Servo press control method and system based on multi-modal perception

Abstract

The invention relates to a servo press control method and system based on multi-modal perception, and the method comprises the steps: collecting and preprocessing multi-modal data reflecting the operation state of a servo press in real time, the multi-modal data comprising stamping tonnage, sliding block position, stamping speed, hydraulic flow, mechanical deformation and safety perception information; multi-modal data fusion is realized by using a mixed strategy combining weighted fusion and a BP neural network; based on the servo press system state reflected by the fusion data, parameters of a PID controller for controlling the current of a motor are adjusted in a self-adaptive mode, and the method comprises the steps that the current operation condition is automatically judged to be no-load, light-load or heavy-load; dynamically adjusting a proportionality coefficient, an integral coefficient and a differential coefficient of a PID (Proportion Integration Differentiation) controller for controlling the current of the motor in combination with a Ziegler-Nichols setting method and a fuzzy control rule; and under the control of the self-adaptive PID controller, an S-shaped acceleration and deceleration curve is adopted to plan the movement track of the sliding rail according to the fused data and a preset stamping process curve.

Description

Technical Field

[0001] This invention relates to the field of electrical control technology for servo presses, and in particular to a servo press control method and system based on multimodal sensing. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Traditional mechanical and hydraulic presses have many inherent drawbacks in the stamping process. Mechanical presses rely on flywheel energy storage, and their slide motion curve is fixed, making it impossible to flexibly adjust according to workpiece material and process requirements. This results in high energy consumption, high noise, and difficulty in achieving precise control. While hydraulic presses can provide constant pressure, their response speed is slow, oil leakage is prone to polluting the environment, and the system efficiency is low. With the rapid development of servo motor technology, power electronics technology, and control theory, servo presses have emerged as a new type of stamping equipment that integrates high precision, high efficiency, and high flexibility.

[0004] Existing servo presses mostly employ a transmission structure of servo motors, ball screws, and crank-connecting rods. The position, speed, or torque of the servo motor is controlled by a programmable logic controller (PLC) or a dedicated motion controller, thereby indirectly controlling the movement of the slide block. However, these systems generally suffer from problems such as limited sensing dimensions, weak data processing capabilities, and rigid control strategies. For example, data collected from a single sensor cannot comprehensively reflect the equipment's operating status, easily leading to control decision deviations due to missing local information, resulting in low stamping accuracy. Furthermore, most systems rely solely on displacement and pressure sensors for closed-loop feedback, ignoring critical status information such as vibration, temperature, and deformation. This results in insufficient robustness when facing complex operating conditions (such as strong load fluctuations and material springback), easily leading to overshoot, oscillation, or even overload shutdown. In addition, traditional control strategies often use fixed PID parameters, unable to adaptively adjust according to real-time operating conditions (no load, light load, heavy load), limiting further improvements in dynamic performance and control accuracy. Summary of the Invention

[0005] To solve the above-mentioned technical problems, or at least partially solve them, the present invention provides a servo press control method and system based on multimodal perception.

[0006] In a first aspect, the present invention provides a servo press control method based on multimodal perception, comprising: real-time acquisition and preprocessing of multimodal data reflecting the operating status of the servo press, wherein the multimodal data includes stamping tonnage, slide position, stamping speed, hydraulic flow, mechanical deformation and safety perception information; Multimodal data fusion is achieved using a hybrid strategy combining weighted fusion and backpropagation neural networks; Based on the servo press system status reflected by the fused data, the parameters of the PID controller controlling the motor current are adaptively adjusted, including: automatically determining whether the current operating condition is no-load, light-load, or heavy-load; and dynamically adjusting the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller controlling the motor current by combining the Ziegler-Nichols tuning method and fuzzy control rules. Under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration and deceleration curve based on the fused data and the preset stamping process curve.

[0007] Furthermore, the servo press includes a machine body, a base at the bottom of the machine body, and a motor and a reducer at the top of the machine body. The output shaft of the motor is connected to the reducer, and the output shaft of the reducer is coupled to a lead screw via a coupling. The lead screw couples to a drive slider, which slides up and down along a guide rail on the machine body. The drive slider is connected to a slider via a composite connecting rod. The composite connecting rod includes two three-link rods hinged to both sides of the machine body. Each three-link rod is also hinged to a first double-link rod and a second double-link rod. The first double-link rod connected to the two three-link rods is hinged to the drive slider, and the second double-link rod connected to the two three-link rods is hinged to the slider. The transmission process of the servo press is as follows: the motor outputs rotational power under servo control, which is amplified by the reducer and matched for torque and speed. The power is then transmitted to the lead screw via a coupling. The lead screw accurately converts the rotational motion into linear motion, directly driving the drive slider to reciprocate up and down along the guide rail. The drive slider drives the composite connecting rod to move the slider up and down, completing the stamping operation.

[0008] Furthermore, the preprocessing of noisy multimodal data includes: noise reduction using wavelet filtering algorithms; using the db4 wavelet basis function for three-level decomposition to filter out high-frequency noise and retain the abrupt change characteristics of the data; using min-max normalization to uniformly map multimodal data of different dimensions to the [0,1] interval; and utilizing 3 The criteria detect outliers and replace them with linear interpolation.

[0009] Furthermore, multimodal data fusion can be achieved using a hybrid strategy combining weighted fusion and backpropagation neural networks, including: For redundant sensors of the same type, any one of the multimodal data collected is used for redundant data acquisition. The data is then weighted and fused based on sensor accuracy. ; in, Data fused from redundant sensors of the same type Let be the weight of the i-th redundant sensor of the same type. This is the raw data of the i-th redundant sensor of the same type.

[0010] Furthermore, the weight optimization process includes: Given n redundant sensors of the same type, for any redundant sensor i, take the measurements from the most recent m times to form a sequence. Then all the data from n redundant sensors of the same type form an m×n data array; Divide the data array of n redundant sensors of the same type into k intervals, and count the frequency of data occurrence in each interval. Let j represent the interval; to optimize weight allocation, calculate the signal entropy of the i-th sensor. : ; Based on the signal entropy of redundant sensors of the same type Determine the weights of each redundant sensor of the same type: .

[0011] Furthermore, the hybrid strategy of combining weighted fusion and BP neural network to achieve multimodal data fusion includes: for heterogeneous multimodal data, a BP neural network is constructed for high-level fusion. The BP neural network combines the measurement results of multiple heterogeneous sensors to obtain a predicted true value. The loss is the error between the true and predicted values. The BP neural network is trained using the gradient descent method and training stops when the error is less than a set threshold. During the weight update process of the BP neural network, a momentum factor is introduced to accelerate convergence.

[0012] Furthermore, under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration and deceleration curve based on the fused data and the preset stamping process curve.

[0013] The formula for the S-shaped acceleration / deceleration curve is as follows:

[0014] in, for The actual displacement of the slider at any given time. This represents the maximum stroke of the slider. For acceleration, The maximum speed during the constant speed phase. These represent the end of acceleration, the end of constant speed, and the total time of motion, respectively.

[0015] Furthermore, the basic parameters of the PID controller are obtained in advance using the Ziegler-Nichols tuning method; the PID controller is set to pure proportional control; under the set operating conditions, the proportional coefficient is gradually increased until critical oscillation occurs, and the corresponding critical gain Ku and oscillation period Tu are recorded; the baseline PID parameters are calculated according to the empirical formula of the Ziegler-Nichols proportional coefficient; the parameter adjustment coefficients under the corresponding operating conditions are matched from the pre-built fuzzy control rule base according to the identified operating conditions, and the baseline PID parameters are adjusted using the parameter adjustment coefficients to obtain the target proportional coefficient, integral coefficient, and derivative coefficient.

[0016] Secondly, the present invention provides a servo press control system based on multimodal perception, implemented based on the aforementioned servo press control method based on multimodal perception, comprising: a transmission module, an execution module, a control module, and a sensing module; The sensing module collects multimodal data reflecting the operating status of the servo press in real time. The transmission module is responsible for efficiently and reliably transmitting the multimodal data collected by the sensing module in real time and for interaction between the various modules. The control module adopts a collaborative control architecture of PLC master station, motion controller and host computer to realize a servo press control method based on multimodal perception; The execution module is responsible for receiving instructions from the control module and completing the corresponding actions.

[0017] Furthermore, the transmission module adopts a hybrid communication of industrial Ethernet and CAN bus; wherein, industrial Ethernet is used to connect the host computer, PLC master station and motion controller, carrying key data streams with high bandwidth and low latency requirements, ensuring rapid interaction of control commands and status feedback; CAN bus is used to connect distributed vibration sensors, safety modules and low-speed I / O devices to form a low-level field network.

[0018] The technical solutions provided in the embodiments of the present invention have the following advantages compared with the prior art: This invention integrates multimodal data from mechanics, kinematics, and electrical systems, employing a hybrid strategy combining weighted fusion and a backpropagation (BP) neural network to achieve precise sensing of the servo press's status, ensuring both perception and control accuracy while balancing precision and real-time performance. Based on the state reflected by the multimodal fusion data, the parameters of the PID controller are adaptively adjusted. Under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration / deceleration curve according to the fused data and a preset stamping process curve. This achieves precise trajectory tracking and adaptive adjustment of the servo press slide, maintaining high stability even under strong load fluctuations and strong environmental interference, providing technical support for the high-performance operation of servo presses. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a servo press control method based on multimodal sensing provided in an embodiment of the present invention; Figure 2 A schematic diagram of a servo press provided in an embodiment of the present invention; Figure 3 A flowchart for the preprocessing of modal data provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a servo press control system based on multimodal perception provided in an embodiment of the present invention.

[0022] The labels and their meanings in the diagram are as follows: 1. Motor, 2. Reducer, 3. Lead screw, 4. Composite connecting rod, 5. Machine body, 6. Slider, 7. Base. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] Example 1 The present invention provides a servo press control method based on multimodal sensing, which is applied to a servo press.

[0026] like Figure 2 As shown, the servo press includes: a machine body 5, a base 7 at the bottom of the machine body 5, a motor 1 and a reducer 2 at the top of the machine body 5, the output shaft of the motor 1 being connected to the reducer 2, the output shaft of the reducer 2 being coupled to a lead screw 3 via a coupling, the lead screw 3 being coupled to a drive slider, and the drive slider sliding up and down along a guide rail provided on the machine body 5. The drive slider is connected to a slider 6 via a composite connecting rod 4. The composite connecting rod 4 includes: two three-link rods hinged to both sides of the machine body, each three-link rod also hinged to a first double-link rod and a second double-link rod, the first double-link rod connected to the two three-link rods being hinged to the drive slider, and the second double-link rod connected to the two three-link rods being hinged to the slider. The transmission process of the servo press is as follows: the motor outputs rotational power under servo control, and after the torque is amplified and the speed is matched by the reducer, the power is transmitted to the lead screw through the coupling; the lead screw accurately converts the rotational motion into linear motion, which directly drives the drive slider to move up and down along the guide rail of the machine body. The drive slider drives the composite connecting rod 4 to move the slider up and down, thereby applying stamping pressure to the workpiece placed on the base and completing the stamping forming operation.

[0027] like Figure 1 As shown, the servo press control method based on multimodal sensing includes: Real-time acquisition and preprocessing of multimodal data reflecting the operating status of the press.

[0028] Multimodal data includes, but is not limited to: stamping tonnage acquired by pressure sensors, i.e., the output force of the slide block of the servo press; real-time slide block position acquired by grating rulers or magnetostrictive displacement sensors; slide block stamping speed calculated by encoders or speed sensors; hydraulic flow rate of the lubrication and cooling hydraulic systems monitored by flow meters; mechanical deformation of key machine body nodes monitored by strain gauges or fiber optic grating sensors; and safety perception information acquired by limit switches, temperature sensors, etc. The data acquisition frequency is dynamically allocated according to signal characteristics and control requirements. Pressure and displacement data, due to their criticality to the control closed loop, are sampled at frequencies up to 1 kHz; vibration data, containing rich mechanical health information, is sampled at 500 Hz; while signals such as temperature and electrical parameters (e.g., motor current, voltage, bus voltage) are sampled at 100 Hz.

[0029] The multimodal data from servo press systems is complex and requires noise reduction and normalization processing. For example... Figure 3 As shown, the multimodal data preprocessing process includes: Denoising is achieved using a wavelet filtering algorithm: the db4 wavelet basis function is selected for three-level decomposition to filter out high-frequency noise and retain the abrupt change characteristics of the data; The min-max normalization method is used to uniformly map multimodal data with different dimensions to the [0,1] interval. The formula is as follows: ; in, This refers to any one of the collected raw multimodal data, such as stamping tonnage, stamping speed, etc. and These are the minimum and maximum values ​​of the data within a set time window, respectively. The data is after normalization; Exploit 3 Criteria for detecting outliers. Specifically, if a data point meets the criteria... in The mean of the data. If the value is less than the standard deviation, it is considered an outlier.

[0030] Outliers are replaced using linear interpolation to ensure the continuity and integrity of the data sequence.

[0031] Multimodal data fusion is achieved by using a hybrid strategy that combines weighted fusion and BP neural network.

[0032] For redundant sensors of the same type, any one of the multimodal data collected is used for redundant data acquisition. The data is then weighted and fused based on sensor accuracy. ; in, Data fused from redundant sensors of the same type Let be the weight of the i-th redundant sensor of the same type. This is the raw data of the i-th redundant sensor of the same type.

[0033] Given n redundant sensors of the same type, for any redundant sensor i, take the measurements from the most recent m times to form a sequence. Then, all the data from n redundant sensors of the same type form an m×n data array; divide the data array of n redundant sensors of the same type into k intervals, and count the frequency of data occurrence in each interval. Let j represent the interval; to optimize weight allocation, calculate the signal entropy of the i-th sensor. : ; Based on the signal entropy of redundant sensors of the same type Determine the weights of each redundant sensor of the same type: ; The above is entirely data-driven and can objectively reflect the information contribution of each sensor. For example, by weighted and fused the measurement data from two pressure sensors that measure the tonnage of stamping, a more reliable fused pressure value can be obtained.

[0034] For heterogeneous multimodal data, such as displacement, pressure, vibration, and temperature, a backpropagation (BP) neural network is constructed for high-level fusion. The example BP neural network takes 8-dimensional preprocessed heterogeneous multimodal data as input, processes it through 16 hidden layer nodes, and outputs the system state after fusion of the 8-dimensional heterogeneous multimodal data. The BP neural network is trained using gradient descent with a learning rate of 0.01, iterating until the error is less than 0.001. To address the slow convergence speed of traditional BP networks, a momentum factor is introduced. (Value is 0.9) Improved weight update formula: in, Let be the weight change from neuron v to neuron u, and s be the number of iterations. For learning rate, Let be the error term of neuron j. The output of the previous layer neuron v, momentum factor This significantly shortens the network convergence time and reduces the fusion error to ±0.2%.

[0035] Table 1 shows the performance comparison of fusion algorithms.

[0036] As shown in Table 1, the hybrid fusion strategy using weighted fusion and BP neural network fusion balances accuracy and real-time performance, with a fusion error ≤ ±0.25% and a processing time ≤ 4ms / sample, meeting the real-time control requirements of the servo press.

[0037] Based on the servo press system status reflected by the fused data, the parameters of the PID controller controlling the motor current are adaptively adjusted, including: automatically determining whether the current operating condition is no-load, light-load, or heavy-load; and dynamically adjusting the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller controlling the motor current by combining the Ziegler-Nichols tuning method and fuzzy control rules.

[0038] In the specific implementation process, the basic parameters of the PID controller are obtained in advance using the Ziegler-Nichols tuning method; the PID controller is set to pure proportional control; under the set operating condition (such as light load condition), the proportional coefficient of the PID controller is gradually increased until critical oscillation occurs, and the corresponding critical gain is recorded. and oscillation period Based on the Ziegler-Nichols proportional coefficient empirical formula, a set of baseline PID parameters is calculated: ; ; ; in, These are the benchmark proportional coefficient, benchmark integral coefficient, and benchmark differential coefficient, respectively.

[0039] Based on the identified operating conditions, the parameter adjustment coefficients for the corresponding operating conditions are matched from the pre-built fuzzy control rule base. The target proportional coefficient, integral coefficient, and derivative coefficient are obtained by adjusting the benchmark PID parameters using the parameter adjustment coefficients.

[0040] The fuzzy control rule base records the relationship between operating conditions and parameter adjustment coefficients. For example, under heavy load conditions, the parameter adjustment coefficient of the proportional coefficient is increased to improve the system's response speed to load changes; under no-load conditions, the parameter adjustment coefficient of the integral coefficient is decreased to avoid overshoot caused by integral accumulation.

[0041] It should be noted that the PID controller in this invention directly regulates the motor current, because all parameters such as pressure and speed ultimately output by the slider are essentially achieved by adjusting the motor current.

[0042] Under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration and deceleration curve based on the fused data and the preset stamping process curve.

[0043] The formula for the S-shaped acceleration / deceleration curve is as follows:

[0044] in, for The actual displacement of the slider at any given time. This represents the maximum stroke of the slider. For acceleration, The maximum speed during the constant speed phase. These represent the end of acceleration, the end of constant velocity, and the total time of motion, respectively. When the slider starts from rest with constant acceleration... During motion, the displacement increases with the square of time. Under the law of uniformly accelerated linear motion, this ensures a smooth start-up of the press without causing excessive electrical shock. When the slider reaches its maximum speed... Then it begins to run at a constant speed, at which point the displacement increases linearly with time, ensuring the processing efficiency of the press; when the slider decelerates at a constant speed... When decelerating to a stop, the displacement expression passes through the maximum stroke. Reverse derivation ensures accurate positioning of the slider's endpoint and avoids overshoot. Experiments show that the velocity curve of this trajectory is a continuous and smooth trapezoidal wave, and the acceleration curve is smooth without abrupt changes. This effectively reduces vibration and impact during slider movement, providing a guarantee for the accurate acquisition of multimodal sensing data and the stable execution of control strategies.

[0045] In actual operation, the system first selects or defines an ideal stamping curve (i.e., pressure-displacement or pressure-time curve) on the host computer based on the material and shape of the workpiece. After startup, the sensing layer begins to collect data at different frequencies. For example, in a typical stamping cycle, the pressure sensor records the complete pressure curve from zero, rapid rise after contact with the workpiece, holding pressure for a period of time, and then releasing. At the same time, the displacement sensor records the precise motion trajectory of the slider following an S-shaped acceleration and deceleration law.

[0046] Through the aforementioned series of intelligent processing and control, the system of this invention exhibited superior performance in experiments: the static pressure control error remained stable within ±0.3%, the displacement repeatability accuracy reached ±0.01 mm, and the dynamic response time was shortened to 0.12 seconds. During a continuous 24-hour load test, the system showed no downtime due to malfunctions, while the comparative traditional system experienced two motor overload alarms under the same conditions, with a maximum pressure control error reaching ±0.9%. This fully demonstrates that the present invention represents a significant improvement of over 50% in stability, anti-interference capability, and overall performance compared to the traditional system.

[0047] Example 2 like Figure 4 As shown, this embodiment of the invention provides a servo press control system based on multimodal perception. The system integrates multimodal data such as mechanical, kinematic, and electrical data of the servo press by constructing a complete "perception-transmission-control-execution" closed-loop architecture. By utilizing data preprocessing, hybrid fusion, and adaptive control strategies, it achieves precise, stable, and efficient control of the stamping process, significantly improving the overall performance of the servo press.

[0048] The servo press control system based on multimodal perception includes: an execution module, a sensing module, a transmission module, and a control module; The sensing module collects multimodal data reflecting the operating status of the servo press in real time. The multimodal data includes stamping tonnage, slide position, stamping speed, hydraulic flow, mechanical deformation, and safety perception information.

[0049] The transmission module is responsible for the efficient and reliable real-time transmission of multimodal data collected by the sensing module and the interaction between various modules. To balance the requirements of high speed, high bandwidth, and high reliability, the transmission module of this invention adopts a hybrid communication architecture of industrial Ethernet and CAN bus. Industrial Ethernet (such as EtherNet / IP, Profinet, or ModbusTCP) is used to connect the host computer, PLC master station, and motion controller, carrying critical data streams with high bandwidth and low latency requirements, such as pressure, displacement, and temperature, ensuring rapid interaction of control commands and status feedback. The CAN bus, due to its excellent anti-interference capability and multi-master communication characteristics, is used to connect distributed vibration sensors, safety modules, and low-speed I / O devices, forming a robust underlying field network. The communication protocol between the PLC master station and various sensors and servo drives is based on the ModbusTCP and CANopen standards, ensuring interoperability between devices from different manufacturers.

[0050] The control module is the decision-making unit of the entire system. It employs a collaborative control architecture of a PLC master station, motion controller, and host computer to implement a servo press control method based on multimodal perception. The host computer runs a modular software interface developed using Visual Studio 2022 and based on the Windows 7 operating system, providing advanced functions such as human-machine interaction, process parameter setting, data storage, and historical traceability. The PLC master station handles tasks such as logic control, safety interlocking, fault diagnosis, and program sequential execution, ensuring the safety and reliability of the system. The motion controller focuses on high-performance real-time motion control, receiving multidimensional data from the sensing module, analyzing, processing, and fusing the data to generate precise control commands that are sent to the execution module. The control module's workflow is a high-speed closed loop: first, it acquires real-time data from the sensing module; second, it preprocesses, fuses, and identifies the state of the data; then, it dynamically adjusts the PID controller parameters for PID control; finally, it sends action commands to the execution module and continuously monitors the execution effect, forming a recurring cycle.

[0051] The execution module is responsible for receiving instructions from the control module and completing corresponding actions. The execution module includes a servo press and a feeding robotic arm; the servo press includes a machine body, a base at the bottom of the machine body, and a motor and a reducer at the top of the machine body. The output shaft of the motor 1 is connected to the reducer, and the output shaft of the reducer is coupled to a lead screw via a coupling. The lead screw couples to a drive slider, which slides up and down along a guide rail on the machine body 5. The drive slider is connected to the slider via a composite connecting rod.

[0052] In the embodiments provided by this invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, structures, or units, and may be electrical, mechanical, or other forms.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0054] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0055] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A servo press control method based on multimodal sensing, characterized in that, include: Real-time acquisition and preprocessing of multimodal data reflecting the operating status of the servo press; multimodal data includes stamping tonnage, slide position, stamping speed, hydraulic flow, mechanical deformation and safety perception information. Multimodal data fusion is achieved using a hybrid strategy combining weighted fusion and backpropagation neural networks; Based on the servo press system status reflected by the fused data, the parameters of the PID controller controlling the motor current are adaptively adjusted, including: automatically determining whether the current operating condition is no-load, light-load, or heavy-load; and dynamically adjusting the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller controlling the motor current by combining the Ziegler-Nichols tuning method and fuzzy control rules. Under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration and deceleration curve based on the fused data and the preset stamping process curve.

2. The servo press control method based on multimodal sensing according to claim 1, characterized in that, The servo press includes a body (5), a base (7) at the bottom of the body (5), a motor (1) and a reducer (2) at the top of the body (5), the output shaft of the motor (1) is connected to the reducer (2), the output shaft of the reducer (2) is coupled to a lead screw (3) through a coupling, the lead screw (3) is coupled to a drive slider, the drive slider slides up and down along the body guide rail set on the body (5); the drive slider is connected to a slider (6) via a composite connecting rod (4); the composite connecting rod (4) includes: two three-bar linkages hinged to both sides of the body, each three-bar linkage is also hinged to The first double auxiliary rod and the second double auxiliary rod, the first double auxiliary rod connected to the two triple auxiliary rods are hinged to the drive slider, and the second double auxiliary rod connected to the two triple auxiliary rods are hinged to the slider (6); the transmission process of the servo press is as follows: the motor outputs rotational power under servo control, and after the torque is amplified and the speed is matched by the reducer, the power is transmitted to the lead screw through the coupling; the lead screw accurately converts the rotational motion into linear motion, directly driving the drive slider to move up and down along the machine body guide rail, and the drive slider drives the composite connecting rod to move the slider up and down to complete the stamping operation.

3. The servo press control method based on multimodal sensing according to claim 1, characterized in that, Preprocessing of noisy multimodal data includes: noise reduction using wavelet filtering algorithm; three-level decomposition using db4 wavelet basis function to filter out high-frequency noise and retain the abrupt change characteristics of the data; min-max normalization to uniformly map multimodal data of different dimensions to the [0,1] interval; and using 3 The criteria detect outliers and replace them with linear interpolation.

4. The servo press control method based on multimodal sensing according to claim 1, characterized in that, Multimodal data fusion is achieved using a hybrid strategy combining weighted fusion and backpropagation (BP) neural networks, including: For redundant sensors of the same type, any one of the multimodal data collected is used for redundant data acquisition. The data is then weighted and fused based on sensor accuracy. ； in, Data fused from redundant sensors of the same type Let be the weight of the i-th redundant sensor of the same type. This is the raw data of the i-th redundant sensor of the same type.

5. The servo press control method based on multimodal sensing according to claim 4, characterized in that, The weight optimization process includes: Given n redundant sensors of the same type, for any redundant sensor i, take the measurements from the most recent m times to form a sequence. Then all the data from n redundant sensors of the same type form an m×n data array; Divide the data array of n redundant sensors of the same type into k intervals, and count the frequency of data occurrence in each interval. Let j represent the interval; to optimize weight allocation, calculate the signal entropy of the i-th sensor. : ； Based on the signal entropy of redundant sensors of the same type Determine the weights of each redundant sensor of the same type: 。 6. The servo press control method based on multimodal sensing according to claim 1, characterized in that, The hybrid strategy of combining weighted fusion and BP neural network to achieve multimodal data fusion includes: for heterogeneous multimodal data, a BP neural network is constructed for high-level fusion. The BP neural network combines the measurement results of multiple heterogeneous sensors to obtain a predicted true value. The loss is the error between the true and predicted values. The BP neural network is trained using the gradient descent method and training stops when the error is less than a set threshold. During the weight update process of the BP neural network, a momentum factor is introduced to accelerate convergence.

7. The servo press control method based on multimodal sensing according to claim 1, characterized in that, Under the control of the adaptive PID controller, the motion trajectory of the slide rail is planned using an S-shaped acceleration and deceleration curve based on the fused data and the preset stamping process curve. The formula for the S-shaped acceleration / deceleration curve is as follows: in, for The actual displacement of the slider at any given time. This represents the maximum stroke of the slider. For acceleration, The maximum speed during the constant speed phase. These represent the end of acceleration, the end of constant speed, and the total time of motion, respectively.

8. The servo press control method based on multimodal sensing according to claim 1, characterized in that, The basic parameters of the PID controller are obtained in advance using the Ziegler-Nichols tuning method; the PID controller is set to pure proportional control; under the set operating conditions, the proportional coefficient is gradually increased until critical oscillation occurs, and the corresponding critical gain Ku and oscillation period Tu are recorded; the baseline PID parameters are calculated according to the empirical formula of the Ziegler-Nichols proportional coefficient; the parameter adjustment coefficients under the corresponding operating conditions are matched from the pre-built fuzzy control rule base according to the identified operating conditions, and the baseline PID parameters are adjusted using the parameter adjustment coefficients to obtain the target proportional coefficient, integral coefficient, and derivative coefficient.

9. A servo press control system based on multimodal sensing, implemented based on the servo press control method based on multimodal sensing as described in any one of claims 1-8, characterized in that, include: Transmission module, execution module, control module, and sensing module; The sensing module collects multimodal data reflecting the operating status of the servo press in real time. The transmission module is responsible for efficiently and reliably transmitting the multimodal data collected by the sensing module in real time and for interaction between the various modules. The control module adopts a collaborative control architecture of PLC master station, motion controller and host computer to realize a servo press control method based on multimodal perception; The execution module is responsible for receiving instructions from the control module and completing the corresponding actions.

10. The servo press control system based on multimodal sensing according to claim 9, characterized in that, The transmission module adopts a hybrid communication of industrial Ethernet and CAN bus. The industrial Ethernet is used to connect the host computer, PLC master station and motion controller, carrying key data streams with high bandwidth and low latency requirements to ensure rapid interaction of control commands and status feedback. The CAN bus is used to connect distributed vibration sensors, safety modules and low-speed I / O devices to form the underlying field network.

Images