A USB intelligent production and assembly method and system

Abstract

The present invention discloses a method and system for intelligent USB production and assembly. The method comprises: placing USB terminals on an assembly platform containing a non-uniform gradient magnetic field to generate a pre-positioned terminal group with spatial orientation characteristics; pressing the pre-positioned terminal group into the elastically deformed inner cavity of a USB shell, automatically correcting the terminal axial offset, and outputting a geometrically aligned semi-finished assembly; applying pulsed mechanical vibration to the semi-finished assembly, collecting the cavity resonant frequency spectrum within the assembly through a microphone array, triggering UV glue to instantly cure, and generating a physically locked USB finished product; applying swept frequency excitation to the finished USB product, extracting full-band voiceprint feature vectors, and inputting them into a pre-trained defect classification model to achieve automatic sorting of finished products. Using embodiments of the present invention, high-precision topology matching and dynamic stress optimization can be achieved in USB assembly, improving product consistency and defect detection efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of USB production technology, and in particular to a USB intelligent production and assembly method and system. Background Art

[0002] Traditional USB assembly processes generally rely on mechanical fixtures or visually assisted calibration, which can lead to low positioning accuracy, complex processes, and inconsistent yield rates. Especially with the trend toward miniaturization, microscopic deformation of the terminals and housing, as well as assembly stress, can easily lead to poor contact or mechanical fatigue. Existing technologies use magnetic positioning primarily for macroscopic parts, while acoustic inspection is typically limited to screening for external defects in finished products, making it difficult to achieve dynamic closed-loop control of the assembly process. Furthermore, curing processes often rely on fixed timing triggers and are unable to adapt to internal component stress states, compromising product reliability. Summary of the Invention

[0003] The purpose of the present invention is to provide a USB intelligent production and assembly method and system to address the deficiencies in the prior art, achieve high-precision topology matching and dynamic stress optimization for USB assembly, and improve product consistency and defect detection efficiency.

[0004] An embodiment of the present application provides a USB intelligent production and assembly method, the method comprising:

[0005] The USB terminals are placed on an assembly platform containing a non-uniform gradient magnetic field. The distribution difference of the magnetic material of the terminals is utilized to make the terminals rotate autonomously to a topological matching posture, thereby generating a pre-positioned terminal group with spatial orientation characteristics.

[0006] The pre-positioned terminal group is pressed into the elastically deformed inner cavity of the USB shell, and a directional deformation feedback force is generated by the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, which automatically corrects the axial deviation of the terminal and outputs a geometrically aligned semi-finished component;

[0007] Apply pulsed mechanical vibration to the semi-finished component, and collect the resonance frequency spectrum of the internal cavity of the component through a microphone array. When the main resonance peak is within a preset range, the UV glue is triggered to instantly cure, generating a physically locked USB finished product;

[0008] A frequency sweep excitation is applied to the finished USB product, and a full-band voiceprint feature vector is extracted and input into a pre-trained defect classification model. Automatic sorting of the finished product is achieved based on the cluster distribution of the feature vector in the acoustic space.

[0009] Optionally, placing the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, utilizing the difference in magnetic material distribution of the terminals to autonomously rotate the terminals to a topologically matching position, and generating a pre-positioned terminal group with spatial orientation characteristics, includes:

[0010] Activate the three sets of orthogonal electromagnetic coils at the bottom of the assembly platform, generate a rotating magnetic field through alternating current with a phase difference of 120 degrees, and output a dynamic magnetic field distribution field with spatial spiral gradient characteristics;

[0011] A high-sensitivity Hall sensor array is used to scan the magnetic flux distribution of the NdFeB magnetic coating on the terminal surface. The magnetic moment direction offset angle is calculated based on the dynamic magnetic field distribution field, and a real-time pose matrix containing X / Y / Z three-axis offset data is output.

[0012] The pose matrix is ​​input into the magnetic field controller, which dynamically adjusts the current intensity of the three coils to form a correction torque of 15-25mN·m between the magnetic field gradient direction and the magnetic moment offset angle, outputting a directional terminal array that completes autonomous rotation.

[0013] The angle between the terminal contact and the housing slot is measured by a confocal laser displacement sensor. When the angle is detected to be ≤0.3°, the electromagnetic locking mechanism is activated to output a pre-positioned terminal group with spatial orientation characteristics.

[0014] Optionally, the pre-positioned terminal group is pressed into the elastically deformed inner cavity of the USB shell, and a directional deformation feedback force is generated by the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, thereby automatically correcting the axial offset of the terminal and outputting a geometrically aligned semi-finished assembly, including:

[0015] Applying 40°C constant temperature hot air to the shell to reduce its elastic modulus to a preset threshold, outputting the shell in a critical state of plastic deformation;

[0016] The contact pressure distribution between the terminal and the tapered micro-rib is collected in real time through the micro-force sensor array, and a pressure gradient cloud map of the contact point is output;

[0017] Based on the pressure gradient cloud map, the pressure abnormality area is identified, and the axial offset angle is calculated by combining the micro-rib structure deformation model, and the offset correction vector parameters are output;

[0018] The offset correction vector parameters are input into the six-degree-of-freedom press-fitting robot, which dynamically adjusts the pressing trajectory and outputs a semi-finished product with an axial deviation of ≤10μm.

[0019] X-ray fluoroscopy is used to scan the fit between the terminal pins and the housing guide grooves to verify the geometric alignment and obtain a geometrically aligned semi-finished component.

[0020] Optionally, applying pulsed mechanical vibration to the semi-finished component, collecting the internal cavity resonance frequency spectrum of the component through a microphone array, and triggering instant curing of the UV glue when the main resonance peak is within a preset range to generate a physically locked USB finished product, including:

[0021] A 0.5-5kHz linear frequency sweep pulse is applied to the semi-finished component through a piezoelectric ceramic exciter, outputting a mechanical vibration spectrum containing 50 characteristic frequency points;

[0022] An 8-channel MEMS microphone array distributed in a ring is used to collect the internal sound pressure signal of the component. The first three resonance peak frequencies and Q factors are extracted through FFT, and the resonance frequency eigenvector is output.

[0023] When the resonance characteristic vector satisfies: main peak F1∈[1850,2150]Hz and Q1>55, a UV trigger signal with a 5ms delay is generated;

[0024] Through an array of multiple UV optical fibers, 365nm UV light is released within 3ms after the delay signal is triggered. The light intensity distribution is dynamically adjusted according to the vibration spectrum to output a physically locked USB product.

[0025] Optionally, applying a sweep frequency excitation to the finished USB product, extracting a full-band voiceprint feature vector and inputting it into a pre-trained defect classification model, and implementing automatic sorting of the finished product based on the cluster distribution of the feature vector in the acoustic space, includes:

[0026] Using a pneumatic sound source to sweep the frequency in the range of 20Hz-20kHz at a rate of 10Hz / ms, the vibration response of the finished product is synchronously collected and a time-stamped stimulus-response soundprint sample library is output;

[0027] Perform 6-layer db4 wavelet packet decomposition on the voiceprint sample, extract the energy entropy and kurtosis coefficient of 16 frequency bands, and output a 128-dimensional acoustic feature vector;

[0028] The feature vector is input into a pre-trained 3-layer convolutional autoencoder, which reduces the dimension to an interpretable 3D acoustic feature space and outputs a distribution topology map containing good / defective clusters.

[0029] Perform DBSCAN density clustering based on the distribution topology graph, combine it with the pre-trained SVM classifier to identify the defect patterns of outlier clusters, and output defect type labels and confidence scores;

[0030] The three-degree-of-freedom pneumatic sorting arm is controlled according to the defect type label. Good products are sent to the packaging area, defective products are returned for repair, and products with structural cracks are scrapped, completing the automatic sorting of finished products.

[0031] Another embodiment of the present application provides a USB intelligent production and assembly system, the system comprising:

[0032] The rotation module is used to place the USB terminal on an assembly platform containing a non-uniform gradient magnetic field. The difference in the distribution of the magnetic material of the terminal is used to make the terminal rotate autonomously to a topological matching posture, thereby generating a pre-positioned terminal group with spatial orientation characteristics;

[0033] A correction module is used to press the pre-positioned terminal group into the elastically deformed USB shell cavity, generate a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, automatically correct the axial deviation of the terminal, and output a geometrically aligned semi-finished assembly;

[0034] A curing module is used to apply pulsed mechanical vibration to the semi-finished component, collect the resonance frequency spectrum of the internal cavity of the component through a microphone array, and trigger the UV glue to instantly cure when the main resonance peak is within a preset range, thereby generating a physically locked USB finished product;

[0035] The sorting module is used to apply sweep frequency excitation to the USB finished product, extract the full-band voiceprint feature vector and input it into the pre-trained defect classification model, and realize automatic sorting of the finished product according to the cluster distribution of the feature vector in the acoustic space.

[0036] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0037] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0038] Compared with the prior art, the present invention provides a USB intelligent production and assembly method, which places USB terminals on an assembly platform containing a non-uniform gradient magnetic field to generate a pre-positioned terminal group with spatial orientation characteristics; presses the pre-positioned terminal group into the elastically deformed inner cavity of the USB shell, automatically corrects the axial offset of the terminals, and outputs a geometrically aligned semi-finished component; applies pulsed mechanical vibration to the semi-finished component, collects the cavity resonant frequency spectrum inside the component through a microphone array, triggers UV glue to instantly cure, and generates a physically locked USB finished product; applies swept frequency excitation to the USB finished product, extracts the full-band voiceprint feature vector, and inputs it into a pre-trained defect classification model to realize automatic sorting of finished products, thereby achieving high-precision topology matching and dynamic stress optimization of USB assembly, improving product consistency and defect detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A hardware structure block diagram of a computer terminal for a USB intelligent production and assembly method provided by an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a flow chart of a USB intelligent production and assembly method provided by an embodiment of the present invention;

[0041] Figure 3This is a structural diagram of a USB intelligent production and assembly system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0043] The embodiment of the present invention first provides a USB intelligent production and assembly method, which can be applied to electronic equipment, such as computer terminals, specifically ordinary computers, etc.

[0044] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal of a USB intelligent production and assembly method provided by an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0045] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any one of the USB intelligent production and assembly methods.

[0046] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0047] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any USB intelligent production and assembly method.

[0048] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0049] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0050] See also Figure 2 The embodiment of the present invention provides a USB intelligent production and assembly method, which may include the following steps:

[0051] S201, placing the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, utilizing the difference in magnetic material distribution of the terminals to autonomously rotate the terminals to a topologically matching position, thereby generating a pre-positioned terminal group with spatial orientation characteristics;

[0052] Specifically, the three sets of orthogonal electromagnetic coils at the bottom of the assembly platform can be activated to generate a rotating magnetic field through alternating current with a phase difference of 120 degrees, outputting a dynamic magnetic field distribution field with spatial spiral gradient characteristics;

[0053] Magnetic Field Generation Mechanism: Three mutually perpendicular (orthogonal) cylindrical electromagnetic coils are embedded in the base of the assembly platform. Each coil is wound with 300 turns of high-purity oxygen-free copper wire and has a coil diameter of 50 mm. The control system feeds sinusoidal alternating current at a frequency of 50 Hz to each of the three coils, with a strict phase difference of 120 degrees (one-third of a cycle). For example, when the current phase of coil A is 0 degrees, that of coil B is 120 degrees, and that of coil C is 240 degrees. This phase difference ensures that the magnetic field vectors generated by the three coils, when superimposed in space, form a rotating magnetic field. The magnetic flux lines (MFLs) are not stationary but rotate around the vertical axis at a speed of 50 revolutions per second. Due to the orthogonal arrangement of the coils, the magnetic flux lines exhibit a spiral gradient in three dimensions: the magnetic field intensity is highest near the center of the platform (approximately 0.5 Tesla, T) and decreases radially toward the outer edges (to 0.05 T). This non-uniform gradient provides directional guidance for the magnetic field.

[0054] Precision Control of the Dynamic Magnetic Field: The stability of the rotating magnetic field is ensured by a digital signal processor (DSP). The DSP outputs three phase-shifted current signals through a high-precision digital-to-analog converter (DAC). These signals are then driven by a power amplifier (with a peak output current of 20 amperes). A real-time current monitoring module (Hall current sensor with an accuracy of ±0.1%) provides feedback on the actual current value. If a phase shift exceeding ±1 degree or an amplitude fluctuation exceeding ±2% is detected, the DSP immediately adjusts the output signal to compensate. This closed-loop control ensures that the generated dynamic magnetic flux distribution field maintains the preset spiral gradient characteristics at any point in space, providing a controllable magnetic environment for the subsequent autonomous rotation of the terminal.

[0055] Physical effects of magnetic field characteristics: The core function of the spiral gradient magnetic field is to exert directional torque on magnetic objects. When a USB terminal (with magnetic coating) is placed in this field, its internal magnetic moment is affected by two key forces:

[0056] Gradient force: points to the area with higher magnetic field intensity (center of the platform), causing the terminals to gather towards the positioning area;

[0057] Rotational torque: Due to the angle between the rotation direction of the magnetic field and the direction of the terminal magnetic moment, a continuous torque is generated to drive the terminal to rotate.

[0058] For example, a neodymium iron boron (NdFeB) coated terminal is forced to rotate 50 times per second in this field until the direction of its magnetic moment is aligned with the axis of rotation of the magnetic field, achieving preliminary orientation.

[0059] A high-sensitivity Hall sensor array is used to scan the magnetic flux distribution of the NdFeB magnetic coating on the terminal surface. The magnetic moment direction offset angle is calculated based on the dynamic magnetic field distribution field, and a real-time pose matrix containing X / Y / Z three-axis offset data is output.

[0060] Magnetic Flux Distribution Scanning Technology: 256 miniature Hall Effect sensors (±0.1 millitesla (mT) sensitivity) are integrated into an 8×32 array 1 mm below the assembly platform surface. When a terminal is placed on the platform, the array scans the magnetic flux distribution (MFD) of the projected area beneath it at a rate of 1000 times per second. Each sensor measures the magnetic flux density (in Tesla) perpendicular to its sensing surface and transmits it to a field-programmable gate array (FPGA). Because the NdFeB coating is 50 microns (μm) thick in the contact area of ​​the terminal and only 10 μm in the non-contact area, this thickness difference results in a characteristic magnetic flux distribution (approximately 15% higher in the contact area), which directly reflects the physical orientation of the terminal.

[0061] Principle of deviation angle calculation: FPGA compares the scanned magnetic flux data with the pre-stored dynamic magnetic field distribution model (including the theoretical magnetic field direction and intensity at each point in space). By solving the following physical relationship:

[0062] The actual magnetic moment direction of the terminal (determined by the peak position of the magnetic flux);

[0063] The theoretical magnetic field direction at the current position (given by the rotating magnetic field model).

[0064] The angle between the two spatial vectors is the magnetic moment deviation angle (Magnetic Moment Deviation Angle). This is calculated using the Direction Cosine Matrix (DCM) algorithm: the actual magnetic moment vector and the theoretical magnetic field vector are projected onto the X / Y / Z axes, and the inverse cosine function is used to calculate the deviation component (unit: degree) for each axis.

[0065] Generation and output of the pose matrix: The offset angle calculation result is formatted as a real-time pose matrix (Real-timePoseMatrix), which is an example of a 3-row × 1-column data structure:

[0066] [ ΔX_angle ] / / X-axis offset angle, range: -5° to +5°;

[0067] [ΔY_angle] / / Y-axis offset angle, range: -5° to +5°;

[0068] [ΔZ_angle] / / Z-axis (rotation axis) tilt angle, ranging from -2° to +2°.

[0069] Matrix data is transmitted to the magnetic field controller via Industrial Ethernet at a frequency of 200 times per second, providing a basis for real-time correction. For example, detecting ΔY_angle = +3.2° means that the terminal is deflected 3.2 degrees clockwise around the Y axis.

[0070] The pose matrix is ​​input into the magnetic field controller, which dynamically adjusts the current intensity of the three coils to form a correction torque of 15-25mN·m between the magnetic field gradient direction and the magnetic moment offset angle, outputting a directional terminal array that completes autonomous rotation.

[0071] Closed-loop control of the corrective torque: After receiving the pose matrix, the magnetic field controller (based on a real-time operating system (RTOS)) generates current regulation commands based on a preset proportional-integral-derivative (PID) control algorithm. The control objective is to stabilize the torque applied to the terminal within a range of 15-25 millinewton-meters (mN·m)—a torque value that has been experimentally verified to overcome the static friction between the terminal and the platform (approximately 10 mN·m) while avoiding terminal oscillation caused by excessive torque. For example, when an offset angle ΔX_angle of +4° is detected, the PID algorithm calculates that the X-axis coil current needs to be increased by 8%, causing the magnetic field gradient to deflect counterclockwise, thereby applying a negative corrective torque.

[0072] Dynamic adjustment of current intensity: The controller outputs three pulse width modulation (PWM) signals to the power amplifier to adjust the current intensity of the three sets of coils (adjustment range: 0-25 A, resolution ±0.01A). The adjustment strategy follows:

[0073] Orthogonal superposition principle: X-axis coil current dominates horizontal plane rotation correction;

[0074] Gradient enhancement principle: The Z-axis coil current enhances the magnetic field gradient in the central area, improving positioning stability.

[0075] For example, when correcting a +4° offset, the X-coil current is increased from 12.5 A to 13.5 A, and the Y / Z-coil current is fine-tuned by ±0.2 A to compensate for coupled interference.

[0076] Orientation Completion: When the pose matrix indicates a three-axis deviation angle of less than ±0.5° for 500 milliseconds (ms), the terminal is considered to have completed autonomous rotation and entered the Oriented Terminal Array state. At this point, the magnetic moment direction of all terminals deviates from the magnetic field rotation axis by ≤0.5°, and contact orientation consistency reaches 99.7%, providing precise input for the press-fit process.

[0077] The angle between the terminal contact and the housing slot is measured by a confocal laser displacement sensor. When the angle is detected to be ≤0.3°, the electromagnetic locking mechanism is activated to output a pre-positioned terminal group with spatial orientation characteristics.

[0078] High-precision angle measurement: A confocal laser displacement sensor is deployed on the side of the platform. Its core indicators are:

[0079] Wavelength: 650 nanometer (nm) red laser; Measurement accuracy: ±0.05 micrometer (μm); Sampling rate: 10 kilohertz (kHz).

[0080] The sensor transmits a laser beam at a 30-degree angle toward the terminal contact area. By analyzing the reflected light's spot position offset and applying triangulation, it calculates the angle (in degrees) between the contact plane and the ideal slot plane. For example, a detected spot offset of 15 microns corresponds to an angle of 0.25°.

[0081] Working mechanism of the electromagnetic locking mechanism: There are 12 sets of electromagnetic locking mechanisms embedded inside the platform, each set includes: electromagnet: generating a 1.2 T magnetic field when energized; soft magnetic alloy locking pin: with a diameter of 2 mm.

[0082] When the sensor detects an angle ≤0.3° (threshold) three times in a row, the controller sends a 12 volt (V) DC voltage to the corresponding electromagnet. The locking pin extends and presses the edge of the terminal within 0.1 seconds, applying a holding force of 50 millinewtons (mN), completely eliminating displacement caused by micro-vibration.

[0083] Output of the pre-positioned terminal group: After locking is completed, the system outputs the pre-positioned terminal group (Pre-positioned Terminal Group), whose key features are:

[0084] The spatial posture error of all terminals is ≤0.3°; the contact height difference is ≤10 microns; the standard deviation of the magnetic moment direction is ≤0.4°.

[0085] This state triggers the next process via an industrial bus (such as EtherCAT) and releases the current of the electromagnetic coil to save energy.

[0086] S202, pressing the pre-positioned terminal assembly into the elastically deformed inner cavity of the USB housing, generating a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the housing and the contact points of the terminals, automatically correcting the axial offset of the terminals, and outputting a geometrically aligned semi-finished assembly;

[0087] Specifically, a constant temperature hot air of 40°C may be applied to the shell to reduce its elastic modulus to a preset threshold value, thereby outputting a shell in a critical state of plastic deformation;

[0088] On the USB assembly line, the USB housing undergoes a heat softening treatment before the pre-positioned terminal assembly enters the press-fitting station. This process is performed by a constant-temperature hot air generation system, which includes a high-precision temperature control module, a circular air duct array, and an infrared temperature monitoring unit. The hot air generator first passes filtered compressed air into a ceramic heating chamber, where it uses a PID (proportional-integral-derivative) temperature control algorithm to stabilize the airflow at 40°C (±0.5°C). The key rationale for choosing 40°C is that this temperature represents the glass transition starting point of the engineering plastics used (such as PC / ABS alloy). (The Tg starting point is approximately 65-85°C, but 40°C significantly reduces the resistance to molecular chain motion.) At this point, the material's elastic modulus (EM), which characterizes its resistance to deformation, drops from 2200 MPa at room temperature to approximately 850 MPa, reaching the predetermined "critical plastic deformation threshold." This means that the material still retains elastic recovery, but its yield strength (YS) drops to a level sufficient for controlled plastic deformation. A circular air duct design ensures that hot air evenly coats the housing surface, preventing localized overheating and deformation. An infrared thermometer scans the housing surface 20 times per second, providing real-time temperature distribution data to the temperature control module for dynamic adjustment of heating power.

[0089] The key to the thermal softening process lies in precisely controlling the material state. When the housing surface temperature reaches 40°C, the internal polymer chain mobility increases, weakening intermolecular forces and causing the elastic modulus (EM) to drop to a preset threshold range (800-900 MPa). This threshold was determined through preliminary material mechanical testing: below 800 MPa may result in permanent deformation, while above 900 MPa results in insufficient deformation feedback. The duration of the hot air application is strictly controlled to 3.5 seconds (±0.2 seconds), timed by a PLC (Programmable Logic Controller). The housing is radially constrained in the fixture, with only the inner cavity open for terminal press-in. The mapping between temperature and elastic modulus is stored in a database. When the infrared thermometer confirms that the overall housing temperature uniformity error is ≤1°C and the duration meets the specified threshold, the system outputs a "critical plastic deformation state" signal. The housing in this state exhibits two key characteristics: first, the tapered micro-ribs on the inner wall allow elastic bending under minimal pressure; second, it retains sufficient resilience to ensure the effectiveness of subsequent deformation feedback.

[0090] To achieve the critical state of plastic deformation, it is necessary to overcome environmental interference. The production line is equipped with an ambient temperature and humidity sensor (monitoring range 0-50°C, accuracy ±0.1°C). When the ambient temperature exceeds 25°C, the hot air temperature is automatically reduced (such as to 38°C) to prevent the superposition effect from causing overheating of the material. The batch differences of the shell are compensated by an online laser thickness gauge: if the wall thickness is detected to increase by 0.1 millimeters (mm), the heating time is extended by 0.15 seconds to ensure the depth of heat conduction. The shell that has completed the heat treatment is transferred by the robot to the pressing station, and an insulation cover is used to maintain temperature stability during the transfer process. The entire process data (temperature curve, material batch, environmental parameters) is uploaded to the MES (manufacturing execution system) for quality traceability and process optimization.

[0091] The contact pressure distribution between the terminal and the tapered micro-rib is collected in real time through the micro-force sensor array, and a pressure gradient cloud map of the contact point is output;

[0092] After the press-fit process begins, a servo motor drives the pre-positioned terminal assembly into the heat-softened housing cavity. Embedded within the housing are 120 spirally arranged tapered micro-ribs (0.3 mm high, 60-degree taper angle). These guide terminal positioning and detect contact anomalies. Integrated into the base of the press-fit fixture is a micro-force sensor array—composed of 8×8 piezoresistive sensing units (1×1 mm²), each corresponding to a specific area at the base of the ribs. When the terminal contact (typically a gold-plated copper sheet) contacts the micro-ribs, the ribs undergo micrometer-scale elastic bending, and stress changes at the base are captured by the underlying sensing units. The sensor operates at a sampling frequency of 5000 Hz and a resolution of 0.01 Newton (N), generating real-time pressure values ​​(PVs) at 64 discrete points.

[0093] After amplification and filtering by the signal conditioning circuit, the raw pressure data is fed into the pressure distribution reconstruction module. This module uses the bicubic interpolation algorithm to expand the 64 discrete points into a two-dimensional pressure distribution matrix of 512×512 pixels. Each pixel in the matrix represents the virtual pressure intensity at that location, ranging from 0 to 2 Newtons (N), and is color-mapped from blue (low pressure) to red (high pressure). The "Contact Pressure Gradient Cloud Map" (CPGCM) generated from this matrix contains two key pieces of information: the absolute pressure distribution (reflecting the degree of contact tightness) and the gradient of the pressure change rate (identifying areas of abnormal stress concentration). For example, during normal press assembly, the cloud map should appear uniformly green (pressure 1.0±0.2N). However, if a red patch (pressure >1.5N) appears in a region accompanied by a high-gradient boundary (pressure difference per unit distance >0.3N / mm), it indicates a risk of axial misalignment.

[0094] The cloud map generation process incorporates a dynamic compensation mechanism. Due to the thermal softening of the outer shell, the stiffness of the micro-ribs drifts with temperature (sensitivity decreases by 0.8% for every 1°C increase in temperature). The system uses real-time temperature feedback to calibrate the raw pressure value (for example, multiplying by a compensation factor of 1.06 at 40°C). A motion artifact suppression algorithm is also employed: when the press speed reaches 5 mm / s, sensor data is synchronized with a high-speed vision positioning system (200 frames / second) to eliminate signal jitter caused by mechanical vibration. The final output pressure gradient cloud map is refreshed at a 50 millisecond (ms) update cycle, providing a high-refresh rate input for subsequent offset correction.

[0095] Based on the pressure gradient cloud map, the pressure abnormality area is identified, and the axial offset angle is calculated by combining the micro-rib structure deformation model, and the offset correction vector parameters are output;

[0096] Pressure anomaly detection is performed by the edge computing unit. This unit loads a pre-trained abnormal region identification model (based on the U-Net convolutional neural network architecture) to perform semantic segmentation on the real-time pressure gradient cloud image. The model first divides the cloud image into 5×5 mm grids and extracts four features from each grid: mean pressure (MP), pressure standard deviation (PSD), maximum pressure gradient (MPG), and high-temperature zone flag (HTZF). A grid is marked as a "Pressure Anomaly Region" (PAR) if it simultaneously meets the following conditions: MP > 1.4N, PSD > 0.25, and MPG > 0.3N / mm. Detection sensitivity can be adjusted using the confidence threshold (default 0.9) to avoid false triggers.

[0097] For each PAR, the system uses the Micro-Rib Deformation Model (MRDM) to calculate the offset. This model is a simplified finite element model that incorporates material constitutive equations (describing the stress-strain relationship of plastics at 40°C) and geometric constraints (rib taper angle, height, and wall thickness). The input parameters are the pressure distribution data for the abnormal area. The core calculation steps include: first, decomposing the pressure distribution into a normal component (causing rib bending) and a tangential component (causing shear at the rib root); second, calculating the bending deflection curve based on the rib cantilever beam characteristics; and third, fitting the angle between the ideal plane and the actual plane of the terminal contact surface using the least squares method. The final output is the "Axial Offset Angle" (AOA), with an accuracy of 0.01 degrees and a range of ±2 degrees. For example, if the average pressure detected at a PAR is 1.6N, the model may output an AOA of 0.75 degrees (biased toward the positive Y-axis).

[0098] The offset angle must be converted into a "Correction Vector Parameter" (CVP) operable by the actuator. This parameter is an array consisting of three elements: the X / Y-axis plane offset (in microns), the Z-axis rotation compensation angle (in degrees), and the dimensionless press force compensation coefficient. The conversion rule is: for every 0.1-degree increase in AOA, the plane offset increases by 15 microns (a linear relationship). If the angle between the AOA direction and the press speed direction is greater than 45°, additional Z-axis rotation compensation is applied (0.2° rotation for every 30 microns of offset). The press force compensation coefficient is adjusted based on the PAR area ratio (for every 10% increase in the ratio, the coefficient decreases by 0.05). All parameters are smoothed using a Kalman filter before output to prevent high-frequency jitter in the robot arm.

[0099] The offset correction vector parameters are input into the six-degree-of-freedom press-fitting robot, which dynamically adjusts the pressing trajectory and outputs a semi-finished product with an axial deviation of ≤10μm.

[0100] The calibration vector parameters (CVP) are transmitted via real-time Ethernet to the control system of a six-degree-of-freedom (6-DOF) press-fitting robot. The robot consists of three linear motion modules (X / Y / Z axes) and three rotary joints (A / B / C axes), achieving a repeatability of ±2 microns (μm). The control system employs a feedforward-feedback hybrid control strategy: the feedforward channel directly generates position compensation commands based on the plane offsets in the CVP (e.g., X +120μm, Y -80μm). The feedback channel uses a laser interferometer to monitor the actual terminal position in real time (sampling rate 1kHz), compares it with the ideal trajectory, and generates a PID (proportional-integral-derivative) correction. These two values ​​are then combined to drive the linear motion module, ensuring the terminal is pressed along the corrected trajectory.

[0101] Dynamic trajectory adjustment requires resolving kinematic conflicts. When the CVP requires simultaneous translation (e.g., +100μm on the X-axis) and rotation compensation (e.g., 0.5° rotation around the Z-axis), the robot's motion planner uses a pose separation algorithm: first, the terminal's posture is adjusted via the A / B rotational joint (taking ≤30ms), and then the X / Y module performs translation (taking ≤50ms). The press-fitting speed is controlled in stages based on deformation feedback: initially, it approaches quickly at 3mm / s, decreasing to 0.8mm / s at 0.5mm from the contact point. After contact, the speed is adjusted based on the press-fitting force compensation coefficient (0.6mm / s at a coefficient of 0.9, and 1.0mm / s at a coefficient of 1.1). The end effector is integrated with a micro-force closed-loop module to ensure that the total press-fitting force remains stable within the range of 45±3N.

[0102] Precision verification and output control. After press-fitting is complete, the robotic arm maintains the terminal position for 50ms to allow material stress relaxation. A high-precision capacitive displacement sensor (resolution 0.1μm) scans the terminal end face and measures its parallelism with the housing reference plane. When the axial deviation (AD) is confirmed to be ≤10μm (i.e., the tangent of the angle between the terminal center axis and the theoretical housing axis is ≤0.00017), the system outputs a "qualified semi-finished product" signal and unlocks the fixture. If the deviation exceeds the standard (e.g., AD > 12μm), a secondary correction is triggered: the robotic arm retracts 0.3mm, recalculates the CVP based on the latest pressure contour, and performs compensatory press-fitting (repeated up to two times). All semi-finished products are marked with a laser traceability code to record key parameters of the press-fitting process (maximum pressure value, number of corrections, and final deviation value).

[0103] X-ray fluoroscopy is used to scan the fit between the terminal pins and the housing guide grooves to verify the geometric alignment and obtain a geometrically aligned semi-finished component.

[0104] Geometric alignment is verified using a Micro-Focus X-ray Inspection System (MFXIS). Semi-finished components are transported by conveyor into a shielded lead room. An X-ray source (50 kilovolts (kV), 80 microamperes (μA)) emits a cone beam through the component. The detector is a 2048 × 2048 pixel flat-panel sensor (50 micron pixel size), with an exposure time of 80 milliseconds (ms). The resulting 3D volume data contains contrasting images of the housing plastic (low-density, dark gray) and the metal terminals (high-density, bright white), with sufficient resolution to distinguish gaps as small as 10 microns.

[0105] Fit analysis is performed using a dedicated algorithm. Material segmentation is performed first: metal / plastic regions are separated using the grayscale histogram bimodal thresholding method (Otsu's method). Key features are then extracted: the terminal pin edges are located (Canny edge detection) and the housing guide slot outlines (Hough transform line detection). The core verification metric is the "pin-slot gap" (PSG). The algorithm measures the distance from each side of the pin to the guide slot wall (ideally 25±5μm) by taking a cross-section every 0.1mm along the length of the pin. Three consecutive cross-sections with a PSG greater than 35μm or less than 15μm are marked as poorly fitted areas.

[0106] Final judgment and data integration. The system calculates three key metrics: Maximum Local Gap (MLG), Average Contact Ratio (ACR), and Axis Curvature (AC). Acceptance criteria are: MLG ≤ 40μm, ACR ≥ 92%, and AC ≤ 0.03mm / m. Qualified components are marked as "geometrically aligned semi-finished components" and move on to the next process. Rejected components are diverted based on failure mode (excessive gaps for repair, bent pins for scrap). All X-ray images and measurement data are linked to a traceability code and stored for ten years for quality analysis.

[0107] S203, applying pulsed mechanical vibration to the semi-finished component, collecting the resonance frequency spectrum of the internal cavity of the component through a microphone array, and triggering the instant curing of the UV adhesive when the main resonance peak is within a preset range to produce a physically locked USB finished product;

[0108] Specifically, a 0.5-5kHz linear frequency sweep pulse can be applied to the semi-finished component through a piezoelectric ceramic exciter to output a mechanical vibration spectrum containing 50 characteristic frequency points;

[0109] Precise Piezoelectric Actuator Drive: The system utilizes a high-response piezoelectric actuator (PZA) as the vibration source. This actuator is constructed from stacked layers of lead zirconate titanate (PZT) ceramic sheets and is driven by a variable voltage between 0 and 100 volts (V). In response to commands from the control system, a digital signal generator (DSG) generates a series of linear sweep pulses (LSPs) within a strictly controlled frequency range of 0.5 kHz to 5 kHz, with a sweep rate set at 200 points per second. The sweep signal is amplified by a high-voltage amplifier (HVA) and applied to the PZT. This generates precise mechanical vibrations by leveraging the inverse piezoelectric effect of piezoelectric materials—where voltage changes induce microscopic deformation of the ceramic sheets. Vibration energy is transferred to the metal housing of the semi-finished component via a titanium alloy probe (with a contact area of ​​only 2 square millimeters), ensuring efficient energy injection without damaging the product.

[0110] Spectral Feature Extraction and Frequency Calibration: Vibration responses are collected in real time by a triaxial accelerometer (TAC) integrated into the shaker base. The sampling frequency is set to 20 kHz to cover the target frequency range. The raw vibration signal is input into a fast Fourier transform (FFT) processor, which uses a Hanning window function to reduce spectral leakage and generate a spectrum with a resolution of 90 Hz. The system extracts 50 feature frequency points (FFPs) from the spectrum, using the following selection criteria:

[0111] Amplitude threshold screening: only peaks with amplitudes 10 decibels (dB) higher than the background noise are retained;

[0112] Evenly distributed frequency band: 40 frequencies are evenly spaced within the range of 0.5-5kHz, with an additional 10 frequencies densely spaced in the resonance sensitive area (1.8-2.2kHz);

[0113] Harmonic Correlation: Automatically mark related pairs of fundamental frequency and second harmonic (such as 2kHz and 4kHz).

[0114] The final output Mechanical Vibration Spectrum (MVS) is stored in the form of a two-dimensional matrix, which contains four types of data: frequency point number, frequency value, amplitude, and phase angle.

[0115] Dynamic Environmental Compensation: To eliminate environmental interference (such as inherent vibration of the equipment), the system employs a dual-channel differential acquisition strategy: one sensor contacts the component under test, while the other is mounted on an adjacent fixture. The coherence function of the two-channel signals is calculated in real time, and a retest is triggered if it falls below 0.95. Furthermore, the exciter has a built-in temperature compensation chip that automatically reduces the drive voltage by 5% when the PZT temperature exceeds 45 degrees Celsius (°C), preventing thermal drift from affecting frequency accuracy.

[0116] An 8-channel MEMS microphone array distributed in a ring is used to collect the internal sound pressure signal of the component. The first three resonance peak frequencies and Q factors are extracted through FFT, and the resonance frequency eigenvector is output.

[0117] Optimized acoustic sensing array layout: An 8-channel MEMS microphone array (Micro-Electro-Mechanical Systems Microphone Array) is deployed in a circular pattern around the periphery of the component, with the array diameter precisely matching the dimensions of the USB casing (approximately 12 mm). Each microphone utilizes a back-pole capacitor structure, with a sensitivity of -38 decibels (dBV / Pa) and a frequency response range of 20 Hz to 10 kHz. The microphones are pointed at a 45-degree angle toward the component joints, utilizing a beamforming algorithm to enhance the directional acquisition of internal cavity sound pressure signals. Sound waves are transmitted through a 0.3 mm pressure relief hole in the casing. The array's synchronous sampling rate is set to 48 kHz, ensuring resolution of high-frequency harmonics above 5 kHz.

[0118] Extracting the core parameters of the resonance characteristics: After preprocessing the raw sound pressure signal (including DC offset correction and 0.3-6kHz bandpass filtering), it is input into the parallel processing Fast Fourier Transform (FFT) module. This module uses the Radix-2 algorithm to complete an 8192-point FFT calculation within 10 milliseconds and output the amplitude spectrum. The system extracts key parameters through the following steps:

[0119] Peak detection: Identify local maximum points in the amplitude spectrum and sort them in descending order of amplitude;

[0120] Main peak screening: select the first three significant peaks (amplitude must be greater than 20dB of background noise) and record them as F1 (main peak), F2, and F3;

[0121] Q factor calculation: For each peak, the quality factor is calculated based on the -3dB bandwidth (Bandwidth at -3dB, BW) according to the formula Q=F / BW (for example, F1=2kHz, BW=36Hz, then Q1≈55.6).

[0122] Finally, the resonance frequency feature vector (RFV) is generated, and the data structure is [F1, Q1, F2, Q2, F3, Q3].

[0123] Anti-interference and signal enhancement technology: To address background noise on the production line (such as the low-frequency hum of fans), the system implements triple suppression:

[0124] Spatial filtering: Utilizes the spatial correlation of microphone arrays to suppress incoherent noise;

[0125] Adaptive notch: real-time identification of 60Hz / 120Hz power frequency interference and generation of inverse phase sound waves to cancel it out;

[0126] Coherent averaging: Performs time domain averaging on three consecutive sweep results to improve the signal-to-noise ratio.

[0127] When the signal-to-noise ratio (SNR) of any microphone channel falls below 40dB, a 20% gain boost is automatically triggered.

[0128] When the resonance characteristic vector satisfies: main peak F1∈[1850,2150]Hz and Q1>55, a UV trigger signal with a 5ms delay is generated;

[0129] Real-time decision logic for eigenvectors: The control system has a built-in resonance decision unit (RDU) to verify the RFV of each component item by item:

[0130] Frequency range verification: The main peak frequency F1 must be strictly between 1850Hz and 2150Hz (this range corresponds to the first-order axial vibration mode of the internal cavity of a good USB);

[0131] Q factor threshold verification: Q1 must be greater than 55 (a high Q value indicates low energy loss, indicating a uniform internal adhesive layer with no bubbles or cold solder joints);

[0132] Auxiliary peak correlation check: F2 / F3 must be an approximate integer multiple of F1 (to verify the consistency of the resonance mode).

[0133] When the above conditions are met simultaneously, the RDU outputs a logic level of "1" (pass), otherwise it outputs "0" (fail).

[0134] The design principle of the delayed trigger mechanism: When the judgment is passed, the system generates a UV trigger signal (Delayed UV Trigger Signal, DUTS) with a 5ms delay. The delay design is based on the following key considerations:

[0135] Mechanical vibration attenuation: reserve 3ms to allow the residual vibration of the piezoelectric exciter to decay to an amplitude of less than 0.1g (to avoid interference with curing);

[0136] Colloid leveling time: 2ms is reserved to allow the UV glue to recover and stabilize after the slight rheological changes caused by vibration and extrusion;

[0137] Photoelectric synchronization accuracy: The error between the rising edge of the trigger signal and the response time of the UV light source must be less than 0.1ms.

[0138] The delay is implemented by the Hardware Timer (HT), the clock source uses a 40MHz crystal oscillator, and the timing error is ±0.01ms.

[0139] Exception handling and process security:

[0140] If three consecutive components are judged to have failed, the system automatically performs the following actions:

[0141] Shaker self-test: Run a standard frequency response test (input a 1kHz sine wave and verify the output acceleration ±5% tolerance).

[0142] Microphone calibration: Play a 2kHz standard sound source and adjust the gain consistency of each channel to ±0.5dB;

[0143] Glue path diagnosis: UV intensity probe detects whether UV glue is expired (requires light intensity > 80mW / cm 2 ).

[0144] All diagnostic results are uploaded to the Manufacturing Execution System (MES) in real time, triggering maintenance work orders.

[0145] Through an array of multiple UV optical fibers, 365nm UV light is released within 3ms after the delay signal is triggered. The light intensity distribution is dynamically adjusted according to the vibration spectrum to output a physically locked USB product.

[0146] Precise control of the UV light delivery system: The UV Fiber Array (UFA) consists of 16 quartz optical fibers arranged in a grid pattern (4mm spacing), each terminated by a high-power 365nm UV LED (1.5 watts per LED). Upon receiving the rising edge of the DUTS signal, the LED driver circuit (in constant current mode, 500 mA) reaches full power within 0.5ms. After being transmitted through the optical fiber, the UV light is focused by a microlens into a 1.5mm diameter spot, evenly covering the UV adhesive layer (50 microns thick) at the seam of the USB casing. Key performance indicators include:

[0147] Time accuracy: Response time from triggering to light intensity reaching the target ≤ 1ms;

[0148] Spectral purity: 365nm main peak half width <10nm (to avoid side reactions caused by stray light);

[0149] Spatial uniformity: Light intensity fluctuation within the irradiated area is less than ±15%.

[0150] Adaptive adjustment strategy for light intensity distribution: The system dynamically adjusts the UV light intensity distribution based on the previously collected mechanical vibration spectrum (MVS). The strategy is as follows:

[0151] Enhanced curing in high-amplitude areas: If the amplitude of a certain frequency point exceeds the average by 20% (such as 2.1kHz), it is determined that there is a micro-gap in the corresponding area, and the fiber power at that location is increased by 20%;

[0152] Extended exposure in low-Q areas: If the local Q1 value is close to the threshold of 55, extend the pulse width from 50ms to 70ms (total energy increased by 40%).

[0153] Resonant mode matching: When the ratio of F2 / F3 to F1 deviates from the theoretical value (e.g. F2 / F1 < 1.8), the light intensity in the entire area is increased by 10% to compensate for the weak rigidity of the structure.

[0154] Light intensity adjustment data is sent to the multi-channel LED driver via a serial peripheral interface (SPI) with a resolution of 1024 levels.

[0155] Curing effect verification and finished product output: After curing is completed, the system ensures physical locking through three verifications:

[0156] Acoustic retest: Second frequency sweep test F1 deviation <±20Hz (glue layer rigidity meets the standard);

[0157] Thermal imaging inspection: infrared camera monitors the curing exothermic peak temperature rise of 40±5°C (reaction sufficiency verification);

[0158] Micro-torque test: The manipulator applies a torsional force of 0.2 Newton-meter (N·m), and the terminal displacement is less than 5 microns.

[0159] Components that pass verification are marked as USB Finished Products and sent to the next workstation via a conveyor belt. Simultaneously, all process parameters (spectrum, light intensity curve, and Q value) are written into the product's digital twin archive.

[0160] S204, applying sweep frequency excitation to the finished USB product, extracting full-band voiceprint feature vectors and inputting them into a pre-trained defect classification model, and realizing automatic sorting of finished products according to the cluster distribution of the feature vectors in the acoustic space.

[0161] Specifically, a pneumatic sound source can be used to sweep the frequency at a rate of 10 Hz / ms in the range of 20 Hz-20 kHz, synchronously collect the vibration response of the finished product, and output a time-stamped stimulus-response voiceprint sample library;

[0162] Pneumatic Acoustic Source Sweep Excitation Mechanism: The pneumatic acoustic source (PAS) is the core excitation device, employing high-pressure nitrogen to drive a precision piezoelectric ceramic diaphragm. When the control system issues a sweep command, a pneumatic valve adjusts the nitrogen flow rate with millisecond precision, causing the diaphragm to produce mechanical vibrations with continuously varying frequencies. The sweep frequency range covers the full audible frequency range (20 Hz to 20 kHz), with a sweep rate of 10 Hz per millisecond (10 Hz / ms), meaning the frequency increases by 10 Hz per millisecond. For example, starting at 20 Hz, the frequency increases to 30 Hz after 1 millisecond, and reaches 1020 Hz after 100 milliseconds. This high-speed sweep excites all potential resonance points within the finished USB device. A sound wave deflector is installed at the outlet of the pneumatic acoustic source to focus the acoustic energy on a 5 mm diameter circular area on the finished product's housing, ensuring that the excitation energy is concentrated and not diffused.

[0163] Vibration Response Synchronous Acquisition System: A triaxial MEMS vibration sensor (TMVS) is positioned opposite the pneumatic sound source. Its three sensing axes correspond to the X, Y, and Z directions of the spatial rectangular coordinate system. The sensor simultaneously acquires the USB product's surface acceleration signals (units: meters per square second) at a sampling rate of 500,000 times per second (500 kHz). Simultaneously, a microelectret microphone (MEM) is embedded within the product's internal cavity to capture the resonant sound pressure (units: Pascals) of the cavity's air column. All sensor signals are time-stamped with microsecond precision by a time synchronization controller (TSC) to ensure strict alignment of the stimulus and response signals.

[0164] The logic behind constructing the voiceprint sample library: A single frequency sweep lasts approximately two seconds (20 Hz to 20 kHz, calculated at a rate of 10 Hz per millisecond). Each acquisition generates three sets of data: the actual output spectrum of the aerodynamic sound source (recording the frequency-sound pressure curve), a triaxial vibration acceleration waveform, and a cavity sound pressure waveform. Each data set contains approximately one million sampling points, which are aligned by timestamp and stored as an Excitation-Response Acoustic Sample (ERAS). Each USB product tested generates a separate ERAS file, with a header recording the product serial number, test time, and ambient temperature and humidity (provided by a temperature and humidity sensor). All ERAS files are uploaded to the Acoustic Sample Database (ASD) in real time, forming a traceable sample library.

[0165] Perform 6-layer db4 wavelet packet decomposition on the voiceprint sample, extract the energy entropy and kurtosis coefficient of 16 frequency bands, and output a 128-dimensional acoustic feature vector;

[0166] The preprocessing process for wavelet packet decomposition is as follows: The original voiceprint sample (ERAS) is first filtered through a bandpass filter bank (BPFB) to remove infrasound below 20 Hz and ultrasonic noise above 20 kHz. The filtered signal is then input into the wavelet packet decomposition engine (WPDE), which decomposes it using the Daubechies 4 wavelet basis. The db4 wavelet is a compactly supported orthogonal wavelet with a 4th-order vanishing moment. Its waveform resembles an impulse signal and is suitable for capturing transient vibration characteristics. The decomposition level is set to 6, dividing the full frequency band (20 Hz to 20 kHz) into 2 to the power of 6, or 64 equal-width subbands (each with a width of approximately 312.5 Hz). For example, the first level decomposition produces two segments: 0-10 kHz and 10-20 kHz; the sixth level ultimately outputs 64 refined subbands.

[0167] Dual extraction strategy of feature parameters:

[0168] Two types of features are calculated for the wavelet coefficients of each sub-band:

[0169] Energy Entropy (EE): This reflects the degree of disorder in the energy distribution within a frequency band. EE is calculated by first taking the sum of the squares of the wavelet coefficients in the frequency band as the total energy. The ratio of this total energy to the sum of the total energies of all 64 subbands is then used as the probability value, which is then substituted into the Shannon entropy formula. Higher EE values ​​indicate a more dispersed energy distribution (possibly implying a loose structure).

[0170] Kurtosis Coefficient (KC): A measure of the sharpness of a signal pulse. It is calculated as the ratio of the fourth-order central moment to the square of the second-order central moment. A kurtosis greater than 3 indicates a sharp impulse (such as crack vibration), while a kurtosis less than 3 indicates a smoother signal (such as uniform curing of a colloid).

[0171] The logic for generating high-dimensional feature vectors: One energy entropy and one kurtosis coefficient are extracted from each subband, resulting in a total of 128 parameters across the 64 subbands. Since the effective frequency band is concentrated between 20 Hz and 20 kHz (approximately covering the first 61 subbands), the last three ultra-high frequency subbands are set to zero. These 128 parameters are then arranged in frequency band order to form a 128-dimensional acoustic feature vector (128-D Acoustic Feature Vector, 128-D AFV). This vector also includes quality labels: these were manually annotated during early training as "good quality," "poor contact," or "structural crack."

[0172] The feature vector is input into a pre-trained 3-layer convolutional autoencoder, which reduces the dimension to an interpretable 3D acoustic feature space and outputs a distribution topology map containing good / defective clusters.

[0173] Convolutional Autoencoder Structural Design: The encoder part of the 3-Layer Convolutional Autoencoder (3L-CAE) includes:

[0174] Input layer: receives a 128-dimensional feature vector (reshaped into an 8×16 matrix);

[0175] First layer: 16 5×1 convolution kernels, outputting 16 feature maps;

[0176] Second layer: 8 3×1 convolution kernels, outputting 8 feature maps;

[0177] The third layer: 3 1×1 convolution kernels, outputting a 3D latent space vector.

[0178] The decoder is stacked symmetrically in reverse, reconstructing the input through deconvolution. The training objective is to minimize the mean squared error (MSE) between the input vector and the reconstructed vector, forcing the latent space to retain core features.

[0179] Pre-training and feature space interpretability: 3L-CAE was pre-trained using 100,000 historical voiceprint samples (including manual annotations). After training, the decoder was discarded, retaining only the encoder. When a new 128-dimensional feature vector was input, the encoder output a 3D latent space vector. Its three dimensions, after manual analysis, have physical meaning:

[0180] The first dimension: Structural Stiffness Index (SSI), which is related to the shell deformation;

[0181] The second dimension: Contact Integrity Coefficient (CIC), which reflects the terminal welding quality;

[0182] The third dimension: Material Homogeneity (MH), which is related to the uniformity of colloid curing.

[0183] Dynamic generation of distribution topology maps: The 3D latent space vectors (SSI, CIC, MH) of each finished USB product are projected into a three-dimensional coordinate system. The origin of the coordinate system is determined by the cluster center of 10,000 sets of good product data. The system then generates an Acoustic Feature Topology Map (AFTM) in real time:

[0184] Good product cluster: concentrated in a sphere with a radius of 0.5 near the origin (coordinate range: SSI: -0.3~0.3, CIC: -0.2~0.4, MH: -0.1~0.2);

[0185] Poor contact cluster: spreading along the negative CIC (typical coordinate: CIC < -0.8);

[0186] Structural crack clusters: shifted along the positive SSI direction (typical coordinates: SSI > 1.0).

[0187] The topology map uses 3D Kernel Density Estimation (3D-KDE) to render the probability density of different areas, forming a gradient cloud of red (good product), yellow (poor contact), and blue (structural cracks).

[0188] Perform DBSCAN density clustering based on the distribution topology graph, combine it with the pre-trained SVM classifier to identify the defect patterns of outlier clusters, and output defect type labels and confidence scores;

[0189] DBSCAN density clustering parameter configuration: DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm. Key parameters include:

[0190] Neighborhood radius (Epsilon, EPS): set to 0.7 (empirical value) to control the clustering range;

[0191] Minimum number of points (MinPts): Set to 15, requiring at least 15 samples around the core point.

[0192] Algorithm execution process:

[0193] Draw a sphere with radius EPS centered at each data point;

[0194] If the number of samples in the sphere is ≥MinPts, it is marked as a core point, otherwise it is an edge point;

[0195] Adjacent core points are merged into clusters, edge points are assigned to adjacent clusters, and isolated points are considered as noise.

[0196] Decision-making mechanism of SVM classifier:

[0197] The Support Vector Machine (SVM) classifier uses the Radial Basis Function Kernel (RBF) with a penalty coefficient C set to 10. The training data comes from the outliers (i.e., non-good cluster samples) marked by DBSCAN in the topology graph. Classification process:

[0198] Input: 3D coordinates of outliers (SSI, CIC, MH);

[0199] Feature mapping: The RBF kernel is used to increase the 3D space to infinite dimensions and maximize the class interval;

[0200] Decision function: Calculate the signed distance from the sample to the classification hyperplane. A positive value indicates "poor contact" and a negative value indicates "structural cracking".

[0201] Defect label and confidence output:

[0202] Each outlier sample gets two types of output:

[0203] Defect Type Label (DTL): "CL" (Contact Loss): poor contact (SVM output > 0); "SC" (Structural Crack): structural crack (SVM output < 0).

[0204] Confidence Score (CS): Calculated as the absolute distance (in standard deviation) between the sample and the SVM classification hyperplane, normalized to a range of 0–100%. For example, a distance of 2.5 gives a CS of 92% (empirical threshold: >85% indicates high confidence). All results are written to a Defect Report (DR), which includes topological coordinates, labels, confidence scores, and the original voiceprint ID.

[0205] The three-degree-of-freedom pneumatic sorting arm is controlled according to the defect type label. Good products are sent to the packaging area, defective products are returned for repair, and products with structural cracks are scrapped, completing the automatic sorting of finished products.

[0206] Motion planning of pneumatic sorting arms:

[0207] 3-DOF Pneumatic Sorting Arm (3DOF-PSA) includes:

[0208] Rotational freedom (range 0°~270°, accuracy ±0.5°);

[0209] Telescopic freedom (stroke 500 mm, repeatability ±0.1 mm);

[0210] Clamping freedom (opening and closing angle 0°~90°).

[0211] After receiving the sorting instructions, the motion controller plans a Bézier curve trajectory to avoid obstacles within the workspace. For example, the path from the inspection position (coordinates X0, Y0, Z0) to the repair area (X1, Y1, Z1) is defined by four control points, and the motion time is optimized to less than 0.8 seconds.

[0212] Adaptive switching of end effectors:

[0213] The end of the sorting arm is equipped with a Quick-Change Tool Head (QCTH), which automatically switches according to the defect label:

[0214] Good quality → Vacuum Suction Cup (VSC): 20mm diameter silicone suction cup, -80 kPa vacuum pressure, gently grasps the shell without leaving any marks;

[0215] Defective contact parts → Three-Finger Soft Gripper (TFSG): Inflatable flexible fingers with an adjustable gripping force of 5 to 15 Newtons to avoid damage to repairable parts.

[0216] Components with structural cracks → Electromagnetic Gripper (EMG): Power is applied to generate a 120N magnetic attraction force, allowing for quick removal to the scrap bin.

[0217] Closed-loop control of the sorting system:

[0218] The entire process is closed in seconds:

[0219] The moment the voiceprint detection is completed, the defect report is pushed to the sorting controller;

[0220] The pneumatic arm selects the tool head and plans the path according to the label (taking ≤ 0.2 seconds);

[0221] Execute sorting action (takes 0.8~1.5 seconds);

[0222] Once in place, the placement accuracy is verified using a laser distance sensor (LDS) (error < 1 mm).

[0223] Feedback sorting results to the central database to update product status.

[0224] The system has a production capacity of 45 pieces per minute and an error rate of <0.05% (verified by regular X-ray inspections).

[0225] It can be seen that placing the USB terminal on an assembly platform containing a non-uniform gradient magnetic field generates a pre-positioned terminal group with spatial directional characteristics; pressing the pre-positioned terminal group into the elastically deformed inner cavity of the USB shell automatically corrects the axial offset of the terminal and outputs a geometrically aligned semi-finished component; applying pulsed mechanical vibration to the semi-finished component, collecting the internal cavity resonant frequency spectrum of the component through the microphone array, triggering the instant curing of the UV glue, and generating a physically locked USB finished product; applying swept frequency excitation to the USB finished product, extracting the full-band voiceprint feature vector and inputting it into the pre-trained defect classification model to realize automatic sorting of finished products, thereby achieving high-precision topology matching and dynamic stress optimization of USB assembly, and improving product consistency and defect detection efficiency.

[0226] Another embodiment of the present invention provides a USB intelligent production and assembly system, see Figure 3 , the system may include:

[0227] The rotation module 301 is used to place the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, and utilize the difference in the distribution of the magnetic material of the terminals to make the terminals rotate autonomously to a topological matching posture, thereby generating a pre-positioned terminal group with spatial orientation characteristics;

[0228] Correction module 302, configured to press the pre-positioned terminal assembly into the elastically deformed inner cavity of the USB housing, generate a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the housing and the contact points of the terminals, automatically correct the axial deviation of the terminals, and output a geometrically aligned semi-finished assembly;

[0229] The curing module 303 is configured to apply pulsed mechanical vibration to the semi-finished component, collect the resonance frequency spectrum of the internal cavity of the component through a microphone array, and trigger the instant curing of the UV adhesive when the main resonance peak is within a preset range, thereby generating a physically locked USB finished product;

[0230] The sorting module 304 is used to apply sweep frequency excitation to the finished USB product, extract the full-band voiceprint feature vector and input it into the pre-trained defect classification model, and realize automatic sorting of the finished product according to the cluster distribution of the feature vector in the acoustic space.

[0231] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0232] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps:

[0233] S201, placing the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, utilizing the difference in magnetic material distribution of the terminals to autonomously rotate the terminals to a topologically matching position, thereby generating a pre-positioned terminal group with spatial orientation characteristics;

[0234] S202, pressing the pre-positioned terminal assembly into the elastically deformed inner cavity of the USB housing, generating a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the housing and the contact points of the terminals, automatically correcting the axial offset of the terminals, and outputting a geometrically aligned semi-finished assembly;

[0235] S203, applying pulsed mechanical vibration to the semi-finished component, collecting the resonance frequency spectrum of the internal cavity of the component through a microphone array, and triggering the instant curing of the UV adhesive when the main resonance peak is within a preset range to produce a physically locked USB finished product;

[0236] S204, applying sweep frequency excitation to the finished USB product, extracting full-band voiceprint feature vectors and inputting them into a pre-trained defect classification model, and realizing automatic sorting of finished products according to the cluster distribution of the feature vectors in the acoustic space.

[0237] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0238] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0239] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0240] S201, placing the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, utilizing the difference in magnetic material distribution of the terminals to autonomously rotate the terminals to a topologically matching position, thereby generating a pre-positioned terminal group with spatial orientation characteristics;

[0241] S202, pressing the pre-positioned terminal assembly into the elastically deformed inner cavity of the USB housing, generating a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the housing and the contact points of the terminals, automatically correcting the axial offset of the terminals, and outputting a geometrically aligned semi-finished assembly;

[0242] S203, applying pulsed mechanical vibration to the semi-finished component, collecting the resonance frequency spectrum of the internal cavity of the component through a microphone array, and triggering the instant curing of the UV adhesive when the main resonance peak is within a preset range to produce a physically locked USB finished product;

[0243] S204, applying sweep frequency excitation to the finished USB product, extracting full-band voiceprint feature vectors and inputting them into a pre-trained defect classification model, and realizing automatic sorting of finished products according to the cluster distribution of the feature vectors in the acoustic space.

[0244] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A USB intelligent production and assembly method, characterized in that: The method comprises: The USB terminals are placed on an assembly platform containing a non-uniform gradient magnetic field. The distribution difference of the magnetic material of the terminals is utilized to make the terminals rotate autonomously to a topological matching posture, thereby generating a pre-positioned terminal group with spatial orientation characteristics. The pre-positioned terminal group is pressed into the elastically deformed inner cavity of the USB shell, and a directional deformation feedback force is generated by the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, which automatically corrects the axial deviation of the terminal and outputs a geometrically aligned semi-finished component; Apply pulsed mechanical vibration to the semi-finished component, and collect the resonance frequency spectrum of the internal cavity of the component through a microphone array. When the main resonance peak is within a preset range, the UV glue is triggered to instantly cure, generating a physically locked USB finished product; A frequency sweep excitation is applied to the finished USB product, and a full-band voiceprint feature vector is extracted and input into a pre-trained defect classification model. Automatic sorting of the finished product is achieved based on the cluster distribution of the feature vector in the acoustic space.

2. The method according to claim 1, characterized in that The method comprises placing the USB terminals on an assembly platform containing a non-uniform gradient magnetic field, utilizing the difference in the distribution of the magnetic material of the terminals to make the terminals rotate autonomously to a topological matching position, and generating a pre-positioned terminal group with spatial orientation characteristics, including: Activate the three sets of orthogonal electromagnetic coils at the bottom of the assembly platform, generate a rotating magnetic field through alternating current with a phase difference of 120 degrees, and output a dynamic magnetic field distribution field with spatial spiral gradient characteristics; A high-sensitivity Hall sensor array is used to scan the magnetic flux distribution of the NdFeB magnetic coating on the terminal surface. The magnetic moment direction offset angle is calculated based on the dynamic magnetic field distribution field, and a real-time pose matrix containing X / Y / Z three-axis offset data is output. The pose matrix is ​​input into the magnetic field controller, which dynamically adjusts the current intensity of the three coils to form a correction torque of 15-25mN·m between the magnetic field gradient direction and the magnetic moment offset angle, outputting a directional terminal array that completes autonomous rotation. The angle between the terminal contact and the housing slot is measured by a confocal laser displacement sensor. When the angle is detected to be ≤0.3°, the electromagnetic locking mechanism is activated to output a pre-positioned terminal group with spatial orientation characteristics.

3. The method according to claim 2, characterized in that The method comprises pressing the pre-positioned terminal group into the elastically deformed inner cavity of the USB shell, generating a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, automatically correcting the axial offset of the terminal, and outputting a geometrically aligned semi-finished component, including: Applying 40°C constant temperature hot air to the shell to reduce its elastic modulus to a preset threshold, outputting the shell in a critical state of plastic deformation; The contact pressure distribution between the terminal and the tapered micro-rib is collected in real time through the micro-force sensor array, and a pressure gradient cloud map of the contact point is output; Based on the pressure gradient cloud map, the pressure abnormality area is identified, and the axial offset angle is calculated by combining the micro-rib structure deformation model, and the offset correction vector parameters are output; The offset correction vector parameters are input into the six-degree-of-freedom press-fitting robot, which dynamically adjusts the pressing trajectory and outputs a semi-finished product with an axial deviation of ≤10μm. X-ray fluoroscopy is used to scan the fit between the terminal pins and the housing guide grooves to verify the geometric alignment and obtain a geometrically aligned semi-finished component.

4. The method according to claim 3, characterized in that The method applies pulsed mechanical vibration to the semi-finished component, collects the resonance frequency spectrum of the internal cavity of the component through a microphone array, and triggers the instant curing of the UV glue when the main resonance peak is within a preset range to generate a physically locked USB finished product, including: A 0.5-5kHz linear frequency sweep pulse is applied to the semi-finished component through a piezoelectric ceramic exciter, outputting a mechanical vibration spectrum containing 50 characteristic frequency points; An 8-channel MEMS microphone array distributed in a ring is used to collect the internal sound pressure signal of the component. The first three resonance peak frequencies and Q factors are extracted through FFT, and the resonance frequency eigenvector is output. When the resonance characteristic vector satisfies: main peak F1∈[1850,2150]Hz and Q1>55, a UV trigger signal with a 5ms delay is generated; Through an array of multiple UV optical fibers, 365nm UV light is released within 3ms after the delay signal is triggered. The light intensity distribution is dynamically adjusted according to the vibration spectrum to output a physically locked USB product.

5. The method according to claim 4, characterized in that The method applies a sweep frequency excitation to the finished USB product, extracts a full-band voiceprint feature vector, inputs the feature vector into a pre-trained defect classification model, and realizes automatic sorting of the finished product according to the cluster distribution of the feature vector in the acoustic space, including: Using a pneumatic sound source to sweep the frequency in the range of 20Hz-20kHz at a rate of 10Hz / ms, the vibration response of the finished product is synchronously collected and a time-stamped stimulus-response soundprint sample library is output; Perform 6-layer db4 wavelet packet decomposition on the voiceprint sample, extract the energy entropy and kurtosis coefficient of 16 frequency bands, and output a 128-dimensional acoustic feature vector; The feature vector is input into a pre-trained 3-layer convolutional autoencoder, which reduces the dimension to an interpretable 3D acoustic feature space and outputs a distribution topology map containing good / defective clusters. Perform DBSCAN density clustering based on the distribution topology graph, combine it with the pre-trained SVM classifier to identify the defect patterns of outlier clusters, and output defect type labels and confidence scores; The three-degree-of-freedom pneumatic sorting arm is controlled according to the defect type label. Good products are sent to the packaging area, defective products are returned for repair, and products with structural cracks are scrapped, completing the automatic sorting of finished products.

6. A USB intelligent production and assembly system, characterized in that: The system comprises: The rotation module is used to place the USB terminal on an assembly platform containing a non-uniform gradient magnetic field. The difference in the distribution of the magnetic material of the terminal is used to make the terminal rotate autonomously to a topological matching posture, thereby generating a pre-positioned terminal group with spatial orientation characteristics; A correction module is used to press the pre-positioned terminal group into the elastically deformed USB shell cavity, generate a directional deformation feedback force through the conical micro-rib structure preset on the inner wall of the shell and the contact point of the terminal, automatically correct the axial deviation of the terminal, and output a geometrically aligned semi-finished assembly; A curing module is used to apply pulsed mechanical vibration to the semi-finished component, collect the resonance frequency spectrum of the internal cavity of the component through a microphone array, and trigger the UV glue to instantly cure when the main resonance peak is within a preset range, thereby generating a physically locked USB finished product; The sorting module is used to apply sweep frequency excitation to the USB finished product, extract the full-band voiceprint feature vector and input it into the pre-trained defect classification model, and realize automatic sorting of the finished product according to the cluster distribution of the feature vector in the acoustic space.

7. The system according to claim 6, characterized in that The rotation module is used to: Activate the three sets of orthogonal electromagnetic coils at the bottom of the assembly platform, generate a rotating magnetic field through alternating current with a phase difference of 120 degrees, and output a dynamic magnetic field distribution field with spatial spiral gradient characteristics; A high-sensitivity Hall sensor array is used to scan the magnetic flux distribution of the NdFeB magnetic coating on the terminal surface. The magnetic moment direction offset angle is calculated based on the dynamic magnetic field distribution field, and a real-time pose matrix containing X / Y / Z three-axis offset data is output. The pose matrix is ​​input into the magnetic field controller, which dynamically adjusts the current intensity of the three coils to form a correction torque of 15-25mN·m between the magnetic field gradient direction and the magnetic moment offset angle, outputting a directional terminal array that completes autonomous rotation. The angle between the terminal contact and the housing slot is measured by a confocal laser displacement sensor. When the angle is detected to be ≤0.3°, the electromagnetic locking mechanism is activated to output a pre-positioned terminal group with spatial orientation characteristics.

8. The system according to claim 7, characterized in that The correction module is specifically used to: Applying 40°C constant temperature hot air to the shell to reduce its elastic modulus to a preset threshold, outputting the shell in a critical state of plastic deformation; The contact pressure distribution between the terminal and the tapered micro-rib is collected in real time through the micro-force sensor array, and a pressure gradient cloud map of the contact point is output; Based on the pressure gradient cloud map, the pressure abnormality area is identified, and the axial offset angle is calculated by combining the micro-rib structure deformation model, and the offset correction vector parameters are output; The offset correction vector parameters are input into the six-degree-of-freedom press-fitting robot, which dynamically adjusts the pressing trajectory and outputs a semi-finished product with an axial deviation of ≤10μm. X-ray fluoroscopy is used to scan the fit between the terminal pins and the housing guide grooves to verify the geometric alignment and obtain a geometrically aligned semi-finished component.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 5.

Images