A design method and experimental system of a magnetostrictive phononic crystal beam adaptive regulation device based on deep learning

Abstract

This invention discloses a design method and experimental system for an adaptive control device of a magnetostrictive phononic crystal beam based on deep learning, relating to the field of intelligent elastic wave metamaterials and adaptive active control technology. The method includes the following steps: S1: Generating a dataset through multiphysics finite element simulation; S2: Training a deep learning model using the dataset to obtain a forward prediction model and a reverse design model; S3: Calling the reverse design model according to the target elastic wave characteristics to calculate the optimal multi-field parameters, adjusting the physical field state of the corresponding device according to the optimal multi-field parameters, and simultaneously measuring the actual elastic wave response under the physical field state to verify the consistency between the actual elastic wave response and the target elastic wave characteristics, thus forming a closed-loop control. This method achieves dynamic, adaptive, and precise control of the bandgap position, width, and defect state frequency of the phononic crystal beam's elastic wave through an intelligent closed loop of "perception-decision-execution".

Description

Technical Field

[0001] This invention belongs to the field of intelligent elastic wave metamaterials and structure adaptive active control technology, specifically involving a design method and experimental system for an adaptive control device for magnetostrictive phononic crystal beams based on deep learning. Background Technology

[0002] In practical engineering and daily life, widespread mechanical vibrations propagate primarily in the form of elastic waves, which can cause structural fatigue and even induce resonance, posing risks to equipment safety and human health. Therefore, developing effective vibration control technologies is of significant practical importance.

[0003] Phononic crystals, as artificial periodic metamaterials, offer a novel solution to the aforementioned problems. Their most prominent feature is the presence of an elastic wave bandgap, meaning vibrations within a specific frequency range cannot propagate, thus enabling vibration isolation and filtering. Introducing structural defects (such as point defects) within the bandgap frequency range creates highly localized defect states, concentrating elastic wave energy near the defects and providing a physical basis for vibration control and energy harvesting. However, once traditional phononic crystals are fabricated, the frequency and location of their bandgap and defect states are fixed, making them unsuitable for adapting to changing external environments and diverse functional requirements, significantly limiting their application in scenarios requiring dynamic responses. While existing technologies (such as Chinese patent CN107968599A) utilize the localization of defect states for energy harvesting, their non-adjustable structural parameters limit energy capture to a fixed single frequency and location, restricting both application flexibility and energy harvesting efficiency.

[0004] Chinese patent CN119171775A discloses a vibration energy harvesting device based on magneto-electro-elasticity. Its core principle is to adjust the distance between magnet blocks using a stepper motor, thereby changing the magnetic field strength and controlling the band gap and defect states of a phononic crystal beam to harvest vibration energy. The device comprises a unit cell consisting of a magnetostrictive column, a piezoelectric sheet, and an aluminum strip. However, the control method is manual or simple. This harvesting device is difficult to implement with rapid dynamic response and reverse engineering in complex multi-field coupling environments. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention aims to provide a design method and experimental system for an adaptive control device for magnetostrictive phononic crystal beams based on deep learning, so as to solve the technical problem that existing traditional phononic crystals and their control methods are difficult to achieve dynamic, adaptive and reverse design of elastic wave characteristics (such as band gap, defect state) in complex environments.

[0006] To achieve the above objectives, the present invention employs the following technical solution: This invention provides a design method for an adaptive control device for a magnetostrictive phonon crystal beam based on deep learning, comprising the following steps: S1: Generate a dataset through multiphysics finite element simulation. The dataset includes multiple field parameters and elastic wave responses corresponding to the multiple field parameters. The multiple field parameters include at least magnetic field strength, axial stress, and temperature. S2: The dataset is used to train a deep learning model to obtain a forward prediction model and a reverse design model; the forward prediction model realizes the mapping from multi-field parameters to elastic wave characteristics, and the reverse design model realizes the reverse solution from the target elastic wave characteristics to the optimal multi-field parameters; S3: Based on the target elastic wave characteristics, the reverse design model is invoked to calculate the optimal multi-field parameters. The physical field state of the deep learning-based magnetostrictive phononic crystal beam adaptive control device is adjusted according to the optimal multi-field parameters. At the same time, the actual elastic wave response under the physical field state is measured to verify the consistency between the actual elastic wave response and the target elastic wave characteristics. Based on the deviation, optimization adjustment is performed to form a closed-loop control.

[0007] Preferably, in step S1, the multiphysics finite element simulation includes coupled simulation of magnetic field, axial stress and temperature field, used to simulate the elastic wave response of magnetostrictive materials under the action of magnetic field, axial stress and temperature field.

[0008] Preferably, in step S2, the deep learning model is trained under supervision using the dataset to establish a nonlinear mapping relationship between multiple field parameters and elastic wave response.

[0009] Preferably, in step S2, the deep learning model is a deep neural network.

[0010] Preferably, in step S3, the optimal multi-field parameters include any one or more of magnetic field strength, axial stress, and temperature.

[0011] Preferably, in step S3, the method for adjusting the physical field state of the deep learning-based magnetostrictive phononic crystal beam adaptive control device includes any one or more of the following: driving the magnetic field control component to adjust the magnetic field strength, driving the mechanical loading component to adjust the axial stress, and driving the temperature control component to adjust the temperature.

[0012] Preferably, in step S3, the actual elastic wave response is obtained by measuring the transmission spectrum or displacement cloud map of the deep learning-based magnetostrictive phononic crystal beam adaptive control device using a laser vibrometer or accelerometer.

[0013] Preferably, in step S3, the invocation of the reverse design model and optimization adjustment are executed through the intelligent control unit. The intelligent control unit integrates a graphical user interface and a deep learning model to set the target elastic wave characteristics and display the control results. The intelligent control unit performs real-time analysis of the measurement results based on the deep learning model, automatically calculates the optimal control parameters corresponding to the target elastic wave characteristics, and instructs the magnetic field control component to adjust the local magnetic field strength of the unit cell, so that the supercell can achieve adaptive closed-loop control of the elastic wave bandgap and defect state frequency under any one or more coupling conditions of magnetic field, stress and temperature.

[0014] Preferably, in step S1, the elastic wave response includes any one or more of band structure, band gap, and defect state frequency.

[0015] This invention also provides an experimental system for implementing the above-mentioned design method of a deep learning-based adaptive control device for a magnetostrictive phonon crystal beam. The system includes a deep learning-based adaptive control device for a magnetostrictive phonon crystal beam, comprising a phonon crystal beam containing a plurality of sequentially arranged unit cells in a periodic distribution. Magnetic field control components are symmetrically arranged on both sides of the phonon crystal beam, and an excitation and measurement component is arranged on one side of the phonon crystal beam. An intelligent control unit is arranged on the phonon crystal beam, integrating a graphical user interface and a deep learning model. The phonon crystal beam, magnetic field control component, excitation and measurement component, intelligent control unit, graphical user interface, and power supply circuit closed loop are described.

[0016] Preferably, the phononic crystal beam comprises no fewer than 10 periodically distributed unit cells, each unit cell consisting of a magnetostrictive material portion and an elastic substrate portion, wherein the magnetostrictive material portion is perpendicularly attached to the elastic substrate portion.

[0017] More preferably, the magnetostrictive material is a magnetostrictive column or patch, and the elastic substrate portion is an aluminum strip.

[0018] More preferably, the magnetostrictive column is arranged in a direction perpendicular to the aluminum strip (i.e., the thickness direction) so that the applied magnetic field can act along the easy magnetization axis of the aluminum strip, thereby maximizing the effect of the physical field and achieving efficient control of the unit cell.

[0019] The phononic crystal beam is fixed above the optical platform. Magnetic field control components are symmetrically arranged on both sides of the phononic crystal beam. Each magnetic field control component contains several sets of magnetic field control units, each corresponding to a unit cell, used to independently control the magnetic field strength of the unit cell. Each magnetic field control component includes a connecting plate, and each magnetic field control unit includes a permanent magnet block and a stepper motor. Several through holes are opened on the connecting plate, and lead screws are inserted through these holes. The permanent magnet block is sleeved on the lead screw closer to the square plate, and the stepper motor is sleeved on the lead screw farther from the square plate. By rotating the stepper motor, the lead screw is driven, thereby driving the permanent magnet block to move vertically, changing the distance between the permanent magnet block and the corresponding unit cell in a non-contact manner, achieving continuous and precise adjustment of the magnetic field strength of the unit cell.

[0020] The intelligent control unit is connected to the PLC via a communication interface for writing programs and issuing instructions; the PLC is connected to the driver via electrical signals, and the driver is connected to the magnetic field control component. The PLC controls the magnetic field control component to execute corresponding instructions through the driver to adjust the magnetic field strength; the driver drives a stepper motor to move the permanent magnet block, thereby achieving precise adjustment of the magnetic field strength.

[0021] More preferably, the driver is a stepper motor driver.

[0022] The excitation and measurement components include an exciter and a data acquisition device. The exciter is disposed on one side of the phononic crystal beam and is used to apply elastic wave excitation to a supercell composed of several unit cells. The data acquisition device is connected to the intelligent control unit through a sensing interface and is used to measure the actual elastic wave response and feed the measurement data back to the intelligent control unit for verification and optimization, forming a closed-loop feedback. More preferably, the data acquisition device is an oscilloscope; the sensing interface is a USB or a data acquisition card. More preferably, the optical platform serves as the base of the entire device, supporting and fixing other components; the phonon crystal beam is connected to the top of the optical platform via a fixing component; the magnetic field control components are symmetrically arranged above and below the phonon crystal beam and connected to the optical platform via supporting components. More preferably, the number of unit cells is 15, wherein at least one unit cell can be designated as a defect unit for independent magnetic field control. More preferably, 15 groups of stepper motors and permanent magnet blocks are symmetrically arranged. Further preferably, the phonon crystal beam is in a suspended state, and the distance between the phonon crystal beam and the permanent magnet block is not less than 1 cm to avoid contact and ensure the uniformity and adjustability of the magnetic field distribution. More preferably, the initial distance between the phonon crystal beam and the permanent magnet block is approximately 10 cm. Further preferably, the connecting plate and the components for fixing the phonon crystal beam are made of non-magnetic materials (such as acrylic) to avoid interference with the controlled magnetic field. Furthermore, the fixing component includes a fixing clamp, a vertical rod, and a horizontal rod, used to adjust and fix the spatial position of the phonon crystal beam.

[0023] A further preferred embodiment of the above design method includes the following steps: 1) Forward data-driven model construction: Based on multiphysics finite element simulation, a numerical model of a magnetostrictive phononic crystal beam under magnetic-force-thermal coupling is established. Parametric scanning is used to generate models covering different magnetic fields. or ), axial stress ( ) and temperature ( A large-scale dataset combining multiple field parameters and corresponding elastic wave responses, wherein the input is the multiple field parameters and the output is the elastic wave response, which includes band structure, band gap, and defect state frequencies. 1) Any one or more of the following: 2) Deep learning model training and deployment: Using the dataset from step 1), train a deep neural network to obtain a forward prediction model and a reverse design model. The forward prediction model achieves fast and high-precision mapping from multi-field parameters to elastic wave characteristics. The training of the reverse design model achieves mapping from target elastic wave characteristics (such as target...) ) Reverse solve the optimal multi-field parameter combination; deploy the trained forward prediction model and reverse design model in the software platform of the intelligent control unit; 3) Adaptive closed-loop control execution: First, the target input: set the specified defect state frequency through the graphical user interface; Secondly, intelligent decision-making: using a software platform to call the reverse design model, automatically calculating the required optimal multi-field parameters (such as the magnetic field of the defect element). Background magnetic field ,stress ,temperature ); Then comes precise execution: the intelligent control unit converts the optimal parameters into control instructions, drives the stepper motor through the programmable logic controller (PLC) and driver, and adjusts the position of the corresponding permanent magnet block to adjust the deep learning-based magnetostrictive phononic crystal beam adaptive control device to the physical field state corresponding to the optimal parameters; Finally, there is the verification feedback: the actual transmission spectrum of the physical field state is obtained through the excitation and measurement unit, the defect state frequency is verified to be consistent with the transmission spectrum, and optimization adjustment is made based on the deviation to achieve closed-loop control.

[0024] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a design method for an adaptive control device for a magnetostrictive phonon crystal beam based on deep learning. It constructs a complete intelligent control loop by integrating three core steps: generating a dataset through multi-physics finite element simulation, training a deep learning model to obtain a reverse design model, and obtaining optimal multi-field parameters through the reverse design model to achieve closed-loop control. This enables the system to automatically, quickly, and accurately solve for and apply the required multi-physics combination based on arbitrarily set target elastic wave characteristics, fundamentally overcoming the limitations of traditional phonon crystals with fixed performance and control relying on manual trial and error.

[0025] To address the highly nonlinear problem caused by strong magnetic-mechanical-thermal coupling, this invention proposes a dataset-driven reverse design method, which avoids the problems of high computational cost and difficulty in convergence of traditional physical model-based optimization methods, and provides an efficient solution for the reverse design of complex multi-field coupled systems.

[0026] Furthermore, by specifically defining the multiphysics finite element simulation as a coupled simulation of magnetic field, axial stress, and temperature field, it is ensured that the generated dataset can more accurately simulate the behavior of magnetostrictive materials in real complex environments, providing a high-fidelity data foundation for subsequent deep learning model training, thereby enhancing the reliability of the entire system's prediction and control.

[0027] Furthermore, it was clarified that deep learning models establish nonlinear mapping relationships through supervised training, which enhances the model's ability to understand and learn the complex causal relationships between multiple field parameters such as magnetic field, axial stress, and temperature field and elastic wave response, thus ensuring the accuracy of intelligent adaptive dynamic control.

[0028] Furthermore, by specifically defining deep learning models as deep neural networks, we can leverage the powerful feature extraction and function approximation capabilities of deep neural networks. This helps to handle higher-dimensional and more complex nonlinear problems, thereby obtaining more accurate forward prediction and reverse design results.

[0029] Furthermore, it was clarified that the optimal multi-field parameters can include any one or more of magnetic field strength, axial stress, and temperature, giving the system great control flexibility. Users can select a single field or a combination of fields for control as needed, achieving dynamic adjustment of elastic wave characteristics over a wide range and for multiple objectives, significantly improving the system's practicality.

[0030] Furthermore, by defining specific means of controlling the physical field state, the optimal parameters calculated by the intelligent algorithm are transformed into actual physical operations, ensuring that the theoretical design can be accurately realized on the physical entity.

[0031] Furthermore, the requirement to obtain the actual elastic wave response through high-precision measurement equipment such as laser vibrometers or accelerometers is beneficial for providing reliable data feedback for verification and optimization, ensuring the final effect of the system's adaptive control.

[0032] Furthermore, by integrating a graphical interface and a deep learning model into the intelligent control unit, the system achieves a unified approach to user intent input, intelligent algorithm decision-making, and result visualization.

[0033] Furthermore, by specifically defining the elastic wave response as the core characteristics of phononic crystals such as band structure, band gap, and defect state frequency, the goal of the entire adaptive control process becomes clearer and more measurable.

[0034] The present invention also provides an experimental system for implementing the above design method, realizing a complete innovation from method to entity, making the system, including algorithm, actuator, sensor and control unit, into an organic whole, and finally constructing a complete technical closed loop of software and hardware collaboration, providing an innovative implementation scheme for the next generation of intelligent acoustic devices. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the structure of the present invention; wherein: 1-threaded rod; 2-upright rod; 3-lead screw; 4-stepper motor; 5-first connecting plate; 6-permanent magnet block; 7-square plate; 8-crossbar; 9-vibrator; 10-second connecting plate; 11-optical platform; 12-magnetostrictive column; 13-knob; 14-fixing clamp; 15-aluminum strip; Figure 2 This is a schematic diagram of the first connecting plate and stepper motor structure of the present invention; Figure 3 This is a schematic diagram of the magnetostrictive column and aluminum strip structure of the present invention; Figure 4 This is a diagram of the square plate and its fixing device according to the present invention; Figure 5 This is a partial structural diagram of the fixing clip of the present invention; Figure 6This invention presents a graph showing the variation of defect frequency under different external magnetic fields and temperatures, under the same defect magnetic field and stress conditions. Figure 7a This invention provides a GUI interface for the positive prediction of single-cell and supercell band structure diagrams, transmission loss diagrams, and magnetic flux density variation diagrams. Figure 7b The GUI interface of this invention is designed in reverse engineering to visualize the external field parameters of the single cell and supercell. Figure 7c This is a schematic diagram of the GUI interface of the present invention, which realizes self-filling data and self-starting stepper motor through automated control; Figure 8a The images show the predicted band structure, transmission loss diagram, and experimental transmission loss diagram obtained by reverse engineering the external field parameters under the condition of 10250Hz input in the GUI according to the present invention. Figure 8b The images show the predicted band structure, transmission loss diagram, and experimental transmission loss diagram obtained by reverse engineering the external field parameters under the condition of 10500Hz input in the GUI according to the present invention. Figure 9 This is a schematic diagram of the connection of the adaptive control device of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 like Figure 1-5As shown, this invention proposes a design method and experimental system for an adaptive magnetostrictive phononic crystal beam based on deep learning, comprising: an optical platform 11, on which two uprights 2 and four threaded rods 1 are threadedly mounted, the uprights 2 and the threaded rods 1 are arranged parallel to each other, and a cross-shaped set screw clamp 14 is slidably mounted on the outside of each upright 2, the cross-shaped set screw clamp 14 is parallel to an acrylic square plate 7, and hexagonal knobs 13 are respectively installed at both ends of the two cross-shaped set screw clamps 14. The uprights 2 and the clamps 14 are vertically distributed, and a cross bar 8 is connected inside each cross-shaped set screw clamp 14. The two cross bars 8 are respectively perpendicularly inserted through both ends of the acrylic square plate 7 to fix the acrylic square plate 7, so that the acrylic square plate 7 is in a suspended state; an excitation and measurement assembly is connected to one side of the acrylic square plate 7, the excitation and measurement assembly including an exciter 9 and an oscilloscope, the exciter 9 being connected to a power supply; The acrylic square plate 7 is symmetrically provided with a first magnetic field control component and a second magnetic field control component on its upper and lower sides. The first magnetic field control component includes a first connecting plate 5 and a first magnetic field control unit. The first magnetic field control unit includes a lead screw 3, a stepper motor 4, and a permanent magnet block 6. The first connecting plate 5 has several through holes, and a lead screw 3 passes through each through hole. The permanent magnet block 6 is installed at the end of the lead screw 3 near the square plate 7, and the stepper motor 4 is installed at the end of the lead screw 3 away from the square plate 7. There are 15 first magnetic field control units and 15 second magnetic field control units. The stepper motor 4 can adjust the vertical movement distance of the permanent magnet block 6. The second magnetic field control assembly includes a second connecting plate 10 and a second magnetic field control unit. The second magnetic field control unit includes a lead screw 3, a stepper motor 4, and a permanent magnet block 6. The second connecting plate 10 has several through holes, and a lead screw 3 passes through each through hole. The permanent magnet block 6 is installed at the end of the lead screw 3 near the square plate 7, and the stepper motor 4 is installed at the end of the lead screw 3 away from the square plate 7. The first connecting plate 5 and the second connecting plate 10 are both made of acrylic material. Two threaded rods 1 are connected to the two ends of the first connecting plate 5 and the second connecting plate 10 respectively for fixing the first connecting plate 5 and the second connecting plate 10.

[0039] according to Figure 3As shown, a phononic crystal beam is placed inside the acrylic square plate 7, and an intelligent control unit is set on the phononic crystal beam. The intelligent control unit integrates a graphical user interface (GUI) and a deep learning model. The phononic crystal beam consists of 15 unit cells. Each unit cell consists of two magnetostrictive pillars 12 and an aluminum strip 15. The two magnetostrictive pillars 12 are symmetrically and vertically arranged on the upper and lower surfaces of the aluminum strip 15. By applying the magnetic field strength of the permanent magnet block 6, the defect of the corresponding unit cell of the permanent magnet block 6 can be introduced, thereby obtaining the band gap and defect state frequency at each unit cell. When the magnitude of the magnetic field strength corresponding to each unit cell is adjusted, it will also affect the change of the band gap and defect state frequency of the corresponding unit cell. When the frequency of the corresponding defect state is applied to the exciter 9, the energy will be locally concentrated at the defect position of the beam, and a clear defect state transmission peak will appear in the transmission loss diagram. The frequency of the transmission peak is different under different external field conditions. Therefore, the GUI is designed through deep learning to achieve adaptive control, thereby realizing the prediction and realization of defect states under different external field conditions and frequencies. The phonon crystal beam, magnetic field control component, excitation and measurement component, intelligent control unit, graphical user interface, and power supply circuit closed loop are described.

[0040] Furthermore, the experimental system for the design method of the aforementioned deep learning-based magnetostrictive phononic crystal beam adaptive control device includes the following steps: 1) Construction of Multiphysics Forward Dataset A multiphysics finite element numerical model of an adaptive control device for a magnetostrictive phononic crystal beam based on deep learning under magnetic-force-thermal coupling is established. The phononic crystal beam is composed of several periodic unit cells, each consisting of a magnetostrictive material part and an elastic substrate part. The middle unit cell, i.e. the eighth unit cell, is designated as the defect element.

[0041] In the numerical model, multiple field parameters such as applied magnetic field strength, axial stress, and temperature are parametrically scanned to generate a large-scale training dataset. The dataset takes the combination of multiple field parameters as input and the elastic wave response as output. The elastic wave response includes at least band structure, band gap range, and defect state frequency characterization quantities.

[0042] 2) Deep learning model training and deployment (forward prediction and reverse design) Using the dataset obtained in step 1), a deep neural network is trained to obtain a model that includes at least the following functions: a forward prediction model, used to achieve fast and high-precision mapping from multi-field parameters (magnetic field, stress, temperature) to elastic wave characteristics (energy band, band gap, defect state frequency or transmission spectrum characteristics); and a reverse design model, used to solve in reverse the target elastic wave characteristics (e.g., target defect state frequency, target band gap location or target transmission spectrum characteristics) to obtain the optimal combination of multi-field parameters that satisfies the target.

[0043] The trained forward prediction model and reverse design model are deployed in the software platform of the intelligent control unit, enabling it to have the ability to input targets, calculate parameters, and generate control commands.

[0044] 3) Adaptive closed-loop control execution First, target input is performed: target control indicators are set through a graphical user interface, and the target control indicators include at least the target defect state frequency or the target transmission spectrum characteristics.

[0045] Secondly, intelligent decision-making is performed: the intelligent control unit calls the reverse design model and automatically outputs the required optimal multi-field parameter configuration. The parameter configuration includes at least the defect element magnetic field and the background element magnetic field, and may further include axial stress and temperature parameters.

[0046] Then, precise execution is performed: the intelligent control unit converts the optimal parameter configuration into control commands, which drive the magnetic field control components to operate through the programmable logic controller and driver. This allows each magnetic field control unit to adjust the spacing between the permanent magnet block and the corresponding unit cell, thereby achieving independent, continuous, and precise adjustment of the local magnetic field strength of each unit cell. Simultaneously, the dynamic loading component and temperature control component can regulate the corresponding stress and temperature fields, enabling the magnetostrictive phononic crystal beam to reach the target multi-field state. The stress field is the residual stress generated during the preparation of the magnetostrictive material. The magnetostrictive material selected in this invention has good performance, with zero residual or pre-stress. The temperature field control can be achieved by controlling the aluminum heating plate with a digitally controlled DC power supply. Finally, verification feedback is performed: controllable excitation is applied to the magnetostrictive phononic crystal beam through the excitation and measurement unit, and vibration response or transmission spectrum is collected. The measured results are compared with the target index. If the target is not met, the deviation is used as feedback input to the intelligent control unit to update the parameters and perform regulation again until the target index is met, thus realizing closed-loop adaptive optimization.

[0047] Example 2 like Figure 6 As shown, to verify the synergistic regulation of the defect state frequency by the external magnetic field and temperature under magnetic-force-thermal coupling conditions, under the conditions of a defect magnetic field of 700 Oe and an axial stress of 0 MPa, the frequency at the defect increases with the increase of the external magnetic field strength. At the same time, under the same defect magnetic field and stress conditions, the defect frequency at an external temperature of 45℃ is generally higher than that at 25℃, indicating that the temperature field and the external magnetic field can work together to achieve coupled regulation of the defect state frequency, thus providing a basis for establishing a data-driven mapping relationship of "multi-field parameters - elastic wave characteristics" in subsequent deep learning models.

[0048] GUI-driven forward prediction-reverse design-automatic execution closed-loop implementation: To achieve the integrated implementation of the design method and experimental system of the adaptive magnetostrictive phononic crystal beam based on deep learning in the software interface of this invention, and to complete the closed-loop linkage from "rapid prediction and evaluation" to "parameter inverse solution" and then to "automatic execution", as shown in Figures 7(a) to 7(c), Figure 7(a) is a schematic diagram of the GUI forward prediction result. Under the given multi-field parameter input conditions, the system quickly predicts and visualizes the elastic wave response of the unit cell and supercell. The displayed content includes at least the band diagram, transmission loss diagram and magnetic flux density change diagram, which are used to quickly evaluate the band gap range and defect state frequency change trend; Figure 7(b) is a schematic diagram of the GUI reverse design. The user inputs the target elastic wave characteristics. After the indicators are determined, the system calls the deep learning reverse design model to automatically solve for the combination of external field parameters that meet the target. The combination of external field parameters includes at least the external magnetic field and defect magnetic field parameters, and may further include stress and temperature parameters, which serve as the control targets for the actuator adjustment. Figure 7(c) is a schematic diagram of GUI automated control. The automated control process includes: 1) inputting parameters; 2) predicting the external magnetic field and defect magnetic field; 3) clicking the prediction button to trigger the calculation; 4) the system realizes parameter self-filling and automatically starts the device to drive the stepper motor to perform adjustment, so that the magnetic field control component automatically adjusts the position of the permanent magnet corresponding to each unit cell, thereby completing the closed-loop linkage of "target input - model calculation - parameter distribution - automatic execution", providing a foundation for subsequent adaptive control combined with measurement feedback.

[0049] Verification of Adaptive Control of Target Frequency: To verify that the present invention can achieve rapid inverse design of external field parameters through deep learning under different target defect state frequencies, and to realize closed-loop control of the corresponding transmission spectrum characteristics and defect state frequencies in the experiment, as shown in Figures 8(a) and 8(b), the target frequencies of 10250Hz and 10500Hz were input into the GUI, respectively. The deep learning inverse design model output the corresponding combination of external field parameters and generated the corresponding predicted band diagram and predicted transmission loss diagram. After the external field parameters were sent to the control system to drive the magnetic field control component to complete the adjustment, the experimental transmission loss diagram was obtained through the excitation and measurement unit and compared with the prediction results. The results show that the prediction and experiment have good consistency under different target frequency conditions, thus proving that the closed-loop adaptive control strategy of the present invention of "target input - model inverse solution - automatic execution - measurement verification" can stably realize the control of the target defect state frequency and related transmission spectrum characteristics.

[0050] like Figure 9As shown, to illustrate the hardware connection relationship and the "perception-decision-execution" closed-loop configuration of the adaptive control system of the present invention, the device includes a power supply, a GUI, an intelligent control unit, an exciter, and an oscilloscope. The GUI is communicatively connected to the intelligent control unit to realize target input, model calculation, and parameter distribution. The intelligent control unit drives the magnetic field control component to control the stepper motor to achieve precise adjustment of the magnetic field per unit cell. The exciter is used to apply controllable excitation to the phononic crystal beam. The oscilloscope is used to display and record the measurement signal and use the measurement result for feedback verification, thereby constituting an adaptive closed-loop control system based on measurement feedback.

[0051] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A design method for an adaptive control device for a magnetostrictive phonon crystal beam based on deep learning, characterized in that, Includes the following steps: S1: Generate a dataset through multiphysics finite element simulation. The dataset includes multiple field parameters and elastic wave responses corresponding to the multiple field parameters. The multiple field parameters include at least magnetic field strength, axial stress, and temperature. S2: Use the dataset to train a deep learning model to obtain a forward prediction model and a reverse design model; S3: Based on the target elastic wave characteristics, the reverse design model is invoked to calculate the optimal multi-field parameters. The physical field state of the deep learning-based magnetostrictive phononic crystal beam adaptive control device is adjusted according to the optimal multi-field parameters. At the same time, the actual elastic wave response under the physical field state is measured to verify the consistency between the actual elastic wave response and the target elastic wave characteristics. Based on the deviation, optimization adjustment is performed to form a closed-loop control.

2. The design method of the deep learning-based adaptive control device for magnetostrictive phonon crystal beams according to claim 1, characterized in that, In step S1, the multiphysics finite element simulation includes the coupled simulation of magnetic field, axial stress and temperature field, which is used to simulate the elastic wave response of magnetostrictive materials under the action of magnetic field, axial stress and temperature field.

3. The design method of the deep learning-based adaptive control device for magnetostrictive phonon crystal beams according to claim 1, characterized in that, In step S2, the deep learning model is trained under supervision using the dataset to establish a nonlinear mapping relationship between multiple field parameters and elastic wave response.

4. The design method of the magnetostrictive phonon crystal beam adaptive control device based on deep learning according to claim 1, characterized in that, In step S2, the deep learning model is a deep neural network.

5. The design method of the magnetostrictive phonon crystal beam adaptive control device based on deep learning according to claim 1, characterized in that, In step S3, the optimal multi-field parameters include any one or more of magnetic field strength, axial stress, and temperature.

6. The design method of the deep learning-based adaptive control device for magnetostrictive phonon crystal beams according to claim 5, characterized in that, In step S3, the method for adjusting the physical field state of the deep learning-based magnetostrictive phononic crystal beam adaptive control device includes any one or more of the following: driving the magnetic field control component to adjust the magnetic field strength, driving the mechanical loading component to adjust the axial stress, and driving the temperature control component to adjust the temperature.

7. The design method of the magnetostrictive phonon crystal beam adaptive control device based on deep learning according to claim 1, characterized in that, In step S3, the actual elastic wave response is obtained by measuring the transmission spectrum or displacement cloud map of the deep learning-based magnetostrictive phononic crystal beam adaptive control device using a laser vibrometer or accelerometer.

8. The design method of the magnetostrictive phonon crystal beam adaptive control device based on deep learning according to claim 1, characterized in that, In step S3, the invocation of the reverse design model and the optimization adjustment are executed through the intelligent control unit, which integrates a graphical user interface and a deep learning model to set the target elastic wave characteristics and display the control results.

9. The design method of the deep learning-based adaptive control device for magnetostrictive phonon crystal beams according to claim 1, characterized in that, In step S1, the elastic wave response includes any one or more of the band structure, band gap, and defect state frequency.

10. An experimental system for implementing the design method of the deep learning-based adaptive control device for magnetostrictive phonon crystal beams according to any one of claims 1-9, characterized in that, The invention includes a deep learning-based adaptive control device for a magnetostrictive phonon crystal beam. The device comprises a phonon crystal beam containing a plurality of sequentially arranged unit cells in a periodic distribution. Magnetic field control components are symmetrically arranged on both sides of the phonon crystal beam, and an excitation and measurement component is arranged on one side of the phonon crystal beam. An intelligent control unit is mounted on the phonon crystal beam, integrating a graphical user interface and a deep learning model. The phonon crystal beam, magnetic field control component, excitation and measurement component, intelligent control unit, graphical user interface, and power supply circuit closed loop are described.

Images