A high-precision assembly intelligent monitoring method and system for CDU liquid cooling equipment
By collecting and analyzing end-face gap images and torque data of CDU liquid cooling equipment, generating assembly posture curves and calculating compensation torque, the problems of off-center load and stress transmission in the pipeline assembly of CDU liquid cooling equipment were solved, and the sealing reliability and stability of the equipment were improved.
Patent Information
- Application Number
- CN202611347497.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-02
- Publication Date
- 2026-11-17
Smart Images

Figure CN123116085A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent measurement and control equipment manufacturing technology, specifically to a high-precision assembly intelligent monitoring method and system for CDU liquid cooling equipment. Background Technology
[0002] As data center computing power continues to expand, the heat density of server chips is rising sharply, making liquid cooling technology the mainstream standard for heat dissipation in computing infrastructure. The Cooling Distribution Unit (CDU), as the core hub of the liquid cooling system, is responsible for the precise distribution and circulation control of the coolant. Its internal piping is intricate and operates under high-pressure fluid conditions. The assembly quality of the CDU directly affects the sealing integrity of the cooling medium; any minor misalignment of pipe joints or assembly defects can lead to leaks and short circuits, potentially causing serious accidents such as server downtime. Therefore, high-precision monitoring of the piping assembly process of the CDU liquid cooling equipment is a fundamental prerequisite for ensuring the safe and stable operation of computing centers.
[0003] In actual pipeline assembly and construction, stainless steel joints have strong surface reflections and visual blind spots in their internal sealing structures. Existing inspection methods typically involve static pressure testing after the entire assembly is complete, or independent verification of the mechanical torque and static two-dimensional profile of the current working node during tightening. This approach not only fails to capture the eccentric compression deformation of the internal sealing ring during screwing but also completely ignores the rigid topological constraints of the overall pipeline. During the tightening of complex multi-branch pipeline networks, the nonlinear tilting force generated at the current assembly node is rigidly transmitted to both sides along the main pipeline, exerting a mechanical prying effect on adjacent tightened joints, leading to micro-loosening at their mating surfaces. Existing methods cannot dynamically track and quantify this spatial force eccentricity and stress transmission phenomenon, making the equipment highly susceptible to micro-leakage faults at hidden loosening points after long-term alternating operation, failing to meet the assembly requirements for zero-defect manufacturing of liquid-cooled equipment. Summary of the Invention
[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a high-precision assembly intelligent monitoring method and system for CDU liquid cooling equipment to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a high-precision assembly intelligent monitoring method for CDU liquid cooling equipment, comprising the following steps: S1, synchronously acquiring end-face gap image sequences and flange assembly torque data during the tightening cycle of the CDU water pipe joint, and extracting the axial relative coordinates and gap width values of the upper and lower flanges in the end-face gap image sequence; S2, mapping the axial relative coordinates and gap width values to a three-dimensional coordinate system on a unified time axis, generating an assembly pose curve containing radial runout and axial compression, and extracting the time change rate of the flange assembly torque data to generate a torque ramp-up gradient; S3, extracting the radial runout from the assembly pose curve. When an abnormal time point occurs where the runout changes abruptly, the torque ramp-up gradient corresponding to the abnormal time point is read. When there is a synchronous sudden increase in the torque ramp-up gradient, an off-center load interference feature containing a three-dimensional offset vector is generated based on the axis relative coordinate offset corresponding to the abnormal time point. S4. Extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints. Calculate the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span. Map the rigid transmission torque to the flange loosening gap of adjacent assembled joints. Generate a re-tightening compensation torque for adjacent assembled joints through the flange loosening gap.
[0006] Preferably, the specific process of synchronously acquiring the end face gap image sequence and flange assembly torque data during the tightening cycle of the CDU water pipe joint is as follows: receiving the engagement trigger signal from the servo tightening spindle, and synchronously activating the polarization industrial vision camera and the servo flange torque sensor according to the engagement trigger signal; extracting multi-angle field frames of the CDU water pipe joint mating surface continuously captured by the polarization industrial vision camera, filtering out the high-brightness metal mirror reflection layer in the multi-angle field frames to generate the end face gap image sequence; extracting the motor current feedback characteristics of the servo flange torque sensor during the tightening cycle, converting the motor current feedback characteristics into real-time torque values to generate flange assembly torque data; extracting the first acquisition timestamp of the end face gap image sequence and the second acquisition timestamp of the flange assembly torque data, and performing hardware-level clock alignment processing on the end face gap image sequence and flange assembly torque data according to the first timestamp and the second timestamp.
[0007] Preferably, the specific process for extracting the axis-to-axis relative coordinates and gap width value of the upper and lower flanges of the joint from the end face gap image sequence is as follows: perform an edge operator extraction operation on the end face gap image sequence to obtain the metal contour of the upper and lower flanges of the joint; perform a sub-pixel level ellipse fitting operation on the metal contour, and extract the geometric center point in the ellipse fitting result to generate the axis-to-axis relative coordinates of the upper and lower flanges of the joint; scan the metal contour along the parallel direction of the line connecting the axes of the upper and lower flanges of the joint, and extract the pixel span between the boundaries of the upper and lower flanges of the joint; combine the pixel span with the calibration intrinsic parameters of the polarization industrial vision camera to map it into a spatial dimension to generate the gap width value.
[0008] Preferably, the specific process of mapping the relative coordinates of the axis center and the gap width value to a three-dimensional coordinate system on a unified time axis to generate an assembly pose curve containing radial runout and axial compression is as follows: extract the geometric center of the stationary base of the CDU main water pipe to construct the absolute origin of the three-dimensional coordinate system; project the relative coordinates of the axis center on the unified time axis to the horizontal reference plane of the three-dimensional coordinate system, and extract the horizontal free trajectory of the relative coordinates of the axis center deviating from the absolute origin on the horizontal reference plane to generate radial runout; project the gap width value on the unified time axis to the vertical reference axis of the three-dimensional coordinate system, and extract the attenuation difference of the gap width value on the vertical reference axis relative to the initial gap width to generate axial compression; fuse the radial runout and axial compression on the unified time axis, and fit to generate an assembly pose curve characterizing the three-dimensional precession process of the joint space.
[0009] Preferably, the specific process for extracting the time change rate of flange assembly torque data to generate the torque ramp gradient is as follows: The flange assembly torque data is truncated according to a preset time window to obtain local torque signal segments; a moving average filtering process is performed on the local torque signal segments to filter out high-frequency mechanical vibration noise generated by mechanical gear meshing, thus obtaining a smooth torque sequence; the data increments of adjacent sampling nodes in the smooth torque sequence are extracted, and the data increments are mapped to the time span of adjacent sampling nodes to obtain the time change rate; the time change rates are spliced according to the sliding order of the time window to generate the torque ramp gradient of the flange assembly torque data.
[0010] Preferably, the specific process for extracting the abnormal time point where the radial runout in the assembly pose curve exhibits a step change and reading the torque ramp-up gradient corresponding to the abnormal time point is as follows: Extract the time series of the radial runout contained in the assembly pose curve on a unified time axis; perform first-order forward differencing on the radial runout time series to obtain the radial free velocity change curve; extract the abrupt data peaks in the radial free velocity change curve that exceed the baseline smooth fluctuation range; mark the timestamps corresponding to the abrupt data peaks on the unified time axis as the abnormal time points where the radial runout exhibits a step change; retrieve the corresponding torque sampling index nodes on the unified time axis based on the abnormal time points; extract the data points at the torque sampling index node positions in the torque ramp-up gradient to obtain the torque ramp-up gradient corresponding to the abnormal time points.
[0011] Preferably, when there is a synchronous sudden increase in the torque ramp gradient, the specific process of generating the off-center load interference feature containing a three-dimensional offset vector based on the shaft center relative coordinate offset corresponding to the abnormal time point is as follows: compare the difference between the torque ramp gradient amplitude corresponding to the abnormal time point and the average amplitude of the torque gradient during the historical smooth period. When the average amplitude difference exceeds the sudden increase judgment threshold set in the reference smooth period, it is confirmed that there is a synchronous sudden increase in the torque ramp gradient; extract the current shaft center relative coordinates of the upper and lower flanges of the joint in the three-dimensional coordinate system corresponding to the abnormal time point, and the initial shaft center relative coordinates of the upper and lower flanges of the joint in the initial zero stress state within the same assembly cycle; perform a spatial vector projection operation on the current shaft center relative coordinates to the initial shaft center relative coordinates, and extract the direction angle deviation and geometric modulus increment generated by the spatial vector projection operation to synthesize a three-dimensional offset vector; combine the time sequence identifier of the abnormal time point and the spatial state information of the three-dimensional offset vector to generate the off-center load interference feature.
[0012] Preferably, the specific process of extracting the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints, and calculating the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span is as follows: Import the structural topology model of the CDU main distribution pipe, extract the spatial path length along the stainless steel pipeline from the current assembly node to the adjacent assembled joint in the structural topology model to generate the rigid pipe segment span; combine the material bending stiffness parameters of the CDU main distribution pipe, map the three-dimensional offset vector contained in the off-center load interference feature into the end off-center load force in the horizontal plane of the three-dimensional coordinate system; use the end off-center load force as the end excitation source, and the rigid pipe segment span as the spatial lever arm length to construct the cantilever beam mechanical transmission equivalent function; input the end off-center load force and the rigid pipe segment span into the cantilever beam mechanical transmission equivalent function to perform a spatial vector product mapping operation, and extract the bending moment tensor acting on the joint surface of adjacent assembled joints output by the spatial vector product mapping operation to generate the rigid transmission torque.
[0013] Preferably, the specific process of mapping the rigid transmission torque to the flange loosening gap of adjacent assembled joints, and generating the re-tightening compensation torque for adjacent assembled joints through the flange loosening gap is as follows: Extract the elastic deformation hysteresis coefficient of the O-ring seal inside the adjacent assembled joints and the mechanical stiffness coefficient of the fastening thread; load the rigid transmission torque into the strain function of the metal mating surface composed of the elastic deformation hysteresis coefficient and the mechanical stiffness coefficient, and extract the asymmetric micro-warping deformation output of the metal mating surface strain function to generate the flange loosening gap of the adjacent assembled joints; extract the maximum opening azimuth angle and opening geometry of the flange loosening gap at the mating surface of the adjacent assembled joints, and inversely map the opening geometry to the thread reset screw-in angle using preset thread lead parameters; perform closed-loop gain adjustment on the basic torque control baseline of the servo tightening spindle according to the thread reset screw-in angle to generate the re-tightening compensation torque for adjacent assembled joints.
[0014] A high-precision assembly intelligent monitoring system for CDU liquid cooling equipment, used to execute the aforementioned high-precision assembly intelligent monitoring method for CDU liquid cooling equipment, includes: a data acquisition module, used to simultaneously acquire end-face gap image sequences and flange assembly torque data during the tightening cycle of the CDU water pipe joint, and extract the axial relative coordinates and gap width values of the upper and lower flanges in the end-face gap image sequence; a pose construction module, used to map the axial relative coordinates and gap width values to a three-dimensional coordinate system on a unified time axis, generating an assembly pose curve containing radial runout and axial compression, and extracting the time change rate of the flange assembly torque data to generate a torque creep gradient; and an off-center load identification module, used to extract the assembly... The abnormal time point where the radial runout in the alignment curve shows a step jump is identified. The torque ramp-up gradient corresponding to the abnormal time point is read. When there is a synchronous sudden increase in the torque ramp-up gradient, an off-center load interference feature containing a three-dimensional offset vector is generated based on the axis relative coordinate offset corresponding to the abnormal time point. The re-tightening compensation module is used to extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints. Based on the rigid pipe segment span, the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints is calculated. The rigid transmission torque is mapped to the flange loosening gap of adjacent assembled joints. The re-tightening compensation torque for adjacent assembled joints is generated through the flange loosening gap.
[0015] The technical effects and advantages of this invention are as follows: (1) In the technical solution described in this invention, the end face gap image sequence and flange assembly torque data of the CDU water pipe joint tightening cycle are collected simultaneously. The relative coordinates of the axial centers of the upper and lower flanges of the joint and the gap width value are mapped to a three-dimensional coordinate system to generate an assembly pose curve containing radial runout and axial compression. The time change rate of the flange assembly torque data is extracted to generate the torque rise gradient. By coordinating the reconstruction of the multi-dimensional spatial precession trajectory and the underlying mechanical torque state on a unified time axis, the interference of the highly reflective environment on the single visual feature is eliminated, and the characterization accuracy of the micro-motion trend and stress state of the joint is improved.
[0016] (2) In the technical solution described in this invention, the abnormal time point where the radial runout of the assembly posture curve undergoes a step change is extracted, and a three-dimensional offset vector with off-center interference characteristics is generated based on the relative coordinate offset of the axis when the torque climbing gradient increases synchronously; then, the rigid pipe section span between it and the adjacent assembled joint is extracted, the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of the adjacent assembled joint is calculated, and it is mapped to the flange loosening gap and a re-tightening compensation torque is generated. This process effectively addresses the secondary loosening problem caused by the implicit stress transmission in multi-pipeline systems, realizes the quantification of assembly off-center loading phenomenon and the dynamic compensation correction of adjacent nodes, and improves the sealing reliability of the overall pipeline connection.
[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description
[0018] Figure 1 This is a flowchart of a high-precision assembly intelligent monitoring method for CDU liquid cooling equipment according to the present invention; Figure 2 This is a schematic diagram illustrating the relationship between the assembly pose curve and the three-dimensional offset vector in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the verification of cross-modal feature synchronization surge determination in an embodiment of the present invention; Figure 4 This is a flowchart of a high-precision assembly intelligent monitoring system for CDU liquid cooling equipment according to the present invention. Detailed Implementation
[0019] Example 1; please refer to Figure 1 This invention provides a technical solution: a high-precision intelligent monitoring method for CDU liquid cooling equipment assembly, comprising the following steps: S1, synchronously acquiring end-face gap image sequences and flange assembly torque data during the tightening cycle of the CDU water pipe joint, and extracting the axial relative coordinates and gap width values of the upper and lower flanges in the end-face gap image sequence; S2, mapping the axial relative coordinates and gap width values to a three-dimensional coordinate system on a unified time axis, generating an assembly pose curve containing radial runout and axial compression, and extracting the time change rate of the flange assembly torque data to generate a torque ramp-up gradient; S3, extracting the radial runout from the assembly pose curve. At the abnormal time point where the quantity shows a step jump, read the torque rise gradient corresponding to the abnormal time point. When there is a synchronous sudden increase in the torque rise gradient, generate an off-center load interference feature containing a three-dimensional offset vector based on the axis relative coordinate offset corresponding to the abnormal time point; S4, extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints, calculate the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span, map the rigid transmission torque to the flange loosening gap of adjacent assembled joints, and generate a re-tightening compensation torque for adjacent assembled joints through the flange loosening gap.
[0020] In this implementation scheme, step S1 performs hardware-level data synchronization acquisition and image basic feature extraction. During the tightening cycle of the flange bolts, the control unit synchronously records continuous images captured by the side-facing industrial camera and mechanical sensor values fed back by the servo spindle to obtain the raw data source. The end-face gap image sequence refers to a set of continuous images taken from a horizontal perspective along the side of the fluid pipeline flange, which changes over time and is used to record the dynamic changes in the gap between the upper and lower flanges as they gradually tighten the sealing ring. The axis relative coordinates refer to the translation and deflection deviation values between the upper pipeline center baseline and the lower base center baseline calculated in the two-dimensional plane of the image. The control system locates the metal contours of the upper and lower flanges using edge operators, calculates the straight-line pixel span between the two end face boundaries to obtain the gap width value, and transforms the actions of the assembly execution stage into a visual digital matrix and stress baseline data that can be parsed by the program.
[0021] Step S2 fuses the two-dimensional image features with one-dimensional mechanical data and reconstructs them into a three-dimensional spatial precession model. On a unified time axis aligned with the visual frame timestamps and torque sampling periods, the system combines the equipment base reference point to map the relative coordinates of the axis center with the gap width value to synthesize an assembly pose curve reflecting the joint's spiraling state in space. Radial runout reflects the non-concentric sway distance of the flange assembly surface in the horizontal plane, while axial compression records the specific depth at which the flange vertically presses the sealing ring along the pipeline's central axis. The system extracts the time increment of continuous flange assembly torque data and calculates the torque rise gradient, which represents the transient growth rate of tightening resistance applied to the joint per unit time. By aligning the three-dimensional spatial displacement parameters with the torque resistance growth rate on the time coordinate, the system objectively quantifies the resistance behavior and spatial trajectory of the flange joint during the spiraling compression process.
[0022] Step S3 determines the jamming state inside the joint by comparing the temporal correlation between spatial pose changes and torque gradients across modalities. The system traverses and analyzes the pose curves to find abnormal time points where the radial runout value undergoes nonlinear deflection within a very short time window. A step jump characterizes the instantaneous spatial misalignment of the flange joint caused by uneven force on the sealing ring or flange surface tilt during threading. After locating this time point, the system reads the torque ramp gradient state corresponding to the same time point. If there is a synchronous sudden increase in the torque ramp gradient at this point, it indicates that the joint end face is not in a normal flexible compression state but has encountered a rigid assembly obstacle. The system generates a three-dimensional offset vector based on the distance of the axis from the design centerline at this moment, and sets this spatiotemporal state data into off-center load interference features to lock in internal force off-center load events caused by unilateral flange overpressure or pipeline tilt.
[0023] Step S4 calculates the stress damage impact of local assembly anomalies on the overall rigid pipe network connection structure and generates corresponding end-point correction commands. The rigid pipe segment span refers to the spatial path length of the stainless steel pipe between the current branch pipe joint experiencing eccentric loading and the adjacent, already tightened joint on the main pipe. Under the strong rigid constraint of the CDU metal pipe network, the three-dimensional offset vector generated by the tilting and jamming of the current joint will transmit lever bending stress outwards. This mechanical component transmitted along the main rigid pipe wall to the side nodes is quantified as a rigid transmission torque. This torque will pry the adjacent, already compliant joint mating surface in the opposite direction. The system calculates the flange loosening gap caused by this prying force passively opening the end face of the adjacent joint based on the stress and strain parameters of the mating surface. The control system converts this gap deviation value into a thread compensation screw-in angle, issues a re-tightening compensation torque execution command, and drives the servo tightening spindle to perform a secondary supplementary tightening action on the adjacent joint that has been loosened by stress.
[0024] Specifically, the process of synchronously acquiring the end face gap image sequence and flange assembly torque data during the tightening cycle of the CDU water pipe joint is as follows: The engagement trigger signal from the servo tightening spindle is received, and the polarization industrial vision camera and servo flange torque sensor are synchronously activated based on the engagement trigger signal; multi-angle field-of-view frames of the CDU water pipe joint mating surface continuously captured by the polarization industrial vision camera are extracted, and the high-brightness metallic mirror reflection layer in the multi-angle field-of-view frames is filtered out to generate the end face gap image sequence; the motor current feedback characteristics of the servo flange torque sensor during the tightening cycle are extracted, and the motor current feedback characteristics are converted into real-time torque values to generate flange assembly torque data; the first acquisition timestamp of the end face gap image sequence and the second acquisition timestamp of the flange assembly torque data are extracted, and hardware-level clock alignment processing is performed on the end face gap image sequence and flange assembly torque data based on the first and second timestamps.
[0025] In this implementation scheme, the system's bottom-level I / O module captures the level transition command issued by the PLC as a meshing trigger signal, directly driving the hardware synchronization clock board to send exposure strobe pulses to the external trigger pin of the polarization industrial vision camera, and simultaneously sending a data latch command to the analog-to-digital conversion channel of the servo flange torque sensor. The multi-angle field-of-view frames output by the industrial camera undergo pixel-level polarization degree calculation in the image buffer. Pixels with polarization degree indices higher than a set threshold are identified as high-brightness metallic mirror reflective layers, and their grayscale values are forcibly set to zero, generating an end-face gap image sequence. After extracting the motor current feedback characteristics of the servo flange torque sensor during the tightening cycle, inverse dynamics calculation is performed to convert it into a real-time torque value. The calculation process is as follows: ;in, This represents the real-time torque value during the nth sampling period; This represents the electromagnetic torque coefficient of the spindle motor; This represents the motor current feedback characteristics during the nth sampling period; This represents the moment of inertia of the rotor on the tightening spindle; This represents the principal axis angular velocity during the nth sampling period; This represents the principal axis angular velocity during the (n-1)th sampling period; This indicates the hardware sampling time interval of the torque sensor; This represents the viscous friction coefficient of the mechanical transmission components. Due to the clock discrepancy between the visual exposure frame rate and the torque sensor sampling frequency, the system extracts the first acquisition timestamp of the end face gap image sequence to construct a reference time axis, and uses linear interpolation to perform hardware-level clock alignment processing on the discrete flange assembly torque data. ;in, This represents the flange assembly torque value after being perfectly aligned with the m-th frame end face gap image; This represents the first acquisition timestamp of the m-th frame image; Indicates that it is located in the absolute time stream The second most recent collection timestamp on the left; Indicates that it is located in the absolute time stream The second most recent collection timestamp on the right; and These represent the known torque sample values corresponding to the time points.
[0026] Specifically, the process of extracting the relative coordinates of the axes of the upper and lower flanges of the joint and the gap width value from the end face gap image sequence is as follows: An edge operator extraction operation is performed on the end face gap image sequence to obtain the metal contour of the upper and lower flanges of the joint; a sub-pixel level ellipse fitting operation is performed on the metal contour, and the geometric center point in the ellipse fitting result is extracted to generate the relative coordinates of the axes of the upper and lower flanges of the joint; the metal contour is scanned along the direction parallel to the line connecting the axes of the upper and lower flanges of the joint, and the pixel span between the boundaries of the upper and lower flanges of the joint is extracted; the pixel span is combined with the calibration intrinsic parameters of the polarization industrial vision camera and mapped to a spatial dimension to generate the gap width value.
[0027] In this implementation scheme, before performing edge operator extraction on the end face gap image sequence, it is necessary to calculate the dynamic gradient extraction threshold in the image space: ;in, Indicates the dynamic gradient extraction threshold; This represents the average magnitude of the grayscale gradient of all pixels in the end-face gap image sequence; The standard deviation of the grayscale gradient of a pixel; This represents the ambient light interference weighting coefficient. The method for determining this coefficient is as follows: Count the total number of overexposed pixels in a single frame of the image where the grayscale value reaches the upper limit of the camera's sensor saturation range; calculate the area ratio of the total number of overexposed pixels to the total number of pixels in that frame; and multiply this area ratio by a pre-calibrated base exposure compensation constant. Extracting grayscale gradients from an image exceeding... The coordinate points are used to construct the metal profile. A least-squares method is used to construct the coordinate residual objective function. Sub-pixel-level elliptic equation approximation calculations are performed on the discrete metal profile points to obtain the image coordinates of the geometric center point, generating the axis-to-axis relative coordinates of the upper and lower flanges of the joint. Using these axis-to-axis relative coordinates as a reference, multiple digital scan lines are arrayed and emitted along the parallel direction connecting the geometric centers of the two flanges. The intersection points of each digital scan line at the boundaries of the upper and lower flanges are recorded. The pixel difference between the two intersection points is calculated, and the arithmetic mean is taken after removing extreme values to generate the pixel span. Finally, an inverse perspective projection model is input to map the pixel span to spatial dimensions, generating the gap width value. ;in, Indicates the gap width value; This indicates the object distance between the optical center of the polarization industrial vision camera lens and the mating surface of the CDU water distribution pipe connector; This indicates the camera's calibrated intrinsic focal length. This indicates the pixel span between the upper and lower flange boundaries of the joint; This indicates the size mapping ratio of a single pixel unit in a camera's image sensor. This parameter represents the spatial angle deviation between the camera's principal optical axis and the plane normal to the line connecting the connector's axis.
[0028] Specifically, the process of mapping the relative coordinates of the axis center and the gap width value to a three-dimensional coordinate system on a unified time axis to generate an assembly pose curve containing radial runout and axial compression is as follows: Extract the geometric center of the stationary base of the CDU main water pipe to construct the absolute origin of the three-dimensional coordinate system; project the relative coordinates of the axis center on the unified time axis to the horizontal reference plane of the three-dimensional coordinate system, and extract the horizontal free trajectory of the relative coordinates of the axis center deviating from the absolute origin on the horizontal reference plane to generate radial runout; project the gap width value on the unified time axis to the vertical reference axis of the three-dimensional coordinate system, and extract the attenuation difference of the gap width value relative to the initial gap width on the vertical reference axis to generate axial compression; fuse the radial runout and axial compression on the unified time axis, and fit to generate an assembly pose curve characterizing the three-dimensional precession process of the joint space.
[0029] In this implementation scheme, the system reads the three-dimensional absolute coordinate set of the stationary base of the CDU main water pipe in the hardware coordinate system of the assembly platform, and extracts the geometric center point coordinate package of the absolute coordinate set to construct the absolute origin of the three-dimensional coordinate system. The axis-center relative coordinates of the p-th sampling node on the unified time axis are vertically projected onto the horizontal reference plane of the three-dimensional coordinate system. The radial runout is generated by comparing the spatial distance between the projected coordinates and the absolute origin. The calculation process is as follows: ;in, This represents the radial runout of the p-th sampling node; This represents the x-coordinate of the current flange axis projection point on the horizontal reference surface. The x-coordinate of the absolute origin on the horizontal reference plane; This represents the ordinate of the current flange axis projection point on the horizontal reference surface; The ordinate of the absolute origin on the horizontal reference plane; This represents the mechanical tilt offset angle of the flange face relative to the horizontal reference plane. The method for determining this mechanical tilt offset angle is as follows: The inverse cosine angle of the dot product of the flange end face normal vector and the horizontal reference plane reference normal vector during the static pre-assembly stage is read, and then algebraically added together with the tilt constant of the polarization industrial vision camera mounting bracket. Projecting the gap width values of the same sampling node on a unified time axis onto the vertical reference axis of the three-dimensional coordinate system generates the axial compression: ;in, This represents the axial compression of the p-th sampling node; This indicates the initial clearance width calibration value under the condition that the flange thread is not screwed in and is under stress; This represents the vernier value of the projection gap width of the p-th sampling node on the vertical reference axis; This indicates the thermal expansion correction coefficient for CDU stainless steel piping material; This indicates the real-time temperature measurement data fed back by the environmental monitoring probe at the assembly and construction site; This represents the standard manufacturing environment temperature for component processing and forming. Following the time sequence of a unified time axis, the radial runout set is used as the radial yaw component of the spatial cylindrical coordinate system, and the axial compression set is used as the vertical depth feed component. Multidimensional coordinate integration matrix splicing calculation is performed to fit and generate an assembly pose curve that characterizes the continuous spiraling and pressing trajectory of the flange joint in three-dimensional space.
[0030] Specifically, the process of extracting the time change rate of flange assembly torque data to generate the torque ramp gradient is as follows: The flange assembly torque data is truncated according to a preset time window to obtain local torque signal segments; a moving average filter is performed on the local torque signal segments to remove high-frequency mechanical vibration noise caused by gear meshing, thus obtaining a smooth torque sequence; the data increments of adjacent sampling nodes in the smooth torque sequence are extracted, and the data increments are mapped to the time span of adjacent sampling nodes to obtain the time change rate; the time change rates are spliced according to the sliding order of the time window to generate the torque ramp gradient of the flange assembly torque data.
[0031] In this implementation scheme, a sampling truncation window with a fixed time width is set, and an overlapping truncation operation is performed on the full-time acquisition axis of the flange assembly torque data according to a preset unidirectional sliding step size to obtain multiple independent local torque signal segments. For the high-frequency mechanical vibration noise caused by gear meshing in the local torque signal segments, a weighted moving average filter function is constructed to perform noise reduction and reconstruction. ; where, the smoothed torque value of the v-th sampling node in the filtered and reconstructed smoothed torque sequence is used as Indicates the total number of samples collected within a single smoothing operation within the local torque signal segment. Indicates; the original torque measurement value offset by the r-th step inward from the filter window in the historical time series is used Indicates; the numerical attenuation weighting coefficient corresponding to the offset position is used The method for determining the numerical attenuation weighting coefficient is as follows: The basic mechanical meshing frequency is obtained by analyzing the number of gear teeth and the current speed of the servo spindle reducer; a Gaussian bell-shaped probability density function with the time offset variable as the independent variable is constructed; the variance of this function is set according to the basic mechanical meshing frequency; and the mapped output value of the function at the discrete offset node r is extracted as... The time rate of change is generated by calculating the difference between the data increment and time span of adjacent observation data based on the reconstructed smooth torque sequence. ;in, This represents the torque ramp-up gradient corresponding to the position of the v-th sampling node; This represents the number of discrete span steps in which the difference calculation is performed in the smoothed torque sequence; This represents the smoothed torque value at the backtracking node position in the smoothed torque sequence; This represents the duration of a single analog-to-digital conversion of the torque sensor; and it is constrained. Based on the monotonically increasing sliding index of the truncated window on a unified time axis, the time change rate of each local torque signal segment is aligned in the time domain, duplicate gradient mapping values in the overlapping areas of adjacent truncated windows are removed, and the torque ramp-up gradient sequence covering the entire assembly cycle is generated by splicing them together in chronological order.
[0032] Specifically, the process of extracting abnormal time points where radial runout in the assembly pose curve exhibits a step change and reading the torque ramp-up gradient corresponding to these abnormal time points is as follows: Extract the time series of radial runout contained in the assembly pose curve along a unified time axis; perform first-order forward differencing on the radial runout time series to obtain the radial free ionization rate change curve; extract abrupt data peaks in the radial free ionization rate change curve that exceed the baseline smooth fluctuation range; mark the timestamps corresponding to these abrupt data peaks on the unified time axis as abnormal time points where radial runout exhibits a step change; retrieve the corresponding torque sampling index nodes on the unified time axis based on the abnormal time points; and extract the data points at the torque sampling index node positions in the torque ramp-up gradient to obtain the torque ramp-up gradient corresponding to the abnormal time points.
[0033] In this implementation scheme, the system extracts the radial runout time series contained in the assembly pose curve on a unified time axis, and performs calculus derivation on this time series using a discrete-time step forward difference algorithm to obtain the radial free ionization rate change curve. The specific calculation process is as follows: ;in, Represents the radial ionization rate of the h-th sampling node; This represents the radial runout of the (h+1)th sampling node; This represents the radial runout of the h-th sampling node; This represents the hardware clock timestamp of the (h+1)th sampling node; The timestamp of the hardware clock for the h-th sampling node is represented; h represents the sequence number of the discrete sampling node on the unified time axis; the time sampling interval is guaranteed to be strictly greater than zero. This represents the differential amplification suppression constant. The method for determining this differential amplification suppression constant is as follows: read the rated mechanical vibration fundamental frequency from the servo spindle nameplate parameters, take the reciprocal of this rated mechanical vibration fundamental frequency as the low-pass cutoff time parameter, and then map it to obtain... After obtaining the radial free ionization rate change curve, the system performs adaptive statistical boundary monitoring to identify abrupt data peaks exceeding the baseline smoothed fluctuation range. The calculation logic for the upper limit of the baseline smoothed fluctuation range is as follows: ;in, This represents the upper limit of the calculated baseline smoothed fluctuation range; This represents the arithmetic mean of the continuous radial ionization rates during the historical smoothing period; represents the total number of data samples in the sliding window during the historical smoothing period; c represents the historical backtracking offset index number within the window; This represents the radial ionization rate offset by c nodes from the historical data. This represents the statistical discrete scaling factor. The method for determining this statistical discrete scaling factor is as follows: Gaussian normal distribution is fitted to the noise floor data of the CDU pipeline under stress-free idling conditions, and the upper bound of the probability density integral with a confidence level of 99.7% is extracted as the [database name missing]. When the calculated real-time free velocity is significantly greater than the upper limit, an interrupt flag is attached to the corresponding timestamp memory address, confirming it as an abnormal time point where a step change occurs in the radial runout. The system uses the underlying data bus to retrieve the torque sampling index node that is hardware-synchronized with this abnormal time point, directly extracts the cached data block mapped by that index node, and obtains the torque ramp-up gradient corresponding to this abnormal time point.
[0034] Specifically, when there is a synchronous sudden increase in the torque ramp-up gradient, the specific process of generating off-center load interference features containing a three-dimensional offset vector based on the shaft center relative coordinate offset corresponding to the abnormal time point is as follows: compare the difference between the torque ramp-up gradient amplitude corresponding to the abnormal time point and the average amplitude of the torque ramp-up gradient during the historical smooth period. When the average amplitude difference exceeds the sudden increase judgment threshold set by the benchmark smooth period, it is confirmed that there is a synchronous sudden increase in the torque ramp-up gradient; extract the current shaft center relative coordinates of the upper and lower flanges of the joint in the three-dimensional coordinate system corresponding to the abnormal time point, as well as the initial shaft center relative coordinates of the upper and lower flanges of the joint in the initial zero stress state within the same assembly cycle; perform a spatial vector projection operation on the current shaft center relative coordinates to the initial shaft center relative coordinates, and extract the direction angle deviation and geometric modulus increment generated by the spatial vector projection operation to synthesize a three-dimensional offset vector; combine the time sequence identifier of the abnormal time point and the spatial state information of the three-dimensional offset vector to generate off-center load interference features.
[0035] In this implementation scheme, the system constructs a gradient comparison function to analyze the difference between the torque ramp gradient magnitude corresponding to abnormal time points and the average magnitude of the torque gradient during historical smooth periods. The calculation process is as follows: ;in, This indicates the average magnitude difference of the torque ramp-up gradient; This represents the instantaneous torque ramp-up gradient magnitude corresponding to the abnormal time point; This represents the total number of torque gradient samples compared during the historical smoothing period; d represents the traversal and accumulation index of the historical torque gradient. This represents the torque gradient sample point during the d-th historical smoothing period. When the calculated average amplitude difference is significantly greater than the set sudden increase threshold... At that time, the state machine confirms a synchronous sudden increase in the torque ramp-up gradient. This sudden increase is determined by a threshold. The determination method is as follows: Call the CDU material attribute database, extract the maximum elastic deformation resistance change rate of the fluororubber sealing ring inside the pipeline, and multiply it by the mechanical transmission efficiency constant of the servo tightening spindle reducer to obtain the result. After confirming the sudden increase, the system extracts the current relative coordinates of the upper and lower flanges of the joint at the abnormal time point in the three-dimensional coordinate system and constructs them as the current spatial vector; simultaneously, it retrieves the initial relative coordinates of the pipeline under zero stress and constructs them as the initial spatial vector. Algebraic geometric projection is then performed on the current spatial vector onto the initial spatial vector to calculate the orientation angle deviation and geometric modulus increment. ; ;in, This represents the angular deviation generated by the projection operation, and it is known that the product of the magnitudes of the two vectors is greater than zero when the offset occurs. This represents the current axis-relative coordinate space vector extracted at the abnormal time point; This represents the initial axis-to-axis relative coordinate space vector recorded under zero stress conditions; This represents the optical distortion compensation angle of the vision system. The method for determining this optical distortion compensation angle is as follows: during the device initialization phase, a Zhang's calibration checkerboard matrix is introduced, and the eigenvalues are obtained by solving the simultaneous equations of nonlinear radial and tangential distortion. . Indicates the geometric modulus increment; This represents the scaling factor for the rigid connection of the structure under the stuck state of the metal flange. The system performs matrix orthogonal synthesis of the calculated orientation angle deviation and geometric modulus increment according to the basis vector rules of the Cartesian coordinate system to generate a three-dimensional offset vector describing the pipe tilt and stuck state. Finally, the hexadecimal timing identifier of the abnormal time point in the underlying clock crystal is extracted, and it is combined bit-by-bit with the spatial three-dimensional tensor data of the three-dimensional offset vector according to the preset data frame structure. This is then encapsulated to generate a data packet characterizing the off-center load interference features and written into the database of the intelligent monitoring system to trigger the downstream stress interference analysis mechanism. Figure 2 As shown, in a three-dimensional coordinate system composed of the horizontal reference plane X, the horizontal reference plane Y, and the axial compression, the assembly posture curve during a normal tightening cycle exhibits a smooth, spiraling downward trend. When interference occurs at an abnormal time point, the flange experiences a step yaw in the projection direction of the horizontal reference plane, causing a nonlinear bend in the assembly posture curve. Extracting the starting calibration coordinates and the offset stability point coordinates of this bend distortion interval, the directed line segment formed by these two coordinate points in space is the three-dimensional offset vector shown in the figure. (See also...) Figure 3 The graph validates the synchronous surge in cross-modal features using a dual Y-axis structure, with the horizontal axis representing a unified time axis. The solid line represents the radial ionization rate change curve calculated from the image sequence, while the dashed line represents the torque gradient based on feedback from the torque sensor. During normal assembly, both curves remain within the baseline smooth fluctuation range. When off-center loading interference occurs, not only does the ionization rate in the visual dimension exceed its upper limit, but the torque gradient in the mechanical dimension also experiences a simultaneous step change, forming a corresponding synchronous surge peak on the overlapping vertical line of the unified time axis. This graph visually verifies the strong coupling correspondence between the two modal features at abnormal time points.
[0036] Specifically, the process of extracting the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints, and calculating the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span is as follows: Import the structural topology model of the CDU main distribution pipe, extract the spatial path length along the stainless steel pipeline from the current assembly node to the adjacent assembled joint in the structural topology model to generate the rigid pipe segment span; Combine the material bending stiffness parameters of the CDU main distribution pipe, map the three-dimensional offset vector contained in the off-center load interference feature into the end off-center load force in the horizontal plane of the three-dimensional coordinate system; Use the end off-center load force as the end excitation source and the rigid pipe segment span as the spatial lever arm length to construct the cantilever beam mechanical transmission equivalent function; Input the end off-center load force and the rigid pipe segment span into the cantilever beam mechanical transmission equivalent function and perform a spatial vector product mapping operation, extract the bending moment tensor acting on the joint surface of adjacent assembled joints output by the spatial vector product mapping operation to generate the rigid transmission torque.
[0037] In this implementation scheme, the system parses the 3D structural engineering drawings of the CDU main and branch water pipes via the manufacturing execution system bus, and loads the topological connection relationships of each pipe joint in 3D space. The system extracts the 3D coordinates of the assembly node currently experiencing off-center loading and the 3D coordinates of adjacent assembled joints from the structural topology model. It then extracts a discrete coordinate micro-element sequence along the central axis of the stainless steel main pipe, performs spatial curve integration on the discrete coordinate micro-element sequence to obtain the wiring length, and generates the rigid pipe segment span. In the mechanical transmission domain, the system treats the 3D offset vector contained in the off-center loading interference characteristics as a small forced displacement occurring at the end of the pipe. Combining this with the material bending stiffness parameters of the CDU main and branch water pipes, it maps this to an off-center loading force at the end of the pipe on the horizontal plane of the 3D coordinate system. The calculation process is as follows: ;in, This represents the calculated end-eccentric load force; This indicates the elastic modulus of CDU stainless steel piping material; The moment of inertia of the cross section of the stainless steel pipe; This represents the span of the rigid pipe section generated by the space curve integral; The geometric modulus scalar represents the three-dimensional bias vector; the mechanical shear deformation correction coefficient is... The method for determining the mechanical shear deformation correction coefficient is as follows: Read the difference between the inner and outer diameters of the CDU pipeline, calculate the geometric ratio of the pipeline wall thickness to the span of the rigid pipe section, and substitute this geometric ratio into the pre-calibrated short beam shear stress attenuation polynomial to obtain the result. The system uses the end-load force as a three-dimensional vectorization source, converts the rigid pipe segment span into a spatial lever arm direction vector pointing to the adjacent assembled joint, and constructs an equivalent function for the mechanical transmission of the cantilever beam to perform a spatial vector product mapping operation. ;in, The bending moment tensor, i.e., the rigid transmission torque, acts on the mating surfaces of adjacent assembled joints. The spatial lever arm direction vector representing the span of a rigid pipe section; This represents a three-dimensional spatial tensor of the end-load force. The system outputs this cross product tensor to objectively quantify the rigid lever effect caused by the current abnormal assembly action on adjacent pipe nodes.
[0038] Specifically, the process of mapping the rigid transmission torque to the flange loosening gap of adjacent assembled joints, and generating the re-tightening compensation torque for adjacent assembled joints based on the flange loosening gap is as follows: Extract the elastic deformation hysteresis coefficient of the O-ring seal inside the adjacent assembled joints and the mechanical stiffness coefficient of the fastening thread; load the rigid transmission torque into the strain function of the metal mating surface composed of the elastic deformation hysteresis coefficient and the mechanical stiffness coefficient, and extract the asymmetric micro-warping deformation output of the metal mating surface strain function to generate the flange loosening gap of the adjacent assembled joints; extract the maximum opening azimuth angle and opening geometry of the flange loosening gap at the mating surface of the adjacent assembled joints, and inversely map the opening geometry to the thread reset screw-in angle based on the preset thread lead parameters; perform closed-loop gain adjustment on the basic torque control baseline of the servo tightening spindle according to the thread reset screw-in angle to generate the re-tightening compensation torque for adjacent assembled joints.
[0039] In this implementation scheme, the system queries the CDU material assembly attribute database to extract the elastic deformation hysteresis coefficient of the fluororubber O-rings inside adjacent assembled joints under rated compression, and reads the mechanical stiffness coefficient of the stainless steel fastening threads under the current engagement state. The system establishes the local force differential equation of the flange mating surface, applies the calculated rigid transmission torque to the strain function of the metal mating surface composed of the above two stiffness attributes, solves for the asymmetric micro-warping deformation of adjacent flange faces when subjected to lateral bending moment prying, and generates the flange loosening clearance. ;in, This indicates the flange loosening clearance dimension calculated from the output. Represents the absolute scalar magnitude of the rigid transmission torque; This represents the elastic deformation hysteresis coefficient of the O-ring seal; This indicates the equivalent radius of action of the flange face subjected to eccentric prying stress; Indicates the mechanical stiffness coefficient of the fastening thread; This represents the asymmetric deformation deflection compensation base. The method for determining this asymmetric deformation deflection compensation base is as follows: Analyze the spatial distribution topology of the bolt holes of adjacent assembled flange joints, and calculate the base by multiplying the cosine of the spatial angle between the vector direction of the rigid transmission torque and the nearest stressed bolt hole by the Poisson's ratio of the flange stainless steel material. The system locates the strain extreme points of the mating surface based on the three-dimensional strain distribution cloud map, extracts the maximum opening azimuth angle and opening geometry of the flange loosening gap at the mating surface, and inversely maps the opening geometry to the thread reset screw-in angle by combining the pipe joint thread lead parameters. ;in, This indicates the thread reset advance angle generated by the mapping; Indicates the thread lead parameter of the pipe fitting; This represents the complete radian constant of a single turn of the thread; This indicates the maximum opening azimuth angle. Based on the obtained thread reset advance angle, the system performs closed-loop gain adjustment by superimposing a feedforward control component onto the basic torque control baseline of the servo tightening spindle PLC controller. ;in, This indicates the generated re-tightening compensation torque; This indicates the base torque control baseline locked in the underlying program of the servo tightening spindle; This indicates the closed-loop gain adjustment coefficient set by the system. After receiving a communication data frame containing the re-tightening compensation torque, the servo tightening spindle drives the actuator to perform a second micro-motion clockwise rotation on the adjacent assembled joint, eliminating the micro-connection loosening caused by rigid transmission at the surface level.
[0040] Example 2; please refer to Figure 4 A high-precision assembly intelligent monitoring system for CDU liquid cooling equipment is disclosed, used to execute the high-precision assembly intelligent monitoring method for CDU liquid cooling equipment described in the embodiments. The system includes: a data acquisition module for simultaneously acquiring end-face gap image sequences and flange assembly torque data during the tightening cycle of the CDU water pipe joint, and extracting the axial relative coordinates and gap width values of the upper and lower flanges in the end-face gap image sequence; a pose construction module for mapping the axial relative coordinates and gap width values to a three-dimensional coordinate system on a unified time axis, generating an assembly pose curve containing radial runout and axial compression, and extracting the time change rate of the flange assembly torque data to generate a torque ramp-up gradient; and an off-center load identification module for... The system extracts abnormal time points in the assembly pose curve where radial runout exhibits a step change, reads the torque creep gradient corresponding to these abnormal time points, and when a synchronous surge occurs in the torque creep gradient, generates an off-center load interference feature containing a three-dimensional offset vector based on the axis-to-axis coordinate offset corresponding to the abnormal time point. A re-tightening compensation module is used to extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints. Based on the rigid pipe segment span, it calculates the rigid transmission torque generated by the three-dimensional offset vector at the mating surface of adjacent assembled joints, maps the rigid transmission torque to the flange loosening gap of adjacent assembled joints, and generates a re-tightening compensation torque for adjacent assembled joints through the flange loosening gap.
[0041] In this implementation scheme, the data acquisition module is responsible for the coordinated scheduling and basic parameter extraction of the underlying sensing hardware at the assembly site. This module listens to the engagement trigger signal of the servo tightening spindle and synchronously triggers the exposure shutter of the polarization industrial vision camera and the analog-to-digital conversion channel of the servo flange torque sensor. After acquiring multi-angle field-of-view frames, the module performs pixel-level polarization calculation to remove the reflective layer of the high-brightness metal mirror, outputting a sequence of end-face gap images, and simultaneously converting the motor current feedback characteristics into real-time torque values. Subsequently, the module completes hardware-level clock alignment processing based on the acquisition timestamps of the dual-channel data, calls edge operators and sub-pixel-level ellipse fitting algorithms to parse the metal contours and geometric center points of the upper and lower flanges of the joint, and then scans and calculates the pixel span between the boundaries along the parallel direction of the axis connection. Finally, it combines the calibrated intrinsic parameters to map and output the relative coordinates of the axis center and the gap width value.
[0042] The pose construction module is responsible for processing multi-dimensional sensing data and reconstructing the motion trajectory of the joint during the screwing process. This module constructs a three-dimensional coordinate system with the geometric center of the stationary base of the CDU main water pipe as the absolute origin. It projects the axis-center relative coordinates extracted on a unified time axis onto a horizontal reference plane to calculate the radial runout, and simultaneously projects the gap width value onto a vertical reference axis to calculate the axial compression. In the time domain, this module fuses and stitches these two sets of spatial displacement components, fitting and outputting an assembly pose curve characterizing the three-dimensional precession process of the joint. When processing mechanical feedback, this module uses a preset time sliding window to segment and filter the flange assembly torque data using a moving average. It extracts the data increments between adjacent sampling nodes and maps them to a time span, calculating the discrete time change rate, and sequentially stitching them into a complete torque ramp-up gradient sequence.
[0043] The off-center load identification module is responsible for cross-modal synchronous discrimination of visual displacement data and mechanical resistance state. This module performs first-order forward difference processing on the radial runout time series in the assembly pose curve to obtain the radial free velocity change curve, locate abrupt data peaks exceeding the baseline smooth fluctuation range, and mark their corresponding timestamps as abnormal time points. Based on this timestamp, the module retrieves the corresponding torque sampling index node in a unified time axis, extracts the instantaneous torque rise gradient, and compares it with the average amplitude difference of historical smoothing period data. When a synchronous sudden increase in the torque rise gradient is confirmed, the module extracts the current relative coordinates of the flange shaft center and the relative coordinates of the shaft center under the initial zero stress state, performs a spatial vector projection operation, calculates the orientation angle deviation and geometric modulus increment, and packages them to generate off-center load interference features containing a three-dimensional offset vector.
[0044] The re-tightening compensation module is responsible for quantifying the mechanical impact of local assembly anomalies on the overall pipeline connection structure and issuing end-point hardware control commands. This module imports the structural topology model of the CDU main distribution pipe, calculates the spatial path length between the current assembly node and adjacent assembled joints along the pipeline route, and generates the rigid pipe segment span. Combining material bending stiffness parameters, this module converts the three-dimensional offset vector contained in the off-center load interference characteristics into an end-point off-center load force, and substitutes it, along with the rigid pipe segment span, into the constructed cantilever beam mechanical transmission equivalent function, performing a spatial vector product mapping operation to output a rigid transmission torque. This torque data is loaded into the strain function of the metal interface, and combined with the stiffness properties of the sealing ring and threads, calculates the flange loosening clearance of adjacent assembled joints, reverse-analyzes its opening geometry into the thread reset screw-in angle, and based on this angle, superimposes closed-loop gain adjustment parameters on the underlying control baseline to generate and send a re-tightening compensation torque to the servo tightening spindle.
[0045] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A high-precision assembly intelligent monitoring method for CDU liquid cooling equipment, characterized in that, Includes the following steps: S1. Synchronously acquire the end face gap image sequence and flange assembly torque data of the CDU water pipe joint tightening cycle, and extract the axial relative coordinates and gap width values of the upper and lower flanges in the end face gap image sequence. S2. Map the relative coordinates of the shaft center and the gap width value to the three-dimensional coordinate system on a unified time axis to generate an assembly pose curve containing radial runout and axial compression, and extract the time change rate of flange assembly torque data to generate torque ramp-up gradient. S3. Extract the abnormal time point in the assembly pose curve where the radial runout changes abruptly, read the torque creep gradient corresponding to the abnormal time point, and when there is a synchronous sudden increase in the torque creep gradient, generate the off-center interference feature containing the three-dimensional offset vector based on the axis relative coordinate offset corresponding to the abnormal time point. S4. Extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints. Calculate the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span. Map the rigid transmission torque to the flange loosening gap of adjacent assembled joints. Generate the re-tightening compensation torque for adjacent assembled joints through the flange loosening gap.
2. The intelligent monitoring method for high-precision assembly of CDU liquid cooling equipment according to claim 1, characterized in that: The specific process of synchronously acquiring the end face gap image sequence and flange assembly torque data during the tightening cycle of the CDU water distribution pipe joint is as follows: Receive the engagement trigger signal from the servo tightening spindle, and synchronously activate the polarization industrial vision camera and the servo flange torque sensor according to the engagement trigger signal; Extract multi-angle field-of-view frames of the CDU water pipe joint joint surface captured continuously by a polarized industrial vision camera, and filter out the high-brightness metal mirror reflective layer in the multi-angle field-of-view frames to generate an end face gap image sequence. Extract the motor current feedback characteristics of the servo flange torque sensor during the tightening cycle, and convert the motor current feedback characteristics into real-time torque values to generate flange assembly torque data. The first acquisition timestamp of the end face gap image sequence and the second acquisition timestamp of the flange assembly torque data are extracted. Hardware-level clock alignment processing is performed on the end face gap image sequence and the flange assembly torque data based on the first timestamp and the second timestamp.
3. The high-precision assembly intelligent monitoring method for CDU liquid cooling equipment according to claim 1, characterized in that: The specific process for extracting the relative coordinates of the axial centers of the upper and lower flanges and the gap width value from the end face gap image sequence is as follows: Edge operator extraction is performed on the end face gap image sequence to obtain the metal contours of the upper and lower flanges of the joint; Perform subpixel-level ellipse fitting on the metal profile, and extract the geometric center point from the ellipse fitting result to generate the relative coordinates of the axis centers of the upper and lower flanges of the joint. Scan the metal profile along the parallel direction of the line connecting the centers of the upper and lower flanges of the joint, and extract the pixel span between the boundaries of the upper and lower flanges of the joint. The pixel span is combined with the calibration intrinsic parameters of the polarized industrial vision camera to generate a gap width value by mapping it to spatial dimensions.
4. The intelligent monitoring method for high-precision assembly of CDU liquid cooling equipment according to claim 1, characterized in that: The specific process of mapping the relative coordinates of the axis center and the gap width value to a three-dimensional coordinate system on a unified time axis to generate an assembly pose curve that includes radial runout and axial compression is as follows: Extract the geometric center of the stationary base of the CDU main water distribution pipe to construct the absolute origin of the three-dimensional coordinate system; Project the axis-center relative coordinates on a unified time axis onto a horizontal reference plane of a three-dimensional coordinate system, and extract the horizontal free trajectory of the axis-center relative coordinates deviating from the absolute origin on the horizontal reference plane to generate radial runout. The gap width value on the unified time axis is projected onto the vertical reference axis of the three-dimensional coordinate system, and the attenuation difference between the gap width value on the vertical reference axis and the initial gap width is extracted to generate the axial compression. By integrating radial runout and axial compression on a unified time axis, an assembly pose curve characterizing the three-dimensional precession process of the joint space is generated through fitting.
5. The intelligent monitoring method for high-precision assembly of CDU liquid cooling equipment according to claim 1, characterized in that: The specific process for extracting the time rate of change of flange assembly torque data to generate the torque ramp-up gradient is as follows: The flange assembly torque data is truncated according to a preset time sliding window to obtain a local torque signal segment; A smooth torque sequence is obtained by performing a moving average filter on the local torque signal segment to remove high-frequency mechanical vibration noise caused by mechanical gear meshing. Extract the data increments of adjacent sampling nodes in the smooth torque sequence, and map the data increments to the time span of adjacent sampling nodes to obtain the rate of change over time; The torque ramp-up gradient of the flange assembly torque data is generated by stitching together the time change rate according to the sliding order of the time window.
6. The high-precision assembly intelligent monitoring method for CDU liquid cooling equipment according to claim 1, characterized in that: The specific process for extracting the abnormal time points where the radial runout in the assembly pose curve exhibits a step change and reading the torque ramp-up gradient corresponding to the abnormal time points is as follows: Extract the time series of radial runout contained in the assembly pose curve on a unified time axis, and perform first-order forward difference processing on the radial runout time series to obtain the radial free rate change curve. Extract the abrupt data peaks that exceed the baseline smooth fluctuation range in the radial free ionization rate change curve, and mark the timestamps corresponding to the abrupt data peaks on a unified time axis as the abnormal time points when the radial jump occurs. Based on the abnormal time point, the corresponding torque sampling index node is retrieved on a unified time axis, and the data point at the position of the torque sampling index node in the torque rise gradient is extracted to obtain the torque rise gradient corresponding to the abnormal time point.
7. The intelligent monitoring method for high-precision assembly of CDU liquid cooling equipment according to claim 1, characterized in that: When there is a synchronous sudden increase in the torque ramp-up gradient, the specific process of generating the off-center interference feature containing the three-dimensional offset vector based on the shaft center relative coordinate offset corresponding to the abnormal time point is as follows: By comparing the torque ramp gradient magnitude corresponding to the abnormal time point with the average magnitude of the torque gradient during the historical smooth period, a synchronous surge in the torque ramp gradient is confirmed when the average magnitude difference exceeds the surge judgment threshold set for the baseline smooth period. Extract the current relative coordinates of the upper and lower flanges of the joint in the three-dimensional coordinate system at the abnormal time point, as well as the initial relative coordinates of the upper and lower flanges of the joint in the initial zero stress state within the same assembly cycle; Perform a spatial vector projection operation from the current axis-center relative coordinates to the initial axis-center relative coordinates, and extract the orientation angle deviation generated by the spatial vector projection operation and synthesize the geometric magnitude increment into a three-dimensional offset vector. The temporal identifiers of the combined abnormal time points and the spatial state information of the three-dimensional bias vector are used to generate the off-center interference feature.
8. The high-precision assembly intelligent monitoring method for CDU liquid cooling equipment according to claim 1, characterized in that: The specific process of extracting the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints, and calculating the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints based on the rigid pipe segment span is as follows: Import the structural topology model of the CDU main distribution pipe, and extract the spatial path length from the current assembly node to the adjacent assembled joint along the stainless steel pipeline to generate the rigid pipe segment span. Combining the material bending stiffness parameters of the CDU main water distribution pipe, the three-dimensional offset vector contained in the off-center load interference characteristics is mapped to the end off-center load force on the horizontal plane of the three-dimensional coordinate system. Using the end-eccentric load as the end excitation source and the span of the rigid pipe section as the spatial lever arm length, an equivalent function for the mechanical transmission of the cantilever beam is constructed. The end-load force and the rigid pipe segment span are input into the cantilever beam mechanical transmission equivalent function and a spatial vector product mapping operation is performed. The bending moment tensor acting on the joint surface of adjacent assembled joints is extracted from the spatial vector product mapping operation to generate rigid transmission torque.
9. The intelligent monitoring method for high-precision assembly of CDU liquid cooling equipment according to claim 1, characterized in that: The specific process of mapping the rigid transmission torque to the flange loosening clearance of adjacent assembled joints, and generating the re-tightening compensation torque for adjacent assembled joints through the flange loosening clearance is as follows: Extract the elastic deformation hysteresis coefficient of the O-ring seal inside adjacent assembled joints and the mechanical stiffness coefficient of the fastening thread. The rigid transmission torque is applied to the strain function of the metal joint surface composed of the elastic deformation hysteresis coefficient and the mechanical stiffness coefficient. The asymmetric micro warping deformation output of the strain function of the metal joint surface is extracted to generate the flange loosening gap of adjacent assembled joints. Extract the maximum opening azimuth angle and opening geometry of the flange loosening gap at the mating surface of the adjacent assembled joint, and in combination with the preset thread lead parameters, reverse map the opening geometry into the thread reset screw-in angle. The basic torque control baseline of the servo tightening spindle is adjusted in a closed loop according to the thread reset screw-in angle to generate a re-tightening compensation torque for adjacent assembled joints.
10. A high-precision assembly intelligent monitoring system for CDU liquid cooling equipment, used to execute the high-precision assembly intelligent monitoring method for CDU liquid cooling equipment as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to simultaneously acquire the end face gap image sequence and flange assembly torque data of the CDU water pipe joint tightening cycle, and extract the axial relative coordinates and gap width values of the upper and lower flanges of the joint from the end face gap image sequence. The pose construction module is used to map the relative coordinates of the axis center and the gap width value to the three-dimensional coordinate system on a unified time axis, generate an assembly pose curve containing radial runout and axial compression, and extract the time change rate of flange assembly torque data to generate torque ramp-up gradient. The off-center load identification module is used to extract abnormal time points in the assembly pose curve where the radial runout changes abruptly, read the torque creep gradient corresponding to the abnormal time point, and generate off-center load interference features containing a three-dimensional offset vector based on the relative coordinate offset of the axis center corresponding to the abnormal time point when there is a synchronous sudden increase in the torque creep gradient. The re-tightening compensation module is used to extract the three-dimensional offset vector of the off-center load interference feature and the rigid pipe segment span between adjacent assembled joints. Based on the rigid pipe segment span, it calculates the rigid transmission torque generated by the three-dimensional offset vector at the joint surface of adjacent assembled joints, maps the rigid transmission torque to the flange loosening gap of adjacent assembled joints, and generates the re-tightening compensation torque for adjacent assembled joints through the flange loosening gap.
