A closed-loop system and method for icing state identification and acoustic de-icing control.

Abstract

The application discloses an icing state recognition and sound wave deicing control closed loop system and method thereof, and relates to the technical field of icing monitoring and deicing control. The system has a detection and deicing dual mode, can output a frequency response signal in the detection mode, and output a sound wave vibration signal to be applied to a structure surface in the deicing mode; a fiber Bragg grating is arranged to collect a structure strain time sequence and output a current strain value; temperature, humidity, wind and other parameters are provided; a multi-source feature fusion and domain adaptive recognition model FCD-Net is deployed to intelligently classify the icing state; an excitation control signal of different power levels is output according to the recognition result; the deicing effect is monitored and the termination output is controlled, and a whole process closed loop control is formed. The application can realize intelligent icing recognition and adaptive deicing control of a power transmission conductor and other structures, has the advantages of high recognition precision, low energy consumption, timely response and the like, and is suitable for anti-icing applications in the fields of electric power, wind power, communication and the like.

Description

Technical Field

[0001] This invention relates to the field of icing monitoring and de-icing control technology, and in particular to a closed-loop system and method for icing status identification and acoustic de-icing control. Background Technology

[0002] In critical infrastructure exposed to outdoor environments, such as power transmission lines, wind turbine blades, and communication base stations, icing is a common and serious environmental hazard that significantly impacts operational safety and structural reliability. To address this, existing technologies have proposed various monitoring methods based on environmental parameter sensing and diverse de-icing techniques, including temperature and humidity sensor-triggered heating devices, image recognition-assisted manual judgment, AC current heating, infrared excitation, and mechanical de-icing. These solutions have improved icing response efficiency to some extent and achieved partial automation.

[0003] However, the aforementioned existing technologies still have the following technical problems that urgently need to be solved: First, most current systems rely mainly on environmental threshold judgment or image processing for icing identification, lacking in-depth perception of the structural response, making it difficult to accurately determine whether icing has actually occurred and its load status, resulting in misidentification and response delays; Second, the monitoring signal dimensions are limited, failing to comprehensively consider the correlation between structural strain, acoustic changes, and environmental conditions, and failing to form the feature loop required for accurate judgment; Third, the control logic generally adopts preset rules, lacking the ability to dynamically respond and adaptively adjust for different icing stages, resulting in high energy consumption and low efficiency in the ice melting process, and the inability to automatically terminate execution after the ice layer is cleared, posing a risk of energy waste and structural thermal damage.

[0004] Therefore, there is an urgent need to propose an ice-recognition and de-icing system and method with multi-source information fusion, high recognition accuracy, intelligent power adjustment and automatic feedback control capabilities. Summary of the Invention

[0005] This invention aims to solve the technical problems of existing icing monitoring systems, such as inaccurate identification of icing status, untimely control response, high energy consumption for de-icing, and lack of automatic feedback termination mechanism. It provides a closed-loop system and method for icing status identification and acoustic de-icing control, which realizes real-time detection, intelligent identification, power adaptive control, and automatic closed-loop feedback of structural icing status.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] On one hand, the present invention provides a closed-loop system for icing state identification and acoustic de-icing control, comprising:

[0008] The acoustic detection and execution module executes either the detection mode or the ice-melting mode according to the control signal. In the detection mode, it outputs an acoustic frequency response signal, and in the ice-melting mode, it outputs an acoustic vibration signal. The vibration energy is then transferred to the surface of the structure through a transducer area set on the monitored structure to mechanically disturb the attached ice.

[0009] The strain monitoring module is used to collect the time series of structural strain from multiple directions of the monitored structure and output the current strain value.

[0010] The environmental acquisition module is used to acquire temperature, humidity and wind speed information around the monitored structure and output environmental parameter data.

[0011] The data processing module, connected to the acoustic detection and execution module, strain monitoring module and environmental acquisition module, is used to fuse the acquired acoustic frequency response signal, current strain value and environmental parameter data, and intelligently classify the icing state of the monitored structure based on the icing state identification model FCD-Net constructed by multi-source feature fusion and domain adaptive mechanism, and output icing state label signal.

[0012] A power control module, connected to the data processing module and the acoustic detection and execution module, is used to output a corresponding excitation power control signal according to the icing status tag signal;

[0013] The feedback judgment module compares the current strain value and acoustic frequency response signal with the preset reference value. When the de-icing criterion is met, it outputs a termination control signal, shuts down the acoustic detection and execution module, and completes closed-loop control.

[0014] A further improvement of the present invention is that the acoustic detection and execution module includes:

[0015] The frequency detection unit includes an interdigital transducer mounted on a piezoelectric substrate. In detection mode, it receives reflected signals and outputs its acoustic frequency response signal. The value of the acoustic frequency response signal shifts with the change in the icing state of the structure surface.

[0016] The acoustic excitation unit, in the ice-melting mode, responds to the power control module to output an excitation power control signal, driving the interdigital transducer to generate Rayleigh-type surface acoustic waves with a frequency of 32.5MHz and a vibration amplitude of 5-10nm on a lithium niobate substrate cut at 128°YX.

[0017] The transducer coupling unit includes an adhesive layer between the piezoelectric substrate and the surface of the monitored structure. The adhesive layer is composed of polyimide or epoxy resin with a thickness of 10–50 μm and is used to ensure the mechanical coupling of surface acoustic waves between the substrate and the structure.

[0018] A further improvement of the present invention is that the strain monitoring module includes:

[0019] The multi-point acquisition unit includes at least four fiber Bragg grating sensors uniformly arranged along the circumference of the cross section of the monitored structure. The sensors are respectively set at different orientation positions to synchronously acquire the axial strain values ​​at the corresponding positions of the structure and output the current strain value group.

[0020] The demodulation unit is used to perform spectral demodulation on the reflection wavelength of each channel in the current strain value group and output strain value vectors corresponding to multiple deployment positions.

[0021] The asymmetric strain identification unit is used to calculate the difference between the maximum and minimum strain values ​​of the strain value vector and compare it with a preset strain difference threshold. When the difference exceeds the preset strain difference threshold, it outputs a strain difference signal to indicate that there is an off-center loading state in the circumferential direction of the structure.

[0022] The asymmetric strain identification unit uses a maximum-minimum difference discrimination model, which is suitable for suspended structures such as power conductors. When "C-shaped" icing occurs at the bottom of the conductor, the strain difference typically exceeds 200. This signal can be used as a basis for identifying the off-center load condition.

[0023] A further improvement of the present invention is that the environmental acquisition module includes:

[0024] The environmental parameter acquisition unit includes a temperature sensor, a relative humidity sensor, and a wind speed sensor, which respectively collect the temperature around the monitored structure. ,humidity and wind speed It also outputs the corresponding environmental parameter signals;

[0025] The parameter construction unit is used to combine the environmental parameter signals to form a standard three-dimensional environmental state vector. The three-dimensional environmental state input vector is output to the data processing module as one of the inputs for icing state recognition.

[0026] A further improvement of the present invention is that the data processing module includes an icing state recognition model FCD-Net constructed based on a multi-source feature fusion and domain adaptation mechanism, wherein FCD-Net includes:

[0027] The feature encoding unit, including a unified encoder structure, is used to process the acquired acoustic frequency response signal, structural strain time series, and structural strain increment sequence. The environmental state parameter vector is concatenated to form a multi-source feature input vector, which is then embedded and encoded to output a fused feature vector.

[0028] The state contrast modeling unit includes a contrast learning module, which is used during the training phase to construct an inter-state embedding distance discriminant model using labeled icing state samples and generate a feature separation loss.

[0029] The classification output unit includes a convolutional neural network classifier, whose input is connected to the feature encoding unit, and outputs an icing state label signal representing the icing state category.

[0030] The domain feature alignment unit, including the maximum mean difference loss module, is used to calculate the distribution deviation based on the feature vectors of the source domain and the target domain during the training phase, and to adjust the parameters of the classifier's front layer through the domain alignment loss.

[0031] The feature mask subunit is used to shield invalid input features through a mask matrix to maintain the integrity of the model input structure when sensor data is missing or abnormal.

[0032] The attention visualization unit is used to generate state recognition heatmaps based on convolutional feature maps, which enhances the interpretability and position sensitivity of the model's classification output.

[0033] A further improvement of the present invention is that the FCD-Net model is constructed using a combined loss function during the training phase, including:

[0034] The state-discrimination loss term, used to enhance the separability of embedded features between ice-free and ice-covered states, is expressed as follows:

[0035] ;

[0036] in: and These represent the feature embedding vectors collected under icy and non-iced conditions, respectively. The embedded feature vectors of the candidate samples used for comparison are derived from the set of all iced or uniced samples in the current training batch, with an index. Excluding in the current training batch All external samples, The cosine similarity between features is represented. This is the temperature scaling factor;

[0037] It should be noted that this loss function A normalized contrastive learning mechanism is adopted to improve the separability of features of different icing states by narrowing the embedding distance of similar samples and widening the similarity between dissimilar samples.

[0038] The domain-adaptive loss term minimizes the distributional difference between the embedded features of the source and target domains. Its maximum mean difference loss is:

[0039] ;

[0040] Where: index and These represent the sample numbers in the source and target domains, respectively. , Let represent the embedded feature vectors of the source and target domain samples, respectively. This indicates the number of samples in the source domain. This indicates the number of samples in the target domain. For kernel function mapping;

[0041] Combined training objective function, including state recognition cross-entropy loss term Overall objective function The expression is:

[0042] ;

[0043] in: , , These are hyperparameter weight coefficients, which are adjusted according to different training tasks.

[0044] A further improvement of the present invention is that the data processing module further includes an ice layer dynamic discrimination unit, comprising:

[0045] The strain time window sampling unit is used to receive the structural strain time series output by the strain monitoring module, and calculate the difference between adjacent time points within a preset sliding time window to form a structural strain increment sequence. ;

[0046] The variability calculation unit calculates the information entropy value based on the probability density distribution estimation results in the sequence set. The expression is:

[0047] ;

[0048] in: For the incremental value to fall into the first The probability density of the distribution interval For interval numbers;

[0049] The stage identification logic unit is used to identify stages based on entropy values. Output icing development stage indicator signal:

[0050] when If so, it is determined to be in the rapid freezing stage;

[0051] when If so, it is determined to be in the stable freezing stage;

[0052] when If so, it is determined to be in the natural melting stage.

[0053] A further improvement of the present invention is that the power control module includes:

[0054] The pattern recognition unit receives the output icing status tag signal and uses it to determine whether the current state is in the initial icing stage, the stable icing stage, or other states.

[0055] The power setting unit is used to set the excitation power control signal parameter values ​​according to the icing stage.

[0056] When the excitation power control signal indicates the initial icing stage, the first set power is output. ;

[0057] When the indication is in the stable icing stage, output the second set power. ;

[0058] The signal conversion unit is used to convert the set power value into a drive excitation signal with corresponding amplitude and frequency parameters, and send it to the acoustic excitation unit to drive the surface acoustic wave actuator to work.

[0059] A further improvement of the present invention is that the feedback determination module includes:

[0060] The data comparison unit is used to receive the current strain value output by the strain monitoring module. and the acoustic frequency response signal output by the acoustic detection and execution module. And respectively compared with the preset reference strain value and reference frequency value Compare;

[0061] The threshold judgment unit is used to output a de-icing completion feedback signal when both of the following criteria are met simultaneously:

[0062] ;

[0063] The feedback command output unit, after determining that de-icing has been completed, sends a termination control signal to the power control module to shut down the acoustic vibration function of the acoustic detection and execution module, thus completing the closed-loop control process.

[0064] The feedback determination module is used to determine the recovery status of the current structure in real time during the ice melting process. When the FBG strain recovers to the ice-free reference state and the acoustic frequency response signal detected by the piezoelectric transducer also recovers to the preset reference frequency range, the system determines that the ice layer has been basically removed and terminates the power output to avoid energy waste.

[0065] On the other hand, the present invention provides a closed-loop method for icing state identification and acoustic de-icing control, which applies the closed-loop system for icing state identification and acoustic de-icing control as described above, and includes the following steps:

[0066] Step 1: Set the acoustic detection and execution device to detection mode and acquire its output acoustic frequency response signal to reflect the icing state of the monitored structure surface;

[0067] Step 2: Collect the time series of structural strain at multiple locations of the monitored structure, extract the current strain value, and simultaneously collect ambient temperature, humidity and wind speed information to generate environmental parameter data;

[0068] Step 3: Input the acoustic frequency response signal, current strain value and environmental parameter data into the icing state recognition model to identify whether icing exists and its current state, and output the corresponding icing state label signal;

[0069] Step 4: Determine whether the initial icing or stable icing stage has been entered based on the icing status tag signal, and generate low-power or high-power excitation power control signals accordingly.

[0070] Step 5: Send the excitation power control signal to the acoustic detection and execution equipment, switch to the de-icing mode, output the acoustic vibration signal, and apply the vibration energy to the surface of the structure through the transducer area in contact with the monitored structure to disturb and remove the ice.

[0071] Step 6: Obtain the current strain value and acoustic frequency response signal, and compare them with the preset reference value to determine whether the preset de-icing conditions are met;

[0072] Step 7: Once it is determined that the ice layer has been removed, a termination control signal is generated to shut down the acoustic excitation output and complete the current ice melting control.

[0073] The beneficial effects of this invention are as follows: By integrating modules such as acoustic detection, structural strain sensing, environmental condition acquisition, and multi-source intelligent recognition, high-precision identification and adaptive control of structural icing states can be achieved. Utilizing the acoustic frequency response signal output by the acoustic detection and execution module, the frequency shift caused by icing on the structural surface can be sensitively captured, thereby timely identifying the ice formation process. The strain monitoring module collects structural strain time series through multi-point deployed fiber Bragg grating sensors, obtaining the structure's response under ice load, further improving the ability to identify ice thickness and distribution. The environmental acquisition module supplements real-time environmental state parameters such as temperature, humidity, and wind, providing basic support for multi-dimensional information fusion. The data processing module deploys the FCD-Net recognition model based on multi-source feature fusion and domain adaptive mechanisms, which can comprehensively process acoustic, structural, and environmental data to achieve intelligent judgment of icing states. The recognition results drive the power control module to output excitation power control signals of different levels, enabling the acoustic detection and execution module to operate in low-power anti-icing or high-power de-icing modes at different icing stages. The feedback judgment module compares the current strain value with the frequency response to determine whether the ice layer has been removed. Upon meeting preset conditions, the excitation output is automatically terminated, thus constructing a closed-loop control process of response, execution, judgment, and termination. The overall system possesses advantages such as high responsiveness, high recognition accuracy, and low energy consumption. It can achieve adaptive control of the entire process of structural anti-icing and de-icing, improving operational reliability and possessing significant engineering practical value. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0075] Figure 1 This is a system modular structure diagram of an embodiment of the present invention;

[0076] Figure 2 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0078] Terminology Explanation:

[0079] Acoustic frequency response signal: refers to the electrical response signal of the piezoelectric transducer after receiving the surface acoustic wave reflected back from the structural surface in detection mode. Its resonant frequency shifts with the icing state.

[0080] Current strain value: refers to the axial strain measurement value obtained by the FBG sensor at a certain sampling time.

[0081] Structural strain time series: refers to a set of strain data continuously collected by multi-point FBG sensors and arranged according to time variation. It is often used to calculate structural variability and dynamic ice load trends.

[0082] Icing state label signal: The classification result label output by the FCD-Net model indicates that the structure is currently in a state such as "no icing", "initial icing", "stable icing" or "natural melting".

[0083] Excitation power control signal: refers to the radio frequency excitation control signal output by the power control module, which includes parameters such as frequency, amplitude and duration, and is used to regulate the transducer's operating mode.

[0084] FCD-Net model: refers to a deep neural network model that integrates acoustic, strain and environmental features from multiple sources, and combines contrastive learning and domain adaptation mechanisms. It is suitable for intelligent identification of ice conditions.

[0085] Feedback Judgment Module: This module determines the completion status of de-icing by comparing the current strain value with the frequency signal and its ice-free reference value in real time. It constitutes the endpoint of the system's closed-loop control logic.

[0086] like Figure 1 As shown, this is an embodiment of the present invention, which provides a closed-loop system for icing state identification and acoustic de-icing control, comprising:

[0087] (1) Acoustic detection and execution module

[0088] The system executes either detection mode or de-icing mode according to the control signal. In detection mode, it outputs an acoustic frequency response signal, and in de-icing mode, it outputs an acoustic vibration signal. The vibration energy is then transferred to the surface of the structure through a transducer area set on the monitored structure to mechanically disturb the attached ice.

[0089] In this embodiment, the acoustic detection and execution module achieves a closed-loop linkage between detection and ice-melting modes in terms of structural layout and functional coordination.

[0090] The frequency detection unit receives the surface acoustic wave reflection signal returned from the structural surface through an interdigital transducer mounted on the piezoelectric substrate in detection mode, and outputs its resonant frequency value in real time as the acoustic frequency response signal. This frequency value is affected by the state of ice on the structural surface and undergoes a measurable shift, which is usually between 0.3-0.8MHz, reflecting the evolution trend of the degree of icing. It serves as one of the main inputs for the data processing module to perform icing identification criteria.

[0091] The acoustic excitation unit operates in the ice-melting mode. In response to the excitation power control signal output by the power control module, it drives the interdigital transducer to generate Rayleigh-type surface acoustic waves on a lithium niobate substrate with a 128° YX crystal chamfer. The frequency of the surface wave is set to 32.5MHz, and the vibration amplitude is controlled within the range of 5–10nm, balancing excitation efficiency and structural stability. The excitation unit can be further adapted to the resonant characteristics of different types of structures (such as wires, supports, etc.) by adjusting the frequency bandwidth.

[0092] The transducer coupling unit provides the necessary mechanical medium in the acoustic wave transmission path. It consists of an adhesive layer made of polyimide or epoxy resin with a thickness of 10–50 μm, used to firmly bond the piezoelectric substrate to the monitored structure. Through a hot-press bonding process, the adhesive layer acquires stable acoustic impedance matching characteristics after curing, ensuring efficient coupling of acoustic wave energy between the substrate and the structural surface, improving energy transmission efficiency and suppressing vibration attenuation.

[0093] (2) Strain monitoring module

[0094] Used to collect the time series of structural strain from multiple directions of the monitored structure and output the current strain value;

[0095] In this embodiment, the strain monitoring module achieves multi-point synchronous sensing of the strain state of the entire cross section of the structure through a distributed Bragg grating (FBG) array.

[0096] Specifically, the multi-point acquisition unit consists of no fewer than four FBG sensors, which are equally spaced along the circumferential direction of the cross-section of the monitored structure. For example, they are attached to typical poles or suspended structures such as power lines, tower beams, or wind turbine blades in a 90° symmetrical layout to simultaneously acquire the axial strain response of each direction at the same time point. Under structural stress changes or ice loads, the central reflected wavelength of each FBG sensor will undergo a slight shift, reflecting the strain value at its location.

[0097] The demodulation unit employs a broadband spectral demodulation method to perform wavelength-to-strain conversion on the multi-channel FBG reflection signal, outputting strain value vectors corresponding to multiple azimuth positions at the current time. This forms a real-time strain distribution map of the structure, which serves as the basis for subsequent asymmetric identification and ice load inference.

[0098] The asymmetric strain identification unit performs a maximum-minimum strain difference calculation on the strain numerical vector, that is:

[0099] ;

[0100] This was then compared to a preset threshold. For typical C-shaped icing patterns (such as under-conductor eccentric loading), experiments showed that the strain difference typically exceeded 200. The value is much higher than the normal range under natural load fluctuations. Therefore, when the criterion is triggered, the system can output the "circumferential off-center load" strain difference signal as a high-confidence identification basis for local icing of the structure.

[0101] (3) Environmental data acquisition module

[0102] Used to acquire temperature, humidity and wind speed information around the monitored structure and output environmental parameter data;

[0103] In this embodiment, the environmental acquisition module is used to provide timely and accurate external meteorological information input, providing auxiliary feature support for intelligent identification of icing conditions.

[0104] - The environmental parameter acquisition unit includes an integrated meteorological micro-station component, which is installed at the end of a bracket or on the surface of the equipment housing adjacent to the structure being monitored, and is equipped with temperature sensors (measuring range -40°C). Up to +80 Accuracy ±0.5 The system uses a relative humidity sensor (measurement range 0–100%, accuracy ±3%RH) and an ultrasonic anemometer (measurement range 0–30m / s, accuracy ±0.3m / s) to collect external environmental parameters once per second and record them with a synchronous timestamp.

[0105] The parameter construction unit performs a structured packaging operation on the collected raw temperature, humidity, and wind data to form a standard three-dimensional environmental state vector: The data, along with the current structural strain value and acoustic frequency response signal, is input to the data processing module as an auxiliary dimension in the input features of the FCD-Net model. The construction process supports redundant channel masking and missing data interpolation to ensure the integrity of the environmental input.

[0106] In particular, in the critical freezing temperature region (0±1) Within this range, when the ambient humidity is greater than 90% and the wind speed is less than 1 m / s, the system automatically increases the sensitivity threshold of the icing recognition model, thereby effectively reducing the probability of missed detection in such weak wind and high humidity weather scenarios and enhancing the model's ability to distinguish icing boundary states.

[0107] (4) Data processing module

[0108] Connected to the acoustic detection and execution module, strain monitoring module and environmental acquisition module, it is used to fuse the acquired acoustic frequency response signal, current strain value and environmental parameter data, and to intelligently classify the icing state of the monitored structure and output the icing state label signal based on the icing state identification model FCD-Net constructed based on the multi-source feature fusion and domain adaptive mechanism.

[0109] In one embodiment, the data processing module includes an icing state recognition model FCD-Net constructed based on multi-source feature fusion and domain adaptation mechanisms, wherein FCD-Net includes:

[0110] The feature encoding unit, including a unified encoder structure, is used to process the acquired acoustic frequency response signal, structural strain time series, and structural strain increment sequence. The environmental state parameter vector is concatenated to form a multi-source feature input vector, which is then embedded and encoded to output a fused feature vector.

[0111] The state contrast modeling unit includes a contrast learning module, which is used during the training phase to construct an inter-state embedding distance discriminant model using labeled icing state samples and generate a feature separation loss.

[0112] The classification output unit includes a convolutional neural network classifier, whose input is connected to the feature encoding unit, and outputs an icing state label signal representing the icing state category.

[0113] The domain feature alignment unit, including the maximum mean difference loss module, is used to calculate the distribution deviation based on the feature vectors of the source domain and the target domain during the training phase, and to adjust the parameters of the classifier's front layer through the domain alignment loss.

[0114] The feature mask subunit is used to shield invalid input features through a mask matrix to maintain the integrity of the model input structure when sensor data is missing or abnormal.

[0115] The attention visualization unit is used to generate state recognition heatmaps based on convolutional feature maps, which enhances the interpretability and position sensitivity of the model's classification output.

[0116] In one specific embodiment, the FCD-Net model is constructed using a combined loss function during the training phase, including:

[0117] The state-discrimination loss term is used to enhance the separability of embedded features between ice-free and ice-covered states, in contrast to the loss function. The expression is:

[0118] ;

[0119] in: and These represent the feature embedding vectors collected under icy and non-iced conditions, respectively. The embedded feature vectors of the candidate samples used for comparison are derived from the set of all iced or uniced samples in the current training batch, with an index. In the current training batch, except All external samples, The cosine similarity between features is represented. This is the temperature scaling factor;

[0120] Domain-adaptive loss term, used to minimize the distribution difference between the embedded features of the source and target domains, maximum mean difference loss. The formula is:

[0121] ;

[0122] Where: index and These represent the sample numbers in the source and target domains, respectively. , Let represent the embedded feature vectors of the source and target domain samples, respectively. This indicates the number of samples in the source domain. This indicates the number of samples in the target domain. For kernel function mapping;

[0123] Combined training objective function, including state recognition cross-entropy loss term Overall objective function The expression is:

[0124] ;

[0125] in: , , These are hyperparameter weight coefficients, which are adjusted according to different training tasks.

[0126] The data processing module also includes an ice layer dynamic discrimination unit, including:

[0127] The strain time window sampling unit is used to receive the structural strain time series output by the strain monitoring module, and calculate the difference between adjacent time points within a preset sliding time window to form a structural strain increment sequence. ;

[0128] The variability calculation unit calculates the information entropy value based on the probability density distribution estimation results in the sequence set. The expression is:

[0129] ;

[0130] in: For the incremental value to fall into the first The probability density of the distribution interval For interval numbers;

[0131] The stage identification logic unit is used to identify stages based on entropy values. Output icing development stage indicator signal:

[0132] when If so, it is determined to be in the rapid freezing stage;

[0133] when If so, it is determined to be in the stable freezing stage;

[0134] when If so, it is determined to be in the natural melting stage.

[0135] It is important to emphasize that the data processing module in this embodiment constructs three types of heterogeneous feature vectors that fuse acoustic frequency response, structural strain sequence, and environmental state parameters. It then employs a composite training strategy integrating contrast loss, domain difference loss, and recognition loss to build the FCD-Net icing state recognition model, which is driven by a multi-source feature fusion and domain adaptation mechanism. This model possesses the following significant advantages:

[0136] Enhanced feature layer discriminability: Compared with existing methods that rely solely on temperature or single sensor data for discrimination, the FCD-Net model utilizes a contrastive learning module to improve the separability of embedded features for ice / free-ice states, significantly enhancing classification robustness under ambiguous boundary conditions.

[0137] Strong adaptability to inter-domain differences: The domain alignment strategy constructed by combining maximum mean difference (MMD) can effectively mitigate the distribution offset problem between training data and actual deployment scenarios, and avoid the model being "unsuitable" for the actual deployment scenario;

[0138] The dynamic process of icing can be identified: by incorporating the structural strain increment entropy index as a means of analyzing the time series change trend, the system can not only identify whether icing has occurred, but also determine its development stage (icing / stabilization / melting).

[0139] It has a feature missing fault tolerance mechanism: it automatically masks abnormal channels through the feature mask submodule, realizes fault tolerance identification under partial sensor failure or signal loss, and enhances system stability.

[0140] Therefore, the data processing module outperforms existing ice condition recognition technologies based on single-source models or non-adversarial training in terms of multi-dimensional feature fusion processing capabilities, cross-environment adaptability, and process stage discrimination capabilities, and has a higher level of intelligence and practical application value.

[0141] (5) Power control module

[0142] Connected to the data processing module and the acoustic detection and execution module, it is used to output a corresponding excitation power control signal according to the icing status tag signal;

[0143] In this embodiment, the power control module is used to perform graded regulation of the acoustic excitation output based on the icing state identification result, so as to achieve a dynamic balance between energy efficiency control and de-icing effect.

[0144] The pattern recognition unit receives the icing status label signal output by FCD-Net and determines the current stage of the structural surface based on the label content, including three categories: initial icing stage, stable icing stage, and non-icing state. The status label signal is represented in coded form, with initial icing corresponding to the code "01", stable icing corresponding to the code "10", and the rest being "00" or "11" to indicate non-icing or melting state.

[0145] The power setting unit selects the output power range based on the current stage:

[0146] If the status label is "01" (initial icing stage), the output excitation power control signal is the first set power. To maintain continuous perturbation excitation with low energy consumption, the initial ice crystal attachment process is disrupted;

[0147] If the status label is "10" (stable icing stage), the output excitation power control signal is the second set power. Increase the vibration amplitude and apply stronger mechanical stress to accelerate the removal of ice.

[0148] The signal conversion unit maps the set power to a modulation signal with corresponding amplitude and frequency parameters, and drives the SAW excitation transducer to work through the radio frequency controller. The signal type is 32.5MHz sine wave envelope modulation, and the vibration pulse duty cycle is adaptively adjusted between 30% and 60%.

[0149] To prevent energy waste or structural damage caused by accidental high-power output, the system has a switching threshold protection mechanism. When the ambient temperature is higher than 2°C or the wind speed is greater than 5m / s, even if the label is "10", it will maintain low power output and issue a system alarm to ensure system robustness.

[0150] (6) Feedback Judgment Module

[0151] It is used to compare the current strain value and acoustic frequency response signal with the preset reference value. When the de-icing criterion is met, it outputs a termination control signal, shuts down the acoustic detection and execution module, and completes closed-loop control.

[0152] The feedback determination module includes:

[0153] The data comparison unit is used to receive the current strain value output by the strain monitoring module. and the acoustic frequency response signal output by the acoustic detection and execution module. And respectively compared with the preset reference strain value and reference frequency value Compare;

[0154] The threshold judgment unit is used to output a de-icing completion feedback signal when both of the following criteria are met simultaneously:

[0155] ;

[0156] The feedback command output unit, after determining that de-icing has been completed, sends a termination control signal to the power control module to shut down the acoustic vibration function of the acoustic detection and execution module, thus completing the closed-loop control process.

[0157] The feedback determination module is used to determine the recovery status of the current structure in real time during the ice melting process. When the FBG strain recovers to the ice-free reference state and the acoustic frequency response signal detected by the piezoelectric transducer also recovers to the preset reference frequency range, the system determines that the ice layer has been basically removed and terminates the power output to avoid energy waste.

[0158] like Figure 2 As shown, another embodiment of the present invention provides a closed-loop method for icing state identification and acoustic de-icing control, applying the closed-loop system for icing state identification and acoustic de-icing control as described above, including the following steps:

[0159] Step 1: Set the acoustic detection and execution device to detection mode and acquire its output acoustic frequency response signal to reflect the icing state of the monitored structure surface;

[0160] Step 2: Collect the time series of structural strain at multiple locations of the monitored structure, extract the current strain value, and simultaneously collect ambient temperature, humidity and wind speed information to generate environmental parameter data;

[0161] Step 3: Input the acoustic frequency response signal, current strain value and environmental parameter data into the icing state recognition model to identify whether icing exists and its current state, and output the corresponding icing state label signal;

[0162] Step 4: Determine whether the initial icing or stable icing stage has been entered based on the icing status tag signal, and generate low-power or high-power excitation power control signals accordingly.

[0163] Step 5: Send the excitation power control signal to the acoustic detection and execution equipment, switch to the de-icing mode, output the acoustic vibration signal, and apply the vibration energy to the surface of the structure through the transducer area in contact with the monitored structure to disturb and remove the ice.

[0164] Step 6: Obtain the current strain value and acoustic frequency response signal, and compare them with the preset reference value to determine whether the preset de-icing conditions are met;

[0165] Step 7: Once it is determined that the ice layer has been removed, a termination control signal is generated, the acoustic excitation output is turned off, and after completing the current ice melting control process, Step 1 is executed again to achieve continuous monitoring and closed-loop control of the icing state.

[0166] In summary, this invention constructs a closed-loop monitoring system comprising six functional modules: acoustic detection, structural strain monitoring, environmental data acquisition, intelligent identification, power regulation, and feedback judgment. This forms a fully automated icing prevention and control system covering the entire process from data acquisition and status identification to excitation control and feedback termination. The system introduces the multi-source fusion identification model FCD-Net, improving identification accuracy and robustness in complex environments. It possesses precise icing stage judgment capabilities and dynamic power adjustment strategies, while also achieving automatic judgment and termination response after icing meltdown, effectively reducing energy consumption and structural thermal load risks. The technical solution provided by this invention can significantly improve the safe operation level of key structures such as transmission lines, wind power equipment, and communication towers under extreme climatic conditions, possessing high engineering practical value and promising prospects for widespread application.

[0167] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0168] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0169] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An ice-coated state identification and acoustic wave ice-melting control closed loop system, characterized in that, The application relates to an ice detection and removal system for a monitored structure, which comprises the following modules: An acoustic detection and execution module, which executes a detection mode or an ice melting mode according to a control signal, outputs an acoustic frequency response signal in the detection mode, outputs a sound wave vibration signal in the ice melting mode, and transmits vibration energy to a structure surface through a transducer area arranged on the monitored structure, so as to mechanically disturb the attached ice; A strain monitoring module, which is used for collecting a multi-azimuth structural strain time sequence of the monitored structure and outputting a current strain value; An environment acquisition module, which is used for acquiring temperature, humidity and wind speed information around the monitored structure and outputting environment parameter data; A data processing module, which is used for fusing the acquired acoustic frequency response signal, current strain value and environment parameter data, constructing an ice state recognition model FCD-Net based on a multi-source feature fusion and a domain adaptive mechanism, intelligently classifying the ice state of the monitored structure, and outputting an ice state label signal; A power control module, which is connected with the data processing module and the acoustic detection and execution module, and is used for outputting a corresponding excitation power control signal according to the ice state label signal; A feedback judgment module, which is used for comparing the current current strain value and acoustic frequency response signal with preset reference values, outputting a termination control signal when a deicing criterion is met, and closing the acoustic detection and execution module to complete closed-loop control.

2. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 1, characterized in that, The acoustic detection and execution module comprises: A frequency detection unit, which comprises an interdigital transducer arranged on a piezoelectric substrate, receives a reflected signal and outputs an acoustic frequency response signal in the detection mode, and the acoustic frequency response signal is offset with the change of the icing state of the structure surface; A sound wave excitation unit, which drives the transducer to output a sound wave vibration signal in the ice melting mode in response to the excitation power control signal, and drives the interdigital transducer to generate a Rayleigh surface acoustic wave with a frequency of 32.5 MHz and a vibration amplitude of 5-10 nm on a 128°YX cut lithium niobate substrate; A transduction coupling unit, which comprises an adhesive layer between the piezoelectric substrate and the structure surface, and the adhesive layer is composed of polyimide or epoxy glue with a thickness of 10-50 mu m, and is used for mechanical coupling of the surface acoustic wave between the substrate and the structure.

3. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 1, wherein, The strain monitoring module comprises: A multi-point acquisition unit, which comprises at least four fiber Bragg grating sensors arranged uniformly along the circumferential direction of the cross section of the monitored structure, and the fiber Bragg grating sensors are arranged at different azimuth positions respectively, and are used for synchronously collecting axial strain values of corresponding positions of the structure and outputting a current strain value group; A demodulation unit, which is used for performing spectrum demodulation on each channel reflected wavelength in the current strain value group, and outputting a strain value vector corresponding to a plurality of arrangement positions; An asymmetric strain recognition unit, which is used for calculating a difference value between a maximum strain value and a minimum strain value of the strain value vector, and comparing the difference value with a preset strain difference threshold value, and outputting a strain difference value signal for indicating that there is a bias state in the circumferential direction of the structure when the difference value is greater than the preset strain difference threshold value.

4. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 1, wherein, The environment acquisition module comprises: The environmental parameter collecting unit comprises a temperature sensor, a relative humidity sensor and a wind speed sensor, which respectively collect the temperature , humidity and wind speed around the monitored structure and output corresponding environmental parameter signals; a parameter construction unit, configured to combine the environment parameter signals to form a standard three-dimensional environment state vector: and output the three-dimensional environment state input vector to the data processing module.

5. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 1, wherein, The data processing module comprises an ice state recognition model FCD-Net constructed based on a multi-source feature fusion and a domain adaptive mechanism, and the FCD-Net comprises: The feature encoding unit comprises a unified encoder structure for splicing the collected acoustic frequency response signal, structure strain time series, structure strain increment series and an environmental state parameter vector to form a multi-source feature input vector, and performs embedded coding to output a fusion feature vector. The state contrast modeling unit comprises a contrast learning module, configured to construct an inter-state embedding distance discrimination model using labeled icing state samples in a training phase, and generate a feature separation loss; The classification output unit comprises a convolutional neural network classifier, connected to the feature encoding unit at an input end, and configured to output an icing state label signal representing an icing state category; The domain feature alignment unit comprises a maximum mean discrepancy loss module, configured to calculate a distribution deviation based on source domain and target domain feature vectors in the training phase, and adjust the parameters of the front layer of the classifier through a domain alignment loss; The feature mask subunit is configured to mask invalid input features through a mask matrix when sensor data is missing or abnormal; The attention visualization unit is configured to generate a state recognition heat map based on a convolutional feature map.

6. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 5, wherein, The FCD-Net model is constructed using a combined loss function in the training phase, comprising: state classification loss term, contrastive loss function The expression is: ； wherein: and and denote the feature embedding vectors collected under the non-icing and icing conditions, respectively, denote the candidate sample embedding feature vectors for comparison, from the set of all icing or non-icing samples in the current training batch, indexed by all samples in the current training batch except denote the cosine similarity between features, is the temperature scaling coefficient; domain adaptation loss term, maximum mean discrepancy loss The formula is: ； wherein: indices and denote the numbering of the samples in the source and target domains, respectively, , denote the embedding feature vectors of the source and target domain samples, respectively, denotes the number of samples in the source domain, denotes the number of samples in the target domain, is a kernel function mapping; The combined training objective function includes a state recognition cross-entropy loss term , the overall objective function The expression is: ； wherein: , , are hyperparameter weight coefficients, adjusted according to different training tasks.

7. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 5, wherein, The data processing module further comprises an ice layer dynamic discrimination unit, comprising: A strain time window sampling unit is configured to receive the structural strain time series output by the strain monitoring module, and calculate the difference between adjacent time points within a preset sliding time window to form a structural strain increment series ; a variability calculation unit that calculates an information entropy value based on the probability density distribution estimation result in the sequence set , the expression is: ； wherein: is the probability density of the increment value falling into the th distribution interval, is the number of intervals; a stage recognition logic unit for recognizing a stage of ice accretion development based on the entropy value an output ice accretion development stage indication signal When then the fast icing phase is determined; When then the stable icing phase is determined. When then the natural thawing phase is determined.

8. The closed loop system of ice accretion detection and acoustic wave de-icing control according to claim 2, wherein, The power control module comprises: The pattern recognition unit receives the output icing state label signal, and is configured to determine whether the current icing state is in an initial icing stage, a stable icing stage or other states; The power setting unit is configured to set the parameter value of the excitation power control signal according to the icing stage: outputting a first set power when the excitation power control signal indicates an initial icing stage ; outputting the second set power when indicated as the stable icing phase ; The signal conversion unit is configured to convert the set power value into a driving excitation signal with corresponding amplitude and frequency parameters, and send it to the acoustic excitation unit.

9. The closed loop system of ice accretion detection and acoustic wave de-icing control of claim 1, wherein, The feedback determination module comprises: a data comparison unit configured to receive the current strain value outputted by the strain monitoring module and the acoustic frequency response signal outputted by the acoustic detection and execution module and compare them with the preset reference strain value and the reference frequency value respectively. The threshold determination unit is configured to output a de-icing completion feedback signal when the following two criteria are met simultaneously: ； The feedback instruction output unit sends a termination control signal to the power control module after determining that de-icing is completed, to turn off the acoustic vibration function of the acoustic detection and execution module, and complete the closed-loop control process.

10. A closed-loop method of ice-coating state recognition and acoustic wave de-icing control, applying the closed-loop system of ice-coating state recognition and acoustic wave de-icing control according to any one of claims 1-9, characterized in that, The method comprises the following steps: Step 1: set the acoustic detection and execution device to detection mode, and obtain the output acoustic frequency response signal; Step 2: collect the structural strain time series at multiple positions of the monitored structure, extract the current strain value, and collect temperature, humidity and wind speed information to generate environmental parameter data; Step 3: input the acoustic frequency response signal, current strain value and environmental parameter data into the icing state recognition model to identify whether icing exists and its state, and output the corresponding icing state label signal; Step 4: determine whether to enter the initial icing or stable icing stage according to the icing state label signal, and generate an excitation power control signal; Step 5: send the excitation power control signal to the acoustic detection and execution device, switch to the de-icing mode, and output the acoustic vibration signal to the structure surface through the transducer area in contact with the monitored structure; Step 6: obtain the current strain value and acoustic frequency response signal, and compare them with the preset reference value to determine whether the preset de-icing condition is met; Step 7: when it is determined that the ice layer has been removed, a termination control signal is generated to turn off the acoustic excitation output, and after completing the current de-icing control process, step 1 is executed again to realize continuous monitoring and closed-loop control of the icing state.

Images