface of the cable when closed; the other end of each flap is a free end and has an outwardly protruding arc-shaped curved surface. The arc-shaped curved surfaces at the free ends of each flap together form a horn-shaped structure. The opening diameter of the horn mouth of the horn-shaped structure is larger than the diameter of the cable, which is convenient for the cable to be inserted from the horn mouth; the curvature of the arc-shaped curved surface matches the surface of the cable to guide the cable into the inside of each flap along its axial direction, ensuring that each flap is coaxial with the cable when closed; a rectangular groove array is provided on the pressing curved surface; A strain gauge carrier plate array, which corresponds to the rectangular grooves one by one. Flexible strain gauges are pasted on the surface of each strain gauge carrier plate, and each strain gauge carrier plate is respectively embedded in the corresponding rectangular groove; Multiple groups of shape memory alloy wires or springs, with both ends of each shape memory alloy wire or spring connected to adjacent flaps respectively; An axial guide rail, which is fixed to the movable carrying platform and is parallel to the axis of the cable; A slider, which is arranged on the circumferential surface of the mounting seat and matches the axial guide rail, and is provided with a locking device thereon, which can fix the slider and the axial guide rail; Multiple anti-rotation pins, which penetrate the radial screw holes of the mounting seat and are tightened against the surface of the cable to circumferentially fix the mounting seat and the cable to prevent relative rotation between the two; An electrothermal control unit, which outputs pulsed current to activate the shape memory alloy wires or springs.

[0045] Usage method of the self-locking fixture module: 1) Installation preparation stage Select a suitable mounting seat gap according to the cable diameter: when the cable diameter is 50 - 100 mm, the gap is configured to be 1.0 - 1.5 mm; when the cable diameter is 100 - 200 mm, the gap is 1.5 - 2.0 mm; when the cable diameter exceeds 200 mm, the gap is 2.0 - 3.0 mm. Fix the axial guide rail to the movable carrying platform (such as a drone), ensuring that the guide rail is parallel to the axis of the cable. Install the slider on the circumferential surface of the mounting seat and match it with the guide rail, and the locking device is in the unlocked state.

[0046] 2) Cable Body Positioning and Insertion Operate the drone to fly to the target position of the cable body. Use the bell mouth structure at the free end of the petal body to guide the cable body to insert along the axial direction: the opening diameter of the bell mouth is larger than the diameter of the cable body, and the curvature of the arc-shaped curved surface matches the surface of the cable body to ensure that the cable body slides into the petal body centered. When the cable body completely enters the central area of the mounting seat through the bell mouth, stop the movement of the drone.

[0047] 3) Clamp Closing and Self-locking Start the electrothermal control unit and apply a pulsed current to the shape memory alloy wire (typical parameters: 5V / 3A, duration 3s). The shape memory alloy shrinks when heated, pulling the adjacent petal bodies to close radially around the cable body. The pressing curved surface inside the petal body closely fits the surface of the cable body to form a uniform pressure. Immediately after closing, tighten the anti-rotation pin through the radial screw hole of the mounting seat to press against the surface of the cable body to prevent relative rotation between the mounting seat and the cable body.

[0048] 4) Strain Gauge Activation and Locking After the petal body is completely closed, the strain gauge carrier plate embedded in the rectangular groove of the pressing curved surface is in full contact with the surface of the cable body. The flexible strain gauge array is powered on to start collecting strain data. Synchronously lock the locking device on the slider to fix the mounting seat and the axial guide rail to eliminate the risk of axial displacement.

[0049] 5) Data Acquisition and Monitoring Keep the clamp in the locked state and start the laser Doppler vibrometry module and the infrared thermal imaging module for synchronous measurement. The typical acquisition duration is 3 - 5 minutes. During this period, the shape memory alloy maintains the contracted state (temperature stabilized at 60 ± 5°C) to ensure a constant fitting pressure.

[0050] 6) Unlocking and Disassembly After the measurement is completed, cut off the current of the electrothermal control unit. The shape memory alloy cools to room temperature (about 30s), the contraction force is released, and the petal body automatically springs open to the initial open state under the action of the hinge torsion spring. Loosen the anti-rotation pin and the slider locking device, and operate the drone to move the mounting seat out along the axial guide rail to make the cable body exit from the bell mouth. The whole process does not require tool assistance, and the single operation time is ≤8s.

[0051] 7) Transfer and Repeated Operation When it is necessary to change the measurement position, keep the clamp in the open state, and control the drone to fly to the next cable body position. Repeat steps 2) - 6) for continuous measurement. Under typical working conditions (cable body with a diameter of 120mm), the full process time for a single point is ≤45s, meeting the requirements of high-altitude rapid inspection.

[0052] The closest prior art is the adhesive fixing method: it is necessary to manually paste strain gauges on the surface of the cable one by one. When working at high altitudes, the positioning accuracy is poor due to the influence of wind, and it is easy to shift before the adhesive layer cures. Due to the curvature change of the curved cable, the fitting in some areas is not solid, and air bubbles and gaps exist at 20% of the measured points. In addition, when disassembling, it is necessary to use a solvent to remove the adhesive residue, and the single operation takes more than 30 minutes. In this embodiment, three-point synchronous coating is realized through a split self-locking structure, the memory alloy trigger ensures uniform pressing around the circumference, the guide rail positioning eliminates the manual centering error, and it can automatically spring open when the current is released during disassembly.

[0053] The cable force test device includes a laser Doppler vibrometry module, an infrared thermal imaging module and a flexible strain gauge array. The laser Doppler vibrometry module and the infrared thermal imaging module are integrated on a movable carrying platform. The laser Doppler vibrometry module non-contact collects vibration signals V (t), the infrared thermal imaging module synchronously obtains the temperature field T (x, t), and the flexible strain gauge array fits on the surface of the cable to collect strain data ε (x, t). The edge computing terminal is built-in with a signal processing unit, a temperature processing unit, a strain processing unit and an LSTM neural network model; the signal processing unit extracts the first-order modal frequency f 1 and the second-order modal frequency f 2; the temperature processing unit calculates the average temperature T avg and the temperature gradient ▽ T ; the strain processing unit calculates the maximum absolute value of the strain data ε max . The edge computing terminal is additionally provided with a dynamic compensation unit and a cable force prediction error calculation unit: the dynamic compensation unit corrects the linear density T avg in real time according to μ and the elastic modulus E ; the cable force prediction error calculation unit calculates the theoretical tension value μ , E and the vibration frequency f n ; when the deviation between the LSTM prediction value T phys and T pred exceeds the threshold T phys , the LSTM model parameters are updated through backpropagation and the corrected value δ is output. .

[0054] Specifically, the edge computing terminal is also built-in with: a dynamic compensation unit, configured to correct the linear density according to T avg ​μ With the elastic modulus E , the corrected linear density μ eff ( T )、elastic modulus E eff ( T ) the formula is as follows: μ eff ( T ) = μ 0 / [1 + α L × ( T avg - T ref )] E eff ( T ) = E 0 × [1 + α E × ( T avg - T ref )] Wherein, μ 0 is the reference linear density, α L is the coefficient of thermal expansion, E 0 is the reference elastic modulus, α E is the thermosensitive coefficient (because the elastic modulus E decreases with the increase of temperature, when actually used α E input a negative value, such as -1.5 × 10 −4 / ℃), T ref is the reference temperature; The cable force prediction error calculation unit, which is based on μ eff ( T ) and E eff ( T ) calculates the theoretical tension value T phys , the formula is as follows: Wherein, n is the harmonic order, n = 1 or 2; f n represents the n th vibration frequency; n The value rule of is as follows: Default value is n = 1 (the first-order mode corresponding to the fundamental frequency); For special working conditions, take n = 2 (the second-order mode); When the length of the cable L < 20 m (short cable) or environmental noise causes f 1 signal-to-noise ratio to be insufficient, the second-order mode frequency is preferentially used f 2.

[0055] If both f 1 and f 2 are input, calculate in parallel T phys ( n = 1) and T phys ( n = 2), and select the result with the smallest deviation from T pred as the verification benchmark.

[0056] n The value of n corresponds strictly to the order of the actually extracted vibration frequency. The mode with the highest energy is preferentially used (usually n = 1). For short cables or strong noise conditions, switch to

[0057] = 2. f 1 to calculate the tension ( T = 4 μL ² f 1²), which is essentially a simplified model that ignores higher-order terms (i.e., fixing n = 1). In the formula of this solution, f n allows the selection of different-order frequencies to calculate the theoretical tension value T phys , thereby improving the completeness of the physical model.

[0058] Compare T pred with T phys . When | T pred - T phys | > δ , based on T pred and T phys calculate the loss function: Among them,δ is a preset threshold value; Update the weight parameters of the LSTM neural network through backpropagation: Among them, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the loss function The gradient of the parameter θ ; η is the learning rate; Output the corrected predicted cable force value , The expression is: When | T pred - T phys | ≤ δ , output T pred as the final cable force value; When | T pred - T phys | > δ , output as the final cable force value.

[0059] The closest prior art is the static parameter vibration frequency method, which calculates the cable force using the fixed linear density μ and elastic modulus E . For example, the traditional device extracts the fundamental frequency f 1 and substitutes it into the formula T = 4 μL ² f 1², but does not correct the material parameters according to the real-time temperature. When the environmental temperature changes, the actual linear density μ of the material and the elastic modulus E significantly deviate from the reference value, resulting in the continuous accumulation of temperature drift errors. This embodiment corrects the μ and E parameters dynamically and introduces a cross-validation mechanism between the theoretical model and the prediction results to correct the output deviation in real time and eliminate the systematic influence of temperature drift on cable force prediction.

[0060] Example 2 In the cable force test of a suspension bridge in a certain canyon, the environmental temperature suddenly rises by 15°C from the reference temperature of 21°C. Using this device to measure a steel cable with a diameter of 90 mm and a length of 180 m: It is measured that f 1 = 1.18 Hz,f f = 2.37Hz, T avg T = 36.2°C, ε max ε = 165 με . The dynamic compensation unit calculates based on T avg the corrected material parameters and then T phys F = 2150 kN. The initial prediction of the LSTM T pred F0 = 2080 kN (| T pred - T phys | = 70 kN > δ = 50 kN), and after triggering the backpropagation update, the output F1 = 2140 kN. After calibration by the hydraulic sensor, the actual cable force is 2180 kN, and the relative error after correction is 1.8%.

[0061] Comparative Example 2 Under the same working conditions, the dynamic compensation unit is disabled: the input of the LSTM model is the original feature vector [1.18, 2.37, 165, 36.2, 0.9, 180, 90], and the output T pred F0 = 2080 kN. Failure to correct the material parameters results in the predicted value being lower than the calibrated value of 2180 kN, with a relative error of 4.6%.

[0062] Effect: In Example 2, by dynamically correcting the material parameters and cross - validating, the error is controlled within 1.8%. In Comparative Example 2, due to ignoring the drift effect of temperature on μ and E , the error increases to 4.6%. This verifies the effect of online parameter correction on suppressing the temperature drift error.

[0063] The cable force test device includes a laser Doppler vibrometer module, an infrared thermal imaging module, and a flexible strain gauge array. The laser Doppler vibrometer module and the infrared thermal imaging module are integrated on a movable carrier platform. The laser Doppler vibrometer module non - contact collects the cable vibration signal V V(t), the infrared thermal imaging module synchronously obtains the surface temperature field of the cable T T(x, t), and the flexible strain gauge array adheres to the cable surface to collect strain data ε ε(x, t). The edge - computing terminal built - in signal processing unit performs variational mode decomposition on V(t) to extract the first - order modal frequency f f1 and the second - order modal frequency f f2. The temperature processing unit calculates the average temperature T T̅ based on T avg T(x, t) and the temperature gradient ∇ T, the strain processing unit calculates the maximum absolute strain value on the surface of the cable based on ε (x, t) ε max . The LSTM neural network model uses the feature vector f 1, f 2, ε max , T avg , ▽ T , L , D as the input to output the predicted tension value T pred , where L is the length of the cable, D is the diameter of the cable. The visualization terminal is wirelessly connected to the edge computing terminal and receives T pred , V (t), T (x, t) and ε (x, t) data streams, and based on T (x, t) it renders the axial temperature thermal distribution cloud map of the cable in real time, and based on V (t) it generates the vibration displacement-frequency modal curve and triggers an audible and visual alarm signal when the tension is abnormal.

[0064] The closest prior art is the single vibration frequency method device, which only outputs the tension numerical result and does not provide the visualization function of the temperature field thermal map or the vibration modal curve. The operator cannot intuitively view the local temperature abnormality or the vibration spectrum details of the cable, resulting in the hidden danger diagnosis relying on offline analysis and the response delay exceeding 10 minutes. This embodiment locates the temperature gradient hot spot through the real-time thermal distribution cloud map, combines the vibration modal curve to identify the frequency offset, and improves the manual diagnosis efficiency by more than 60%, and shortens the alarm response time to within 30 seconds.

[0065] The cable force test device includes a laser Doppler vibrometer module, an infrared thermal imaging module and a flexible strain gauge array. The laser Doppler vibrometer module and the infrared thermal imaging module are integrated on a movable carrying platform. The time synchronization control module provides a microsecond-level precision time reference signal through the GPS timing unit or the oven-controlled crystal oscillator unit, and synchronously sends a hardware trigger pulse to the laser Doppler vibrometer module, the infrared thermal imaging module and the flexible strain gauge array, so that the three start data acquisition at the same time. The collected vibration signal V (t), the temperature field T (x, t) and the strain data ε (x, t) are all embedded with the time stamp at the trigger moment t k . The edge computing terminal is built-in with a time domain alignment unit, based on the time stamp t k toV ( t k ), T (x, t k ), ε (x, t k ) resampled to a unified time axis. The signal processing unit processes the resampled vibration signal V ( t k ) performs variational mode decomposition to extract the time-aligned first-order modal frequencies f 1( t k ) and the second-order modal frequency f 2( t k ). The temperature processing unit is based on the resampled temperature field T (x, t k ) Calculate the time-aligned average temperature T avg ( t k ) and temperature gradient ▽ T ( t k ). The strain processing unit is based on the resampled strain data ε (x, t k ) Calculate the maximum absolute strain value of the cable surface aligned with time ε max ( t k ). LSTM neural network model with time-aligned feature vectors[ f 1( t k ), f 2( t k ), ε max ( t k ), T avg ( t k ), ▽ T ( t k ), L , D ] is input and output is predicted tension value T pred ( t k ).

[0066] Specifically, the time synchronization control module is configured as follows: Provide a time reference signal with microsecond-level accuracy through a GPS timing unit or an oven-controlled crystal oscillator unit integrated in the movable bearing platform; Send hardware trigger pulses to the laser Doppler vibrometer module, the infrared thermal imaging module, and the flexible strain gauge array synchronously, so that the three start data acquisition at the same moment; For the collected vibration signal V (t), temperature field T (x, t) and strain data ε (x, t) embed the time stamp of the trigger moment t k ; The edge computing terminal also has built-in: A time domain alignment unit, which is configured to resample t k the V ( t k ), T (x, t k ), ε (x, t k ) to a unified time axis and output a time-aligned feature vector f 1( t k ), f 2( t k ), ε max ( t k ), T avg ( t k ), ▽ T ( t k ); where: The signal processing unit is configured to perform variational mode decomposition on the resampled V ( t k ) to extract the first-order modal frequency f 1( t k ) and the second-order modal frequency f 2( t k ); The temperature processing unit is configured to calculate the average temperature T (x, t k ) based on the resampled Tavg ( t k ) and temperature gradient ▽ T ( t k ); The strain processing unit is configured to calculate the maximum absolute strain value at all positions on the surface of the cable body during the measurement period based on the resampled ε (x, t k ) ε max ( t k ); The LSTM neural network model takes the time-aligned feature vectors f 1( t k ), f 2( t k ), ε max ( t k ), T avg ( t k ), ▽ T ( t k ), L , D as inputs and outputs the predicted tension value T pred ( t k ).

[0067] The closest prior art is a software-synchronized multi-sensor system, which relies on the operating system timestamp to achieve data synchronization, and the accuracy is only at the millisecond level. For example, when traditional devices collect vibration and strain data, due to the difference in sensor response delays, the time deviation between the vibration peak and the strain peak exceeds 500 ms. This deviation causes the extracted modal frequency to mismatch the actual strain state, resulting in the failure of feature fusion. In this embodiment, the three sensors are forced to start synchronously by a hardware trigger pulse, and combined with microsecond-level timestamp alignment and resampling, the timing misalignment under dynamic conditions is eliminated, ensuring the physical consistency of vibration, temperature, and strain characteristics.

[0068] Example 3 In the cable force test of a certain cross-sea suspension bridge, the ambient temperature is 28 °C and the wind speed is 6 m / s to simulate dynamic interference. The device is used to measure a steel cable with a diameter of 100 mm and a length of 210 m: The time synchronization module sends a hardware pulse to trigger the three sensors. The laser vibration measurement module measures f 1( t k) = 1.08 Hz, f 2( t k ) = 2.16 Hz; The infrared thermal imaging module measures T avg ( t k ) = 30.5 °C, ▽ T ( t k ) = 1.2 °C / m; The flexible strain gauge array outputs ε max ( t k ) = 162 με . The edge computing terminal inputs the time-aligned feature vector [1.08, 2.16, 162, 30.5, 1.2, 2l0, 100] into the LSTM model and outputs T pred = 2400 kN. After calibration by the hydraulic sensor, the actual cable force is 2450 kN, and the relative error is 2.04%.

[0069] Comparative Example 3 Disable the time synchronization module under the same working conditions: Manually trigger the three sensors, and the acquisition time deviation is about 500 ms. The laser vibration measurement module extracts f 1 = 1.07 Hz (fluctuating under wind speed interference), and the infrared thermal imaging module outputs T avg = 30.5 °C, ▽ T = 1.2 °C / m; The flexible strain gauge array fails to capture the strain peak due to hysteresis and outputs ε max = 155 με . The LSTM inputs the non-aligned vector [1.07, 2.14, 155, 30.5, 1.2, 210, 100], and the output tension is 2320 kN. The error relative to the actual cable force of 2450 kN reaches 5.3%.

[0070] Effect: In Example 3, through hardware triggering and timestamp alignment, the error is controlled within 2.04% under dynamic working conditions. In Comparative Example 3, due to time asynchronization, the vibration frequency drifts and the strain peak is missed, and the error increases to 5.3%. The time synchronization mechanism improves the feature fusion accuracy by 3.26 percentage points, verifying the effectiveness of microsecond-level synchronization in suppressing dynamic interference.

[0071] The cable force testing device includes a flexible strain gauge array and a strain transfer compensation unit built into the edge computing terminal. After obtaining the original strain data collected by the flexible strain gauge array, this unit extracts the high-frequency harmonic components in the frequency band of 50 to 200 Hz and calculates their energy amplitudes. When the energy amplitude is lower than the preset threshold, it triggers the medium type identification mechanism: if the attenuation amplitude in the frequency band of 70 to 100 Hz is significantly greater than that in other frequency bands, it is determined as the grease layer; if the attenuation amplitude in the frequency band of 150 to 200 Hz is significant, it is determined as the water film layer; if the attenuation is uniform across all frequency bands, it is determined as the oxide layer. Select the compensation coefficient according to the identification result: the compensation coefficient for the grease layer is 1.15, the compensation coefficient for the water film layer is 1.08, and the compensation coefficient for the oxide layer is 1.05. Multiply the compensation coefficient by the original strain data to output the compensated strain value, and calculate the maximum absolute strain value based on this.

[0072] Specifically, the edge computing terminal also has built-in: A strain transfer compensation unit, which is signal-connected to the flexible strain gauge array and configured to: Obtain the original cable body surface strain data collected by the flexible strain gauge array ε (x, t); Extract the strain data ε (x, t) of the high-frequency harmonic components in the frequency band of 50 - 200 Hz; Calculate the energy amplitude of the high-frequency harmonic components A h ; When A h is lower than the preset threshold A th it triggers the medium type identification: Grease layer identification: the attenuation amplitude in the frequency band of 70 - 100 Hz is greater than that in other frequency bands; Water film layer identification: the attenuation amplitude in the frequency band of 150 - 200 Hz is greater than that in other frequency bands; Oxide layer identification: uniform attenuation across all frequency bands; Select the compensation coefficient according to the identified medium type k : Grease layer: k = 1.15; Water film layer: k = 1.08; Oxide layer: k = 1.05; Output the compensated strain data: ε comp (x, t) = k × ε (x, t); Based on the compensated strain data ε comp(x, t) to calculate the maximum absolute strain value at all positions on the surface of the cable during the measurement period ε max . During the measurement period, traverse all position coordinates x and all time points t to determine the compensated strain data ε comp the maximum value of the absolute value of (x, t) ε max .

[0073] The closest prior art is the traditional strain gauge measurement method, which directly uses the original strain data to deduce the cable force. For example, under the grease contamination condition, the traditional device records a strain value of 142 με , but fails to recognize that the signal attenuation in the 70 to 100 Hz frequency band reaches 25%, resulting in the cable force prediction value being 11.1% lower than the true value. This embodiment identifies the medium type through the high-frequency harmonic attenuation mode and dynamically compensates to eliminate the strain signal distortion caused by grease or water film.

[0074] Example 4 In the cable force test of a certain suspension bridge, a device is used to measure a steel cable with a diameter of 100 mm and a length of 210 m. The ambient temperature is 28 °C, and there is grease attached to the surface of the cable. The laser vibration measurement module extracts f 1 = 1.08 Hz, f 2 = 2.16 Hz; the infrared thermal imaging module measures T avg = 30.5 °C, ▽ T = 1.2 °C / m. The strain transfer compensation unit detects that the high-frequency harmonic energy is lower than the threshold. After identifying the characteristics of the grease layer, the compensation coefficient 1.15 is applied, and the original strain value of 130 με is corrected to 150 με . The edge computing terminal inputs the feature vector [1.08, 2.16, 150, 30.5, 1.2, 210, 100] into the LSTM model and outputs T pred = 2450 kN. After calibration by the hydraulic sensor, the actual cable force is 2500 kN, and the relative error is 2.0%.

[0075] Comparative Example 4 Disable the strain transfer compensation unit under the same working conditions: directly use the original strain value of 130 με . The LSTM model inputs the feature vector [1.08, 2.16, 130, 30.5, 1.2, 210, 100], and outputs a tension of 2350 kN. Due to the signal attenuation caused by grease not being corrected, the predicted value is significantly lower than the actual cable force of 2500 kN, and the relative error is 6.0%.

[0076] Effect: In Example 4, through medium type identification and strain compensation, the error was controlled within 2.0%. In Comparative Example 4, due to the neglect of strain attenuation caused by grease, the error increased to 6.0%. It verifies the effective suppression of surface medium pollution by the high-frequency harmonic energy detection and dynamic compensation mechanism.

[0077] The infrared thermal imaging module of the cable force test device includes a reference blackbody plate fixed on a movable bearing platform. This plate is within the field of view of the module and parallel to the surface of the cable. The surface emissivity is constantly 0.97. The environmental radiation compensation unit performs the following operations: Obtain the environmental temperature by photographing the reference blackbody plate through the infrared thermal imaging module T amb , and synchronously collect the original temperature measurement value of the cable surface T meas ; Select the emissivity according to the surface characteristics of the cable observed manually - the emissivity of the silver - white metallic luster surface is set to 0.85, the emissivity of the gray - black non - plated and non - rusty surface is set to 0.88, the emissivity of the red - brown oxidized rust spot surface is set to 0.92, the emissivity of the PE sheath surface is 0.94, and the emissivity of the anti - corrosion coating surface is 0.91; Based on the principle of radiative heat balance, use the environmental temperature data of the reference blackbody plate T amb and the emissivity of the cable to correct the true temperature of the cable surface. The temperature processing unit calculates the average temperature T avg and the temperature gradient ▽ T .

[0078] Specifically, the infrared thermal imaging module includes: A reference blackbody plate, which is fixed on a movable bearing platform, within the field of view of the infrared thermal imaging module, parallel to the surface of the cable, and its emissivity is fixed at 0.97; An environmental radiation compensation unit, which is configured to: Obtain the environmental temperature by photographing the reference blackbody plate through the infrared thermal imaging module T amb , unit K ; Obtain the temperature value of the original measurement of the cable surface by the infrared thermal imaging module T meas , unit K ; Obtain the type of the cable surface judged manually and determine the emissivity ε obj , and the specific method is as follows: If the surface of the cable is the surface of a newly galvanized steel cable, and the surface characteristic is silver - white metallic luster, then ε obj =0.85; If the surface of the cable body is the surface of an ordinary steel cable, and the surface features are a grayish-black carbon steel substrate without coating and without rust, then ε obj = 0.88; If the surface of the cable body is the surface of a rusty steel cable, and the surface features are reddish-brown oxidation rust spots, then ε obj = 0.92; If the surface of the cable body is a PE sheath, then ε obj = 0.94; If the surface of the cable body is an anti-corrosion coating, then ε obj = 0.91; The true temperature of the cable body surface is corrected by the following formula: , unit K ; The temperature processing unit is configured to calculate the average temperature T corr based on the corrected temperature field T avg and the temperature gradient ▽ T .

[0079] The closest prior art is an infrared temperature measurement scheme without a reference benchmark, which directly uses the original measurement value T meas to calculate the temperature gradient. For example, in the case of strong sunlight in a traditional device, the measured value of the cable body surface temperature deviates by 3.2 °C due to ignoring the environmental radiation interference, and then the calculation error of the temperature gradient exceeds 40%. In this embodiment, the environmental temperature benchmark is provided in real time by referring to the blackbody plate T amb , and combined with emissivity classification correction, the temperature measurement distortion caused by the reflected radiation of the metal surface is eliminated.

[0080] Example 5 In the test of rusty steel cables of a cross-sea bridge, the environmental temperature is 38 °C and there is strong sunlight interference. The original measurement value of the infrared thermal imaging module T meas = 41.3 °C; the measured environmental temperature of the reference blackbody plate T amb = 35.6 °C; it is manually judged as a rusty steel, and the emissivity is determined to be 0.92. After radiation compensation, the corrected temperature of 39.8 °C is output. The temperature processing unit calculates the average temperature T avg = 39.5 °C, the temperature gradient ▽ T = 1.3 °C / m. Combining the vibration frequencies f 1 = 1.12 Hz, f 2 = 2.24 Hz and the strain extreme value εmax = 168 με The LSTM model outputs a tension of 2280 kN. The actual cable force calibrated by the hydraulic sensor is 2320 kN, with a relative error of 1.7%.

[0081] Comparative Example 5 Remove the reference blackbody plate under the same working conditions: directly use the infrared original measurement value T meas = 41.3 °C, calculate the average temperature of 41.0 °C, and the temperature gradient ▽ T = 1.6 °C / m. Input the feature vector [1.12, 2.24, 168, 41.0, 1.6, 190, 110] into the LSTM model, and the output tension is 2410 kN. Since the environmental radiation is not corrected, T meas It is 1.5 °C higher than the true value of 39.8 °C, resulting in a prediction tension error of up to 3.9%.

[0082] Effect: In Example 5, the environmental radiation is compensated by the reference blackbody, the temperature measurement error is reduced from 1.5 °C to 0.3 °C, and the cable force prediction error is controlled at 1.7%. In Comparative Example 5, due to the radiation interference not being eliminated, the calculation deviation of the temperature gradient ▽ T reaches 23%, and finally the cable force error increases to 3.9%. This verifies the effectiveness of the radiation compensation mechanism for high-temperature and strong-light working conditions.

[0083] The movable bearing platform of the cable force test device uses a drone platform. The split clamp body of the self-locking clamp module is provided with an annular mounting seat, which is coaxially sleeved outside the cable body, and an adjustable gap is formed between the inner diameter of the mounting seat and the diameter of the cable body. When the cable body diameter is 50 - 100 mm, the gap is set to 1.0 - 1.5 mm; when the cable body diameter is 100 - 200 mm, the gap is 1.5 - 2.0 mm; when the cable body diameter exceeds 200 mm, the gap is 2.0 - 3.0 mm. The two annular planes of the mounting seat are perpendicular to the axis, and at least three rigid flaps are evenly distributed circumferentially. One end of the flap is hinged to the annular plane through a rotating shaft, and the inner pressing surface fits the surface of the cable body when closed, and the free end protrudes outward to form a flared structure to guide the cable body to insert. A rectangular groove array is opened on the pressing surface to embed the flexible strain gauge carrier plate. Shape memory alloy wires connect adjacent flaps, an axial guide rail is fixed on the drone platform, and a slider with a locking device fixes the position of the mounting seat. The anti-rotation pin penetrates through the mounting seat and presses against the surface of the cable body to prevent rotation, and the electrothermal control unit outputs a pulsed current to activate the shape memory alloy to achieve self-locking.

[0084] The closest existing technology is a fixed-gap fixture, which uses a uniform 2mm gap between the mounting base and the cable body. For example, when installing an 80mm cable body, the excessive gap in conventional devices causes the petal body to wobble by over 5mm, while insufficient gap can cause jamming when installing a 200mm cable body. This embodiment utilizes a stepped gap design to accommodate diameter variations from 50 to 300mm, eliminating installation position deviation.

[0085] The LSTM neural network model in the cable tension test device adopts a two-layer deep architecture. After receiving the input feature vector, the first LSTM layer, consisting of 128 LSTM units, performs primary time series modeling of multi-source features such as vibration frequency, strain extremes, and temperature fields, extracting nonlinear relationships in the time dimension. The second LSTM layer, also configured with 128 LSTM units, inherits the hidden state sequence output by the first layer and further explores long-term dependencies between features through a gating mechanism. The final hidden state is input into the fully connected output layer, which uses a linear activation function to directly map it to the tension prediction value. The entire model is deployed locally on the edge computing terminal, and forward inference takes less than 200ms.

[0086] The closest existing technology is a single-layer LSTM prediction model, which consists of only 64 LSTM units and lacks deep time series processing capabilities. For example, when subjected to sudden changes in cable length, this traditional device experiences fluctuations in predicted values exceeding 8% due to insufficient model capacity, making it unsuitable for the variable conditions of long-span structures. This implementation enhances feature abstraction capabilities by stacking two LSTM layers, significantly improving the model's generalization performance for complex physical relationships.

[0087] About the LSTM prediction model of this application: 1. Model structure composition: The first long short-term memory layer: contains 128 memory units, responsible for receiving and preliminarily processing input features.

[0088] The second long short-term memory layer also contains 128 memory units and receives the output results of the first layer for deep feature extraction.

[0089] Fully connected output layer: contains only one neuron and uses a linear activation function to directly output the tension prediction value.

[0090] 2. Input data processing flow: Eigenvector construction: Seven eigenvalues are combined in a fixed order: first-order modal frequency, second-order modal frequency, maximum absolute strain value, average temperature, temperature gradient, cable length, and cable diameter.

[0091] Normalization preprocessing: Scale all feature values to the same order of magnitude (e.g., in the range [0, 1]). The scaling parameters (minimum / maximum or mean / standard deviation) need to be determined in advance through the training data, and the same parameters are used to process new data during actual prediction.

[0092] 3. Prediction calculation steps: Step 1. First-layer feature extraction Input the 7-dimensional normalized feature vector into the first long short-term memory layer. This layer calculates and outputs a 128-dimensional feature encoding vector through an internal gating mechanism (including an input gate, a forget gate, and an output gate).

[0093] Step 2. Second-layer depth processing Use the 128-dimensional vector output from the first layer as the input to the second long short-term memory layer. This layer further extracts high-order feature relationships and finally outputs a new 128-dimensional depth feature vector.

[0094] Step 3. Tension value calculation Input the 128-dimensional vector output from the second layer into the fully connected output layer, and obtain the final result through the following linear calculation: Predicted tension value = (weight matrix × input vector) + bias term Where: The weight matrix is a pre-trained parameter matrix with 1 row and 128 columns; The input vector is the 128-dimensional data output from the second layer; The bias term is a single pre-trained parameter value; The calculation result is the tension prediction value.

[0095] 4. Model training and deployment requirements: Training stage: A large number of sample data containing feature values and true tension labels need to be collected in advance, and the parameters of each layer (including the gating weights of the two memory units, the weight matrix and bias term of the output layer) are optimized through the backpropagation algorithm.

[0096] Deployment requirements: The trained model parameters need to be stored in the edge computing terminal in a fixed manner. During actual prediction, the fixed model is directly called to execute the above three-step calculation.

[0097] 5. Key implementation points: The cascade structure of the two memory units can effectively capture the complex temporal relationships between features; The input features must be arranged in the same order and uniformly normalized; The linear design of the output layer ensures the continuous output of the tension value; All calculations are completed in real time at the edge terminal, and the single prediction time is controlled within 200 ms.

[0098] The cable force test method includes the following steps: The laser Doppler vibrometer module non-contact collects the vibration signal of the cable; the infrared thermal imaging module synchronously obtains the surface temperature field of the cable; the flexible strain gauge array collects the strain data. The edge computing terminal performs signal processing: performs variational mode decomposition on the vibration signal to extract the first-order modal frequency and the second-order modal frequency; calculates the average temperature and the temperature gradient based on the temperature field; calculates the maximum absolute strain value on the surface of the cable based on the strain data. The feature vector includes the modal frequency, the strain extreme value, the average temperature, the temperature gradient, the length and the diameter, inputs into the LSTM neural network model, and outputs the predicted tension value. Based on the average temperature, correct the linear density and elastic modulus parameters, and calculate the theoretical tension value. When the deviation between the predicted tension value and the theoretical tension value exceeds the preset threshold, calculate the loss function and update the weight parameters of the LSTM model through backpropagation, and output the corrected predicted tension value as the final result.

[0099] Specifically, the test method of the cable force test device includes the following steps: Non-contact collect the vibration signal of the cable through the laser Doppler vibrometer module V (t); Non-contact collect the surface temperature field of the cable through the infrared thermal imaging module T (x, t); Collect the surface strain data of the cable through the flexible strain gauge array ε (x, t); Transmit the collected V (t), T (x, t), ε (x, t) to the edge computing terminal; The signal processing unit performs V Variational mode decomposition on (t) and extract the first-order modal frequency f 1 and the second-order modal frequency f 2; The temperature processing unit calculates the average temperature T Based on (x, t) T avg And the temperature gradient ▽ T ; The strain processing unit calculates the maximum absolute strain value at all positions on the surface of the cable during the measurement period based on ε (x, t) ε max ; The feature vector f 1, f 2, ε max , T avg , ▽ T , L , DInput the LSTM neural network model and output the predicted tension value T pred ; where L is the length of the cable body D is the diameter of the cable body Based on T avg the linear density μ and the elastic modulus E are corrected by the dynamic compensation unit. The corrected linear density μ eff () T and the elastic modulus E eff () T are calculated as follows μ eff () T = μ 0 / [1 + α L × ( T avg - T ref )] E eff () T = E 0 × [1 + α E × ( T avg - T ref )] Where μ 0 is the reference linear density α L is the coefficient of thermal expansion E 0 is the reference elastic modulus α E is the thermosensitive coefficient T ref is the reference temperature The theoretical tension value μ eff () T and E eff () T are calculated by the cable force prediction error calculation unit. The formula is as follows T phys : Where n is the harmonic order n = 1 or 2 f n represents then The first-order vibration frequency; n The value-taking rule is as follows: By default, take n = 1 (the first-order mode corresponding to the fundamental frequency).

[0100] For special working conditions, take n = 2 (the second-order mode) When the length of the cable body L < 20 m (short cable) or the environmental noise causes f 1 signal-to-noise ratio to be insufficient, the second-order mode frequency is preferentially adopted f 2.

[0101] If both f 1 and f 2 are input, calculate in parallel T phys ( n = 1) and T phys ( n = 2), and select the result with the smallest deviation from T pred as the verification benchmark.

[0102] n The value of n corresponds strictly to the order of the actually extracted vibration frequency. The mode with the highest energy is preferentially adopted (usually n = 1). For short cables or strong noise conditions, switch to

[0103] = 2.

[0103] Compare T pred with T phys . When | T pred - T phys | > δ , based on T pred and T phys calculate the loss function: where, δ is the preset threshold; Update the weight parameters of the LSTM neural network through backpropagation: where, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the loss function The gradient of the parameter θ ; η is the learning rate; Output the corrected predicted cable force value , The expression is: When | T pred - T phys | ≤ δ , output T pred as the final cable force value; When | T pred - T phys | > δ , output as the final cable force value.

[0104] The closest prior art is the single vibration frequency method, which directly substitutes the extracted fundamental frequency into the physical formula to calculate the tension. For example, in the case of sudden temperature changes in the traditional method, due to the lack of a cross-validation mechanism between the prediction model and the physical model, the error continues to increase to more than 12% under a 20°C temperature change. This embodiment realizes the dynamic coordination of theoretical calculation and machine learning prediction through online parameter correction and adaptive model update.

[0105] Example 6 In the cable force test of a long-span suspension bridge, the ambient temperature suddenly rises from the reference temperature of 20°C to 35°C. The measured vibration frequencies are f 1 = 1.18 Hz, f 2 = 2.37 Hz, the average temperature is 36.2°C, and the extreme strain is 165 με . The initial predicted tension of the LSTM model is 2080 kN. After the dynamic compensation unit corrects the material parameters, the calculated theoretical tension is 2150 kN. Since the deviation of 70 kN exceeds the threshold of 50 kN, the model parameters are updated by backpropagation, and the corrected tension of 2140 kN is output. The actual cable force calibrated by the hydraulic sensor is 2180 kN, and the relative error after correction is 1.8%.

[0106] Comparative Example 6 Under the same working conditions, the traditional vibration frequency method is used: only the vibration signal is extracted to obtain f 1 = 1.18 Hz, and the tension value is calculated through the physical formula (the linear density is taken as the reference value without correction). The output tension is 1940 kN, and the error relative to the actual cable force of 2180 kN reaches 11.0%. Due to the actual decrease of 2.1% in the linear density caused by the uncompensated temperature drift and the lack of a model correction mechanism, the error continues to accumulate.

[0107] Effect: In Example 6, through the cross-validation of theoretical values and predicted values and the online update of the model, the error is controlled within 1.8% under a temperature change of 15°C. In Comparative Example 6, due to the neglect of material parameter drift and the absence of a correction mechanism, the error increases to 11.0%. It verifies the improvement of the adaptability of the dynamic cooperation mechanism to temperature mutation conditions.

[0108] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily achieved. Therefore, without departing from the general concept defined by the claims and the equivalent scope, the present invention is not limited to specific details.

Claims

1. Cable force testing device, characterized in that, Including: A laser Doppler vibrometry module, which is integrated on a movable carrying platform and is used for non-contact acquisition of the vibration signals of a cable body V (t); An infrared thermal imaging module, which is integrated into a movable bearing platform and is used for non-contact acquisition of the surface temperature field of a cable body T (x, t); A flexible strain gauge array that is attached to the surface of the cable body and used to collect strain data on the surface of the cable body ε (x, t); Edge computing terminal, receiving vibration signals V (t), temperature field T (x, t) and strain data ε (x, t); Built into the edge computing terminal: A signal processing unit configured to perform variational mode decomposition on V (t) and extract the first-order modal frequency f 1 and the second-order modal frequency f 2; A temperature processing unit configured to calculate an average temperature based on T (x, t) T avg and a temperature gradient ∇ T ; A strain processing unit configured to calculate the maximum absolute strain value at all positions on the surface of the cable during the measurement period based on ε (x, t) ε max ; LSTM neural network model, whose input feature vector f 1, f 2, ε max , T avg , ▽ T , L , D , and outputs the predicted tension value T pred ; Among them, L is the length of the cable body; D is the diameter of the cable body.

2. The cable force testing device according to claim 1, wherein Also including: A self-locking fixture module for fixing the flexible strain gauge array on the surface of the cable body, which includes: A split fixture body, which includes: A mounting seat, which is circular and coaxially sleeved outside the cable body. The inner diameter of the mounting seat is larger than the diameter of the cable body, and a gap is formed between the two; At least three rigid flaps, which are evenly distributed on a circular plane of the mounting seat along the circumferential direction of the mounting seat. One end of each flap is hinged to the circular plane through a rotating shaft, so that each flap can open and close radially around the cable body; a pressing curved surface is provided on the inner side of each flap, and it fits with the surface of the cable body when closed; the other end of each flap is a free end and has an outwardly protruding arc-shaped curved surface. The arc-shaped curved surfaces at the free ends of each flap together form a horn-shaped structure, and the opening diameter of the horn mouth of the horn-shaped structure is larger than the diameter of the cable body, facilitating the insertion of the cable body from the horn mouth; the curvature of the arc-shaped curved surface matches the surface of the cable body, guiding the cable body to enter the inside of each flap along its axis direction to ensure that each flap is coaxial with the cable body when closed; a rectangular groove array is provided on the pressing curved surface; A strain gauge carrier plate array, which corresponds to the rectangular grooves one by one. Flexible strain gauges are pasted on the surface of each strain gauge carrier plate, and each strain gauge carrier plate is respectively embedded in the corresponding rectangular groove; Multiple groups of shape memory alloy wires or springs, with both ends of each shape memory alloy wire or spring connected to adjacent flaps respectively; An axial guide rail, which is fixed to the movable carrying platform and is parallel to the axis of the cable body; A slider, which is arranged on the circumferential surface of the mounting seat and matches the axial guide rail, and is provided with a locking device on it, and the locking device can fix the slider and the axial guide rail; Multiple anti-rotation pins, which penetrate the radial screw holes of the mounting seat and then press against the surface of the cable body to circumferentially fix the mounting seat and the cable body to prevent relative rotation between the two; An electrothermal control unit, which outputs pulsed current to activate the shape memory alloy wires or springs.

3. The cable force testing device according to claim 1, wherein The edge computing terminal also has built-in: Dynamic compensation unit, configured to be based on T avg Corrected linear density μ And elastic modulus E , Corrected linear density μ eff ( T ), Elastic modulus E eff ( T ), The formula is as follows: μ eff （ T ）= μ 0 / [1+ α L ×( T avg - T ref )] E eff （ T ）= E 0×[1+ α E ×( T avg - T ref )] wherein, μ 0 is the reference line density, α L is the coefficient of thermal expansion, E 0 is the reference elastic modulus, α E is the thermosensitive coefficient, T ref is the reference temperature; Tension prediction error calculation unit, which is based on μ eff ( T ) and E eff ( T ) to calculate the theoretical tension value T phys , and the formula is as follows: Among them, n is the harmonic order, n = 1 or 2; f n represents the n th vibration frequency; Compare T pred with T phys . When | T pred - T phys | > δ , calculate the loss function based on T pred and T phys : Among them, δ is a preset threshold value; Updating the LSTM neural network weight parameters through backpropagation: Among them, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the loss function The gradient of the parameter θ ; η is the learning rate; Output the corrected predicted cable force value , The expression is: When | T pred - T phys | ≤ δ Output T pred as the final cable force value; When | T pred - T phys | > δ output as the final cable force value.

4. The cable force testing device according to claim 3, characterized in that, Also including: A visualization terminal, which is connected to an edge computing terminal through a communication module, is used to receive T pred 、 V (t), T (x, t), ε (x, t), and based on T (x, t) to generate a heat distribution cloud map, and based on V (t) to generate a vibration mode curve. When | T pred - T phys | > δ a deviation alarm signal is triggered.

5. The cable force testing device according to claim 1, characterized in that, Also including: A time synchronization control module, which is configured to: Provide a time reference signal with microsecond-level accuracy through a GPS timing unit or a temperature-controlled crystal oscillator unit integrated on the movable carrying platform; Send hardware trigger pulses to the laser Doppler vibration measurement module, the infrared thermal imaging module and the flexible strain gauge array synchronously, so that the three start data acquisition at the same moment; For the collected vibration signals V (t), temperature field T (x, t) and strain data ε (x, t) embed the time stamp of the trigger moment t k ; The edge computing terminal also has built-in: A time domain alignment unit configured to, based on timestamps t k resample V ( t k ) T (x, t k ) ε (x, t k ) to a unified time axis and output a time-aligned feature vector f 1( t k ) f 2( t k ) ε max ( t k ) T avg ( t k ) ▽ T ( t k )]; where: The signal processing unit is configured to perform variational mode decomposition on the resampled V ( t k ) to extract the first-order modal frequency f 1( t k ) and the second-order modal frequency f 2( t k ); The temperature processing unit is configured to calculate the average temperature based on the resampled T (x, t k ) T avg ( t k ) and the temperature gradient ▽ T ( t k ); The strain processing unit is configured to calculate the maximum absolute strain values at all positions on the surface of the cable body during the measurement period based on the resampled ε (x, t k ); ε max ( t k ); The LSTM neural network model takes the time-aligned feature vectors f 1( t k ), f 2( t k ), ε max ( t k ), T avg ( t k ),▽ T ( t k ), L , D as input and outputs the predicted tension value T pred ( t k )。 6. The cable force testing device according to claim 1, wherein, The edge computing terminal also has built-in: A strain transfer compensation unit, which is signal-connected to the flexible strain gauge array and is configured to: Obtain the original surface strain data of the cable collected by the flexible strain gauge array ε (x, t); Extract strain data ε High-frequency harmonic components in the 50-200 Hz frequency band in (x, t); Calculate the energy amplitude of high-frequency harmonic components A h ; When A h below a preset threshold A th it triggers the identification of the medium type: Grease layer identification: The attenuation amplitude in the 70-100 Hz frequency band is greater than that in other frequency bands; Water film layer identification: The attenuation amplitude in the 150-200 Hz frequency band is greater than that in other frequency bands; Oxide layer identification: Uniform attenuation in the whole frequency band; Select a compensation factor according to the recognized medium type k : Grease layer: k = 1.15; Water film layer: k = 1.08; Oxide layer: k = 1.05; Output compensated strain data: ε comp (x, t) = k × ε (x, t); Based on the compensated strain data ε comp (x, t), calculate the maximum absolute strain value at all positions on the surface of the cable during the measurement period ε max .

7. The cable force testing device according to claim 1, wherein, The infrared thermal imaging module includes: A reference blackbody plate, which is fixed to the movable carrying platform and is within the field of view of the infrared thermal imaging module and is parallel to the surface of the cable body, and its emissivity is fixed at 0.97; An environmental radiation compensation unit, which is configured to: Capture the reference blackbody plate through the infrared thermal imaging module to obtain the ambient temperature T amb ; Obtain the temperature value of the original measurement of the cable body surface by the infrared thermal imaging module T meas ; Obtain the type of the cable body surface determined by manual judgment and determine the emissivity ε obj , and the specific method is as follows: If the surface of the cable body is the surface of a newly galvanized steel cable with a surface characteristic of a silver-white metallic luster, then ε obj = 0.85; If the surface of the cable body is the surface of a common steel cable, and the surface features are a grayish-black carbon steel substrate without coating and without rust, then ε obj = 0.88; If the surface of the cable body is the surface of a rusty steel cable and the surface feature is reddish-brown oxidation rust spots, then ε obj = 0.92; If the surface of the cable body is a PE sheath, then ε obj = 0.94; If the surface of the cable body is an anti-corrosion coating, then ε obj = 0.91; Correct the true temperature of the cable body surface through the following formula: The temperature processing unit is configured to calculate an average temperature based on the corrected temperature field T corr and a temperature gradient ∇ T avg T .​ 8. The cable force testing device according to claim 2, characterized in that The movable bearing platform is a drone; when the diameter of the cable is 50 - 100 mm, the gap between the mounting seat and the cable is 1.0 - 1.5 mm; when the diameter of the cable is 100 - 200 mm, the gap between the mounting seat and the cable is 1.5 - 2.0 mm; when the diameter of the cable is > 200 mm, the gap between the mounting seat and the cable is 2.0 - 3.0 mm.

9. The cable force testing device according to claim 1, characterized in that, The LSTM neural network model includes: Two LSTM layers, each LSTM layer containing 128 LSTM units; A fully connected output layer with a linear activation function; The LSTM neural network model is configured to: Receive the input feature vector f 1, f 2, ε max , T avg , ▽ T , L , D ; Process the input feature vector through two LSTM layers to generate the final hidden state; Input the hidden state into the fully connected output layer to output the predicted tension value T pred 。 10. A testing method using the cable force testing device according to claim 3, characterized in that, Include the following steps: Collect the vibration signals of the cable body non - contact through the laser Doppler vibrometer module V (t); Non-contact acquisition of the surface temperature field of the cable body through the infrared thermal imaging module T (x, t); Collect the surface strain data of the cable through a flexible strain gauge array ε (x, t); Transmit the collected V (t), T (x, t), ε (x, t) to the edge computing terminal; The signal processing unit performs variational mode decomposition on V (t) and extracts the first-order modal frequency f 1 and the second-order modal frequency f 2; The average temperature is calculated by the temperature processing unit based on T (x, t) T avg and the temperature gradient ▽ T ; The strain processing unit calculates the maximum absolute strain values at all positions on the surface of the cable body during the measurement period based on ε (x, t) ε max ; Input the feature vector f 1, f 2, ε max , T avg , ▽ T , L , D into the LSTM neural network model, and output the predicted tension value T pred ; Among them, L is the length of the cable body; D is the diameter of the cable body; Based on T avg , the line density is corrected by the dynamic compensation unit μ and the elastic modulus E , and the corrected line density μ eff ( T ), and the elastic modulus E eff ( T ) are given by the following formulas: μ eff （ T ）= μ 0 / [1+ α L ×( T avg - T ref )] E eff （ T ）= E 0×[1+ α E ×( T avg - T ref )] Among them, μ 0 is the reference line density, α L is the coefficient of thermal expansion, E 0 is the reference elastic modulus, α E is the thermosensitive coefficient, T ref is the reference temperature; The theoretical tension value T phys is calculated by the cable force prediction error calculation unit based on μ eff ( T ) and E eff ( T ), and the formula is as follows: T phys ​ Among them, n is the harmonic order, n = 1 or 2; f n represents the n th vibration frequency; Compare T pred with T phys . When | T pred - T phys | > δ , calculate the loss function based on T pred and T phys : Among them, δ is a preset threshold value; Update the LSTM neural network weight parameters through backpropagation: Among them, θ is the set of all trainable weight parameters of the LSTM neural network, including the weight matrix, bias vector of the LSTM layer, and the weights of the output layer; represents the loss function The gradient of the parameter θ ; η is the learning rate; Output the corrected predicted cable force value , The expression is: When | T pred - T phys | ≤ δ Output T pred as the final cable force value; When | T pred - T phys | > δ , output as the final cable force value.

Images