A laser riveting apparatus and method

Abstract

This invention relates to a laser riveting processing apparatus and method. The apparatus comprises five functional units: a dual-wavelength composite laser source unit mounted on a frame unit, a galvanometer scanning and autofocus unit, a thermoplastic molding unit, and a four-degree-of-freedom precision motion platform. Two lasers, a semiconductor laser and a fiber laser, are coupled into a common optical path via dichroic mirrors. A CCD camera enters the coaxial optical path through a PBS beam splitter B. The coaxial optical path reflects the light information to the galvanometer scanning and autofocus unit via a mirror assembly. The advantages are that, based on the material differences of the material to be processed, a suitable laser source is selected. After softening the material through laser processing, the thermoplastic molding unit and the four-degree-of-freedom precision motion platform complete the molding process. This reduces the number of plastic riveting samples lost during the product debugging stage and improves production efficiency.

Description

Technical Field

[0001] This invention relates to the field of laser processing, specifically to a laser riveting processing apparatus and method for plastic riveting. Background Technology

[0002] Plastic riveting is a commonly used mechanical joining technique, primarily used to permanently or semi-permanently connect plastic parts to other materials (metals, plastics, etc.). Engineering plastics are the mainstream plastics used for riveting, with main types including PC, ABS, and PC+ABS. Plastic riveting technology features a regular riveting surface structure and excellent surface finish, and also boasts advantages such as advanced technology, environmental friendliness, and stable and reliable processing quality. It can be widely used in aerospace, military, electronics, and home appliances, among other fields.

[0003] The current mainstream types of plastic hot riveting technology in China mainly include ultrasonic riveting, electromagnetic riveting, hot air riveting and photothermal radiation riveting. The principles of various process technologies are as follows: (1) Ultrasonic riveting: The plastic parts are thermally melted and formed by the high-frequency vibration of ultrasonic waves acting on the riveting part; (2) Electric heating riveting: The copper pressure head is heated by electric heating, and the plastic is softened and pressed to form a joint through heat conduction; (3) Hot air riveting: It is a precision connection process for thermoplastic plastics and dissimilar materials such as metals and circuit boards. The plastic rivet is softened by non-contact heating with high-temperature hot air, and then stamped with a cold pressure head to achieve a highly reliable connection; (4) Photothermal radiation heating riveting: The connection technology uses the light radiation energy absorbed by the material and converted into heat energy to locally melt or soften the area to be connected, and forms a stable riveting structure under the action of mechanical pressure coupling.

[0004] Existing laser welding equipment has the following disadvantages: (1) Existing single laser light sources are difficult to handle the high-efficiency riveting processing and efficiency requirements of materials of different materials or thicknesses. (2) For different plastic riveting samples, conventional solutions require the development of different fixtures and are prone to damaging the products. (3) Traditional mold pressing units use pre-set pressure for riveting parts. Traditional plastic product riveting cannot extract and provide feedback on force and temperature, and cannot adjust the results of laser softening of plastic in real time. (4) The riveting positions of the parts are not in the same plane. Existing riveting pressing equipment requires the manufacture and design of multiple fixtures or multiple processing times to complete the riveting process, which will reduce the accuracy and efficiency of the entire riveting process. (5) Most equipment only stores data on product processing material type, material color, riveting structure shape, pressing pressure parameters, temperature parameters, laser power parameters, scanning speed and riveting joint image information. It cannot integrate and compare data to establish a process database, and cannot process data based on intelligent algorithms. Product debugging requires a long cycle. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention provides a laser riveting processing device and method.

[0006] This invention is achieved through the following technical solution: A laser riveting processing device is characterized by comprising five functional units, namely a dual-wavelength composite laser source unit installed on the equipment frame unit, a galvanometer scanning and autofocus unit, a thermoplastic molding unit, and a four-degree-of-freedom precision motion platform. The dual-wavelength composite laser source unit includes two lasers mounted on the upper part of the equipment frame unit, a dichroic mirror, a PBS beam splitter A, a PBS beam splitter B, a mirror assembly, a green light source, a CCD camera, and a driver board. Two lasers, a semiconductor laser and a fiber laser, are coupled into a common optical path via dichroic mirrors. The green-band light source enters the coaxial optical path through PBS beam splitter A to provide illumination for the CCD camera. The CCD camera enters the coaxial optical path through PBS beam splitter B. The coaxial optical path reflects the light information to the galvanometer scanning and autofocus unit through a mirror assembly.

[0007] More preferably, a 1064 nm fiber laser and a 1710 nm semiconductor laser are used; The output power of a 1064 nm fiber laser can be adjusted by changing the pump current through the built-in pump power supply, thereby changing the gain. The output power of a 1710 nm semiconductor laser can be adjusted by controlling the current in the driver board and the temperature control of the water-cooling plate. The 1710 nm semiconductor laser is a mid-infrared semiconductor laser. It has strong intrinsic absorption in most thermoplastics, including PMMA, PC, PET, and PP, and can heat the bulk and soften the surface without the need for an absorber. It is mainly used for transparent / thin parts. The 1064 nm fiber laser is a near-infrared laser source. It allows for more concentrated processing and a smaller heat-affected zone. It requires the use of an absorber, but it is prone to ablation at high power. It is mainly used for thick parts / reinforced plastics.

[0008] In a further preferred embodiment, the galvanometer scanning and autofocus unit is installed inside one side of the substrate support, and employs a galvanometer assembly and a Z-axis vertical linear motion module in cooperation; the galvanometer assembly is mounted on the Z-axis vertical linear motion module; a temperature sensor A is installed at the lower part of the galvanometer assembly; The upper part of the substrate support has holes for the light path to pass through. The substrate support is used to connect the dual-wavelength composite laser light source unit, the galvanometer scanning and autofocus unit and the thermoplastic molding unit. The galvanometer assembly moves up and down with the Z-axis vertical linear motion module, working in conjunction with the A and B sets of rotating components within the four-degree-of-freedom precision motion platform to achieve vertical focus. The laser processing focus can determine the distance to the focal point using the visual image sharpness evaluation function of the CCD camera on the computer, and inversely calculate the travel of the Z-axis vertical linear motion module based on the image sharpness using a visual algorithm. The galvanometer assembly contains components that reflect light and can receive laser light generated by the dual-wavelength composite laser source unit. The light-blocking block and bellows cover are used to protect the spatial optical path from dust interference.

[0009] A further preferred embodiment of the thermoplastic molding unit is a combination of synchronous belt, ball screw, pressure sensor, temperature sensor B, and closed-loop control. The thermoplastic molding unit is installed on one side of the base plate support and includes a servo motor, synchronous pulley A, synchronous belt, synchronous pulley B, ball screw assembly, slider mechanism, guide rod, pressure sensor, upper pressure head assembly, guide flange, shock-absorbing spring, guide column, mold pressure head (made of brass), and temperature sensor B connected in sequence; the shock-absorbing spring is sleeved on the outside of the guide column. The servo motor rotates, driving synchronous pulley A to rotate. Synchronous pulley A drives the synchronous belt drive, which in turn drives synchronous pulley B to rotate. Synchronous pulley B drives the ball screw assembly to rotate, converting the rotational motion into linear motion. The ball screw's slider mechanism performs linear motion. The slider mechanism is fixed to the guide rod with screws, and the guide rod is connected to the pressure sensor's connecting rod with threads. The slider mechanism drives the pressure sensor's connecting rod, thereby moving the upper pressure head assembly downwards. The mold pressure head is connected to the upper pressure head assembly via guide pillars, damping springs, and guide flanges. The upper pressure head assembly and the mold pressure head maintain linear motion through four guide pillars, and four sets of damping springs effectively buffer the pressure-bonded plastic riveting, preventing hard damage to the plastic riveting assembly. When the thermoplastic molding unit first comes into contact with the product, the pressure sensor records the change value of the pressure K over time in real time. With the target molding pressure K0 (0.3–0.9 MPa) as the set value, the pressure sensor feeds back the PID adjustment of the servo motor torque, while the temperature sensor B detects the temperature drop trend in real time.

[0010] A further preferred embodiment is a four-degree-of-freedom precision motion platform, installed in the middle of the equipment frame unit 5 and located below the base plate support, including an X-axis horizontal linear motion module, a Y-axis horizontal linear motion module, an A-axis rotary servo motor, a B-axis rotary servo motor, and a flexible gripper. The X-axis horizontal linear motion module includes an X-axis horizontal linear active module and an X-axis horizontal linear follower module, which completes linear motion in the X-axis direction. The linear active module is the power source module. The Y-axis horizontal linear motion module is installed between the X-axis horizontal linear active module and the X-axis horizontal linear follower module via a module connecting plate. The Y-axis horizontal linear motion module is equipped with left and right mounting plates. A B-axis motor fixing plate is installed between the left and right mounting plates via a rotating shaft. An A-axis rotary servo motor is installed on the outside of the right mounting plate. The output shaft of the A-axis rotary servo motor is fixedly connected to the B-axis motor fixing plate via the rotating shaft, thereby driving the B-axis motor fixing plate to rotate a certain angle on the A-axis, and driving the plastic riveting sample to rotate a certain angle on the A-axis. A B-axis rotary servo motor is fixedly installed at the bottom of the B-axis motor mounting plate. A B-axis work rotary table is installed on the top of the B-axis motor mounting plate. The output shaft of the B-axis rotary servo motor is fixedly connected to the B-axis work rotary table, thereby driving it to rotate a certain angle on the B-axis, and driving the plastic riveting sample to rotate a certain angle on the B-axis. The A-axis rotary servo motor provides the rotational motion of the A-axis. The rotational motion is transmitted to the B-axis motor mounting plate through the rotary plate. The B-axis rotary servo motor is fixed on the B-axis motor mounting plate to realize the rotational motion of the B-axis. The B-axis rotary table is fixedly connected to the output shaft of the B-axis rotary servo motor. The plastic riveting sample is placed on the upper part of the B-axis rotary table. Two flexible grippers simultaneously press the plastic riveting sample from both sides. The X-axis horizontal linear active module and the Y-axis horizontal linear motion module adjust the position of the plastic riveting sample on the XY axis, moving the plastic riveting sample directly below the galvanometer assembly. The A-axis rotary servo motor and the B-axis rotary servo motor complete the attitude adjustment of the plastic riveting sample in two rotational directions, adapting to the riveting requirements of riveting parts at multiple angles. The flexible grippers can flexibly adapt to and fix plastic riveting samples of different shapes, minimizing damage to the plastic riveting sample during product pressing.

[0011] Preferably, the suction nozzle is fixed to the left mounting plate via a dust connection plate, and the suction nozzle is able to absorb the dust generated by the dual-wavelength composite laser light source unit in the air; The equipment rack unit provides support and fixation for the equipment at the top, the middle section of the frame is used to store various control unit components, and the bottom caster assembly is used for moving and leveling the equipment.

[0012] The industrial control computer has an integrated electrical assembly control unit installed inside. It uses CNN-LSTM for forward and backward prediction and uses the CNN-LSTM model for model training. CNN is mainly used for the features of riveting images, including the riveting structure shape and surface flatness of the sample. LSTM is used to capture the correlation between process parameters and image features. Process parameters include pressing pressure parameters, laser power parameters, and scanning speed. It can realize forward prediction of riveting joint image information.

[0013] The beneficial effects of this invention are that, based on the material differences of the materials to be processed, a suitable laser light source is selected. After the material is softened by laser processing, a thermoplastic molding unit completes the compression molding operation. Simultaneously, a comprehensive database is constructed, containing information on the product's material type, material color, riveting structure shape, mold pressure parameters, laser power parameters, scanning speed, and riveting image information. This enables forward prediction of riveting image information and reverse prediction of process parameters. This processing technology can effectively reduce the number of plastic riveting samples lost during the product debugging stage, significantly improving production efficiency.

[0014] This application employs five units that work collaboratively and intelligently in the plastic riveting process to jointly complete the riveting and assembly of precision parts, ensuring the accuracy and stability of the processing and assembly. The use of a dual-wavelength composite laser source unit expands the processing range and efficiency for different materials; the flexible grippers adapt to the product's clamping requirements without easily damaging the product; force and temperature can be extracted and fed back, allowing for real-time adjustment of the laser-softened plastic's pressing effect; one device can complete multiple riveting processes on different planes simultaneously; and an intelligent software system reduces the debugging cycle.

[0015] This invention patent aims to realize laser thermoplastic riveting processing of various engineering plastic materials using a single device, and its core structure has the following advantages: (1) The dual-wavelength composite laser source unit is a 1064 fiber laser and a 1710 semiconductor laser, and a dynamic energy distribution module is used to dynamically allocate the proportion of the two beams of energy, realize the collaborative processing of the dual-wavelength composite laser source, and has an automatic focusing function. It is linked with the five-degree-of-freedom linkage system (X-axis + Y-axis + Z-axis + A-axis rotation + B-axis rotation) to ensure focusing accuracy. It can adapt to the processing needs of plastic parts of various materials and thicknesses, and realize precise processing under multiple working conditions.

[0016] (2) The laser riveting forming equipment is a five-degree-of-freedom linkage system motion mechanism, equipped with a lower four-degree-of-freedom precision motion platform, and an upper galvanometer scanning and autofocus unit equipped with a vertical Z-axis degree of freedom, together forming a five-degree-of-freedom linkage system. The equipment contains flexible grippers and pressure sensors, which can support the fixing of plastic riveting samples of different shapes and avoid surface damage. The CCD camera module can automatically correct the posture in linkage with the bottom four-degree-of-freedom precision motion platform. It can support the processing of plastic parts at different angles and multiple riveting points, realize single clamping and one-time forming, and improve processing efficiency and assembly accuracy.

[0017] (3) The thermoplastic molding unit contains a riveting mold unit and a hot pressure closed-loop control module. The riveting mold unit can quickly change the pressure head to adapt to the manufacturing needs of riveting pressure heads of different sizes and shapes. The hot pressure closed-loop control module can collect pressure and temperature in real time and dynamically adjust the pressing parameters.

[0018] (4) Extract core data from the processing process using a computer system, including the product material type (PA, PB, ABS, etc.), material color (black, dark, natural, light, or colored thermoplastics with added infrared absorbers), riveting structure shape (column diameter: 0.8–5 mm, height: 1.0–6 mm), pressing pressure parameters (0.3–0.9 MPa), temperature parameters (30–300℃), laser power parameters (1710 nm energy 5-80W, 1064 nm energy 10-300W), and scanning speed (5–100 rpm). The system constructs a multi-material process database based on the above eight sets of parameters (mm / s) and riveting joint image information. The software system uses a CNN-LSTM model for model training. The CNN is mainly used for the features of the riveting image (such as the riveting structure shape and surface flatness of the sample), while the LSTM is used to capture the correlation between process parameters (pressing pressure parameters, laser power parameters, scanning speed) and image features, which can realize the forward prediction of riveting joint image information. At the same time, a closed-loop iterative reverse prediction model of image features and process parameters is adopted. Combined with the defect identification of plastic riveting process, the process parameters are predicted and iterated in reverse, which effectively reduces the debugging cost, improves the debugging efficiency, and shortens the debugging cycle. Attached Figure Description

[0019] The attached figure is a schematic diagram of the structure of the present invention.

[0020] Figure 1 , Figure 2 The images shown are, in order, a 3D view and a front view of the overall structure of the laser riveting plastic processing equipment; Figure 3 , Figure 4 The images, in order, are a 3D view and a top view of the dual-wavelength composite laser source unit; Figure 5 , Figure 6 , Figure 7 The views shown are, in order, the right view, the left view, and the perspective view of the galvanometer scanning and autofocus unit and the thermoplastic molding unit; Figure 8 , Figure 9 The images, in order, are a 3D view and a front view of a four-degree-of-freedom precision motion platform. Figures 10-12 The image shows a flexible gripper from Yiheda Company, model number NIL21-G18-3. The images are, in order, the front view, left view, and a diagram showing the bending direction and stroke of the gripper's fingertips.

[0021] Figure 13 This is a flowchart of the product's riveting point focusing control process; Figure 14 This is a flowchart of the visual coordinate positioning and riveting process; Figure 15 This is a flowchart of the CNN-LSTM forward prediction process; Figure 16 This is a flowchart of the CNN-LSTM inverse prediction process.

[0022] In the diagram: 1. Dual-wavelength composite laser source unit; 2. Galvanometer scanning and autofocus unit; 3. Thermoplastic molding unit; 4. Four-degree-of-freedom precision motion platform; 5. Equipment frame unit; 1-1. 1064 fiber laser; 1-2. Dichroic mirror; 1-3. PBS beam splitter A; 1-4. PBS beam splitter B; 1-5. Mirror assembly; 1-6. 1710 semiconductor laser; 1-7. Green light source; 1-8. CCD camera assembly; 1-9. Driver board; 2-1. Module reinforcement plate A; 2-2. Z-axis vertical... 1. Linear motion module; 2.3. Galvanometer assembly; 2.4. Bellows cover; 2.5. Module reinforcing plate B; 2.6. Module connecting plate; 2.7. Light blocking block; 2.8. Module fixing base plate; 2.9. Base plate reinforcing rib; 2.10. Base plate support; 2.11. Temperature sensor A; 3.1. Mold pressure head; 3.2. Guide post; 3.3. Shock-absorbing spring; 3.4. Guide flange; 3.5. Pressure sensor; 3.6. Upper pressure head assembly; 3.7. Guide rod; 3.8. Slider mechanism; 3.9. Reinforcing rib plate C; 3.10. Ball screw assembly; 3-11. Servo motor; 3-12. Synchronous belt; 3-13. Synchronous pulley A; 3-14. Module connecting plate B; 3-15. Synchronous pulley B; 3-16. Temperature sensor B; 4-1. X-axis horizontal linear follower module (300 mm); 4-2. Smoke and dust connection plate; 4-3. Dust suction nozzle; 4-4. B-axis motor fixing plate; 4-5. Flexible gripper; 4-6. Rotary plate; 4-7. A-axis rotary servo motor; 4-8. Y-axis horizontal linear motion module (500 mm); 4-9. Module connecting plate; 4-10. X-axis horizontal linear active module (300 mm); 4-11. B-axis work rotary table; 4-12. Rotary motor connecting plate; 4-13. Cable chain; 4-14. B-axis rotary servo motor; 4-15. Plastic riveting sample; 4-16. Mounting plate. Detailed Implementation

[0023] The attached figure shows a specific embodiment of the present invention.

[0024] The laser riveting processing device of the present invention comprises five functional units, such as... Figure 1As shown, the components mounted on the equipment frame unit 5 are: a dual-wavelength composite laser source unit 1, a galvanometer scanning and autofocus unit 2, a thermoplastic molding unit 3, and a four-degree-of-freedom precision motion platform 4. These units work together and intelligently adapt during the plastic riveting process to complete the riveting and assembly of precision parts, ensuring the accuracy and stability of the processing and assembly.

[0025] (1) Dual-wavelength composite laser source unit (1064 nm laser + 1710 nm laser + dynamic energy distribution) The dual-wavelength composite laser source unit 1 includes two lasers mounted on the upper end of the equipment frame unit 5, a dichroic mirror 1-2, a PBS beam splitter A1-3, a PBS beam splitter B1-4, a reflector assembly 1-5, a green light source 1-7, a CCD camera 1-8, and a driver board 1-9.

[0026] The dual-wavelength composite laser source unit employs two lasers: a 1710 nm semiconductor laser (5 W-80 W) 1-6 and a 1064 nm fiber laser (10 W-300 W) 1-1. The 1064 nm fiber laser 1-1 and the 1710 nm semiconductor laser 1-6 are coupled into a common optical path via a dichroic mirror 1-2. A 532 nm green band light source 1-7 enters the coaxial optical path via a PBS beam splitter A1-3 to provide illumination for the CCD camera 1-8. The CCD camera 1-8 enters the coaxial optical path via a PBS beam splitter B1-4. The coaxial optical path reflects the light information to the galvanometer scanning and autofocus unit 2 via a mirror assembly 1-5.

[0027] The system comprises two lasers and includes a dynamic energy distribution module that can dynamically allocate the energy ratio of the two beams. The 1064nm fiber laser 1-1 adjusts its output power by changing the pump current through its internal pump power supply, thereby altering the gain. The 1710nm semiconductor laser 1-6 adjusts its output power by controlling the current in the driver board 1-9 in conjunction with the temperature control of the water-cooled plate.

[0028] The dynamic energy distribution module is the control system for the entire device. The specific signal chain is: industrial computer (central control) → control card / IO module → branch control. Specifically, the branch control is as follows: 1064 nm path, laser control card → optical laser power supply + galvanometer driver; 1710 nm path, digital-analog signal → LD constant current drive source + TEC temperature control.

[0029] This equipment uses a 1064 nm fiber laser (1-1) and a 1710 nm semiconductor laser (1-6). The 1710 nm semiconductor laser (1-6) is a mid-infrared semiconductor laser, which has strong intrinsic absorption in most thermoplastics (PMMA, PC, PET, PP), allowing for bulk heating and surface softening without the need for an absorbent. It is mainly used for transparent / thin parts. The 1064 nm fiber laser is a near-infrared laser source, allowing for more concentrated processing and a smaller heat-affected zone. It requires an absorbent, but at high power, it is prone to ablation and is mainly used for thick parts / reinforced plastics. This laser processing and riveting equipment can dynamically allocate the energy ratio of the two beams, enabling collaborative processing using dual-wavelength composite laser source units.

[0030] (2) Galvanometer scanning and autofocus unit (CCD vision + Z-axis vertical linear motion module) The galvanometer scanning and autofocus unit 2 is installed inside one side of the substrate support 2-10, and is equipped with a galvanometer assembly 2-3 and a Z-axis vertical linear motion module 2-2. The galvanometer assembly 2-3 is mounted on the Z-axis vertical linear motion module 2-2. A temperature sensor A2-11 is installed at the lower part of the galvanometer assembly.

[0031] Temperature sensor A2-11 is a digital infrared temperature sensor.

[0032] The upper part of the substrate support 2-10 has holes for the light path to pass through. The substrate support 2-10 is used to connect the dual-wavelength composite laser light source unit 1, the galvanometer scanning and autofocus unit 2, and the thermoplastic molding unit 3.

[0033] The galvanometer assembly 2-3 is fixed on the module connecting plate 2-6. The module reinforcing plate A (2-1) and the module reinforcing plate B (2-5) are connected to the module connecting plate 2-6 by threads to increase strength. The Z-axis vertical linear motion module 2-2 has a stroke of 150 mm and can move in the vertical direction.

[0034] The Z-axis vertical linear motion module 2-2 can use a motor to drive a ball screw to rotate. A nut is fitted on the screw. The nut cannot rotate but can only perform linear motion. The nut is rigidly connected to the slide, so the slide drives other structures to perform linear motion.

[0035] The galvanometer assembly 2-3 moves up and down with the Z-axis vertical linear motion module 2-2 (i.e., movement in the Z-axis direction). This, combined with the A and B sets of rotating components within the four-degree-of-freedom precision motion platform 4, enables vertical focus. Laser processing focus can be determined by the visual image sharpness evaluation function F0 of the CCD camera 1-8 on the computer, which identifies the distance to the focal point. Based on the image sharpness, a visual algorithm is used to infer the travel distance of the Z-axis vertical linear motion module 2-2. The specific algorithm is described in [reference needed]. Figure 5The control algorithm; the galvanometer assembly 2-3 contains components that reflect light and can receive the laser generated by the dual-wavelength composite laser source unit 1; the light blocking block 2-7 and the bellows cover 2-4 are used to protect the spatial optical path from dust interference; the module fixing base plate 2-8 and the base plate reinforcing rib 2-9 are used to fix the Z-axis vertical linear motion module 2-2.

[0036] (3) Thermoplastic plastic compression molding unit (mold + pressure sensor + temperature sensor B + closed-loop control) The thermoplastic molding unit 3 is installed on one side of the base plate support 2-10 and includes a servo motor 3-11, a synchronous pulley A3-13, a synchronous belt 3-12, a synchronous pulley B3-15, a ball screw assembly 3-10, a slider mechanism 3-8, a guide rod 3-7, a pressure sensor 3-5, an upper pressure head assembly 3-6, a guide flange 3-4, a shock-absorbing spring 3-3, a guide post 3-2, a mold pressure head 3-1 (made of brass), and a temperature sensor B3-16; the shock-absorbing spring 3-3 is sleeved on the outside of the guide post 3-2.

[0037] The temperature sensor B3-16 is model PT-100.

[0038] After the dual-wavelength composite laser source unit 1 softens the plastic through the galvanometer assembly 2-3, the temperature sensor A2-11 detects that the predetermined temperature threshold T0 (T0=180–260℃) has been reached, and the system immediately shuts off. The laser determines the relative position of the product through the coaxial optical path CCD camera 1-8, and the four-degree-of-freedom precision motion platform 4 moves the softened plastic assembly to the predetermined position under the mold pressure head 3-1. Servo motor 3-11 rotates, driving synchronous pulley A3-13 to rotate. Synchronous pulley A3-13 drives synchronous belt 3-12, which in turn drives synchronous pulley B3-15. Synchronous pulley B3-15 drives ball screw assembly 3-10 to rotate. Ball screw assembly 3-10 converts rotational motion into linear motion. The slider mechanism 3-8 of ball screw 3-10 performs linear motion. Slider mechanism 3-8 is fixed to guide rod 3-7 by set screws. Guide rod 3-7 is connected to pressure sensor... The connecting rod of the device 3-5 is connected by a thread. The slider mechanism 3-8 drives the connecting rod of the pressure sensor 3-5, thereby driving the upper pressure head assembly 3-6 to move downward. The mold pressure head 3-1 is connected to the upper pressure head assembly 3-6 through the guide post 3-2, the shock-absorbing spring 3-3, and the guide flange 3-4. The upper pressure head assembly 3-6 and the mold pressure head 3-1 maintain linear motion through the four guide posts 3-2. The four sets of shock-absorbing springs effectively buffer the pressing of the plastic riveting, preventing hard damage to the plastic riveting assembly.

[0039] When the thermoplastic molding unit 3 first comes into contact with the product, pressure sensor 3-5 records the change in pressure K over time in real time. Using the target molding pressure K0 (0.3–0.9 MPa) as the setpoint, the pressure sensor provides feedback to the PID controller to adjust the servo motor torque. Simultaneously, temperature sensor B3-16 monitors the temperature decrease trend in real time. Maintaining constant pressure, the unit enters the pressure holding stage. As the temperature approaches the setting temperature T... d (T) d =80-120℃) and below, after riveting and pressing are completed, the product returns to the initial galvanometer processing position. CCD cameras 1-8 record two-dimensional images of the plastic riveted product, and data two-dimensional images are used to establish a sample database. Through actual products, the riveting graphic information and standards of qualified products are defined manually, and a one-to-one correspondence with the pressure curve of the pressed product is established as the basic identification data of the software system.

[0040] (4) Four-degree-of-freedom precision motion platform (X-axis and Y-axis linear + A-axis and B-axis dual rotation + flexible gripper) The four-degree-of-freedom precision motion platform (4) is installed in the middle of the equipment frame unit 5 and located below the base plate support 2-10. It includes an X-axis horizontal linear motion module, a Y-axis horizontal linear motion module, an A-axis rotary servo motor 4-7, a B-axis rotary servo motor 4-14, and a flexible gripper.

[0041] The flexible gripper currently uses standard parts from Yiheda Company, model number NIL21-G18-3. It is controlled by an air source and solenoid valve, and the fingers are made of silicone. Positive pressure opens it, and negative pressure contracts it. The entire flexible gripper has an internal aluminum alloy base and segmented silicone air chambers, utilizing vacuum negative pressure and silicone rebound to achieve clamping and releasing. See [link / details]. Figures 10-12 The images are, in order, the front view, left view, and the direction and stroke of the fingertip bending movement of the flexible gripper.

[0042] The X-axis horizontal linear motion module includes a 300mm X-axis horizontal linear active module 4-10 and a 300mm X-axis horizontal linear follower module 4-1, which completes linear motion in the X-axis direction. The X-axis horizontal linear active module 4-10 is the power source module, and the X-axis horizontal linear follower module 4-1 is the follower module with a stroke of 300mm.

[0043] Y-axis horizontal linear motion module 4-8, the linear module has a travel of 500 mm, and module connecting plate 4-9 is used to connect 500 mm Y-axis horizontal linear motion module 4-8, 300 mm X-axis horizontal linear active module 4-10 and 300 mm X-axis horizontal linear follower module 4-1.

[0044] The Y-axis horizontal linear motion module 4-8 is equipped with left and right mounting plates 4-16. A B-axis motor fixing plate 4-4 is mounted between the left and right mounting plates via a rotating shaft. An A-axis rotary servo motor 4-7 is mounted on the outside of the right mounting plate. The output shaft of the A-axis rotary servo motor 4-7 is fixedly connected to the B-axis motor fixing plate 4-4 via the rotating shaft, thereby driving the B-axis motor fixing plate 4-4 to rotate a certain angle on the A-axis, which in turn drives the plastic riveting sample 4-15 to rotate a certain angle on the A-axis.

[0045] The B-axis motor mounting plate 4-4 has a B-axis rotary servo motor 4-14 fixedly mounted on its bottom. The B-axis rotary table 4-11 is mounted on the B-axis motor mounting plate 4-4. The output shaft of the B-axis rotary servo motor 4-14 is fixedly connected to the B-axis rotary table 4-11, thereby driving it to rotate a certain angle on the B-axis, which in turn drives the plastic riveting sample 4-15 to rotate a certain angle on the B-axis.

[0046] The A-axis servo motor 4-7 provides the rotational motion of the A-axis. The rotational motion is transmitted to the B-axis motor mounting plate 4-4 through the rotary plate 4-6. The B-axis servo motor 4-14 is fixed on the B-axis motor mounting plate 4-4 to realize the rotational motion of the B-axis. The B-axis rotary table 4-11 is fixed to the output shaft of the B-axis servo motor 4-14 through a threaded connection. The two sets of rotating components A and B are fixed to the slide of the Y-axis horizontal linear motion module 4-8 through the rotary motor connecting plate 4-12 through a threaded connection.

[0047] The suction nozzle 4-3 is fixed to the left mounting plate 4-16 via the dust connection plate 40. The suction nozzle 4-3 can absorb the dust generated by the dual-wavelength composite laser light source unit 1 in the air. The dust connection plate 4-2 is fixed to the rotary motor connection plate 4-12. The suction nozzle 4-3 is fixed to the left mounting plate 4-16 via the dust connection plate 40. Figure 4 The location shown.

[0048] The X-axis horizontal linear motion module and the Y-axis horizontal linear motion module can use a motor to drive a ball screw to rotate. A nut is fitted on the screw. The nut cannot rotate but can only perform linear motion. The nut is rigidly connected to the slide, so the slide drives other structures to perform linear motion.

[0049] The plastic riveting sample 4-15 is placed on the upper part of the B-axis rotary table of 4-11. Two flexible grippers 4-5 simultaneously press the plastic riveting sample 4-15 from both sides, accurately fixing it to the Y-axis motor mounting plate 4-4. The X-axis horizontal linear active module 4-10 and the Y-axis horizontal linear motion module 4-8 adjust the linear movement of the plastic riveting sample 4-15 in the XY directions, moving the plastic riveting sample 4-15 directly below the galvanometer assembly 2-3. The A-axis servo motor 4-7 and the B-axis servo motor 4-14 complete the attitude adjustment of the plastic riveting sample 4-15 in the two rotational directions, adapting to the multi-angle riveting requirements of the riveted parts. The flexible grippers 4-5 can flexibly adapt to and fix plastic riveting samples of different shapes, and can minimize damage to the plastic riveting sample during product pressing.

[0050] (5) Equipment rack unit 5 The equipment frame unit provides support and fixation for the entire upper equipment, the middle section of the frame is used to store various control unit devices, and the bottom caster assembly is used for equipment movement and leveling.

[0051] (6) Integrated electrical assembly control unit 6 (CNN-LSTM forward and backward prediction + database + PID closed loop) The industrial control computer 6 has an integrated electrical assembly control unit installed inside. It uses CNN-LSTM for forward and backward prediction and uses the CNN-LSTM model for model training. CNN is mainly used for the features of riveting images, including the riveting structure shape and surface flatness of the sample. LSTM is used to capture the correlation between process parameters and image features. Process parameters include pressing pressure parameters, laser power parameters, and scanning speed. It can realize forward prediction of riveting joint image information.

[0052] The controller uses an industrial PC, which is a classic 4U rack-mount type, specifically the Advantech IPC-610-H. It supports ATX / Micro-ATX motherboards, is usually paired with an i5 / i7 processor, and provides multiple PCI / PCIe slots, allowing simultaneous insertion of fiber laser control cards and semiconductor laser driver cards.

[0053] The industrial control computer is the core of the overall control system, running host computer software (such as a control interface developed in C# / C++ or commercial laser software), responsible for issuing arbitrary process parameters, timing logic judgments, and human-machine configuration. The motion control card solution is suitable for high-precision motion control applications. A single card can connect to a 1064 nm galvanometer and laser power supply, and can also send "power setting voltage" and "LD enable" signals to a 1710 nm constant current source via a general-purpose AO (analog output) or DO (digital output) port, achieving nanosecond-level synchronization between the dual lasers and the motion axis. A dedicated laser control power supply receives external control signals to adjust light and power, and integrates overcurrent and overtemperature protection. The laser galvanometer control card is a commonly used hardware interface for 1064 nm lasers. It connects directly to the DB head of the fiber laser via a DB25 or extended I / O port, outputting signals such as switching on, frequency, and energy modulation, while simultaneously controlling the deflection of the galvanometer's X and Y axis motors. The LD constant current drive power supply nonlinearly controls the 1710 nm laser by adjusting the drive current (not voltage). The power of the 1710 nm laser must be controlled, and soft start and current limiting protection are required; the TEC temperature controller is also necessary because the 1710 nm laser is very sensitive to temperature and must be equipped with a TEC bidirectional temperature control module. Typically, a PID algorithm is used to control the TEC chip to keep the temperature within ±0.1℃.

[0054] An automatic focusing method for laser riveting processing equipment includes the following steps: (1) Based on the preset CAD model data or process preset parameters, determine the coplanarity of the initial riveting points of the plastic riveting specimen; If it is determined that all the riveting points to be processed are on the same processing plane, the focal length pre-adjustment operation of the galvanometer assembly 2-3 will be performed first to lay the benchmark for subsequent riveting processing; (2) The galvanometer assembly 2-3 moves up and down with the Z-axis vertical linear motion module 2-2. The laser processing focal length focusing can be determined by the image sharpness evaluation function of the CCD camera 1-8 in the computer to determine the distance to the focal point. Based on the image sharpness, the travel distance of the Z-axis vertical linear motion module 2-2 is calculated using an image vision algorithm. The details are as follows: The coaxial optical path CCD camera 1-8 collects the image of the riveting area of the workpiece, and extracts the defocus state information of the vertical Z-axis based on the imaging sharpness of the riveting contour. The system has a built-in sharpness evaluation function (such as gradient square function / Laplacian function), which performs quantitative calculation on the real-time image and outputs the real-time sharpness evaluation value F0; the preset standard sharpness evaluation value is taken as the threshold F, and when the real-time evaluation value F0<F, the automatic focusing process is triggered; the software compares the real-time evaluation value F0 with the preset threshold F, and if F0<F, it is determined that the galvanometer assembly 2-3 has Z-axis defocus, and the focusing displacement H needs to be executed; at the same time, the optimal movement direction of the Z-axis is determined through the hill-climbing search algorithm (step size 0.01mm, accuracy ±0.005mm) to improve the imaging sharpness. The controller accurately issues the focusing instruction to the Z-axis vertical linear motion module 2-2, driving the galvanometer assembly 2-3 to complete the closed-loop motion of displacement H along the Z-axis; when the real-time sharpness evaluation value F0 satisfies |F0-F|≤A (where A is the preset allowable threshold of defocus amount, A=0.02), it is determined that focusing is completed, and the galvanometer assembly stops moving.

[0055] (3) If it is determined that each riveting point is not in the same processing plane, control the A-axis servo motor 4-7 and the B-axis servo motor 4-14 in the four-degree-of-freedom precision motion platform (4) to act in coordination, drive the plastic riveting specimen (4-15) to rotate and adjust around the A-axis and B-axis, with the rotation range of the AB axes being ±90°, so as to adjust the target riveting point of the plastic riveting specimen 4-15 to a horizontal state, ensuring the consistency of the processing reference plane; then perform the operation of step (2).

[0056] Regarding the gradient square function, the published paper: Research on Focus Evaluation Function of Wide-Angle Camera Image, Wang Hongshu, Xu Xiping, School of Optoelectronic Engineering, Changchun University of Science and Technology. Formula (11).

[0057] Regarding the Laplacian function, the published paper: A New Image Sharpness Evaluation Function, Zhu Kongfeng, Jiang Wei, Wang Duanfang, Zhang Jin, Zhou Xian, School of Information Science and Engineering, Shandong University, Formula (1), (2).

[0058] Regarding the hill-climbing search algorithm, the published paper: Research on Camera Autofocus Algorithm Based on Deep Learning and Reinforcement Learning, Wei Xiaolin, Master's Dissertation of Chongqing University. The optimal focal length search method and hill-climbing algorithm are disclosed therein.

[0059] The coaxial optical path CCD camera 1-8 is responsible for implementing the positioning of XY coordinates and the path optimization method, and the specific process is as follows: (1) The CCD camera collects template image information: collect the feature image of the riveting point of the qualified plastic workpiece in advance as the positioning template, which contains the precise contours of the rivet column and the reference hole of the plastic workpiece, and sample the coordinates (X0, Y0) of the riveting point of the template plastic workpiece; (2) Image feature matching for coarse localization: Real-time acquisition of images of plastic workpieces, using template matching algorithm to complete coarse localization of plastic workpieces, and obtaining pixel coordinates (U) of the riveting points. i V i ); (3) Pixel coordinate to physical coordinate transformation: Based on the calibrated coordinate transformation matrix, the pixel image coordinates (U, U, U) of the identified riveting points are transformed into physical coordinates. i V i The resulting transformation matrix H is used to convert the coordinates into physical coordinates (X). i Y i ); (4) Calculate the unit deviation: Calculate the positional deviation Δx = X0 - X i △y=Y0- Y i If the deviation is small, use a galvanometer to compensate for Δx and Δy to eliminate the deviation; if the deviation is large and exceeds the threshold, the machine will alarm and require manual handling. (5) Triggering laser processing: Positioning is completed, and the dual laser riveting process is started. The laser begins to heat and soften the plastic workpiece to be riveted. (6) When the plastic workpiece reaches the softening riveting temperature threshold To, the laser heating is immediately terminated; (7) The four-degree-of-freedom precision motion platform moves the plastic workpiece so that the riveting point is moved to the lower part of the thermoplastic molding unit. The mold head of the thermoplastic molding unit moves downward to apply pressure to the riveting point of the plastic workpiece. According to the preset pressure Ko, the pressure sensor measures the actual pressure value K. If the pressure difference |K-Ko|≤D, the pressure value C is maintained. If the pressure difference is greater than D, the PID adjustment of the servo motor torque is used to increase the pressure. (8) When the temperature of the plastic workpiece drops to ≤ the mold setting temperature T d Then the mold pressure head stops holding pressure and moves upward, leaving the plastic workpiece; (9) The four-degree-of-freedom precision motion platform moves the plastic workpiece so that its riveting point is moved to the set position under the galvanometer, and the CCD camera captures the standard image of the riveting point of the plastic workpiece. (10) If there is another point to be riveted on the plastic workpiece, continue processing; otherwise, end processing.

[0060] When riveting the plastic sample 4-15 to the second riveting point, repeat the above steps.

[0061] Regarding template matching algorithms, real-time acquisition of workpiece images using a coaxial optical path CCD is achieved through template matching algorithms for coarse workpiece positioning. For model matching, the most commonly used algorithm is Normalized Cross-Correlation (NCC).

[0062] Publication paper: Template matching algorithm based on circular projection and radial projection, Chen Xiaoqiang, Chen Lifeng, School of Mechanical and Electrical Engineering, Hunan University of Science and Technology. Formula (3).

[0063] Regarding the transformation matrix H used in the CCD camera's XY coordinate positioning method, H is the homography matrix.

[0064] Public paper: Registration and fusion method of visible light and infrared images of foreign objects in low-illuminance coal flow_Li Jia, Li Junxia, ​​Kou Ziming, Wu Lei, Xu Weipeng, first published online in Coal Science and Technology, Formula (3), Formula (4).

[0065] The forward prediction execution process for the final processed product of plastic riveting samples 4-15 is based on constructing a mathematical mapping relationship between process parameters and processing quality using historical experimental data. When new process parameters are input, the system uses a CNN-LSTM model for training to automatically calculate the expected pressing quality score and the final pressing effect, thereby reducing debugging costs and shortening the debugging cycle. The specific execution steps are as follows: (1) Experimental Design and Data Acquisition: The range of process parameters was clarified, and the feasible ranges for processing material type, material color, riveting structure dimensions, laser power (30 W-150 W), scanning speed, and pressing pressure were determined. Example parameters are as follows: Product material is PC, riveting structure diameter is 6 mm, height is 4 mm, laser power is 30-100 W, scanning speed is 90-160 mm / s, and pressing pressure is 0.3-0.9 mm / s. MPa; Design an experimental scheme, except for the product material, select 30-50 sets of parameter combinations evenly within the feasible range for each process parameter, and conduct at least 2-3 repeated experiments for each parameter combination, and take the average value of the experimental results to reduce random errors; Perform laser softening and pressing experiments according to the preset scheme, and use CCD cameras 1-8 to acquire images of the riveting area and record relevant data before and after each experiment, and simultaneously record seven sets of core process parameters: processing material type, material color, riveting structure size, laser power, scanning speed, pressing pressure, and temperature; Construct an experimental data table, using a row structure to correspond to a single set of experiments, with each row completely recording the process parameters, experimental data, and image information of that set of experiments; (2) Definition of quality scoring criteria: Determine the key indicators affecting riveting quality, mainly including core parameters such as riveting width, flash size, surface flatness, and surface finish. Formulate scoring rules, set ideal standard values ​​and allowable deviation ranges for each key quality indicator, and clarify the criteria for judging whether the indicators meet the standards; establish a quality scoring function, assign reasonable weights to each key quality indicator, and obtain the comprehensive score of riveting quality by weighted summation of the scores of each indicator; carry out manual verification and calibration work, extract 10-15 typical experimental images, first the algorithm automatically completes the quality scoring, and then senior technicians perform manual scoring, compare the differences between the two scores, iteratively adjust the weight coefficients and judgment thresholds of the quality scoring function, so that the automatic scoring results of the algorithm and the manual judgment tend to be consistent, and then complete the batch scoring of all experimental samples based on the calibrated quality scoring function; (3) Prediction Model Construction: The experimental data table and quality score data are integrated to form a standardized dataset. Each row of the dataset contains eight process parameters and their corresponding comprehensive quality scores: material type, material color, riveting structure, laser power, scanning speed, pressing pressure, temperature, and CCD riveting image. The CNN-LSTM model is selected as the prediction model. The CNN is responsible for extracting image feature information (including material type, material color, riveting structure, and CCD image information), and the LSTM is responsible for capturing the temporal / coupling rules between parameters (including laser power, scanning speed, pressing pressure, and temperature). The image features extracted by the CNN and the temporal features of the process parameters extracted by the LSTM are spliced / fused. The material type, material color, riveting structure, laser power, scanning speed, pressing pressure, and temperature of the training set are used as input variables, and the comprehensive quality score of the CCD image riveting is used as the target output variable. The nonlinear mapping relationship between the input and output is fitted through model iterative training. The trained model is used to predict the quality score of the process parameters of the test set. The predicted score is compared with the actual quality score of the test set to determine the deviation, and the process parameters are automatically iterated. (4) Deployment of prediction function: A visual operation interface is built to support users to manually input process parameters such as processing material type, material color, riveting structure, laser power, scanning speed, pressing pressure, and temperature; after the user inputs the parameters, the system first verifies the rationality of the parameters. If the input parameters exceed the preset feasible range, a warning prompt will be automatically issued, while the prediction function is retained, allowing the user to continue to perform the prediction operation; based on the input parameters, the system calculates and outputs the expected pressing quality score and the final pressing effect through the trained prediction model, and at the same time provides corresponding process parameter optimization suggestions; (5) Continuously collect process parameters and processing quality data in the actual production process, and regularly update and iterate the prediction model to optimize the model parameters.

[0066] Regarding the CNN-LSTM model, the system uses the CNN-LSTM model for training and automatically calculates the expected pressing quality score and the final pressing effect.

[0067] Publication paper: Research on laser spot focusing state discrimination algorithm based on time-series prediction, Ren Ziwen, Liu Huaiguang, Sun Wei, Hubei Provincial Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology.

[0068] Regarding the quality scoring function: The quality scoring function is a manually constructed weighted evaluation rule used to calculate a comprehensive quality true value score based on measured indicators such as riveting width, flash size, and surface flatness, thus completing the dataset labeling. The CNN-LSTM model is a data-driven prediction model that takes image features and process time-series coupled features as input and learns the nonlinear mapping relationship between features and quality scores. The two functions are different and their implementation logic is independent, but they work together in the system: the former generates training labels, and the latter achieves quality prediction for unknown process conditions.

[0069] Public Paper: Camera Exposure Optimization and Image Quality Assessment in Structured Light Measurement_Advances in Laser and Optoelectronics, Zhang Wei, Song Jie, Lü Sheng, Zeng Zhilong, Fang Qi, Wang Shenghuai, School of Mechanical Engineering, Hubei University of Automotive Technology. 2.3 Image quality assessment method, for the fluctuation of pixel mean, formula (3).

[0070] The two lasers, and their respective applicable image features and temporal / coupling rules, are as follows: The 1064 nm laser is a near-infrared laser source, providing concentrated processing energy. However, engineering materials have low absorption rates, and the energy is concentrated in the shallow surface layer, easily causing surface burning and melting. It is primarily suitable for black / dark-colored general-purpose engineering plastics, glass fiber reinforced modified plastics, and high-melting-point rigid engineering plastics. Under an independent green light source, the characteristics are as follows: dark-colored pure plastic parts have a uniformly dark background, retaining the original injection-molded texture; plastic rivets, positioning holes, and mating end faces have sharp and regular contours; glass fiber reinforced plastics have a matte gray tone with a uniform and dense glass fiber texture; rigid plastics such as POM and PPS have a dense and smooth surface with high structural molding precision. The entire image shows no laser-induced thermal damage or melting marks, only presenting the original morphology and geometry of the workpiece.

[0071] The 1710 nm laser light source was used for pure plastic laser riveting of transparent / high-transparency amorphous plastics (PC, PMMA, COC), white / light-colored engineering plastics (PP, PE, natural POM), and translucent soft plastics. Under the imaging of an independent green light source CCD camera: transparent plastic parts are bright and transparent overall, with a mirror-smooth surface, sharp riveting post outlines, and are completely translucent; white / light-colored plastic parts appear uniformly light gray or milky white, with fine and dense texture; translucent plastic parts have a soft light texture and slightly softened edges. The entire image shows no thermal damage, only the original morphology and geometry.

[0072] 1710 nm: Primarily targets interface / shallow layer precise softening; 1064 nm: Responsible for deep penetration and heat replenishment, and overall temperature uniformity.

[0073] Timing / Coupling Patterns: 1. Timing Pattern Description: Laser power, scanning speed, pressing pressure, and area temperature are timing-varying parameters that evolve systematically with the three process stages of laser melting, pressing and holding, and cooling and resetting. Laser power and scanning speed start and stop synchronously, maintaining steady-state operation during the heating stage. Pressing pressure lags behind laser loading, and pressure is released after holding in the melting zone. Area temperature rises and falls continuously with laser action, forming a complete temperature timing curve. The timing phase, rate of change, and duration of the four parameters together constitute the process timing features, which are learned by an LSTM network. 2. Coupling Pattern Description: Laser power and scanning speed jointly determine the heat input per unit area, exhibiting an inverse constraint. Power, scanning speed, and area temperature are positively and negatively correlated, respectively, jointly controlling the workpiece's thermal state. Pressing pressure must match the melting temperature window to achieve effective riveting. Simultaneously, static features extracted by the CNN, such as processing material, material color, riveting structure, and CCD image, serve as pre-constraints to limit the value range and matching combination of process parameters. The model combines CNN spatial features with LSTM temporal coupling features to establish a mapping relationship between eight-dimensional parameters and riveting quality scores.

[0074] The reverse prediction execution process for plastic riveting specimen 4-15 is as follows: Inputting an image of the riveting in any final pressing state, the system uses a CNN-LSTM model to automatically derive and output a set of optimal process parameters, covering key parameters such as material type, material color, riveting structure shape, laser power, scanning speed, pressing pressure, and specimen temperature. The specific execution steps are as follows: (1) Construction of process parameter image feature database: Based on the dataset accumulated from historical experiments, a database linking process parameters and riveting image features is constructed; for the final pressing image corresponding to each set of experiments, four core characterization features are extracted and integrated into a one-dimensional feature vector, specifically including: riveting width (unit: pixels or millimeters), flash size (characterized as bright area), surface flatness (characterized as gray scale variance), and surface smoothness (characterized as image texture gradient); corresponding data tables are established simultaneously to ensure that each set of process parameters and the extracted feature vector form a unique and direct mapping relationship; (2) Standardized extraction of query image features: The image is normalized in size and corrected in illumination to eliminate size differences and illumination interference; then, four core features consistent with the database are extracted, and the normalization method is used to uniformly scale the numerical range of all features to the 0-1 range to ensure that each feature has an equal weight contribution in the subsequent retrieval process. (3) Similarity retrieval mechanism construction: The feature vector of the query image after standardization is compared with the feature vector of all historical images in the database one by one. The similarity is quantified by calculating the feature distance between the two. The smaller the feature distance, the higher the image similarity. The retrieval strategy adopts the K-nearest neighbor (K-NN) method, and selects K=3 historical images with the smallest distance as the similar sample set. In the similar sample set, samples with feature distance <0.1 (i.e., highly similar) are given higher weight. If the feature distance of all samples exceeds the preset threshold, it is determined that there are no matching similar cases in the database. When there are no similar cases in the database, the initial parameters are generated by combining the process parameters of similar materials and similar structures in the existing database. After trial processing, the database is automatically learned and updated. (4) Optimal process parameter generation and confidence output: Based on the K most similar historical samples obtained by similarity retrieval (taking K=3 as an example), extract the corresponding three sets of process parameter combinations, and perform fusion calculation by weighted summation. The weight allocation is based on the feature distance between each sample and the query image. The smaller the distance, the higher the weight. Finally, a set of recommended optimal process parameter combinations is generated. When there are no similar cases in the database, the initial parameters are generated by combining the process parameters of similar materials and similar structures in the existing database. After trial processing, the database is automatically learned and updated. (5) Closed-loop optimization, model iterative update and self-learning of new cases: Establish a feedback mechanism for actual processing data in the production process, and continuously collect new riveting images and corresponding process parameters and quality data. Regularly incorporate new data into the feature database and iteratively optimize the process parameter-feature mapping relationship; for new working conditions without similar cases, supplement experimental data and expand the database, while adjusting the similarity calculation threshold and feature weight allocation rules to continuously improve the accuracy and generalization ability of reverse prediction, and realize the closed-loop optimization and long-term iteration of the system.

[0075] Regarding the reverse prediction K-nearest neighbor (K-NN) method, i.e., the K-NN nearest neighbor method, Publicly published paper: Water level outlier detection based on different machine learning algorithms_Hao Bin, Zhu Qinling, Li Qi, Shanxi Provincial Water Resources Development Center.

[0076] The core principle of traditional near-infrared laser processing technology applied to plastic riveting is as follows: Near-infrared lasers are used to locally heat the plastic connection area, causing the plastic to soften or melt. This is then molded into a rivet head under mechanical pressure. After the plastic cools and solidifies, a reliable connection between the components is achieved. Near-infrared laser heating has the following core advantages: it uses non-contact heating, effectively reducing mechanical stress on the plastic parts and lowering the risk of contamination during processing; the heating area can be precisely controlled, effectively minimizing the heat-affected zone and ensuring processing accuracy; it can adapt to the processing needs of plastic parts with complex geometries; and it has excellent material compatibility, making it widely applicable to the riveting of various thermoplastic plastics.

[0077] The key protection points of this invention are: 1. Using two wavelength optical paths for processing, it is possible to process plastics of various colors or thicknesses; 2. The five-degree-of-freedom linkage system can support plastic riveting at multiple positions at different angles, with one clamping and one processing step; 3. The computer can extract the material of the processed product, the shape of the riveted plastic, the mold pressure parameters, the laser power parameters, the scanning speed, and the image information of the riveting joint, and establish a process database for multiple materials. The software system is trained using eight sets of parameters, and can forward predict the image information of the riveting joint based on parameters such as mold pressure parameters, laser power parameters, and scanning speed. At the same time, using the provided image information of the riveting joint, it can reverse predict seven sets of process parameters and perform optimization iterations, saving debugging costs and improving debugging efficiency and time.

[0078] The above embodiments and accompanying drawings are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention do not depart from the spirit of the present invention and should also fall within the protection scope of the claims of the present invention.

Claims

1. A laser riveting processing device, characterized in that: It includes five functional units, namely, a dual-wavelength composite laser light source unit (1) installed on the equipment frame unit (5), a galvanometer scanning and autofocus unit (2), a thermoplastic plastic molding unit (3), and a four-degree-of-freedom precision motion platform (4). The dual-wavelength composite laser source unit (1) includes two lasers installed on the upper end of the equipment frame unit 5, a dichroic mirror (1-2), a PBS beam splitter A (1-3), a PBS beam splitter B (1-4), a mirror assembly (1-5), a green light source (1-7), a CCD camera (1-8), and a driver board (1-9). Two lasers, namely a semiconductor laser (1-6) and a fiber laser (1-1), are coupled into a common optical path through a dichroic mirror (1-2). The green light source (1-7) enters the coaxial optical path through the PBS beam splitter A (1-3) to provide illumination for the CCD camera (1-8). The CCD camera (1-8) enters the coaxial optical path through the PBS beam splitter B (1-4). The coaxial optical path reflects the light information to the galvanometer scanning and autofocus unit (2) through the mirror assembly (1-5).

2. The laser riveting processing device according to claim 1, characterized in that: The light source uses a 1064 nm fiber laser (1-1) and a 1710 nm semiconductor laser (1-6); The 1064 nm fiber laser (1-1) changes the output power by adjusting the pump current of the built-in pump power supply inside the laser and thus changing the gain. The 1710 nm semiconductor laser (1-6) adjusts the output power of the laser by controlling the current in the driver board (1-9) and the temperature control of the water cooling plate. The 1710 nm semiconductor laser (1-6) is a mid-infrared semiconductor laser. Most thermoplastic plastics, including PMMA, PC, PET, and PP, have strong intrinsic absorption. It can heat the bulk and soften the surface without the need for an absorbent. It is mainly used for transparent / thin parts and is suitable for pure plastic laser riveting of transparent / high-transparency amorphous plastics, white / light-colored engineering plastics, and semi-transparent soft plastics. The 1064 nm fiber laser is a near-infrared laser source, which allows for more concentrated processing and a smaller heat-affected zone. It requires the use of an absorber, but it is prone to ablation at high power. It is mainly used for thick parts / reinforced plastics and is suitable for black / dark general-purpose engineering plastics, glass fiber reinforced modified plastics, and high-melting-point rigid engineering plastics.

3. The laser riveting processing device according to claim 1, characterized in that: The galvanometer scanning and autofocus unit (2) is installed inside one side of the substrate support (2-10) and uses the galvanometer assembly (2-3) and the Z-axis vertical linear motion module (2-2) to cooperate with each other; the galvanometer assembly (2-3) is installed on the Z-axis vertical linear motion module (2-2); a temperature sensor A (2-11) is installed at the lower part of the galvanometer assembly. The upper part of the substrate support (2-10) has holes for the light path to pass through. The substrate support is used to connect the dual-wavelength composite laser light source unit (1), the galvanometer scanning and autofocus unit (2) and the thermoplastic molding unit (3). There are light-blocking blocks (2-7) and bellows covers (2-4) on the outside of the optical path. The light-blocking blocks (2-7) and bellows covers (2-4) are installed on the inner side of the substrate support (2-10). The light-blocking blocks (2-7) and bellows covers (2-4) are used to protect the spatial optical path from dust interference. The galvanometer assembly (2-3) moves up and down with the Z-axis vertical linear motion module (2-2) for vertical focus. The laser processing focus can work with the CCD camera (1-8) to determine the distance to the focus. The galvanometer assembly (2-3) contains components that reflect light and are capable of receiving laser light generated by the dual-wavelength composite laser source unit (1).

4. The laser riveting processing device according to claim 1, characterized in that: The thermoplastic molding unit (3) adopts a combination of synchronous belt, ball screw, pressure sensor, temperature sensor B and closed-loop control; The thermoplastic molding unit (3) is installed on one side of the substrate support (2-10), including a servo motor (3-11), synchronous pulley A (3-13), synchronous belt (3-12), synchronous pulley B (3-15), ball screw assembly (3-10), slider mechanism (3-8), guide rod (3-7), pressure sensor (3-5), upper pressure head assembly (3-6), guide flange (3-4), shock-absorbing spring (3-3), guide column (3-2), mold pressure head (3-1), and temperature sensor B (3-16); the shock-absorbing spring (3-3) is sleeved on the outside of the guide column (3-2); After the dual-wavelength composite laser source unit (1) softens the plastic through the galvanometer assembly (2-3), the temperature sensor A (2-11) detects that the predetermined temperature threshold T0 has been reached and immediately shuts off the system; the laser determines the relative position of the product through the coaxial optical path CCD camera (1-8), and the four-degree-of-freedom precision motion platform (4) moves the softened plastic assembly to the predetermined position under the mold pressure head (3-1); The servo motor (3-11) rotates, driving the synchronous pulley A (3-13) to rotate. The synchronous pulley A drives the synchronous belt (3-12), which in turn drives the synchronous pulley B (3-15) to rotate. The synchronous pulley B then drives the ball screw assembly (3-10) to rotate. The ball screw assembly converts the rotational motion into linear motion. The ball screw slider mechanism (3-8) performs linear motion. The slider mechanism is fixed to the guide rod (3-7) with screws, and the guide rod is connected to the pressure sensor (3-5) via threads. The slider mechanism (3-8) drives the connecting rod of the pressure sensor (3-5), thereby driving the upper pressure head assembly (3-6) to move downward; the mold pressure head (3-1) is connected to the upper pressure head assembly (3-6) through the guide column (3-2), the shock-absorbing spring (3-3), and the guide flange (3-4). The upper pressure head assembly and the mold pressure head (3-1) maintain linear motion through the four guide columns (3-2), and the four sets of shock-absorbing springs effectively buffer the pressing of the plastic riveting, preventing hard damage to the plastic riveting assembly; When the thermoplastic molding unit (3) first comes into contact with the product, the pressure sensor (3-5) records the change value of pressure K over time in real time. With the target pressing pressure K0 (0.3–0.9 MPa) as the set value, the pressure sensor feeds back the PID adjustment of the servo motor torque. At the same time, the temperature sensor B (3-16) detects the temperature drop trend in real time. The pressure K is kept constant and the unit enters the pressure holding stage. When the temperature drops to the setting temperature T, the unit enters the pressure holding stage. d After riveting and pressing are completed, the product returns to the initial galvanometer processing position; the CCD camera (1-8) records the two-dimensional image of the plastic riveted product, and establishes a sample database by using the two-dimensional image data. The riveting graphic information and standards of qualified products are defined by human beings through actual products, and a one-to-one correspondence with the pressure curve of the pressed product is established as the basic identification data of the software system.

5. The laser riveting processing device according to claim 1, characterized in that: A four-degree-of-freedom precision motion platform (4) is installed in the middle of the equipment frame unit (5) and located below the base plate support (2-10). It includes an X-axis horizontal linear motion module, a Y-axis horizontal linear motion module (4-8), an A-axis rotary servo motor (4-7), a B-axis rotary servo motor (4-14), and a flexible gripper (4-5). The X-axis horizontal linear motion module includes an X-axis horizontal linear active module (4-10) and an X-axis horizontal linear follower module (4-1) to complete linear motion in the X-axis direction. The X-axis horizontal linear active module (4-10) is the power source module. The Y-axis horizontal linear motion module (4-8) is installed between the X-axis horizontal linear active module (4-10) and the X-axis horizontal linear follower module (4-1) via the module connecting plate (4-9). The Y-axis horizontal linear motion module (4-8) is equipped with left and right mounting plates (4-16). A B-axis motor fixing plate (4-4) is installed between the left and right mounting plates via a rotating shaft. An A-axis rotary servo motor (4-7) is installed on the outside of the right mounting plate. The output shaft of the A-axis rotary servo motor is fixedly connected to the B-axis motor fixing plate (4-4) via the rotating shaft, thereby driving the B-axis motor fixing plate (4-4) to rotate a certain angle on the A-axis, and driving the plastic riveting sample (4-15) to rotate a certain angle on the A-axis. The B-axis motor mounting plate (4-4) is fixedly installed at the bottom of the B-axis rotary servo motor (4-14). The B-axis motor mounting plate has a B-axis rotary table (4-11) on it. The output shaft of the B-axis rotary servo motor (4-14) is fixedly connected to the B-axis rotary table (4-11), thereby driving it to rotate a certain angle on the B-axis, and driving the plastic riveting sample (4-15) to rotate a certain angle on the B-axis. The A-axis rotary servo motor (4-7) provides the rotational motion of the A-axis, which is transmitted to the B-axis motor mounting plate (4-4). The B-axis rotary servo motor (4-14) is fixed on the B-axis motor mounting plate (4-4) to realize the rotational motion of the B-axis worktable (4-11) on the B-axis. Two flexible grippers (4-5) are mounted on the B-axis rotary table (4-11); The plastic riveting sample (4-15) is placed on the upper part of the B-axis rotary table (4-11). Two flexible grippers press the plastic riveting sample from both sides simultaneously. The X-axis horizontal linear active module (4-10) and the Y-axis horizontal linear motion module (4-8) adjust the position of the plastic riveting sample (4-15) on the X and Y axes, moving the plastic riveting sample directly below the galvanometer assembly (2-3). The A-axis rotary servo motor (4-7) and the B-axis rotary servo motor (4-14) complete the posture adjustment of the plastic riveting sample (4-15) in two rotational directions, adapting to the riveting requirements of riveted parts at multiple angles. The flexible grippers can flexibly adapt to and fix plastic riveting samples of different shapes, minimizing damage to the plastic riveting sample during product pressing.

6. The laser riveting processing device according to claim 5, characterized in that: The dust suction nozzle (4-3) is fixed on the left mounting plate (4-16) through a smoke and dust connection plate (40), and the dust suction nozzle can absorb the smoke and dust generated by the dual-wavelength composite laser light source unit (1) in the air; The equipment frame unit (5) provides support and fixation for the entire upper part of the equipment, the middle part of the frame body is used to store various control unit components, and the bottom caster assembly is used for equipment movement and leveling; The industrial control computer is internally equipped with an integrated electrical assembly control unit (6), which adopts CNN-LSTM forward and backward prediction and uses the CNN-LSTM model for model training. CNN is mainly used for the features of riveting images, including the shape of the riveting structure of the sample and surface flatness; LSTM is used to capture the correlation between process parameters and image features. The process parameters include pressing pressure parameters, laser power parameters, and scanning speed, which can realize forward prediction of riveting joint image information.

7. An automatic focusing method for a laser riveting processing apparatus according to claim 1, characterized in that: It includes the following steps: (1) Based on preset CAD model data or preset process parameters, the coplanarity of the initial riveting points of the plastic riveting specimen is determined; If it is determined that all the riveting points to be processed are on the same processing plane, the focal length pre-adjustment operation of the galvanometer assembly (2-3) is preferentially executed to lay a reference for subsequent riveting processing; (2) The galvanometer assembly (2-3) moves up and down with the Z-axis vertical linear motion module (2-2). The focusing of the laser processing focal length can determine the distance from the focal point through the image clarity evaluation function of the CCD camera (1-8) in the computer, and the stroke of the Z-axis vertical linear motion module (2-2) is inversely deduced based on the image clarity using an image vision algorithm; the details are as follows: The coaxial optical path CCD camera (1-8) collects the image of the riveting area of the workpiece, and extracts the defocus state information of the vertical Z-axis based on the imaging clarity of the riveting contour; the system has a built-in image clarity evaluation function that performs quantitative calculation on the real-time image and outputs the real-time clarity evaluation value F0; the preset standard clarity evaluation value is taken as the threshold F, and when the real-time evaluation value F0<F, the automatic focusing process is triggered; The software compares the real-time evaluation value F0 with the preset threshold F. If F0<F, it is determined that the galvanometer assembly (2-3) has Z-axis defocus, and the focusing displacement H needs to be executed; at the same time, through the hill-climbing search algorithm with a step size of 0.01mm and an accuracy of ±0.005mm, the optimal movement direction of the Z-axis is determined to improve the imaging clarity; the controller accurately issues the focusing instruction to the Z-axis vertical linear motion module (2-2), driving the galvanometer assembly (2-3) to complete the closed-loop movement of displacement H along the Z-axis; when the real-time clarity evaluation value F0 satisfies |F0-F|≤A, where A is the preset allowable defocus threshold and A=0.02, it is determined that the focusing is completed, and the galvanometer assembly stops moving; (3) If it is determined that the riveting points are not in the same processing plane, the A-axis servo motor (4-7) and B-axis servo motor (4-14) in the four-degree-of-freedom precision motion platform (4) are controlled to work together to drive the plastic riveting sample (4-15) to rotate and adjust on the A-axis and B-axis. The rotation range of the A-axis is ±90°, and the target riveting point of the plastic riveting sample (4-15) is adjusted to a horizontal state to ensure the consistency of the processing reference surface; then the operation of step (2) is performed.

8. A method for XY coordinate positioning and path optimization using a coaxial optical path CCD camera in a laser riveting processing apparatus according to claim 7, characterized in that: Includes the following steps: (1) CCD camera acquires template image information: The riveting point feature image of qualified plastic workpiece is acquired in advance as a positioning template, which includes the precise outline of the rivet post and reference hole of the plastic workpiece, and the riveting point coordinates (X0, Y0) of the sample template plastic workpiece are obtained. (2) Image feature matching for coarse localization; Real-time image acquisition of plastic workpieces; using a template matching algorithm to perform coarse positioning of the plastic workpieces and obtain the pixel coordinates (U) of the riveting points. i V i ); (3) Pixel coordinate to physical coordinate transformation: Based on the calibrated coordinate transformation matrix, the pixel image coordinates (U, U, U) of the identified riveting points are transformed into physical coordinates. i V i The resulting transformation matrix H is used to convert the coordinates into physical coordinates (X). i Y i ); (4) Calculate the unit deviation: Calculate the positional deviation Δx = X0 - X i △y=Y0- Y i If the deviation is small, use a galvanometer to compensate for Δx and Δy to eliminate the deviation; if the deviation is large and exceeds the threshold, the machine will alarm and require manual handling. (5) Triggering laser processing: Positioning is completed, and the dual laser riveting process is started. The laser begins to heat and soften the plastic workpiece to be riveted. (6) When the plastic workpiece reaches the softening riveting temperature threshold To, the laser heating is immediately terminated; (7) The four-degree-of-freedom precision motion platform moves the plastic workpiece so that the riveting point is moved to the lower part of the thermoplastic molding unit. The mold head of the thermoplastic molding unit moves downward to apply pressure to the riveting point of the plastic workpiece. According to the preset pressure Ko, the pressure sensor measures the actual pressure value K. If the pressure difference |K-Ko|≤D, the pressure value C is maintained. If the pressure difference is greater than D, the PID adjustment of the servo motor torque is used to increase the pressure. (8) When the temperature of the plastic workpiece drops to ≤ the mold setting temperature T d Then the mold pressure head stops holding pressure and moves upward, leaving the plastic workpiece; (9) The four-degree-of-freedom precision motion platform moves the plastic workpiece so that its riveting point is moved to the set position under the galvanometer, and the CCD camera captures the standard image of the riveting point of the plastic workpiece. (10) If there is another point to be riveted on the plastic workpiece, continue processing; otherwise, end processing.

9. A method for forward predictive execution flow in a laser riveting processing apparatus according to claim 1, characterized in that: The core of the forward prediction execution process for the final processed product of plastic riveting samples is to construct a mathematical mapping relationship between process parameters and processing quality based on historical experimental data. When new process parameters are input, the system uses a CNN-LSTM model for model training to automatically calculate the expected pressing quality score and the final pressing effect, thereby reducing debugging costs and shortening the debugging cycle. The specific execution steps are as follows: (1) Experimental design and data acquisition: Clarify the range of process parameters and determine the feasible range of processing material type, material color, riveting structure size, laser power, scanning speed and pressing pressure; Design an experimental scheme, and in addition to the product material, select 30-50 sets of parameter combinations evenly within the feasible range for each process parameter. Each set of parameter combinations should be repeated at least 2-3 times, and the average value of the experimental results should be taken to reduce random errors. Perform laser softening and pressing experiments according to the preset scheme. Before and after each experiment, use a CCD camera to collect images of the riveting area and record relevant data. Simultaneously record seven sets of core process parameters: processing material type, material color, riveting structure size, laser power, scanning speed, pressing pressure, and sample temperature. Construct an experimental data table, using a row structure to correspond to a single set of experiments. Each row should completely record the process parameters, experimental data, and image information of that set of experiments. (2) Definition of quality scoring standards: Determine the key indicators affecting riveting quality, mainly including core parameters such as riveting width, flash size, surface flatness, and surface finish; formulate scoring rules, set ideal standard values ​​and allowable deviation ranges for each key quality indicator, and clarify the criteria for judging whether the indicators meet the standards; establish a quality scoring function, assign reasonable weights to each key quality indicator, and obtain a comprehensive riveting quality score by weighted summation of the scores of each indicator; carry out manual verification and calibration work, extract 10-15 typical experimental images, first automatically complete the quality scoring, and then have senior technicians perform manual scoring, compare the differences between the two scores, iteratively adjust the weight coefficients and judgment thresholds of the quality scoring function, so that the automatic scoring results are consistent with the manual judgment, and then complete the batch scoring of all experimental samples based on the calibrated quality scoring function; (3) Prediction model construction: The experimental data table and quality score data are integrated to form a standardized dataset. Each row of the dataset contains eight process parameters and their corresponding comprehensive quality scores: material type, material color, riveting structure, laser power, scanning speed, pressing pressure, temperature, and CCD riveting image. The CNN-LSTM model is selected as the prediction model. The CNN is responsible for extracting image feature information, including material type, material color, riveting structure, and CCD image information. The LSTM is responsible for capturing the temporal / coupling rules between parameters, including laser power, scanning speed, pressing pressure, and temperature. The image features extracted by the CNN and the temporal features of the process parameters extracted by the LSTM are spliced / fused. The material type, material color, riveting structure, laser power, scanning speed, pressing pressure, and temperature of the training set are used as input variables, and the comprehensive quality score of the CCD image riveting is used as the target output variable. The nonlinear mapping relationship between the input and output is fitted through model iterative training. The trained model is used to predict the quality score of the process parameters of the test set. The predicted score is compared with the actual quality score of the test set to determine the deviation, and the process parameters are automatically iterated. (4) Deployment of prediction function: A visual operation interface is built to support users to manually input process parameters such as processing material type, material color, riveting structure, laser power, scanning speed, pressing pressure, and temperature; after the user inputs the parameters, the system first verifies the rationality of the parameters. If the input parameters exceed the preset feasible range, a warning prompt will be automatically issued, while the prediction function is retained, allowing the user to continue to perform the prediction operation; based on the input parameters, the system calculates and outputs the expected pressing quality score and the final pressing effect through the trained prediction model, and at the same time provides corresponding process parameter optimization suggestions; (5) Continuously collect process parameters and processing quality data in the actual production process, and regularly update and iterate the prediction model to optimize the model parameters.

10. A method for reverse prediction of the execution process in a laser riveting apparatus according to claim 1, characterized in that: The reverse prediction execution process for plastic riveting specimens is as follows: Inputting an image of the riveting specimen in any final pressing state, the system uses a CNN-LSTM model to automatically derive and output a set of optimal process parameters, covering key parameters such as material type, material color, riveting structure shape, laser power, scanning speed, pressing pressure, and specimen temperature. The specific execution steps are as follows: (1) Construction of process parameter image feature database: Based on the dataset accumulated from historical experiments, a database linking process parameters and riveting image features is constructed; for the final pressing image corresponding to each set of experiments, four core characterization features are extracted and integrated into a one-dimensional feature vector, specifically including: riveting width, flash size, surface flatness, and surface finish; corresponding data tables are established simultaneously to ensure that each set of process parameters and the extracted feature vector form a unique and direct mapping relationship; (2) Standardized extraction of query image features: The image is normalized in size and corrected in illumination to eliminate size differences and illumination interference; then, four core features consistent with the database are extracted, and the normalization method is used to uniformly scale the numerical range of all features to the 0-1 range to ensure that each feature has an equal weight contribution in the subsequent retrieval process. (3) Similarity retrieval mechanism construction: The feature vector of the query image after standardization is compared with the feature vector of all historical images in the database one by one. The similarity is quantified by calculating the feature distance between the two. The smaller the feature distance, the higher the image similarity. The retrieval strategy adopts the K-nearest neighbor K-NN method and selects K=3 historical images with the smallest distance as the similar sample set. In the similar sample set, samples with feature distance <0.1 are given higher weight. If the feature distance of all samples exceeds the preset threshold, it is determined that there are no matching similar cases in the database. When there are no similar cases in the database, the initial parameters are generated by combining the process parameters of similar materials and similar structures in the existing database. After trial processing, the database is automatically learned and updated. (4) Optimal process parameter generation and confidence output: Based on the K most similar historical samples obtained by similarity retrieval, extract the corresponding three sets of process parameter combinations, and perform fusion calculation by weighted summation. The weight allocation is based on the feature distance between each sample and the query image. The smaller the distance, the higher the weight. Finally, a set of recommended optimal process parameter combinations is generated. When there are no similar cases in the database, the initial parameters are generated by combining the process parameters of similar materials and similar structures in the existing database. After trial processing, the database is automatically learned and updated. (5) Closed-loop optimization, model iterative update and self-learning of new cases: Establish a feedback mechanism for actual processing data in the production process, continuously collect new riveting images and corresponding process parameters and quality data; regularly incorporate new data into the feature database, iteratively optimize the process parameter-feature mapping relationship; for new working conditions without similar cases, supplement experimental data and expand the database, while adjusting the similarity calculation threshold and feature weight allocation rules, continuously improve the accuracy and generalization ability of reverse prediction, and realize the closed-loop optimization and long-term iteration of the system.

Images