Low-carbon steel cold extrusion forming production line and forming dynamic pressure compensation process thereof

Abstract

The present application relates to the technical field of cold extrusion forming, and particularly relates to a low-carbon steel cold extrusion forming production line and a forming dynamic pressure compensation process thereof. The low-carbon steel cold extrusion forming production line comprises a feeding conveying line, a hydraulic machine main body composed of an upper cross beam, a movable beam and a lower cross beam, a hydraulic system, a main cylinder, an extrusion mechanism, a finished product conveying line and an upper computer, the upper computer is in communication connection with the feeding conveying line, the hydraulic machine main body, the finished product conveying line and the extrusion mechanism, the feeding conveying line and the finished product conveying line are arranged on one side of the hydraulic machine main body, the hydraulic machine main body is internally provided with the extrusion mechanism, and the extrusion mechanism comprises a lower die rotating mechanism, a pressure head switching mechanism and a stress base. The cold extrusion forming dynamic pressure compensation process is used for accurately compensating pressure, realizing high-precision closed-loop control of pressure compensation and enhancing self-adaptive capacity.

Description

Technical Field

[0001] This invention relates to the field of cold extrusion molding technology, and in particular to a low-carbon steel cold extrusion molding production line and its dynamic pressure compensation process. Background Technology

[0002] In the metal processing and manufacturing field, low-carbon steel cold extrusion forming technology is widely used in the production of precision parts in industries such as automobiles, machinery and equipment, and hardware products due to its advantages such as improving billet utilization and enhancing the mechanical properties of parts. However, the currently known low-carbon steel cold extrusion forming operation mode has many prominent problems that restrict the development of the industry.

[0003] In traditional production processes, from loading the billet to unloading the finished product after extrusion molding, a large number of steps rely on manual operation. Workers need to frequently move heavy low-carbon steel billets and load and unload molds, resulting in extremely high labor intensity.

[0004] During cold extrusion, uneven deformation of low-carbon steel billets can easily lead to quality problems. Traditional constant pressure edge force control cannot adapt to the stress requirements of low-carbon steel at different deformation stages. Edge areas often crack due to insufficient pressure or excessive deformation due to excessive pressure, resulting in fluctuations in the dimensional accuracy of parts.

[0005] In terms of production efficiency, the speed and precision of manual operation are limited by physical strength and energy, and problems such as deviation in the placement of blanks lead to equipment idling and waiting, poor process connection, and difficulty in improving the production rhythm of the entire production line, which cannot meet the needs of modern manufacturing for high-efficiency production. Summary of the Invention

[0006] The purpose of this invention is to solve the problems existing in the prior art, and to propose a low-carbon steel cold extrusion forming production line and its forming dynamic pressure compensation process.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a low-carbon steel cold extrusion forming production line, comprising a feeding conveyor line, a hydraulic press body consisting of an upper crossbeam, a movable beam and a lower crossbeam, a hydraulic system and a main cylinder, an extrusion mechanism, a finished product conveyor line and a host computer, wherein the host computer is communicatively connected to the feeding conveyor line, the hydraulic press body, the finished product conveyor line and the extrusion mechanism, characterized in that the feeding conveyor line and the finished product conveyor line are arranged on one side of the hydraulic press body, and the extrusion mechanism is arranged inside the hydraulic press body, the extrusion mechanism comprising a lower die rotation mechanism, a pressure head switching mechanism and a force-bearing base;

[0008] The pressure head switching mechanism includes a push cylinder, a pad plate, a slide plate, a first pressure head, and a second pressure head. The output end of the push cylinder is fixedly connected to one side of the slide plate, and the first and second pressure heads are fixedly connected to the lower surface of the slide plate. The slide plate slides on the lower surface of the pad plate, and the pad plate is fixedly connected to the bottom of the movable beam.

[0009] The lower mold rotation mechanism includes a rotation mechanism and at least one lower mold. The rotation mechanism drives the lower mold to rotate. When the lower mold is pressed, it corresponds to the force-bearing base, the first pressure head, or the second pressure head.

[0010] The pressure head switching mechanism is used to perform the switching action when the lower mold rotates to the target station.

[0011] Furthermore, the lower mold includes a mold cavity, a mold frame, and an ejector rod, with the mold cavity corresponding to the pressure head switching mechanism;

[0012] There are four lower molds, namely the first lower mold, the second lower mold, the third lower mold and the fourth lower mold. The first lower mold, the second lower mold, the third lower mold and the fourth lower mold are arranged in a ring and can be pressed by the pressure head switching mechanism.

[0013] Furthermore, a dynamic pressure compensation process for cold extrusion forming, applied to the aforementioned low-carbon steel cold extrusion forming production line, includes the following process steps:

[0014] S1. Data Acquisition and Preprocessing;

[0015] S2, Generate the theoretical extrusion pressure curve;

[0016] S3. Monitor the actual pressure and calculate the pressure deviation;

[0017] S4. Control the output compensation amount of the fuzzy neural network compensation model;

[0018] S5, PID adaptive parameter dynamic adjustment;

[0019] S6, Dynamic response steps of the hydraulic system S5;

[0020] S7. Evaluation and iterative optimization of compensation effect.

[0021] Furthermore, data acquisition in S1 includes collecting the physical parameters of the billet, including billet density ρ, billet hardness HV, and billet yield strength σ. s The real-time temperature of the billet T, the geometric parameters of the lower die, including the extrusion ratio λ and the die angle α, and the equipment status parameters, including the stiffness K of the hydraulic system. s Pump flow rate Q;

[0022] S1 also includes establishing a constitutive relation model of the billet. Where σ is the flow stress of the billet, σ0 is the strength coefficient of the billet, n is the strain hardening index of the billet, C is the strain rate sensitivity coefficient of the billet, ε is the equivalent strain of the billet, and ε´ is the equivalent strain rate of the billet.

[0023] Furthermore, in S2, the theoretical extrusion pressure P is calculated based on a multi-field coupling model. th The calculation formula is: Where K1 is the triaxial compressive stress correction factor, σ s λ is the yield strength of the billet, λ is the extrusion ratio of the billet, K2 is the nonlinear exponent of the extrusion ratio, μ is the friction coefficient of the contact surface between the lower die and the billet, α is the die angle of the lower die, ΔT is the change in billet temperature, and K3 is the temperature sensitivity coefficient.

[0024] Furthermore, in step S3, the actual pressure P is collected in real time by a pressure sensor embedded in the first or second pressure head. act And calculate the pressure deviation ΔP between the actual pressure and the theoretical extrusion pressure. The calculation formula is: , where P th P represents the theoretical compressive pressure. act This represents the actual pressure; the formula for calculating the rate of change of its pressure deviation is: , where ΔP(t) represents the pressure deviation at the current moment, ΔP(t-1) represents the pressure deviation at the previous moment, and Δt represents the sampling time interval.

[0025] Furthermore, in S4, the pressure deviation ΔP, the pressure deviation change rate ΔP′, and the billet yield strength σ are used as the parameters. s The extrusion ratio λ of the billet and the real-time temperature T of the billet are used as input layer variables of the fuzzy neural network compensation model.

[0026] Furthermore, in step S6, the hydraulic system receives the PID control signal from step S5 and adjusts the valve opening degree. The calculation formula is: ,in, For the initial aperture, K q ΔU is the valve gain, and ΔU is the compensation amount.

[0027] Furthermore, in S7, the compensation effect is judged based on the evaluation index. If the requirements are not met, the model parameters or control strategy are optimized in reverse to form a closed-loop evolution.

[0028] The evaluation metrics include Mean Absolute Percentage Error (MAPE) assessment and dynamic performance index assessment.

[0029] Based on the actual pressure data collected by S3 and the theoretical extrusion pressure curve of S2, the average absolute percentage error is calculated using the following formula:

[0030] When the mean absolute percentage error (MAPE) exceeds a threshold, the model parameter self-learning mechanism is triggered to update the billet constitutive model and neural network weights.

[0031] Where: P thi It is the theoretical extrusion pressure value of the i-th sampling point (from the theoretical extrusion pressure curve in step S2);

[0032] P acti It is the actual extrusion pressure value of the i-th sampling point (collected from the pressure sensor in step S3);

[0033] i is the index of the sampling point (from 1 to N);

[0034] N is the total number of sampling points involved in the calculation (i.e., the number of pressure data collected during the evaluation period).

[0035] Furthermore, the threshold for Mean Absolute Percentage Error (MAPE) is set to 3%. When the MAPE is greater than 3%, the formula for calculating the neural network weights updated by the model parameter self-learning mechanism is as follows:

[0036] ,in,

[0037] Indicates the current weight.

[0038] η represents the learning rate.

[0039] δ j Indicates the error term.

[0040] Indicates input layer variables,

[0041] α represents the learning rate.

[0042] This represents the weight connecting the "i-th neuron in the input layer" and the "j-th neuron in the hidden layer" in the (t+1)-th iteration of the neural network.

[0043] When dynamic performance indicators are not up to standard, prioritize optimizing PID parameters.

[0044] Compared with existing technologies, the low-carbon steel cold extrusion forming production line and its dynamic pressure compensation process provided by this invention have the following advantages:

[0045] 1. High degree of automation: It uses feeding conveyor lines and finished product conveyor lines to transport materials, and four-station cyclic pressing. The pressing head switching mechanism can switch pressing heads at will, saving labor costs.

[0046] 2. Achieve high-precision closed-loop control, with the MAPE threshold accurately covering the minimum error, locking the pressure control accuracy within 3%, thus solving the problem of pressure deviation caused by billet fluctuation and lower die wear;

[0047] 3. Enhanced adaptive capability: Based on real-time collected data on pressure deviation ΔP, deviation change rate ΔP', and billet yield strength σ sThe system dynamically optimizes fuzzy rule weights and PID control parameters based on multiple dimensions such as extrusion ratio λ and billet temperature T, enabling adaptive control during cold extrusion without frequent manual intervention to adapt to different billet specifications.

[0048] 4. High practicality: It adopts a modular design and can directly reuse existing production line hardware such as pressure sensors and electro-hydraulic servo valves. It can achieve intelligent transformation through algorithm upgrades and reduce transformation costs. At the same time, it is equipped with an interactive interface to support manual intervention and rule import, which takes into account both algorithm autonomy and process flexibility, making it easy to promote and apply on existing cold extrusion production lines. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall invention;

[0050] Figure 2 This is a side view of the present invention;

[0051] Figure 3 This is a three-dimensional schematic diagram of the lower mold rotation mechanism of the present invention;

[0052] Figure 4 for Figure 3 Top view;

[0053] Figure 5 This is a side view of the lower mold rotation mechanism of the present invention;

[0054] Figure 6 This is a cross-sectional view of the mold rotation mechanism of the present invention;

[0055] Figure 7 This is a schematic diagram of the gear drive mechanism of the present invention;

[0056] Figure 8 This is a schematic diagram of the pressure head switching mechanism of the present invention;

[0057] Figure 9 This is a schematic diagram of the process flow of the present invention;

[0058] Figure 10 This is a schematic diagram of the framework process of the present invention;

[0059] Figure 11 This is a schematic diagram of the framework process for step S7;

[0060] In the diagram: 2. Host computer; 3. Feeding conveyor line; 4. Feeding robot; 5. Hydraulic press body; 6. Unloading robot; 7. Finished product conveyor line; 9. Lower mold rotation mechanism; 10. Press head switching mechanism; 11. Force-bearing base; 12. Rotating platform; 13. Gear drive mechanism; 131. Motor; 132. Drive gear; 133. Driven push block; 134. Push block cylinder; 135. Driven gear; 136. Driven push block; 137. Connecting gear; 138. Base; 14. Rotary gear disk; 15. First lower mold; 151. Mold cavity; 152. Ejector rod; 153. Pad block; 16. Second lower mold; 17. Third lower mold; 18. Fourth lower mold; 19. Push cylinder; 20. Pad plate; 21. Slide plate; 22. First press head; 23. Second press head. Detailed Implementation

[0061] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. At the same time, the terms "first," "second," etc., are only used to distinguish the names of various components and do not have a primary or secondary relationship. Therefore, they should not be construed as limitations on this invention.

[0063] Example 1: In this example, the present invention provides a low-carbon steel cold extrusion forming production line. Please refer to the accompanying drawings. Figures 1-8 As shown, the system includes a feeding conveyor line 3, a hydraulic press body 5 consisting of an upper crossbeam, a movable beam, a lower crossbeam, a hydraulic system, and a main cylinder, an extrusion mechanism, a finished product conveyor line 7, and a host computer 2. The host computer 2 is communicatively connected to the feeding conveyor line 3, the hydraulic press body 5, the finished product conveyor line 7, and the extrusion mechanism. The feeding conveyor line 3 and the finished product conveyor line 7 are located on one side of the hydraulic press body 5. The extrusion mechanism is installed inside the hydraulic press body 5. The extrusion mechanism includes a lower mold rotation mechanism 9, a pressure head switching mechanism 10, and a force-bearing base 11. The force-bearing base 11 is fixedly installed on the upper plane of the lower crossbeam of the hydraulic press. An ejector cylinder is also provided below the movable beam. The ejector cylinder is used to lift the movable beam away from the lower mold rotation mechanism 9.

[0064] The pressure head switching mechanism 10 includes a push cylinder 19, a pad 20, a slide plate 21, a first pressure head 22, and a second pressure head 23. The output end of the push cylinder 19 is fixedly connected to one side of the slide plate 21. The first pressure head 22 and the second pressure head 23 are fixedly connected to the lower surface of the slide plate 21. The slide plate 21 slides on the lower surface of the pad 20. The pad 20 is fixedly connected below the movable beam.

[0065] The lower mold rotating mechanism 9 includes a rotating mechanism and at least one lower mold. The rotating mechanism drives the lower mold to rotate. When the lower mold is pressed, it corresponds to the force-bearing base 11, the first pressing head 22, or the second pressing head 23.

[0066] The pressure head switching mechanism 10 is used to perform a switching action when the lower mold rotates to the target station.

[0067] The first pressure head 22 and the second pressure head 23 are used alternately to extend the service life of the pressure heads. Different pressure heads can also be replaced to perform pressing operations on different lower molds. The target station is the pressing station inside the hydraulic press body 5. When switching between the first pressure head 22 and the second pressure head 23, the hydraulic cylinder 19 is pushed to push the slide plate 21 forward or backward, thus switching between the first pressure head 22 and the second pressure head 23, so that the first pressure head 22 and the second pressure head 23 correspond to the lower mold for pressing operations.

[0068] The lower mold includes a mold cavity 151, an ejector rod 152, and a mold frame 154. The mold cavity 151 corresponds to the pressure head switching mechanism 10. The ejector rod 152 is provided inside the mold frame 154 and corresponds to the mold cavity 151. A pad block 153 is also provided around the ejector rod. The pad block 153 is placed between the mold frame 154 and the mold cavity 151 to enhance the stability of the mold cavity 151.

[0069] There are four lower molds, such as Figure 4 As shown, the structures are identical, namely the first lower mold 15, the second lower mold 16, the third lower mold 17, and the fourth lower mold 18. The first lower mold 15, the second lower mold 16, the third lower mold 17, and the fourth lower mold 18 rotate sequentially and are pressed by the corresponding pressure head switching mechanism 10. Specifically, during the pressing process, for example, when the first lower mold 15 is pressing, the second lower mold 16 rotates to the demolding mechanism, and at the same time, the fourth lower mold 18 rotates to the side of the feeding robot 4 for feeding.

[0070] On one side of the force-bearing base 11, a demolding mechanism is also fixedly installed. The demolding mechanism includes a demolding cylinder, and the extended end of the demolding cylinder is fixedly connected to the bottom of the ejector rod 152. The demolding mechanism demolds the finished products in the first lower mold 15, the second lower mold 16, the third lower mold 17 and the fourth lower mold 18 respectively.

[0071] In this embodiment, the rotating mechanism includes a rotating platform 12 and a gear drive mechanism 13, such as... Figure 7 As shown, which belongs to the prior art, the rotating platform 12 has multiple worktables for placing lower molds evenly distributed around its periphery. The rotating platform 12 has a rotary gear disk 14 with a gear 1 at its lower part, and the rotary gear disk 14 meshes with the gear drive mechanism 13. The rotating platform 12 rotates and operates in cycles under the drive of the gear drive mechanism 13.

[0072] The gear drive mechanism 13 includes a motor 131, a reducer, a base 138, and a gear set. The gear set is mounted on the base 138 and includes a driving gear 132, a driven gear 135, and a connecting gear 137. The motor 131 is connected to the reducer, and the reducer drive shaft is connected to the driving gear 132. The driving gear 132 meshes with the driven gear 135, and the driven gear 135 meshes with the connecting gear 137.

[0073] The top of the drive gear 132 is fixedly connected to the drive push block 133, and the two ends of the drive push block 133 are fixedly provided with push block cylinders 134. The top of the driven gear 135 is fixedly provided with a driven push block 136.

[0074] Driven gear 135 is an intermittent gear. The intermittent surface of the intermittent gear is aligned with the extension direction of driven push block 136, and the height of the intermittent surface of the intermittent gear is equal to the height of driving gear 132.

[0075] The interval time of driven gear 135 is the time during which the moving beam presses the blank in the lower mold.

[0076] The driven push block 136 is shaped like a "racket". The arc surface of the "racket" handle is adapted to the outer arc surface of the push block cylinder 134. The distance between the end faces of the two push block cylinders 134 is greater than the maximum diameter of the driven push block 136.

[0077] When motor 131 drives the driving gear 132 to rotate, the driving push block 133 will also rotate, and the driven push block 136 will also be driven to rotate by the driven gear 135. Because the direction of the driven push block 136 is the same as the direction of the intermittent surface of the intermittent gear, when the driven push block 136 rotates between the two push block cylinders 134, the intermittent surface of the driven gear 135 will not be meshed with the gear, and the driven gear 135 will stop rotating. The connecting gear 137 will also stop rotating. When it is on the left... When the pusher cylinder 134 rotates to the arc-shaped surface at the root of the driven pusher 136 under the drive gear 132, the driven pusher 136 is pushed by the pusher cylinder 134. As the two pusher cylinders 134 on the drive gear 132 continue to rotate, the driven pusher 136 drives the driven gear 135 to rotate. In this way, the driven gear 135 meshes with the drive gear 132 again, which in turn drives the connecting gear 137 to rotate. The connecting gear 137 then drives the rotary gear disk 14 to rotate. The two pusher cylinders 134 on the drive gear 132 alternately push the driven pusher 136, thus achieving uninterrupted cyclic operation.

[0078] In this embodiment, both the feeding conveyor line 3 and the finished product conveyor line 7 are chain plate conveyor lines. The feeding and unloading can be carried out by robots to save labor costs.

[0079] The implementation steps of this production line are as follows: Start the rotating mechanism; the first lower mold 15, the second lower mold 16, the third lower mold 17, and the fourth lower mold 18, as follows: Figure 2 , Figure 4 As shown, the loading robot 4 loads the blank into the first lower mold 15. The rotating mechanism drives the first lower mold 15 to rotate to the pressing station inside the hydraulic press body 5. The fourth lower mold 18 rotates to the side of the loading robot 4, and the loading robot 4 loads the blank into the fourth lower mold 18. At the same time, the first lower mold 15 is above the force-bearing base 11. The main cylinder drives the movable beam to move downwards, and the movable beam drives the first pressure head 22 to press. After pressing, the ejector cylinder below the movable beam lifts the movable beam, and the rotating mechanism rotates again. The first lower mold 15 rotates clockwise to the position of the second lower mold 16. At the same time, the fourth lower mold 18 rotates to the pressing station inside the hydraulic press body 5 for pressing operations. The rotating mechanism rotates again. The first lower mold 15 rotates to the position of the third lower mold 17, which corresponds to the demolding mechanism. The finished product in the first lower mold 15 is now above the demolding mechanism. The demolding cylinder pushes it out for demolding. The unloading robot 6 takes out the finished product and places it on the finished product conveyor line 7. The rotating mechanism rotates again, and the first lower mold 15 rotates to the side of the loading robot 4. The loading robot 4 puts the blank into the first lower mold 15. At the same time, the fourth lower mold 18 rotates to the position above the demolding mechanism. After demolding, the unloading robot 6 takes out the finished product from the fourth lower mold 18 and places it on the finished product conveyor line 7. The rotating mechanism rotates again to bring the first lower mold 15 into the hydraulic press body 5 for pressing. This process is repeated until all products are pressed.

[0080] Example 2, as Figure 9 and Figure 10 As shown, a dynamic pressure compensation process for cold extrusion molding is applied to the aforementioned low-carbon steel cold extrusion molding production line. To improve the cold extrusion molding effect of low-carbon steel, a host computer 2 and a PLC controller are provided. The host computer and the PLC controller work together to control the compensation process. The PLC controller has a built-in PID algorithm module and includes an input module, a processing module, and an output module. Specifically, the process includes the following steps:

[0081] S1. Data Acquisition and Preprocessing;

[0082] S2, Generate the theoretical extrusion pressure curve;

[0083] S3. Monitor the actual pressure and calculate the pressure deviation;

[0084] S4. Control the output compensation amount of the fuzzy neural network model;

[0085] S5, PID adaptive parameter dynamic adjustment;

[0086] S6, Dynamic response steps of the hydraulic system S5;

[0087] S7. Evaluation and iterative optimization of compensation effect.

[0088] In step S1, data acquisition includes collecting the physical parameters of the billet, including billet density ρ, billet hardness HV, and billet yield strength σ. s The real-time temperature T of the billet is an inherent property of the billet, and can be measured by an infrared thermometer installed on the feeding conveyor line. The geometric parameters of the lower die are collected, including the extrusion ratio λ, die angle α, and equipment status parameters. These geometric parameters are fixed design parameters of the lower die. The data also includes the hydraulic system stiffness K. s Pump flow rate Q; hydraulic system stiffness K s The flow rate Q can be calculated based on the hydraulic cylinder diameter, piston rod diameter, and hydraulic fluid volumetric elastic modulus. The pump flow rate Q is obtained from the output oil circuit or main oil circuit of the main pump. These basic data are pre-set in the host computer and then sent to the PLC controller via the host computer 2.

[0089] This embodiment calculates the extrusion ratio λ for Q235 low-carbon steel shaft parts (initial diameter ϕ40mm, target finished product diameter ϕ20mm). The extrusion ratio is defined as the ratio of the initial cross-sectional area A0 of the billet to the cross-sectional area A1 of the finished product. The calculation formula is as follows: ,

[0090] It adopts a forward cold extrusion process and is formed by a 5000T hydraulic press. The lower die has a conical structure and the lubrication method is MoS2 dry film lubrication.

[0091] Billet density ρ, billet hardness HV, billet yield strength σ s The values ​​can be directly read from the blank properties, and the mold geometry parameters are determined by the design drawings. For example, the lower mold angle α is 45°, and the hydraulic system stiffness K... s The pump flow rate Q is measured by dynamic testing of the hydraulic system (step pressure method), and K is obtained. s =1.2×10 9 N / m,

[0092] For the dynamic response PID algorithm module in step S6, the rated pump flow rate of the hydraulic pump is Q = 60 L / min = 1 × 10⁻⁶. −3 m 3 / s, an infrared thermometer is installed on the side of the lower mold to detect the temperature change of the billet in real time. After the data is collected, it is preprocessed to establish a constitutive model of the billet. Where σ is the flow stress of the billet, σ0 is the strength coefficient of the billet, n is the strain hardening index of the billet, C is the strain rate sensitivity coefficient of the billet, ε is the equivalent strain of the billet, and ε´ is the equivalent strain rate of the billet.

[0093] We assume the strength coefficient σ0 = 500 MPa.

[0094] The strain hardening index n = 0.15, and the strain rate sensitivity coefficient C = 0.03.

[0095] In cold extrusion, the equivalent strain ε is approximately equal to the extrusion ratio ε = lnλ. Therefore, ε = ln4 ≈ 1.386.

[0096] Assuming the extrusion speed v = 10 mm / s and the blank thickness at the die inlet h = 5 mm, then:

[0097] ,

[0098] Finally, these data are substituted into the billet constitutive relation model:

[0099] .

[0100] The methods for obtaining the billet constants in the billet constitutive model include: determining the strain rate of low-carbon steel at 10 by uniaxial tensile testing. −3 ~10 2 s −1The stress-strain curves within the range were fitted using the nonlinear least squares method to obtain the strength coefficient σ0, strain hardening index n, and strain rate sensitivity coefficient C of the billet, where σ0 ranges from 300 to 600 MPa, n is 0.1 to 0.2, and C is 0.01 to 0.05.

[0101] S2. Generate the theoretical extrusion pressure curve. First, calculate the theoretical extrusion pressure P based on a multi-field coupling model. th The calculation formula is:

[0102] Where K1 is the triaxial compressive stress correction factor, σ s λ is the yield strength of the billet, λ is the extrusion ratio of the billet, K2 is the nonlinear exponent of the extrusion ratio, μ is the friction coefficient of the contact surface between the lower die and the billet, α is the die angle of the lower die, ΔT is the change in billet temperature, and K3 is the temperature sensitivity coefficient.

[0103] In cold extrusion, the billet is under triaxial compressive stress (extended from all sides by the lower die), while the yield strength...

[0104] σ s It was measured by a uniaxial tensile test (uniaxial stress). Triaxial compressive stress would make the actual deformation resistance of the billet much greater than σ. s K1 is used to quantify this "strengthening effect of stress state".

[0105] By calculating the flow stress σ of the billet, the correction factor of the theoretical pressure formula can be determined. σ is the "calibration source" for the correction factor. (The last sentence appears to be incomplete and possibly refers to a different calculation method.) s Under triaxial compressive stress, the flow stress σ≈541.5MPa≈2.3σ s (σ) s (≈235MPa), first assume K2=0.7, then substitute into the theoretical extrusion pressure formula to verify, given that P is measured by simulation experiment. th ≈770MPa (theoretical cold extrusion pressure), λ=4, μ=0.1, α=45°, (cotα=1), ΔT=30°, K3=−0.002, from the formula The calculation yields K1=1.2. When K1=1.2, K1*σ s =282MPa, close to σ / 2 (541.5 / 2≈270.75), indicating that K1 only corrects for the partial strengthening of triaxial stress (the remaining strengthening is covered by the correction for strain hardening by K2); K1 is the tube stress state, K2 is the tube strain hardening, and K3 is the tube temperature. This "step-by-step correction" balances the complexity and accuracy of the formula. The calibration of the process correction coefficients K1, K2, and K3 depends on the calculation of the billet constitutive relation model σ.

[0106] The cold extrusion process is simultaneously affected by multiple factors such as triaxial compressive stress, strain hardening, temperature change, and friction. If a single coefficient is used to correct all factors, the formula will be too complex to be applied in engineering. Stepwise correction (K1 tube stress state, K2 tube strain hardening, K3 tube temperature) simplifies the formula and can cover the main influencing factors through the division of labor of each coefficient.

[0107] Substitute K1, K2, and K3 into the formula. The theoretical extrusion pressure can then be calculated.

[0108] In actual cold extrusion production, the operating conditions are diverse (e.g., changing the billet from Q235 to 304 stainless steel, adjusting the extrusion ratio from 4 to 5, and changing the die temperature). If relying solely on "simulation experiments," each change in operating conditions requires retesting (taking several hours or even longer). However, with a calibrated theoretical formula, only the billet yield strength and extrusion ratio need to be replaced, allowing for the calculation of the theoretical extrusion pressure for the corresponding operating condition in seconds, significantly improving engineering efficiency. This makes the theoretical extrusion pressure calculation formula universally applicable to multiple operating conditions.

[0109] The required pressure during cold extrusion changes dynamically, mainly for the following reasons:

[0110] Initial stage: As soon as the billet comes into contact with the lower mold, it begins to undergo plastic deformation, and the deformation resistance rises rapidly from zero.

[0111] Stable extrusion stage: The billet continues to pass through the lower die, and the degree of deformation (equivalent strain ε of the billet) reaches its maximum value and remains stable. However, due to the strain hardening effect of the billet, the flow stress will still increase slowly, causing the required pressure to rise slowly.

[0112] Final stage: The remaining length of the billet is very short, and deformation begins to extend to the entire remaining billet. The shape of the deformation zone changes, causing fluctuations in pressure.

[0113] Changes in friction conditions: In the initial stage of extrusion, the MoS2 dry film may not be evenly spread, resulting in greater friction.

[0114] As extrusion proceeds, the increased temperature of the lower die may cause changes in the properties of the MoS2 dry film, and the frictional force will also change accordingly.

[0115] Changes in contact area: For positive extrusion, as the first pressure head 22 or the second pressure head 23 advances, the contact area between the billet and the lower die also changes, which directly affects the total friction force.

[0116] Therefore, the pressure changes dynamically during the extrusion process. The host computer 2, using the input billet parameters, lower die parameters, and a preset constitutive model σ, calculates and predicts the theoretical extrusion pressure curve—a continuous theoretical extrusion pressure curve—required to complete the entire extrusion process under current conditions, varying with time or displacement. The specific process is as follows:

[0117] 1. The instantaneous strength (flow stress σ) of the billet during dynamic deformation is calculated using a constitutive model of the billet. The calculation formula is as follows: .

[0118] 2. Establish the relationship with displacement / time:

[0119] Equivalent strain ε: During extrusion, the equivalent strain ε has a direct mathematical relationship with the instantaneous billet length or master cylinder displacement S (ε increases from 0 to lnλ). Therefore, ε is a function of displacement S: ε = f(S)

[0120] Equivalent strain rate ε': related to extrusion speed v. When the extrusion speed v remains constant, ε' can be expressed as a function of the main cylinder displacement S.

[0121] 3. Iterative calculation to generate curves:

[0122] The host computer discretizes the entire extrusion stroke (from displacement S=0 to S=maximum value) into tens of thousands of points.

[0123] For each displacement point S i The process of generating the theoretical extrusion pressure curve is as follows:

[0124] a. Calculate the strain ε at this location. i and strain rate ε' i ;

[0125] b. ε i and ε' i Substituting the blank into the constitutive model, the instantaneous flow stress σ at that point is calculated. i ;

[0126] c. σ i Substituting into the theoretical pressure formula, we get: ,

[0127] This gives us the displacement point S. i Corresponding theoretical extrusion pressure value .

[0128] All displacement points S i and the calculated theoretical extrusion pressure value When connected, they form the final theoretical extrusion pressure curve.

[0129] This curve is then sent to the PLC controller as a benchmark for real-time pressure monitoring and deviation calculation in S3.

[0130] In S3, the actual pressure P is collected in real time by the pressure sensor embedded in the first pressure head 22 or the second pressure head 23. act And calculate the pressure deviation ΔP between the actual pressure and the theoretical extrusion pressure. The calculation formula is: , where P th P represents the theoretical extrusion pressure value. act This represents the actual pressure value; the formula for calculating the rate of change of its pressure deviation is: , where ΔP(t) represents the pressure deviation at the current moment, ΔP(t-1) represents the pressure deviation at the previous moment, and Δt is the sampling time interval.

[0131] In this embodiment, the pressure sensor is a strain gauge pressure sensor, installed between the first pressure head 22 or the second pressure head 23 and the contact surface of the mold cavity 151. The strain signal is converted into voltage by a charge amplifier. The PLC controller input module receives the voltage signal after conversion by the charge amplifier and, based on the calibration curve of the pressure sensor (which is the "voltage value - pressure value" correspondence determined after the pressure sensor is calibrated at the factory or on-site), converts the voltage signal into the actual pressure value P. act The input module will input the actual pressure value P. act The data is transmitted to the processing module in the PLC controller. The processing module calculates the pressure deviation ΔP and the pressure deviation change rate ΔP' in real time as the basis for compensation decisions. The output module receives the signal from the processing module and uploads the pressure deviation ΔP and the pressure deviation change rate ΔP' to the host computer 2.

[0132] In S4, the fuzzy neural network compensation model outputs the compensation amount, and the host computer 2 outputs the pressure deviation ΔP, the pressure deviation change rate ΔP′, and the billet yield strength σ. s The extrusion ratio λ of the billet and the real-time temperature T of the billet are used as input layer variables of the fuzzy neural network compensation model, which is run by the fuzzy neural network compensation controller.

[0133] In this embodiment, the fuzzy neural network compensation controller communicates with the host computer 2 via an Ethernet interface. The fuzzy neural network compensation model includes an input layer, a data preprocessing module, a fuzzy inference module, and a rule base management module. The input layer receives data including the pressure deviation ΔP, which reflects the degree of deviation between the actual extrusion pressure and the theoretical pressure value; the pressure deviation change rate ΔP′, which reflects the trend and severity of pressure fluctuations; and the billet yield strength σ, which reflects the critical stress at which the billet begins plastic deformation. sThe extrusion ratio λ, which reflects the degree of billet deformation, and the physical parameter T, which reflects the real-time temperature of the billet during cold extrusion, are normalized and mapped to the [-1,1] interval by the data preprocessing module.

[0134] For example, if the pressure deviation ΔP ranges from [−50, 50] MPa, then after normalization... ,

[0135] If the real-time temperature T of the billet is in the range of [20, 150] °C, then after normalization... T´ represents the normalized temperature value.

[0136] The fuzzy inference module maps the input vector preprocessed by the data preprocessing module into control compensation quantities, including:

[0137] 1. Fuzzification layer: Converts precise values ​​in the data preprocessing module into fuzzy set membership degrees, i.e., the fuzzification result;

[0138] 2. Rule Reasoning Layer: Executes fuzzy rules based on data in the rule base management module to obtain fuzzy conclusions;

[0139] 3. Defuzzification layer: Converts fuzzy conclusions into precise control quantities. Using the centroid method (weighted average method), the "fuzzy conclusions" of multiple activation rules are merged into a precise compensation quantity ΔU.

[0140] The rule base management module includes a fuzzy rule base for matching and comparing with fuzzy set membership degrees. The fuzzy rule base employs a dual-redundant storage architecture (RAM real-time access + Flash persistent backup), storing multiple core fuzzy rules. Each fuzzy rule contains metadata such as a unique ID, prerequisite membership parameters, conclusion membership parameters, weight (0~1), cumulative activation count, and average error contribution. It supports efficient indexing, prioritizing input variables (pressure deviation ΔP, pressure deviation change rate ΔP′, billet yield strength σ). s A binary tree index is established to reflect the real-time temperature T of the billet during cold extrusion. At the same time, high-frequency triggering rules (the first 20 rules) are pre-stored in the fast lookup table to ensure that the rule call delay is ≤50μs, which meets the real-time requirements of 100Hz sampling frequency.

[0141] The rule base management module selects valid fuzzy rules (rules that match the current fuzzification result and have a qualified error contribution) from the fuzzy rule base. The fuzzy inference module performs rule inference based on the fuzzification result and valid fuzzy rules to obtain fuzzy conclusions. Subsequently, the fuzzy inference module uses the centroid method defuzzification algorithm to convert the fuzzy conclusions into precise compensation amount ΔU. Finally, the fuzzy inference module outputs the control compensation amount ΔU to the host computer 2.

[0142] S5. Dynamic adjustment of PID adaptive parameters; the host computer transmits the precise compensation amount ΔU to the PLC controller, and the PLC's built-in PID algorithm module adjusts the PID parameters according to the compensation amount ΔU, dynamically adjusting the PID parameter ratio (K). p ), integral (K) i ), differential (K) d The parameters are used to generate the control signal for the electro-hydraulic servo valve. For example, the calculated precise compensation amount ΔU = +12MPa, ΔU > 10MPa, is then used to calculate the output current signal I. I0 is the intermediate value, K I This is for calibrating the gain.

[0143] S6, Hydraulic system dynamic response step S5; Hydraulic system dynamic response PLC controller, electro-hydraulic servo valve receives control signal from PLC controller and adjusts valve opening degree. This drives the master cylinder pressure to rapidly approach the theoretical extrusion pressure value. The calculation formula is as follows: ,in, For the initial aperture, K q Let ΔU be the valve gain, and ΔU be the compensation amount. In this embodiment, let ΔU be the valve gain. K q Given 2%, ΔU = +2MPa, substitute these values ​​into the formula:

[0144] ,

[0145] As the opening of the electro-hydraulic servo valve increases from 30% to 34%, the hydraulic pump flow increases, and the main cylinder pressure rises rapidly, approaching the theoretical extrusion pressure value.

[0146] Assuming ΔU = -5MPa, calculate The opening of the electro-hydraulic servo valve is reduced from 30% to 20%, the hydraulic pump flow rate is reduced, and the master cylinder pressure drops (to eliminate overpressure).

[0147] The variable pressure sensor collects P every 10ms act The new pressure deviation and deviation change rate are calculated and used as the control input for the next round.

[0148] In the S7, as Figure 11 As shown, the compensation effect evaluation and iterative optimization includes judging the compensation effect based on the evaluation index. If the requirements are not met, the model parameters (bill constitutive relationship model and fuzzy neural network compensation model) or control strategy are optimized in reverse to form a closed-loop evolution. The compensation effect evaluation and iteration are run through the rule base management module, which also includes a rule evaluation module, a rule optimization submodule, a rule update submodule, a rule conflict resolution submodule, and an interaction submodule.

[0149] The rule evaluation module assesses whether each fuzzy rule helps reduce or increases deviation during the control process. It can quantify the effectiveness of fuzzy rules in real time and calculate the error contribution (EC) of each fuzzy rule. i The mean absolute percentage error (MAPE) is calculated to evaluate the fit between the theoretical extrusion pressure curve and the actual pressure curve, and to determine the accuracy of the billet constitutive model. For example, for every 10 products produced, the activation frequency (as a percentage of the total number of triggers) and error contribution EC of each fuzzy rule are calculated. i (The average pressure deviation when the fuzzy rule is triggered) is used to generate a validity score of 0-100 (a score of 80 or above is considered a "valid rule"). In this embodiment, the error contribution EC i The threshold is 8 MPa, and the mean absolute percentage error (MAPE) threshold is set to 3%.

[0150] The rule optimization submodule receives diagnostic results from the evaluation module and performs fine-tuning and optimization on poorly performing fuzzy rule sets. For example, it can streamline the fuzzy rule library, delete rules with validity scores <60, merge similar fuzzy rules (e.g., those with precondition vector similarity >85%) using a cosine similarity algorithm (threshold 0.85), and retain redundant fuzzy rules after weighted averaging. Another function is weight redistribution: normalizing the weights of remaining fuzzy rules according to their validity scores to ensure high error contribution coefficients (ECs). i The weight of rules (such as those with an error of <5MPa) is increased by 10% to 20%. If the performance of the hydraulic system still fails to meet the standards after optimization, a new fuzzy rule generation process is triggered.

[0151] The rule update submodule is used to update and upgrade the billet constitutive model and the fuzzy neural network compensation model.

[0152] For example: Automatic updates are triggered when the following conditions are detected: pressure fluctuation standard deviation > 8 MPa for 3 consecutive batches of products; billet yield strength σs deviation > 15% from historical value (e.g., when changing material grade); manual triggering via host computer (e.g., when changing mold). Incremental learning: Based on 500 sets of sample data (input variables + optimal compensation amount) for new conditions, mapping relationships are extracted through CART decision tree to generate 8-12 new rules. After simulation verification (error < 5 MPa), these rules are added to the fuzzy rule base.

[0153] The rule conflict resolution submodule is responsible for arbitration and ensuring that a unified and reasonable control command is output when multiple fuzzy rules are activated at the same time and their output conclusions contradict each other.

[0154] For example, during real-time monitoring and inference, if the same input vector triggers rules with opposite conclusions (such as ΔU=PB "significantly increased" and ΔU=NB "significantly decreased"), the rule conflict resolution submodule calculates the conflict intensity. ,in, This represents the "most aggressive compensation direction" among all fuzzy rules that trigger the same input. This represents the "most conservative compensation direction" among all triggering fuzzy rules under the same input. and The greater the difference, the more intense the contradiction between the rules. Calculating the conflict intensity transforms the "implicit contradictions" of fuzzy rules into "explicit values," supporting intelligent conflict handling in cold extrusion pressure control.

[0155] For example, during cold extrusion, the billet temperature rises sharply (T=120°C, much higher than normal), and at the same time the pressure deviation ΔP=+30MPa (the actual pressure value is higher than the theoretical extrusion pressure value).

[0156] Fuzzy rule triggering and conflict: Rule 1 and Rule 2 are two specific fuzzy rules stored in the fuzzy rule base.

[0157] Fuzzy Rule 1 (High Temperature + Positive Deviation): Conclusion (A significant reduction in pressure was required);

[0158] Fuzzy Rule 2 (New Lower Mold Rule Not Updated): Conclusion (The error message is "increase pressure");

[0159] The calculated conflict intensity is: C = |−20−15| = 35MPa (exceeding 20% ​​of full scale, triggering a high-conflict strategy).

[0160] When the error contribution EC2 of fuzzy rule 2 is detected to be 25MPa (far higher than the threshold of 8MPa), fuzzy rule 2 is temporarily disabled.

[0161] Only the conclusion of fuzzy rule 1 (ΔU) is executed. min =−20MPa), the pressure quickly dropped back to the theoretical extrusion pressure value;

[0162] After production is completed, the rule update submodule automatically learns the new working conditions and corrects the error in fuzzy rule 2.

[0163] Resolution strategy:

[0164] If the conflict intensity C ≤ 20% of the full scale (e.g., ≤ 20 MPa): take the weighted average compensation amount;

[0165] If the conflict intensity C > 20% at full scale: temporarily disable low-efficiency rules and enable backup fuzzy rules (based on historical best solutions).

[0166] Three consecutive conflicts trigger the rule update submodule to relearn.

[0167] The interaction submodule is used to send alarm information to the host computer 2 and pop up a human-machine interface to request engineers to manually adjust fuzzy rules or PID parameters. The host computer 2's HMI interface supports manual creation / import of fuzzy rules, which directly affect the control logic. It supports importing / exporting XML format rule packages, facilitating the replication of rule libraries from mature technologies. Rules can be added / deleted, and operation audit logs are recorded.

[0168] For example, when a new die for cold extrusion is put into production, the billet is a custom alloy (bill yield strength σ). s (fluctuation up to 20%), the initial algorithm rule controls the fluctuation to ±15MPa;

[0169] Manually add rules through the 2HMI interface on the host computer:

[0170] IF(ΔP = positive deviation, T = high temperature, λ = large deformation), THEN(ΔU = significant decompression, weight = 0.8).

[0171] After the rules were imported, the pressure fluctuation immediately dropped to ±5MPa, shortening the commissioning cycle.

[0172] The evaluation indicators simultaneously perform mean absolute percentage error (MAPE) assessment, fuzzy rule error contribution assessment, and dynamic performance indicator assessment.

[0173] In this embodiment, the threshold for Mean Absolute Percentage Error (MAPE) is set to 3%. When MAPE is greater than 3%, the calculation formula for updating the neural network weights by the self-learning mechanism is as follows:

[0174] ,in,

[0175] Indicates the current weight.

[0176] η represents the learning rate.

[0177] δ j Indicates the error term.

[0178] Indicates input layer variables,

[0179] α represents the learning rate.

[0180] This represents the weight connecting the "i-th neuron in the input layer" and the "j-th neuron in the hidden layer" in the (t+1)-th iteration of the neural network.

[0181] The Mean Absolute Percentage Error (MAPE) is calculated using the following formula:

[0182] When the mean absolute percentage error (MAPE) exceeds a threshold, the model parameter self-learning mechanism is triggered to update the billet constitutive model and neural network weights.

[0183] Where: P thi It is the theoretical extrusion pressure value of the i-th sampling point (from the theoretical extrusion pressure curve in step S2);

[0184] P acti It is the actual extrusion pressure value of the i-th sampling point (collected from the pressure sensor in step S3);

[0185] i is the index of the sampling point (from 1 to N);

[0186] N is the total number of sampling points involved in the calculation (i.e., the number of pressure data collected during the evaluation period).

[0187] When the mean absolute percentage error (MAPE) exceeds the threshold, the model parameter self-learning mechanism is triggered. The model includes a billet constitutive model and a fuzzy neural network compensation model. The fuzzy neural network compensation model also includes a neural network weight layer. Self-learning is achieved by adjusting the neural network weights to optimize the fuzzy rule weights or compensation accuracy and update the billet constitutive model and neural network weights.

[0188] Because the extrusion process is dynamic and continuous (pressure changes with time or stroke) during cold extrusion, it is necessary to collect pressure data at multiple moments (e.g., once every 0.1 seconds, for 10 seconds, then N=100). N determines the statistical range of the mean absolute percentage error (MAPE): the larger N is, the more complete the process stages are covered, and the better the mean absolute percentage error (MAPE) can reflect the overall actual pressure control accuracy.

[0189] Assuming that during the cold extrusion process, the theoretical extrusion pressure at 5 points is:

[0190] (Theoretical pressure curve as a function of stroke);

[0191] The pressure deviation is small, and the following is calculated:

[0192] ,

[0193] 2.49% < 3% (threshold), pressure control accuracy meets the standard, and the self-learning mechanism is not triggered.

[0194] like (The pressure deviation was too large at a certain stage, causing a sudden change in the hardness of the simulated billet).

[0195] ,

[0196] 3.33% > 3% (threshold), indicating insufficient pressure control precision, triggering a self-learning mechanism to update the billet constitutive relation model and neural network weights.

[0197] After the dynamic performance indicators have been evaluated and stabilized, they must meet the following conditions: |ΔP| ≤ 3MPa (qualified), response time ≤ 0.5s, and overshoot ≤ 5% (qualified) to be considered qualified.

[0198] Error contribution EC i ≤ 8MPa is considered acceptable, with an error contribution rate of EC i This represents the impact of a single fuzzy rule on the pressure deviation (unit: MPa). The influence of a single fuzzy rule is determined by its weight in the compensation decision. When multiple fuzzy rules are triggered, the fuzzy rule with the higher weight contributes more to the final compensation amount.

[0199] Once the validated strategies (a set of PID parameters and fuzzy rule weights) are verified as qualified, the fuzzy neural network compensation model saves these configurations for subsequent compensation under similar operating conditions, enabling strategy reuse and optimization.

[0200] If the error contribution EC i If the performance fails to meet the requirements, fuzzy rule optimization or new fuzzy rule generation is triggered. This is because when overall performance is acceptable but individual fuzzy rules perform poorly, the problem lies in the "knowledge base" of the intelligent compensation system itself, requiring fine-tuning. When dynamic performance indicators fail to meet the requirements, PID parameters are optimized first. This is because the PID is a fundamental controller, and its adjustment can usually quickly and effectively solve most common control problems.

[0201] If the optimization fails to meet the requirements after three consecutive attempts, an alarm will be triggered on the host computer. You can then manually adjust the rules or PID parameters.

[0202] The billet constitutive model can be updated using machine learning methods (such as regression analysis, Gaussian process models, etc.) to dynamically correct the parameters of the billet constitutive model.

[0203] Neural network weight updates can be optimized using reinforcement learning algorithms (such as Q-learning and Deep Deterministic Policy Gradient (DDPG)).

[0204] The training samples for the fuzzy neural network compensation model include: the input training vectors include pressure deviation ΔP, pressure deviation change rate ΔP′, and billet yield strength σ. s The extrusion ratio λ of the billet and the real-time temperature T of the billet are used as parameters. The output training vector includes the compensation coefficient K. c The training algorithm employs gradient descent with momentum term, with a learning rate η = 0.01~0.1, momentum factor b = 0.8~0.95, and error function... Where m is the sample size. This represents the compensation coefficient obtained by the model for the i-th sample. This represents the actual compensation coefficient corresponding to the i-th sample.

[0205] This dynamic pressure compensation process for cold extrusion molding can compensate for either the first pressure head 22 or the second pressure head 23, resulting in better forming effect of low carbon steel in cold extrusion molding.

[0206] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A cold extrusion forming dynamic pressure compensation process, comprising the following process steps: S1, data acquisition and preprocessing; S2, generating a theoretical extrusion pressure curve; S3, monitoring the actual pressure and calculating the pressure deviation; S4, controlling the fuzzy neural network compensation model to output the compensation amount; S5, PID adaptive parameter dynamic adjustment; S6, hydraulic system dynamic response step S5; S7, compensation effect evaluation and iterative optimization; Wherein, the data collection in S1 includes collecting the physical parameters of the blank, including the density of the blank ρ, the hardness of the blank HV, the yield strength of the blank σ s , the real-time temperature of the blank T, the geometric parameters of the lower die, including the extrusion ratio of the lower die λ and the die angle of the lower die α, and the device state parameters, including the stiffness of the hydraulic system K s , the pump flow Q; The S1 further comprises establishing a blank constitutive relation model Wherein, b is the flow stress of the blank, b0 is the strength coefficient of the blank, n is the strain hardening index of the blank, C is the strain rate sensitivity coefficient of the blank, ε is the equivalent strain of the blank, and ε' is the equivalent strain rate of the blank. In the S3, the actual pressure P is collected in real time by the pressure sensor embedded in the first pressure head (22) or the second pressure head (23) act , and the pressure deviation ΔP between the actual pressure and the theoretical extrusion pressure is calculated, and the calculation formula is: , wherein P th represents the theoretical extrusion pressure, P act represents the actual pressure; and the pressure deviation change rate calculation formula is: , wherein ΔP(t) represents the pressure deviation at the current time, ΔP(t-1) represents the pressure deviation at the previous time, and Δt represents the sampling time interval. In the S4, the pressure deviation ΔP, the pressure deviation change rate ΔP', the blank yield strength σ s , the extrusion ratio λ of the blank, and the real-time temperature T of the blank are used as the input layer variables of the fuzzy neural network compensation model. In the S6, the hydraulic system receives the PID control signal of the step S5, and adjusts the valve opening degree , the calculation formula is , wherein, is the initial opening degree, K q is the valve gain, and ΔU is the compensation amount.

2. A cold-swaging dynamic pressure compensation process according to claim 1, wherein, In the S2, the extrusion pressure P is calculated based on a multi-field coupling model th , and the calculation formula is , wherein K1 is a three-way pressure stress correction coefficient, σ s is the yield strength of the blank, λ is the extrusion ratio of the blank, K2 is a nonlinear index of the extrusion ratio, μ is the friction coefficient of the contact surface between the lower die and the blank, α is the die angle of the lower die, ΔT is the temperature change of the blank, and K3 is a temperature sensitivity coefficient.

3. A cold-swaging dynamic pressure compensation process according to claim 1, wherein, In S7, the compensation effect is judged based on the evaluation index, and if the requirement is not met, the model parameters or control strategy are optimized in reverse to form a closed loop evolution; The evaluation index includes the mean absolute percentage error MAPE evaluation and the dynamic performance index evaluation, Based on the actual pressure data collected in S3 and the theoretical extrusion pressure curve in S2, the mean absolute percentage error is calculated, and the calculation formula is: When the mean absolute percentage error MAPE exceeds the threshold value, the model parameter self-learning mechanism is triggered to update the blank constitutive model and neural network weights, wherein: P thi is the theoretical extrusion pressure value at the i-th sampling point (theoretical extrusion pressure curve from step S2); P acti is the actual extrusion pressure value of the i-th sampling point (acquired from the pressure sensor in step S3); i is the index of the sampling point (from 1 to N); N is the total number of sampling points participating in the calculation (i.e. the number of pressure data collected in the evaluation period).

4. A cold-swaging dynamic pressure compensation process according to claim 3, wherein, The threshold of the mean absolute percentage error MAPE is set to 3%, and when the mean absolute percentage error MAPE is greater than 3%, the calculation formula of the neural network weight value of the model parameter self-learning mechanism is updated as follows: wherein, denotes the current weight, η represents the learning rate, δ j denotes an error term, Indicates input layer variables, α represents the learning rate, Wt+1ijdenotes the weight of the connection between the "input layer i-th neuron" and the "hidden layer j-th neuron" at the t+1 iteration in the neural network.

5. A low-carbon steel cold extrusion forming production line for the cold extrusion dynamic pressure compensation process of any one of claims 1-4, comprising a feeding conveying line (3), a hydraulic machine main body (5) composed of an upper cross beam, a movable beam and a lower cross beam, a hydraulic system, a main cylinder, an extrusion mechanism, a finished product conveying line (7) and an upper computer (2), the upper computer (2) is in communication connection with the feeding conveying line (3), the hydraulic machine main body (5), the finished product conveying line (7) and the extrusion mechanism, characterized in that, The loading conveying line (3) and the finished product conveying line (7) are arranged on one side of the hydraulic machine main body (5), the hydraulic machine main body (5) is provided with an extrusion mechanism, and the extrusion mechanism comprises a lower die rotating mechanism (9), a pressure head switching mechanism (10) and a stress base (11). The pressure head switching mechanism (10) comprises a pushing oil cylinder (19), a backing plate (20), a sliding plate (21), a first pressure head (22) and a second pressure head (23), the output end of the pushing oil cylinder (19) is fixedly connected to one side of the sliding plate (21), the lower plate surface of the sliding plate (21) is fixedly connected to the first pressure head (22) and the second pressure head (23), the sliding plate (21) slides on the lower plate surface of the backing plate (20), and the backing plate (20) is fixedly connected below the movable beam. The lower die rotating mechanism (9) comprises a rotating mechanism and at least one lower die, the rotating mechanism drives the lower die to rotate, and the lower die corresponds to the stress base (11), the first pressure head (22) or the second pressure head (23) when being pressed. The pressure head switching mechanism (10) is used to perform switching action when the lower die rotates to the target station; The lower die comprises a die cavity (151), a die frame and an ejection rod (152), and the die cavity (151) corresponds to the pressure head switching mechanism (10); The lower die has four, which are a first lower die (15), a second lower die (16), a third lower die (17) and a fourth lower die (18), and the first lower die (15), the second lower die (16), the third lower die (17) and the fourth lower die (18) are sequentially annular and can be pressed by the pressure head switching mechanism (10).

Images