Bionic fish with adjustable rigidity and control method

Abstract

The application discloses a bionic fish with adjustable rigidity and a control method, which comprises a bionic fish body, a silica gel fish tail and a tail fin assembly, a driving series module is arranged in the silica gel fish tail; the driving series module comprises a first elliptical ring, a second elliptical ring and a third elliptical ring, adjustable rigidity structures are fixed between the first elliptical ring and the second elliptical ring and between the second elliptical ring and the third elliptical ring; the adjustable rigidity structure comprises a first supporting plate, a second supporting plate, two SMA springs and two wave connecting plates; an IMU sensor and a depth camera are arranged at the tail end of the silica gel fish tail; the SMA springs are electrically connected with a master control unit and a single-channel driving circuit. The bionic fish with adjustable rigidity and the control method can change the bending rigidity and the swing characteristics according to different swimming frequencies, water flow environments or propulsion requirements, so that the adaptability of the bionic fish tail in a complex underwater environment is improved.

Description

Technical Field

[0001] This invention relates to the field of biomimetic fish technology, specifically to a biomimetic fish with adjustable stiffness and a control method thereof. Background Technology

[0002] Bionic fish are an important research direction in the field of underwater bionic robots. By simulating the periodic swaying of a fish's tail to generate propulsion, they can achieve efficient, low-noise, and flexible underwater movement. Compared with traditional propeller propulsion, bionic fish tail propulsion has significant advantages in terms of less fluid disturbance, better stealth, and stronger maneuverability. Therefore, it has broad application prospects in underwater exploration, environmental monitoring, and operations in complex waters.

[0003] Current biomimetic fish actuation methods are mainly divided into traditional actuators and smart material actuators, with the former accounting for approximately 72% and the latter approximately 28%. Among traditional actuators, servo motors / electric motors are the most widely used, but they typically require transmission mechanisms, sealing structures, and rigid connectors, resulting in a large overall size, complex structure, high sealing difficulty, and mechanical noise during operation. In smart material actuators, piezoelectric ceramic actuators offer the advantage of fast response speed, but their deformation range is limited, and they have high requirements for drive voltage and structural design. These actuation methods generally suffer from problems such as large size, difficult sealing, high noise, high power consumption, fixed structural stiffness, and insufficient environmental adaptability, making it difficult to simultaneously meet the requirements of high response, low power consumption, and variable stiffness control.

[0004] SMA (Synthetic Motion Material) is a smart material capable of reversible deformation or contraction under thermal excitation. Its shape memory effect includes single-pass, two-pass, and full-pass shape memory alloys, featuring compact structure, frictionless actuation, and ease of integration. This helps reduce size, weight, and noise, and enhances complex motion capabilities. By electrically heating SMA, it can induce contraction deformation, thereby driving flexible structures to achieve bending or oscillating motion. On the other hand, metamaterial structures can achieve programmable stiffness, adjustable mechanical properties, and lightweight characteristics through artificial microstructure design. This can overcome the limitations of traditional homogeneous materials in terms of stiffness, deformation modes, and mechanical response, providing controllable dynamic mechanical properties for biomimetic fishtail structures. However, in existing biomimetic fishtail structures, the integration between SMA actuation and adjustable stiffness structures remains insufficient. Most solutions only achieve a single actuation or a single flexible oscillation function, failing to fully utilize the periodic oscillation capability brought by alternating heating on one side of the SMA, and failing to effectively introduce the adjustable stiffness characteristics of metamaterial structures into the deformation control of biomimetic fishtails. Therefore, existing technologies still suffer from problems such as unadjustable tail stiffness, poor swing stability, insufficient precision in motion frequency control, limited propulsion efficiency, and weak adaptability to different underwater environments, making it difficult to meet the needs of micro-bionic fish for efficient, stable, and low-noise propulsion in complex underwater environments. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, this invention provides a biomimetic fish with adjustable stiffness and a control method therein, solving problems such as non-adjustable tail stiffness, poor swing stability, insufficient precision in motion frequency control, limited propulsion efficiency, and weak adaptability to different underwater environments in existing technologies. This invention aims to meet the needs of micro-bionic fish for efficient, stable, and low-noise propulsion in complex underwater environments.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A biomimetic fish with adjustable stiffness and its control method are disclosed. The biomimetic fish body, a silicone fish tail, and a tail fin assembly are included. The biomimetic fish body and the tail fin assembly are fixedly connected via the silicone fish tail. A drive series module is disposed within the silicone fish tail, which drives the tail fin assembly to swing left and right. The drive series module includes a first elliptical ring, a second elliptical ring, and a third elliptical ring. Adjustable stiffness structures are fixed between the first and second elliptical rings and between the second and third elliptical rings. The adjustable stiffness structure includes a first support plate, a second support plate, and two SMA springs, as well as connecting... The first and second support plates have two corrugated connecting plates, which are staggered in an X-shape. The left side of the first support plate is connected to the left side of the second support plate, and the right side of the first support plate is connected to the right side of the second support plate by SMA springs. An IMU sensor and a depth camera are installed at the end of the silicone fish tail. The IMU sensor is used to collect acceleration and angular velocity data during the movement of the silicone fish tail, and the depth camera is used to collect spatial position information of the end of the silicone fish tail or the marker point. The SMA springs are electrically connected to the main control unit and the single-channel drive circuit.

[0007] Furthermore, the bionic fish body includes an upper shell, a lower shell, a dorsal fin, an pelvic fin, and two pectoral fins. The upper shell and the lower shell are sealed and fixed to form the fish body. The rear side of the upper shell and the rear side of the lower shell are sealed and connected to the silicone fish tail. A rubber sheet is fitted at the connection between the upper shell and the lower shell, and waterproof tape is installed at the external connection between the upper shell and the lower shell.

[0008] Furthermore, the tail fin assembly includes a tail fin and a support plate fixedly connected to the tail fin. The tail fin is fixed to the silicone fish tail by the support plate, and the silicone fish tail and the support plate are integrally formed.

[0009] Furthermore, both the first support plate and the second support plate are provided with mounting holes for installing SMA springs. Bolt assemblies are provided in the mounting holes. The SMA springs are clamped and fixed on the first support plate and the second support plate by the bolt assemblies. The bolt assemblies include studs, spring washers, nuts and bolts.

[0010] Furthermore, the first elliptical ring is elliptical in shape and has two symmetrical inward mounting protrusions. The mounting protrusions are provided with limiting grooves for mounting the first support plate or the second support plate.

[0011] Furthermore, the second elliptical ring is shaped like a frustum of an ellipse, the outer surface of the second elliptical ring is shaped like a frustum of a cone, and a second rectangular mounting hole for mounting the first support plate or the second support plate is provided in the middle of the second elliptical ring.

[0012] Furthermore, the third elliptical ring is shaped like a frustum of an ellipse, and the outer surface of the third elliptical ring is shaped like a frustum of a cone. The large end and the small side of the third elliptical ring are provided with insertion slots, and the large end of the third elliptical ring is provided with a third rectangular mounting hole that cooperates with the first support plate or the second support plate.

[0013] A method for controlling a biomimetic fish with adjustable stiffness includes the following steps: S1: Set the swing amplitude, swing frequency, initial phase, and initial PWM duty cycle of the silicone fish tail SMA springs on the left and right sides; S2: The main control unit calculates the target swing angle of the silicone fish tail at each sampling moment based on the swing amplitude, swing frequency and initial phase of the silicone fish tail, combined with the swing angle model, and obtains the swing trajectory of the silicone fish tail. The main control unit controls the silicone fish tail to swing left and right periodically according to the preset swing amplitude and swing frequency based on the swing trajectory of the silicone fish tail. S3: During the swinging of the silicone fish tail, IMU inertial measurement data and depth camera visual data are collected by IMU sensor and depth camera respectively; IMU inertial measurement data includes acceleration and angular velocity data during the movement of silicone fish tail; depth camera visual data includes spatial position information of the end of silicone fish tail or marker point; S4: The main control unit performs synchronization, filtering, normalization and fusion processing on IMU inertial measurement data and depth camera visual data to obtain multi-sensor time series data, and inputs the multi-sensor time series data into the attitude prediction model to obtain the prediction results of the silicone fish tail attitude state, end position and actual swing angle. S5: The main control unit first calculates the swing error by combining the target swing angle and the actual swing angle with the swing error model; then, based on the swing error, the main control unit corrects the PWM duty cycle by combining the corrected PWM duty cycle model, and adjusts the PWM duty cycle of the current drive-side SMA spring; the main control unit further combines the limiting function model to control the PWM duty cycle of the current drive-side SMA spring within the allowable range. S6: The main control unit adjusts the alternating heating cycle and switching frequency of the SMA springs on both sides according to the deviation between the target swing frequency and the actual swing frequency of the silicone fish tail, so that the actual swing frequency of the silicone fish tail gradually approaches the target swing frequency. S7: Repeat steps S2 to S6 to gradually bring the actual swinging state of the silicone fish tail closer to the target swinging state.

[0014] The beneficial effects of this invention are as follows: In this invention, the bionic fish with adjustable stiffness incorporates rubber sheets and waterproof tape between the shell sections, providing excellent sealing and waterproofing performance. The silicone fish tail and tail fin support plate are integrally molded for a secure connection. An adjustable stiffness drive module is constructed internally, consisting of three-stage elliptical rings, X-shaped staggered wave connecting plates, and double-sided SMA springs. The elliptical rings are equipped with limiting grooves, rectangular mounting holes, and hollowed-out weight-reduction holes, facilitating support plate positioning and assembly while reducing overall weight. The SMA springs are secured with bolt assemblies, ensuring stable installation and easy disassembly and maintenance. The segmented adjustable stiffness structure allows for flexible changes in the overall stiffness of the fish tail, relying on the SMA springs to drive the tail fin's movement. The overall design combines advantages such as waterproof sealing, lightweight construction, convenient assembly, adjustable stiffness, and stable and reliable transmission.

[0015] This invention presents a control method for an adjustable stiffness bionic fish tail. By pre-setting parameters such as the amplitude and frequency of the tail's swing, and relying on a swing angle model to generate a standard trajectory, the tail swings periodically according to a preset pattern. Motion and spatial position data are simultaneously acquired via an IMU sensor and a depth camera, fused, and then input into a posture prediction model to accurately obtain the actual posture, end position, and true swing angle of the tail. The swing angle deviation is calculated using an error model, and the SMA spring PWM duty cycle is dynamically adjusted using a correction and limiting model to ensure compliant drive parameters. The swing frequency can also be updated by adjusting the alternating heating cycle of the spring, iteratively generating a new target trajectory. The entire process is a closed-loop cycle that continuously corrects deviations and adaptively adjusts drive parameters, effectively improving the swing control accuracy of the bionic silicone fish tail and ensuring that the actual swing state continuously approaches the preset target state.

[0016] This invention employs a posture prediction method based on collaborative training to predict the position, bending angle, and swing posture of the fish tail in real time, reducing reliance on complex physical models and providing reliable feedback for the alternating heating control of the SMA spring. By adjusting the PWM duty cycle, heating duration, and alternating drive timing with the assistance of posture feedback, the accuracy and motion stability of the fish tail swing control can be improved.

[0017] The biomimetic fish tail drive structure of the present invention enables the fish tail to change its bending stiffness and swing characteristics according to different swimming frequencies, water flow environments or propulsion requirements, thereby improving the adaptability of the biomimetic fish tail in complex underwater environments.

[0018] This invention improves the controllability of the fish tail's swing amplitude and the ability to adjust the propulsion force by adjusting the adjustable stiffness structure of the fish tail's bending deformation and combining it with the single-sided alternating heating drive of the SMA spring.

[0019] This invention forms an internal driving skeleton by connecting the front and rear of a swing unit structure inside the fish tail, creating a stable force transmission and guidance path. This reduces lateral deviation, swaying, and inconsistent trajectory problems during the swinging process, thereby improving the stability and repeatability of the fish tail swing.

[0020] This invention integrates the swing unit structure, SMA spring, and fishtail body into a single integrated design, allowing the drive structure to be compactly arranged inside the fishtail, facilitating assembly and improving structural integration and reliability.

[0021] This invention proposes a semi-supervised learning attitude prediction method based on collaborative training to achieve real-time and accurate prediction of tail position and attitude, and to construct a control closed loop. Attached Figure Description

[0022] Figure 1 This is a front view of the biomimetic fish with adjustable stiffness in this invention. Figure 2 This is a cross-sectional view of the biomimetic fish with adjustable stiffness in this invention. Figure 3 This is a schematic diagram of the biomimetic fish actuator. Figure 4 This is a schematic diagram of a structure with adjustable stiffness. Figure 5 This is a schematic diagram of the structure of the first elliptical ring; Figure 6 Schematic diagram of the second elliptical ring Figure 1 ; Figure 7 Schematic diagram of the second elliptical ring Figure 2 ; Figure 8 Schematic diagram of the third elliptical ring Figure 1 ; Figure 9 Schematic diagram of the third elliptical ring Figure 2 ; Figure 10 This is a schematic diagram of a single-channel drive circuit. Figure 11 This is a comparison chart of the pose prediction RMSE between the Co-Training algorithm in this invention and traditional methods; Figure 12 This is a comparison chart of displacement prediction data in this invention.

[0023] The symbols for each component are as follows: 1. Upper shell; 2. Dorsal fin; 3. Silicone fish tail; 4. Caudal fin; 5. Lower shell; 6. Pelvic fin; 7. Rubber sheet; 8. Pectoral fin; 9. First elliptical ring; 91. Mounting protrusion; 92. Limiting groove; 10. Second elliptical ring; 101. Second rectangular mounting hole; 11. Third elliptical ring; 111. Insert groove; 112. Third rectangular mounting hole; 113. Hollowed-out weight-reducing hole; 12. Support plate; 13. Adjustable stiffness structure; 14. Nut; 15. SMA spring; 16. Double-ended stud; 17. Spring washer; 18. Bolt. Detailed Implementation

[0024] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0025] Example: like Figure 1 and Figure 2 As shown, the adjustable stiffness bionic fish of this invention includes a bionic fish body, a silicone fish tail 3, and a tail fin assembly. The bionic fish body and the tail fin assembly are fixedly connected by the silicone fish tail 3. A drive series module is provided inside the silicone fish tail 3, which is used to drive the tail fin assembly to swing left and right. The bionic fish body includes an upper shell 1, a lower shell 5, a dorsal fin 2, a pelvic fin 6, and two pectoral fins 8. The upper shell 1 and the lower shell 5 are sealed and fixed to form the fish body. The rear side of the upper shell 1 and the rear side of the lower shell 5 are both sealed and connected to the silicone fish tail 3. A rubber sheet 7 is fitted at the connection between the upper shell 1 and the lower shell 5, and a waterproof tape is provided at the external connection between the upper shell 1 and the lower shell 5. The tail fin assembly includes a tail fin 4 and a support plate 12 fixedly connected to the tail fin 4. The tail fin 4 is fixed to the silicone fish tail 3 by the support plate 12, and the silicone fish tail 3 and the support plate 12 are integrally formed. Specifically, the tail fin assembly consists of the support plate 12 and the tail fin 4, realizing power transmission and propulsion output. The support plate 12 is fixed at both ends to the third elliptical ring 11 and the base of the tail fin 4 via the insertion slots 111, forming a rigid transmission chain between the tail fin 4, the third elliptical ring 11, and the internal drive series module. This directly transmits the contraction force of the SMA spring 15 to the tail fin 4, achieving oscillating propulsion. The mating surfaces of the upper shell 1 and the lower shell 5 are sealed with a flexible rubber sheet to ensure shell sealing and assembly compatibility. The pectoral fin 8, pelvic fin 6, and dorsal fin 2 are all made of PLA material and manufactured using 3D printing. They are respectively bonded and fixed to the corresponding outer positions of the fish body with waterproof and high-temperature resistant adhesive, ensuring biomimetic shape and firm fixation. The front end of the silicone fish tail 3 is fitted onto the rear opening of the biomimetic fish body. Multiple layers of waterproof tape are wrapped around the fitting seam to achieve a sealed and tight fit between the silicone fish tail 3 and the biomimetic fish body, preventing underwater water seepage and detachment.

[0026] like Figure 3 and 4As shown, the drive series module includes a first elliptical ring 9, a second elliptical ring 10, and a third elliptical ring 11. Adjustable stiffness structures 13 are fixed between the first elliptical ring 9 and the second elliptical ring 10, and between the second elliptical ring 10 and the third elliptical ring 11. The two adjustable stiffness structures are identical, with a corrugated connecting plate in between, and are spaced apart along the fishtail axis. The adjustable stiffness structure 13 includes a first support plate, a second support plate, and two SMA springs 15, as well as two corrugated connecting plates connecting the first and second support plates. The two corrugated connecting plates are staggered in an X-shape. Each SMA spring 15 is electrically connected to the main control unit and a single-path drive circuit. The left side of the first support plate is connected to the left side of the second support plate, and the right side of the first support plate is connected to the right side of the second support plate, both by SMA springs 15. Mounting holes for the SMA springs 15 are provided on both the first and second support plates, and bolt assemblies are installed in these holes. The SMA springs 15 are clamped and fixed to the first and second support plates by the bolt assemblies. The bolt assemblies include double-ended studs 16, spring washers 17, nuts 14, and bolts 18. All four SMA springs 15 are made of nickel-titanium alloy, have identical specifications, and are arranged symmetrically on each side. Each SMA spring has spring washers pressed at both ends for reliable connection with the studs, bolts, and wires. They shrink when heated by electricity and naturally cool and reset when de-energized.

[0027] like Figure 5 and 6 As shown, the first elliptical ring 9 is in the shape of an elliptical ring surface. The first elliptical ring 9 is provided with two symmetrical mounting protrusions 91 facing inward. The mounting protrusions are provided with limiting grooves 92 for mounting the first support plate or the second support plate.

[0028] like Figure 7 As shown, the second elliptical ring 10 is in the shape of a frustum of an ellipse, and the outer surface of the second elliptical ring 10 is in the shape of a frustum of a cylinder. A second rectangular mounting hole 101 for mounting the first support plate or the second support plate is provided in the middle of the second elliptical ring 10.

[0029] like Figure 8 and 9 As shown, the third elliptical ring 11 is in the shape of a frustum of an ellipse, and the outer surface of the third elliptical ring 11 is in the shape of a frustum of a cone. The large end and the small side of the third elliptical ring 11 are provided with insertion slots 111, which are inserted and fixed to the support plate 12. The large end of the third elliptical ring 11 is provided with a third rectangular mounting hole 112 that cooperates with the first support plate or the second support plate. Two hollow weight reduction holes 113 are provided on the third elliptical ring 11.

[0030] The rigid support assembly consists of three coaxial rigid elliptical rings: a first elliptical ring 9, a second elliptical ring 10, and a third elliptical ring 11. These rings provide radial support, axial positioning, and swing guidance for the drive series module. Their positions and connections are as follows: the first elliptical ring 9 is located at the front end of the drive series module; the second elliptical ring 10 is located between the two sets of adjustable stiffness frames; and the third elliptical ring 11 is located at the end of the drive series module. The inner diameters of the three rings match the outer contour of the series module, and their inner walls are fixed to corresponding positions on the series module via slots. All three are coaxial and arranged axially to ensure the coaxiality of the series module, limit lateral offset, and form a stable swing guidance reference.

[0031] The silicone fish tail is integrally cast using a mold, with a hollow, sealed cavity inside. The shape of the cavity perfectly matches the outer contour of the three circular rings and the series module. During assembly, the first elliptical ring 9, the second elliptical ring 10, and the third elliptical ring 11, which are fixed to the series module, are embedded into the hollow cavity of the silicone fish tail. The inner wall of the silicone fits tightly with the outer wall of the elliptical rings, forming a biomimetic fish tail shape, which has the functions of waterproofing, drag reduction, water pressure resistance, and flexible swing.

[0032] like Figure 10 As shown, the single-channel drive circuit includes an Arduino microcontroller, a current-limiting resistor, an NPN power transistor, a freewheeling diode, an SMA spring, and a DC power supply. The Arduino microcontroller acts as the control signal source, outputting a PWM square wave signal. A 1kΩ resistor is connected between the Arduino output and the transistor base to limit the base current. The NPN power transistor acts as a switching device, controlling the SMA spring's energization or de-energization. The SMA spring serves as the heating drive load. The +18V DC power supply provides heating current to the SMA spring. A 1N4007 diode is connected in reverse parallel across the SMA spring to suppress transient voltages that may occur during power-off, protecting the transistor and control circuit.

[0033] The working process of a single-channel drive circuit: The first stage is when the Arduino outputs a high-level signal, which is the heating stage of the SMA spring. When the Arduino outputs a high-level PWM signal, this signal is current-limited by a 1kΩ resistor and then input to the base of the transistor, allowing the transistor to obtain sufficient base current and enter saturation conduction. At this time, the heating circuit containing the SMA spring is connected, current flows through the SMA spring and generates Joule heating, causing the SMA spring temperature to rise rapidly; when its temperature reaches the phase transition temperature, the SMA spring contracts, thereby pulling the fish tail to deflect to that side. During this process, the 1N4007 diode withstands reverse voltage and is in the cutoff state, not participating in the main current path.

[0034] The second stage is when the Arduino outputs a low level, which is the SMA spring power-off and cooling stage. When the Arduino outputs a low-level signal, the base current of the transistor disappears, and the transistor changes from the conducting state to the cut-off state, cutting off the main current path of the SMA spring. At this time, the SMA spring stops being powered, no longer generates Joule heat, and begins to cool down naturally through the surrounding environment. Simultaneously, at the moment of power-off, the inductive effect in the circuit may generate a reverse induced electromotive force. At this time, the 1N4007 diode connected in reverse parallel conducts, providing a discharge path for the induced current and preventing the reverse high voltage from breaking down the transistor, thus protecting the circuit. As the temperature of the SMA spring gradually decreases, it gradually returns to its original length and, in conjunction with the fishtail structure, returns to its mid-position, preparing for the next swing drive.

[0035] The principle of controlling the swing of the silicone fishtail 3 by alternately driving the two SMA springs on both sides using a single-channel drive circuit is as follows: Each SMA spring on both sides is equipped with an independent drive circuit, and the two PWM signals are controlled by an Arduino in a time-sharing manner to achieve alternating heating and driving of the left and right SMA springs. When the PWM signal of the left circuit is high, the left SMA spring is energized, heats up, and contracts, thus pulling the fishtail to the left; at this time, the PWM signal of the right circuit is low, and the right SMA spring is in a de-energized cooling and reset state. Subsequently, when the PWM signal of the left circuit becomes low, the left SMA spring is de-energized and cools down; simultaneously, the PWM signal of the right circuit becomes high, the right SMA spring is energized, heats up, and contracts, thus pulling the fishtail to the right. By setting the periodic control timing through the Arduino, the two PWM signals alternately output high and low levels, and the left and right SMA springs are energized and contracted sequentially, thereby driving the fishtail to produce a continuous and stable periodic left-right swing. Meanwhile, by adjusting the duty cycle and period of the PWM signal, the heating duration, contraction amount, and on / off frequency of the SMA spring can be controlled, thereby achieving the adjustment of the amplitude and frequency of the fish tail swing.

[0036] A method for controlling the above-mentioned adjustable stiffness biomimetic fish includes the following steps: S1: Set the swing parameters of the silicone fishtail 3; specifically: The main control unit sets the swing amplitude, swing frequency, initial phase, and initial PWM duty cycle of the silicone fish tail 3 and the SMA springs 15 on the left and right sides according to the target swimming speed, propulsion requirements, or preset motion mode of the bionic fish. S2: Generates the swaying trajectory of the silicone fishtail 3; specifically: The main control unit calculates the target swing angle of the silicone fish tail 3 at each sampling time based on the swing amplitude, swing frequency and initial phase of the silicone fish tail 3, combined with the swing angle model, and obtains the target swing trajectory of the silicone fish tail 3. The main control unit controls the silicone fish tail 3 to swing periodically left and right according to the preset swing amplitude and swing frequency based on the target swing trajectory. The swing angle model in step S2 is as follows: ; In the formula, For the first The swaying angle of the silicone fish tail at each sampling time; The amplitude of the swing of the silicone fish tail 3; The oscillation frequency of the silicone fish tail 3; The sampling sequence number; The time interval between two adjacent sampling times; The initial phase of the silicone fishtail 3 swing trajectory; It is a sine function; The silicone fish tail swing angle model is used to generate the expected periodic left and right swing trajectory of the silicone fish tail 3, that is, to determine the target swing angle of the silicone fish tail 3 at each sampling time, providing a reference benchmark for subsequent attitude error calculation and PWM adjustment.

[0037] In step S2, the swing angle model is used to form the desired swing trajectory and serves as the target reference for attitude error calculation, PWM duty cycle correction, and swing frequency adjustment in subsequent steps. The principle behind the main control unit controlling the SMA spring 15 to alternately heat one side, adjusting the bending angle of the adjustable stiffness structure 13, and driving the silicone fishtail 3 to achieve periodic left-right swinging is as follows: The main control unit controls the SMA spring 15 on both sides of the adjustable stiffness structure 13 to be energized and de-energized through two independent drive circuits on the left and right sides respectively. When the left SMA spring 15 is energized and heated, it contracts and generates tension, causing the two series-connected adjustable stiffness structures 13 embedded in the silicone fishtail 3 to bend to the left, causing the silicone fishtail 3 to deflect to the left. At this time, the right SMA spring 15 is de-energized, cools down, and is in a recovery state. When the right SMA spring 15 is energized and heated, it contracts and causes the adjustable stiffness structure 13 to bend to the right, causing the silicone fishtail 3 to deflect to the right. At this time, the left SMA spring 15 is de-energized and cools down. Through the alternating heating of the left and right SMA springs 15, the adjustable stiffness structure 13 forms a continuous reciprocating bending inside the silicone fishtail 3, thereby driving the silicone fishtail 3 to achieve periodic left-right swinging.

[0038] The main control unit adjusts the heating time, PWM duty cycle, and alternating left-right drive frequency of the SMA spring 15 to change the degree of contraction, stiffness state, and bending degree of the adjustable stiffness structure 13, thereby regulating the structural stiffness of the silicone fishtail 3. Specifically, during the heating process of the SMA spring 15, it undergoes a phase change and generates contraction force. Simultaneously, its mechanical properties change with temperature, enabling it to drive the adjustable stiffness structure 13 to bend, thus driving the silicone fishtail 3 to swing and adjusting the local stiffness of the silicone fishtail 3. By adjusting the heating time, PWM duty cycle, and alternating left-right drive frequency of the SMA spring 15, the degree of contraction, stiffness state, and bending degree of the adjustable stiffness structure 13 are changed, allowing the silicone fishtail 3 to obtain corresponding bending and recovery capabilities under different swing frequencies and water flow conditions.

[0039] S3: Collect posture data of silicone fishtail 3: Specifically: During the swinging process of the silicone fish tail 3, IMU inertial measurement data and depth camera visual data are collected by IMU sensor and depth camera respectively; IMU inertial measurement data includes acceleration and angular velocity data during the movement of silicone fish tail 3; depth camera visual data includes spatial position information of the end of silicone fish tail 3 or the marked point. The attitude prediction model in step S3: ; In the formula, For the first The three pose states of the silicone fish tail are estimated at each sampling time. The pose prediction model is based on a multi-sensor temporal fusion state estimation framework consisting of an existing publicly available CNN-LSTM temporal feature extraction model and a KalmanNet state estimation model. For the first The multi-sensor time-series data input to the silicone fishtail pose prediction model at each sampling time point includes IMU inertial measurement data and depth camera vision data; This represents the model parameters of the attitude prediction model; , , The first The estimated silicone fishtail 3-end at each sampling time point , , Position in three directions; For the first The actual swing angle or actual bending angle of the silicone fish tail is estimated at each sampling time. This indicates vector transpose; the 3-position state of the silicone fish tail includes the position of the tail tip and the actual swing angle; The attitude prediction model is used to explain how the main control unit obtains the actual motion state of the fishtail based on sensor data. Due to the thermal hysteresis and nonlinear contraction characteristics of the SMA spring, the silicone fishtail 3 exhibits large deformation and nonlinear motion characteristics. Using an attitude prediction model that integrates IMU inertial measurement data and depth camera visual data helps to improve the prediction accuracy of the actual swing angle and end position of the fishtail.

[0040] The IMU inertial measurement data and depth camera visual data collected in step S3 are used as input data for the attitude prediction model in step S4, which are then used to obtain the end position, actual swing angle and bending posture of the silicone fish tail 3. S4: Estimated actual swaying state of silicone fishtail 3: Specifically: The main control unit performs synchronization, filtering, normalization and fusion processing on IMU inertial measurement data and depth camera visual data to obtain multi-sensor time series data, and inputs the multi-sensor time series data into the attitude prediction model to obtain the prediction results of the silicone fish tail 3 attitude state, end position and actual swing angle. In terms of the algorithm framework, two models learn from the time-series data of the fishtail motion, and generate pseudo-labels through high-confidence prediction results. These models are then used to alternately update model parameters, thereby improving the adaptability of the attitude prediction model to complex swaying conditions. After training, the outputs of the two models can be weighted and fused to obtain the final predicted results of the fishtail end position, bending angle, and swaying posture. In terms of the algorithm framework, this invention adopts a dual-model collaborative training framework consisting of a CNN-LSTM temporal feature extraction model and a KalmanNet state estimation model. The CNN-LSTM temporal feature extraction model and the KalmanNet state estimation model share the same set of input data, namely the preprocessed multi-sensor temporal dataset. The dataset includes the three-axis acceleration and three-axis angular velocity temporal information collected by the IMU, and the three-dimensional spatial coordinate temporal information of the fish tail end collected by the depth camera. The two models independently learn the nonlinear motion temporal data of the fish tail from the two dimensions of temporal features and state recursion. During the training phase, pseudo-labels are generated using their respective high-confidence prediction results for cross-training. The weight parameters of the two models are alternately and iteratively updated to reduce the prediction bias caused by SMA thermal hysteresis and the three major deformations of the silicone fish tail, and to improve the robustness of the model under complex swing conditions. During the real-time inference phase, the two models output their respective predictions in parallel. Then, the two sets of outputs of CNN-LSTM and temporal KalmanNet are weighted and fused to obtain the unified output of the attitude prediction model, which can completely represent the real-time motion state of the silicone fish tail.

[0041] The posture prediction model outputs the fishtail posture state, including the end of the silicone fishtail 3 at... , , The positions in three directions and the actual swing angle of the silicone fish tail 3 are used to reflect the actual bending effect of the adjustable stiffness structure 13 after the SMA spring 15 is alternately heated on one side, and to provide actual state data for the swing error calculation in step S5. S5: Corrects PWM control parameters based on attitude error; specifically: The main control unit first compares the target swing angle generated by the swing angle model in step S2 with the actual swing angle obtained by the attitude prediction model in step S4, and calculates the swing error in combination with the swing error model. Then, based on the swing error, the main control unit corrects the PWM duty cycle in combination with the modified PWM duty cycle model to adjust the PWM duty cycle of the current drive-side SMA spring 15. The main control unit further combines the amplitude limiting function model to control the PWM duty cycle of the current drive-side SMA spring 15 within the allowable range to avoid overheating or insufficient drive of the SMA spring 15. In step S5, the oscillation error model is as follows: ; In the formula, For the first Error of the swaying angle of the silicone fish tail at each sampling time; The swing angle of the silicone fish tail at the Kth sampling time; The actual swing angle of the silicone fishtail 3 output by the attitude prediction model; The oscillation error model is used to determine the deviation between the actual oscillation state and the target oscillation state of the silicone fishtail 3. When the angle is large, it indicates that the actual swing angle of the silicone fishtail 3 deviates significantly from the target angle, requiring correction of the SMA spring's drive parameters; when When the value is small, it indicates that the actual swing state of the silicone fish tail 3 is close to the target state, and the current control parameters can be maintained or finely adjusted.

[0042] In step S5, the PWM duty cycle model is corrected as follows: ; In the formula, For the first The PWM duty cycle of the current drive-side SMA spring 15 at each sampling moment; This is the initial PWM duty cycle; This is the proportional adjustment coefficient; For the first Error of the swaying angle of the silicone fish tail at each sampling time; This represents the limiting function, used to limit the PWM duty cycle within the allowable range to prevent the SMA spring 15 from overheating or under-driving. In step S5, the limiting function model is as follows: ; In the formula, The duty cycle of the PWM to be limited; The minimum allowed PWM duty cycle; The maximum allowed PWM duty cycle; To take the smaller value function; To take the larger value function; The modified PWM duty cycle model and limiting function model are used to explain how the main control unit adjusts the heating intensity of the SMA spring 15 based on the swaying error of the silicone fishtail 3. Let the allowable attitude error threshold be... ,when When the swing amplitude of the silicone fish tail 3 is less than the target swing amplitude, the PWM duty cycle of the corresponding SMA spring 15 is increased or the heating time is extended, so that the bending degree of the adjustable stiffness structure 13 increases; when When the actual swing angle of the silicone fishtail 3 is within the target allowable error range, the main control unit keeps the current PWM duty cycle, heating time, and alternating drive cycle unchanged, so that the silicone fishtail 3 maintains the current stable swing state; when If the silicone fishtail 3 swings beyond the target amplitude, the PWM duty cycle can be reduced, the heating time shortened, or the alternating drive cycle of the SMA springs 15 on both sides adjusted to reduce the bending degree of the adjustable stiffness structure 13. Through the aforementioned adjustment methods, the actual swing state of the silicone fishtail 3 can be gradually made closer to the target swing state.

[0043] S6: Adjust the swaying frequency of the silicone fish tail 3; specifically: The main control unit adjusts the alternating heating cycle and switching frequency of the SMA springs 15 on both sides according to the deviation between the target swing frequency and the actual swing frequency of the silicone fish tail 3, so that the actual swing frequency of the silicone fish tail 3 gradually approaches the target swing frequency. The swing angle model in step S2 is used to generate the target swing trajectory of the silicone fish tail 3, where the target swing frequency is... The main control unit obtains the actual oscillation frequency of the silicone fishtail 3 based on the actual oscillation state of the silicone fishtail 3. The oscillation frequency error is calculated. The oscillation frequency error model is as follows: ; In the formula, For the first The oscillation frequency error of each control cycle; For the first The target oscillation frequency for each control cycle; For the first The actual oscillation frequency of each control cycle.

[0044] The main control unit corrects the target oscillation frequency for the next control cycle based on the oscillation frequency error. The frequency correction model is as follows: ; In the formula, The target oscillation frequency for the next control cycle; This is the frequency adjustment coefficient; when When the actual swing frequency of the silicone fish tail 3 is lower than the target swing frequency, the main control unit increases the target swing frequency of the next control cycle, shortens the alternating heating cycle and switching interval of the SMA springs 15 on both sides, and increases the swing frequency of the silicone fish tail 3. when When the actual swing frequency of the silicone fish tail 3 is consistent with the target swing frequency, the main control unit keeps the current target swing frequency, the alternating heating cycle on one side and the switching frequency unchanged, so that the silicone fish tail 3 maintains the current stable swing state. when When the actual swing frequency of the silicone fish tail 3 is higher than the target swing frequency, the main control unit reduces the target swing frequency of the next control cycle, extends the alternating heating cycle and switching interval of the SMA springs 15 on both sides, and reduces the swing frequency of the silicone fish tail 3. Corrected target oscillation frequency Feedback is sent to step S2 and used as the target swing frequency parameter in the swing angle model of the next control cycle; step S2 regenerates the target swing trajectory of the silicone fish tail 3 according to the new target swing frequency, and then continues to execute steps S3 to S6, thereby forming a closed-loop control process of target trajectory generation, actual attitude feedback, PWM duty cycle correction and swing frequency adjustment.

[0045] The main control unit executes the target swing trajectory generation, attitude prediction, error calculation and PWM parameter correction in a loop according to the above steps, so that the contraction of the SMA springs 15 on the left and right sides and the bending degree of the adjustable stiffness structure 13 are continuously adjusted according to the actual posture of the silicone fish tail 3 during the alternating heating process on one side, thereby improving the control accuracy of the bionic fish tail swing amplitude, trajectory repeatability and underwater motion stability.

[0046] The new oscillation frequency is used as the input parameter of the oscillation angle model to regenerate the target oscillation angle trajectory at the corresponding frequency; when the alternating heating cycle is short, the oscillation frequency of the silicone fish tail 3 increases; when the alternating heating cycle is long, the oscillation frequency of the silicone fish tail 3 decreases. Therefore, the main control unit can change the swing frequency parameters in the swing angle model according to different swimming speeds or propulsion requirements, and further cooperate with the modified PWM duty cycle model in step S5 to correct the PWM duty cycle, so that the silicone fish tail 3 maintains the corresponding swing frequency and swing amplitude under different working conditions. S7: Cyclic execution of auxiliary closed-loop control; Repeat steps S2 to S6 to enable the main control unit to cycle through the generation of the target swing trajectory, sensor data acquisition, attitude prediction, swing error calculation, PWM duty cycle correction, amplitude limiting protection and swing frequency adjustment, thereby achieving swing control of the silicone fish tail 3. During the loop, step S2 calls the swing angle model to generate the target swing angle; steps S3 and S4 call the attitude prediction model to obtain the actual swing state; step S5 calls the swing error calculation method, PWM duty cycle correction method and amplitude limiting function to correct the drive parameters of SMA spring 15; step S6 feeds back the adjusted swing frequency to the swing angle model to update the target trajectory for the next cycle. By cyclically executing steps S2 to S6, the actual swing state of the silicone fishtail 3 gradually approaches the target swing state, thereby improving the swing amplitude control accuracy, swing trajectory repeatability, swing frequency controllability, and underwater motion stability of the silicone fishtail 3.

[0047] Real-time prediction of the tail tip position, bending angle, and swing posture during the movement, stiffness variation, and dynamic swaying of a biomimetic fish is a crucial foundation for achieving alternating heating control of the SMA spring and closed-loop regulation of the tail sway. Due to the thermal response hysteresis, nonlinear contraction, and hysteresis characteristics of the SMA spring, and the continuous deformation of the flexible fish tail during movement, relying solely on traditional physical models, integral calculations, or simple filtering methods easily leads to problems such as error accumulation, weak anti-interference ability, and insufficient adaptability to large flexible deformations. To improve the accuracy of tail posture prediction, this invention sets up a posture prediction algorithm module, employing a multi-sensor fusion posture prediction method based on Co-Training to fuse IMU inertial measurement data and depth camera visual data. This method utilizes a small amount of labeled data and a large amount of unlabeled time-series data for semi-supervised training, reducing reliance on complex physical models. It can predict the tail tip position, bending angle, and swaying trajectory in real time and feed the prediction results back to the main control unit, providing a basis for SMA spring heating timing, PWM duty cycle adjustment, and tail sway amplitude control.

[0048] Regarding sensor configuration, this invention can place an IMU sensor at the end of the bionic fish tail or near the free end of the tail to collect triaxial acceleration and triaxial angular velocity data; simultaneously, a depth camera is set to collect spatial position and depth information of the end of the fish tail or a marker point. The main control unit performs time synchronization, data alignment, and sliding window segmentation on the two types of sensor data to form a multi-sensor time-series dataset.

[0049] in, Figure 11This paper presents a comparison of the RMSE results between the collaborative training method of this invention and the traditional method under different driving modes. Under both single-memory alloy spring drive and dual-memory alloy spring drive conditions, the RMSE corresponding to the collaborative training method of this invention is lower than that of the traditional method, indicating that this attitude prediction method can reduce the prediction error of the fish tail motion state and improve the estimation accuracy of the fish tail end displacement and swing posture.

[0050] Figure 12 This presents a comparison of displacement experimental data and prediction data from different methods. Figure 12 It can be seen that the trend of the co-training prediction data of this invention is closer to that of the displacement experimental data, and the prediction results can better follow the nonlinear motion process of the fish tail end displacement changing with time. In contrast, the traditional method predicts data with a large deviation from the experimental data in the later stage of motion, indicating that the traditional method is not good at representing the nonlinear deformation and posture change during the continuous swinging process of the fish tail.

[0051] Therefore, combining Figure 11 and Figure 12 It can be seen that the present invention improves the prediction accuracy of the attitude prediction model for the actual motion state of the fish tail through collaborative training, and can more accurately obtain the actual attitude, end position and true swing angle of the silicone fish tail 3. The prediction result can be used as feedback information in the control closed loop to calculate the error between the target swing angle and the actual swing angle, and further correct the PWM duty cycle, heating time and left and right alternating drive timing of the SMA spring 15, so that the actual swing state of the silicone fish tail 3 continuously approaches the preset target state.

[0052] During closed-loop control, the main control unit compares the attitude prediction result with the target swing amplitude. When the predicted swing amplitude is less than the target value, the PWM duty cycle of the corresponding SMA spring can be increased or the heating time can be extended; when the predicted swing amplitude is too large or yaw occurs, the PWM duty cycle can be reduced, the heating time shortened, or the alternating drive timing of the left and right SMA springs can be adjusted. These methods improve the consistency, motion stability, and control accuracy of the biomimetic fishtail swing trajectory.

Claims

1. A biomimetic fish with adjustable stiffness, characterized in that, It includes a bionic fish body, a silicone fish tail (3) and a tail fin assembly. The bionic fish body and the tail fin assembly are fixedly connected by the silicone fish tail (3). A drive series module is provided inside the silicone fish tail (3). The drive series module is used to drive the tail fin assembly to swing left and right. The drive series module includes a first elliptical ring (9), a second elliptical ring (10) and a third elliptical ring (11). An adjustable stiffness structure (13) is fixed between the first elliptical ring (9) and the second elliptical ring (10), and between the second elliptical ring (10) and the third elliptical ring (11). The adjustable stiffness structure (13) includes a first support plate, a second support plate and two SMA springs (15), as well as two wave connecting plates connecting the first support plate and the second support plate. The two wave connecting plates are staggered in an X shape. The left side of the first support plate and the left side of the second support plate, and the right side of the first support plate and the right side of the second support plate are connected by SMA springs (15). The end of the silicone fish tail (3) is equipped with an IMU sensor and a depth camera. The IMU sensor is used to collect acceleration and angular velocity data during the movement of the silicone fish tail (3), and the depth camera is used to collect spatial position information of the end of the silicone fish tail (3) or the marker point. Each of the SMA springs (15) is electrically connected to the main control unit and the single-path drive circuit.

2. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The bionic fish body includes an upper shell (1), a lower shell (5), a dorsal fin (2), a pelvic fin (6), and two pectoral fins (8). The upper shell (1) and the lower shell (5) are sealed and fixed to form the fish body. The rear side of the upper shell (1) and the rear side of the lower shell (5) are sealed and connected to the silicone fish tail (3). A rubber sheet (7) is attached to the connection between the upper shell (1) and the lower shell (5). A waterproof tape is provided at the external connection between the upper shell (1) and the lower shell (5).

3. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The tail fin assembly includes a tail fin (4) and a support plate (12) fixedly connected to the tail fin (4). The tail fin (4) is fixed to the silicone fish tail (3) by the support plate (12). The silicone fish tail (3) and the support plate (12) are integrally formed.

4. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The first support plate and the second support plate are both provided with mounting holes for installing SMA springs (15). A bolt assembly is provided in the mounting holes. The SMA springs (15) are clamped and fixed on the first support plate and the second support plate by the bolt assembly. The bolt assembly includes a stud (16), a spring washer (17), a nut (14) and a bolt (18).

5. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The first elliptical ring (9) is in the shape of an elliptical ring surface. The first elliptical ring (9) is provided with two symmetrical mounting protrusions (91) facing inward. The mounting protrusions are provided with limiting grooves (92) for mounting the first support plate or the second support plate.

6. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The second elliptical ring (10) is in the shape of a frustum of an ellipse. The outer surface of the second elliptical ring (10) is in the shape of a frustum of a cylinder. The middle part of the second elliptical ring (10) is provided with a second rectangular mounting hole (101) for mounting the first support plate or the second support plate.

7. The biomimetic fish with adjustable stiffness according to claim 1, characterized in that, The third elliptical ring (11) is in the shape of an elliptical frustum. The outer surface of the third elliptical ring (11) is in the shape of a frustum. The large end and the small side of the third elliptical ring (11) are provided with insertion slots (111). The large end of the third elliptical ring (11) is provided with a third rectangular mounting hole (112) that cooperates with the first support plate or the second support plate.

8. A control method for an adjustable stiffness bionic fish according to any one of claims 1-7, characterized in that, Includes the following steps: S1: Set the swing amplitude, swing frequency, initial phase and initial PWM duty cycle of the silicone fish tail (3) and the SMA springs (15) on the left and right sides; S2: The main control unit calculates the target swing angle of the silicone fish tail (3) at each sampling time based on the swing amplitude, swing frequency and initial phase of the silicone fish tail (3) and the swing angle model, and obtains the swing trajectory of the silicone fish tail (3). The main control unit controls the silicone fish tail (3) to swing periodically left and right according to the swing trajectory of the silicone fish tail (3). S3: During the swinging process of the silicone fish tail (3), IMU inertial measurement data and depth camera visual data are collected by IMU sensor and depth camera respectively; IMU inertial measurement data includes acceleration and angular velocity data during the movement of silicone fish tail (3); depth camera visual data includes spatial position information of the end of silicone fish tail (3) or the marker point; S4: The main control unit performs synchronization, filtering, normalization and fusion processing on the IMU inertial measurement data and depth camera visual data to obtain multi-sensor time series data, and inputs the multi-sensor time series data into the attitude prediction model to obtain the prediction results of the attitude state, end position and actual swing angle of the silicone fish tail (3). S5: The main control unit first calculates the swing error by combining the target swing angle and the actual swing angle with the swing error model; then, based on the swing error, the main control unit corrects the PWM duty cycle by combining the corrected PWM duty cycle model and adjusts the PWM duty cycle of the current drive-side SMA spring (15); the main control unit further combines the limiting function model to control the PWM duty cycle of the current drive-side SMA spring (15) within the allowable range; S6: The main control unit adjusts the alternating heating cycle and switching frequency of the SMA springs (15) on both sides according to the deviation between the target swing frequency and the actual swing frequency of the silicone fish tail (3), so that the actual swing frequency of the silicone fish tail (3) gradually approaches the target swing frequency. S7: Repeat steps S2 to S6 to gradually bring the actual swing state of the silicone fish tail (3) closer to the target swing state.

Images