A method for intelligent regulation of wall thickness of a crucible rotating pressure for piezoelectric single crystal growth

Abstract

The present application relates to the technical fields of intelligent manufacturing and automation control, and discloses a kind of intelligent regulation and control methods for the wall thickness of crucible spin pressing in piezoelectric single crystal growth, comprising: collecting data in the process of spin pressing, and obtaining stable acquisition sequence by preprocessing;Stable acquisition sequence is fused by adaptive weighted fusion and Kalman filtering, and the fused wall thickness is obtained by fusing the laser thickness value and the wall thickness value and compensation value respectively mapped by the spin pressing force, torque and eccentricity;Calculate the comprehensive deviation vector of wall thickness deviation, local wall thickness thinning rate, eccentricity deviation and roundness error;Based on the comprehensive deviation vector and the wall thickness deviation, the fuzzy adaptive PID output correction value and the hierarchical threshold comparison are executed in parallel to generate pass jump and trajectory reconstruction instructions, and the closed-loop cooperative control is formed by issuing the numerical control spin pressing system;After each pass is completed, the wall thickness distribution is obtained by fusion and solution, and the deviation is determined to add compensation pass or trajectory reconstruction, and iteration is carried out until the requirements are met.The present application realizes the online intelligent closed-loop regulation and control of the wall thickness of crucible spin pressing.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and automation control technology, and in particular to an intelligent method for controlling the crucible wall thickness during piezoelectric single crystal growth. Background Technology

[0002] Piezoelectric single crystal materials, such as lithium niobate, lithium tantalate, and lanthanum gallium silicate, are widely used in high-frequency communication filters, ultrasonic sensors, precision actuators, and microwave devices. These piezoelectric single crystals require high stability of the thermal field, uniformity of the crystal growth interface, and consistency of the growth environment during their growth. Therefore, the dimensional accuracy, wall thickness uniformity, shape and position accuracy, and high-temperature structural stability of the crucible used to support the melt or participate in the construction of the high-temperature growth environment directly affect the crystal growth quality.

[0003] In the manufacturing process of crucibles used for piezoelectric single crystal growth, the uniformity of crucible wall thickness distribution is particularly important. If there are localized deviations in wall thickness, roundness, or insufficient uniformity in the forming process, it can easily lead to uneven heat conduction at high temperatures, localized thermal field fluctuations, or decreased structural stability. This affects the stability of the crystal growth process and may result in problems such as crystal cracking, increased defects, poorer diameter uniformity, and reduced yield. Especially for large-size, high-precision crucibles, minute wall thickness errors or shape deviations during the forming process may be further amplified during subsequent high-temperature use, adversely affecting the quality of high-end piezoelectric single crystals.

[0004] Currently, crucible spinning processes typically require consideration of multiple factors, including material plastic deformation, wall thickness variation, contour accuracy, and processing stability. However, due to the large size of the crucible structure, complex wall morphology, and significant differences in deformation states in local areas, problems such as uneven wall thickness distribution, localized thinning, eccentricity, insufficient roundness, and unstable forming quality in transition areas are prone to occur during the spinning process. These problems not only affect the geometric accuracy and reliability of the crucible but also increase the difficulty of subsequent finishing and the loss of precious metal materials.

[0005] Furthermore, the quality of crucible spinning is influenced by a combination of factors, including equipment condition, material batch, process parameters, processing path, and operator experience. In actual processing, insufficient monitoring of the forming state or imprecise process adjustments can easily lead to accumulated processing errors, making it difficult to consistently guarantee consistency between different batches of products. This is especially true in deformation-sensitive areas such as the crucible opening, corner transition zones, and bottom center, where wall thickness and shape / position errors are more likely to concentrate, thus affecting overall forming accuracy.

[0006] Therefore, how to improve the accuracy of wall thickness determination, enhance the ability to control local forming deviations, reduce the accumulation of processing errors, and improve the consistency and stability of large-size high-precision crucible manufacturing during crucible spinning has become an urgent technical problem to be solved in the field of crucible manufacturing for piezoelectric single crystal growth. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent control method for crucible spinning wall thickness in piezoelectric single crystal growth. This method enables high-precision sensing of wall thickness status, real-time compensation of forming deviations, and coordinated control of processing trajectory and process parameters during crucible spinning, thereby improving crucible wall thickness uniformity, roundness accuracy, forming consistency, and manufacturing stability.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent control of crucible spin forming wall thickness for piezoelectric single crystal growth, comprising: Step S1: During the spinning process, data is collected synchronously by a laser thickness sensor, a two-dimensional displacement sensor, a spinning force sensor, a torque sensor, and a grating encoder, and a stable acquisition sequence is obtained after preprocessing. Step S2: Adaptive weighted fusion and Kalman filtering are applied to the stable acquisition sequence to fuse the laser thickness measurement value, the wall thickness value mapped by the rotational pressure, the wall thickness value mapped by the torque, and the compensation value mapped by the eccentricity to obtain the fused wall thickness. The weights are dynamically allocated based on vibration, steady state, and eccentricity. The wall thickness deviation is calculated based on the fused wall thickness and the target wall thickness, and the local wall thickness thinning rate, eccentricity deviation, and roundness error are calculated to generate a comprehensive deviation vector. Step S3, based on the comprehensive deviation vector and the wall thickness deviation, execute in parallel: use fuzzy adaptive PID to output the correction values ​​of the rotary wheel radial feed, spindle speed and spinning force; and according to the comparison between the comprehensive deviation vector and the graded threshold, select to maintain the path, or insert compensation passes and densification paths, or terminate and reconstruct the entire trajectory, and generate pass jump instructions and trajectory reconstruction instructions; Step S4: Send the correction value and instruction to the CNC spinning system to form a multi-closed-loop collaborative control system consisting of a closed loop for the spinning wheel trajectory, a closed loop for process parameters, a closed loop for the process flow, and a closed loop for wall thickness deviation. Step S5: After each pass is completed, a new stable acquisition sequence is obtained by collecting and preprocessing data through each sensor. The fusion calculation as in step S2 is performed sequentially on multiple measurement positions of the crucible to obtain the fused wall thickness of each part to form the wall thickness distribution, and the wall thickness deviation, eccentricity, and roundness error are calculated. If the finished product accuracy is met, the process is complete. If the wall thickness deviation or roundness error of any section exceeds the limit, an additional compensation pass is added and the process returns to step S3. If all deviations exceed the standard, the trajectory is reconstructed and all parameters are corrected before returning to step S3. The process continues until the wall thickness deviation and roundness meet the crucible accuracy requirements.

[0009] Furthermore, step S1 involves preprocessing the collected data to obtain a stable collection sequence, specifically including: The acquired raw signal is subjected to combined filtering, which includes first performing first-order hysteresis filtering to suppress high-frequency random noise, and then performing median filtering with adaptive window length adjustment to remove pulsed abrupt interference caused by stick-slip vibration. The multi-channel signals after the combined filtering are aligned with the same time scale according to the equal time scale synchronization triggering mechanism to ensure the homogeneity of multi-source data at the same sampling time. Based on the plastic rheological stress-strain constitutive model of platinum material spin forming, a reasonable deformation range threshold is set, and abnormal signals in the aligned data that exceed the reasonable deformation range threshold are marked and removed to form the stable acquisition sequence.

[0010] Furthermore, the weights mentioned in step S2 are dynamically allocated based on vibration, steady state, and eccentricity, specifically including: The fluctuation amplitude of the signals collected by the rotary force sensor and the torque sensor is detected in real time. When the fluctuation amplitude exceeds the preset vibration intensity threshold, it is determined that the current vibration is in a strong range. The weight ratio of the wall thickness value mapped by the rotary force and the wall thickness value mapped by the torque is increased, and the weight ratio of the laser thickness measurement value is decreased accordingly. When the fluctuation amplitude is lower than the preset steady-state determination threshold, it is determined that the current state is in the steady-state spinning range, and the dominant weight ratio corresponding to the laser thickness measurement value is increased. When the eccentricity detected by the two-dimensional displacement sensor exceeds the preset eccentricity judgment threshold, it is determined that the current region is prone to eccentricity, and the weight ratio corresponding to the compensation value mapped by the eccentricity is automatically increased.

[0011] Furthermore, the wall thickness value mapped by the spinning force and the wall thickness value mapped by the torque mentioned in step S2 are obtained through the following mapping models: Based on the spinning forming mechanism, a spinning force-wall thickness mapping model is constructed with radial spinning force and axial spinning force as inputs and wall thickness change as output. The spinning force-wall thickness mapping model characterizes the nonlinear analytical relationship between spinning force and wall thickness reduction caused by material flow in the plastic deformation zone. Based on the spinning mechanism, a torque-wall thickness mapping model is constructed with the spindle load torque as input and the wall thickness change as output. The torque-wall thickness mapping model characterizes the correspondence between the material's resistance to plastic deformation and the wall thickness reduction. The compensation value mapped by the eccentricity is obtained through an eccentricity-wall thickness deviation compensation model constructed based on the geometric relationship between eccentricity, spindle rotation angle and wall thickness deviation.

[0012] Furthermore, the step S2, which employs adaptive weighted fusion and Kalman filtering to obtain the fused wall thickness, specifically includes: The laser thickness measurement value, the wall thickness value mapped by the rotational force, the wall thickness value mapped by the torque, and the compensation value mapped by the eccentricity are used as inputs for adaptive weighted fusion, and a preliminary fused wall thickness value is obtained through weighted fusion. The initial fused wall thickness value is used as the measurement value of the Kalman filter. The time update and measurement update mechanism of the Kalman filter are used to perform optimal recursive estimation of the crucible wall thickness state in combination with the system state equation, and the fused wall thickness is output. The system state equation is constructed based on the plastic deformation characteristics of the material during spinning.

[0013] Furthermore, the step S3, which utilizes fuzzy adaptive PID to output correction values ​​for the radial feed of the spinning wheel, the spindle speed, and the spinning force, specifically includes: The wall thickness deviation and the local wall thickness thinning rate are used as inputs to the fuzzy controller. After fuzzification, fuzzy inference and defuzzification, the online adjustment amount of the PID control parameters is output. The proportional, integral, and derivative coefficients of the PID controller are self-tuned in real time using the online adjustment parameters. The PID controller, after online self-tuning, outputs the radial feed correction value of the spinning wheel, the spindle speed correction value, and the spinning force correction value based on the wall thickness deviation.

[0014] Furthermore, step S3, which involves comparing the comprehensive deviation vector with the grading threshold to select a maintained path, insert compensation paths and densified paths, or terminate and reconstruct the entire trajectory, specifically includes: A first deviation threshold and a second deviation threshold are set, wherein the first deviation threshold is less than the second deviation threshold; When the magnitude of the comprehensive deviation vector is less than the first deviation threshold, it is determined that the comprehensive deviation is in the normal range, and the current pass and processing path are maintained to continue processing; When the magnitude of the comprehensive deviation vector is between the first deviation threshold and the second deviation threshold, it is determined that the comprehensive deviation is in the medium deviation range, local compensation paths are automatically inserted, and encrypted compensation paths are generated in the thin wall thickness region and the eccentric region. When the magnitude of the comprehensive deviation vector is greater than the second deviation threshold, it is determined that the comprehensive deviation is in the severe deviation range. The current pass is immediately terminated, and the process is reverted to the designated correction section. The entire spinning trajectory is reconstructed, and the spinneret infeed angle, feed rate, and spinneret-crucible contact area are adjusted synchronously.

[0015] Furthermore, the multi-closed-loop collaborative control described in step S4, which forms a closed loop for the wheel trajectory, a closed loop for process parameters, a closed loop for the flow process, and a closed loop for wall thickness deviation, specifically includes: Control priorities are assigned to the rotary wheel trajectory closed loop, process parameter closed loop, process flow closed loop, and wall thickness deviation closed loop, respectively, wherein the control priority of the process flow closed loop is higher than that of the wall thickness deviation closed loop, and the control priority of the wall thickness deviation closed loop is higher than that of the rotary wheel trajectory closed loop and the process parameter closed loop. When the pass change command or trajectory reconstruction command is triggered, the process closed loop responds first, suspending the execution of the rotary wheel trajectory closed loop and the process parameter closed loop. After the pass change or trajectory reconstruction is completed, the rotary wheel trajectory closed loop and the process parameter closed loop are reactivated with the updated pass parameters and trajectory. During normal spinning, the wall thickness deviation closed loop uses a control period no greater than a preset wall thickness control period threshold as the outer loop, and the spinning wheel trajectory closed loop and the process parameter closed loop use a control period no greater than a preset trajectory control period threshold as the inner loop, forming a cascade-parallel hybrid control architecture. The correction amount output by the wall thickness deviation closed loop is used as the given input of the spinning wheel trajectory closed loop and the process parameter closed loop to achieve rapid convergence of the wall thickness deviation.

[0016] Furthermore, step S5, which involves sequentially performing the fusion calculation as in step S2 on multiple measurement locations of the crucible to obtain the fusion wall thickness at each location to form a wall thickness distribution, specifically includes: The laser thickness sensor and the two-dimensional displacement sensor are controlled to make equal-interval steps along the axial direction of the crucible, and the crucible is driven to rotate at least one revolution at each axial position. Multi-source data at various circumferential angle positions are collected simultaneously to form a measurement path covered by a circumferential-axial spiral. For each measurement point on the measurement path, the adaptive weighted fusion and Kalman filtering are performed to obtain the fused wall thickness value, wall thickness deviation, eccentricity and roundness error of the measurement point. The fused wall thickness values ​​of all measurement points are compared with the target wall thickness, mapped to the wall thickness distribution according to spatial coordinates, and the boundaries and deviation amplitudes of the thinner and thicker wall thickness regions are identified. Record the wall thickness deviation and roundness error of each measurement section after each processing pass, and establish a deviation-processing pass historical dataset.

[0017] Furthermore, in step S5, the determination of whether to add compensation tracks or reconstruct the trajectory is based on the following quantitative criteria, and when adding compensation tracks or reconstructing the trajectory is triggered, learning and adjustment are performed based on the aforementioned deviation-track historical dataset: The absolute value of the wall thickness deviation of each measurement section is compared with the preset local deviation allowable threshold, and the absolute value of the roundness error of each measurement section is compared with the preset local roundness allowable threshold. If the wall thickness deviation or roundness error of any section exceeds the corresponding preset local deviation allowable threshold or preset local roundness allowable threshold, it is determined as a local over-limit, and additional local compensation passes are triggered. When adding local compensation passes, the deviation change trend of multiple consecutive passes in the current defect area is extracted from the deviation-pass history dataset. The required compensation amount is predicted based on the deviation change trend, and the radial feed amount and feed rate of the rotary wheel of the densified compensation path of the local compensation pass are adjusted accordingly. Calculate the root mean square value of the wall thickness deviation and the root mean square value of the roundness error of all measurement sections, and use them as the overall deviation index respectively; if the root mean square value of the wall thickness deviation exceeds the preset overall wall thickness deviation reconstruction threshold, or the root mean square value of the roundness error exceeds the preset overall roundness reconstruction threshold, it is determined that the overall deviation exceeds the standard, and the trajectory reconstruction and full parameter correction are triggered. When trajectory reconstruction is triggered, the deviation improvement rate of multiple consecutive tracks in the entire region is extracted from the deviation-track history dataset. If the deviation improvement rate is lower than the preset convergence threshold, the range of the rotary wheel entry angle adjustment and the magnitude of the feed rate change are increased simultaneously during trajectory reconstruction to accelerate deviation convergence.

[0018] The beneficial effects of this invention are: 1. Improve the accuracy and stability of wall thickness detection: By fusing and solving multi-source detection data, the influence of spinning vibration, surface condition changes and single detection errors on wall thickness judgment can be reduced, thereby improving the reliability of crucible wall thickness detection results.

[0019] 2. Improve the real-time performance and accuracy of wall thickness control: Based on the deviation between the fused wall thickness and the target wall thickness, process parameters such as the radial feed of the swivel, the impeller speed and the swivel force are corrected in real time, which helps to reduce the accumulation of wall thickness deviation and improve the uniformity of crucible wall thickness.

[0020] 3. Enhance the adaptive adjustment capability of the spinning process: By comprehensively judging the wall thickness deviation, local thinning rate, eccentricity deviation and roundness error, the process can choose to maintain processing, add compensation passes or reconstruct the processing trajectory according to different forming states, thereby improving the forming quality of complex areas.

[0021] 4. Improve the overall shape and position accuracy and forming consistency of the crucible: While controlling the wall thickness, the eccentricity and roundness error are taken into account, which can effectively suppress defects such as local thinning, eccentricity and insufficient roundness, and improve the structural stability and batch processing consistency of the crucible.

[0022] 5. Reduce processing and debugging costs and material loss: Through multi-closed-loop collaborative control and pass compensation mechanism, the reliance on manual experience and the number of repeated trial rotations can be reduced, the loss of precious metal materials can be reduced, and the manufacturing efficiency and yield of crucibles for piezoelectric single crystal growth can be improved. Attached Figure Description

[0023] Figure 1 This is a flowchart of the steps of the intelligent control method for crucible spinning wall thickness in the piezoelectric single crystal growth of the present invention. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown in Example 1, this example provides a method for intelligent control of crucible wall thickness during piezoelectric single crystal growth, using a platinum crucible with a diameter of φ280mm and a target wall thickness of 2.0mm as an example. This platinum crucible is used in the growth processes of piezoelectric single crystals such as lithium niobate, lithium tantalate, and lanthanum gallium silicate, and has high requirements for wall thickness uniformity, roundness, coaxiality, and high-temperature structural stability. The target accuracy of this example is: wall thickness deviation ±0.02mm, roundness ≤0.03mm, and coaxiality ≤0.03mm.

[0026] The CNC spinning machine used in this embodiment has a spindle speed range of 0~300 r / min and a positioning accuracy of 0.001 mm. It is equipped with a radial feed servo axis, an axial feed servo axis, a spindle servo system, and a spinning pressure adjustment mechanism. The multi-sensor detection unit includes a laser thickness sensor, a laser two-dimensional displacement sensor, a spinning pressure sensor, a spindle torque sensor, and a grating encoder. The laser thickness sensor adopts a dual-probe structure, with a measurement range covering 1~5 mm and a resolution of not less than 0.001 mm; the laser two-dimensional displacement sensor is used to detect radial runout and axial position changes of the crucible's outer wall, with a resolution of not less than 0.001 mm; the spinning pressure sensor has a range of 0~10 kN and a sampling frequency of not less than 1 kHz; the spindle torque sensor has a range of 0~200 N·m and a sampling frequency of not less than 1 kHz; the grating encoder is used to acquire the position of the spinning wheel, the circumferential angle of the crucible, and the axial coordinates, with a position resolution of 0.001 mm.

[0027] The real-time controller adopts an industrial real-time controller with a control cycle of 1ms, supports multi-closed-loop motion control, and communicates with the sensor acquisition module, servo driver, and CNC spinning system via real-time Ethernet or fieldbus. The processing environment temperature is 20±2℃, the relative humidity is 40%~70%, and the vibration displacement of the equipment foundation is no greater than 0.02mm. Before processing, each sensor is calibrated using standard gauge blocks, standard circular rings, and no-load rotation tests.

[0028] During system initialization, the target wall thickness of 2.0 mm, initial pass planning, initial machining trajectory, spindle speed, rotary wheel radial feed, spinning force control range, and deviation threshold are imported. The initial spindle speed is set to 120 r / min, the initial rotary wheel radial feed is set to 0.02 mm / rev, the initial spinning force control range is 1.5~3.5 kN, and the initial pass planning is 3~5 passes. Measurement positions are set at the crucible opening, middle, corner transition zone, and bottom center zone. At least 16 circumferential measurement points are set for each axial section, with a total of no less than 128 measurement points.

[0029] The deviation thresholds are divided into three levels. Level 1 thresholds are: absolute wall thickness deviation ≤ 0.010 mm, local wall thickness thinning rate ≤ 1.5%, eccentricity deviation ≤ 0.010 mm, and roundness error ≤ 0.015 mm. Level 2 thresholds are: absolute wall thickness deviation > 0.010 mm and ≤ 0.020 mm, local wall thickness thinning rate > 1.5% and ≤ 3.0%, eccentricity deviation > 0.010 mm and ≤ 0.020 mm, or roundness error > 0.015 mm and ≤ 0.030 mm. Level 3 thresholds are: absolute wall thickness deviation > 0.020 mm, local wall thickness thinning rate > 3.0%, eccentricity deviation > 0.020 mm, or roundness error > 0.030 mm. When the overall deviation is within the first-level threshold range, the current processing path is maintained; when any parameter enters the second-level threshold range, a compensation pass is inserted and the encrypted path is executed; when any parameter enters the third-level threshold range and the proportion of the out-of-limit area to all measurement positions is less than 20%, an additional local compensation pass is added; when the proportion of the out-of-limit area to all measurement positions reaches more than 20% of the third-level threshold, or when the average wall thickness deviation still exceeds 0.020 mm after two consecutive passes, the current processing segment is terminated and the entire trajectory is reconstructed.

[0030] In this embodiment, the first-level, second-level, and third-level thresholds are individual deviation thresholds, used to determine whether wall thickness deviation, local wall thickness thinning rate, eccentricity deviation, and roundness error exceed limits, respectively. In subsequent specific embodiments, the first deviation threshold M1 and the second deviation threshold M2 are graded thresholds corresponding to the magnitude of the comprehensive deviation vector, used to comprehensively judge the overall forming state after normalizing multiple individual deviations. The individual deviation thresholds and the comprehensive deviation vector magnitude threshold together constitute a graded threshold system, where the individual deviation thresholds are used to identify specific defect types, and the comprehensive deviation vector magnitude threshold is used to determine the control strategy for maintaining the path, inserting compensation passes, or reconstructing the entire trajectory.

[0031] After processing begins, during the spinning process, data is synchronously acquired using a laser thickness sensor, a 2D displacement sensor, a spinning force sensor, a torque sensor, and a grating encoder. A stable acquisition sequence is obtained after preprocessing. The acquired data includes laser thickness measurement values, 2D displacement values, spinning force values, spindle load torque, spindle angle, radial position of the spinning wheel, and axial position of the spinning wheel. Preprocessing includes time alignment, median filtering, low-pass filtering, outlier removal, coordinate registration, and temperature drift compensation. The stable acquisition sequence is a multi-source data set corresponding to the same measurement position within the same sampling period. Its stability condition is that the wall thickness change is less than 0.005 mm and the radial displacement fluctuation is less than 0.003 mm within five consecutive sampling periods.

[0032] Subsequently, adaptive weighted fusion and Kalman filtering were applied to the stable acquisition sequence to obtain the fused wall thickness. The wall thickness values ​​mapped from laser thickness, rotational pressure, torque, and eccentricity were fused. The rotational pressure-mapped wall thickness was calculated based on the rotational pressure-wall thickness calibration curve, the torque-mapped wall thickness was calculated based on the torque-wall thickness calibration curve, and the eccentricity-mapped compensation value was obtained by calculating the crucible's geometric center offset from the two-dimensional displacement data and then converting it. The fusion weights were dynamically allocated based on vibration, steady-state conditions, and eccentricity: when the radial vibration amplitude was greater than 0.015 mm, the weight of the laser thickness measurement was reduced while the weights of the rotational pressure-mapped and torque-mapped wall thickness values ​​were increased; when the acquired data met the steady-state conditions, the weight of the laser thickness measurement was increased; when the eccentricity deviation was greater than 0.010 mm, the weight of the eccentricity-mapped compensation value was increased. After the fusion result was input into the Kalman filter model, the fused wall thickness at the current measurement position was output.

[0033] The wall thickness deviation is calculated based on the fused wall thickness and the target wall thickness of 2.0 mm. Local wall thickness thinning rate, eccentricity deviation, and roundness error are also calculated to generate a comprehensive deviation vector. The local wall thickness thinning rate is calculated as a percentage of the difference between the previous and current fused wall thicknesses relative to the previous wall thickness. The eccentricity deviation is calculated as the radial distance between the geometric center of the crucible and the rotation center of the spindle. The roundness error is calculated as the difference between the maximum and minimum radii of the same cross-section. The comprehensive deviation vector, including wall thickness deviation, local wall thickness thinning rate, eccentricity deviation, and roundness error, characterizes the overall forming state of the crucible at the current measurement position.

[0034] Based on the comprehensive deviation vector and wall thickness deviation, parameter correction and trajectory decision-making are performed in parallel. On one hand, fuzzy adaptive PID is used to output correction values ​​for the radial feed of the spinning wheel, spindle speed, and spinning force. The inputs of the fuzzy adaptive PID include wall thickness deviation, wall thickness deviation change rate, local wall thickness thinning rate, eccentricity deviation, and roundness error. The outputs include the correction values ​​for the radial feed of the spinning wheel, spindle speed, and spinning force. When the fused wall thickness is less than the target wall thickness and there is a tendency for excessive thinning, the radial feed of the spinning wheel and the spinning force are reduced, and the spindle speed is appropriately increased. When the fused wall thickness is greater than the target wall thickness and the forming is insufficient, the radial feed of the spinning wheel and the spinning force are appropriately increased, and the spindle speed is reduced or maintained. The single correction amount for the radial feed of the spinning wheel is controlled within ±0.005mm / rev, the single correction amount for the speed is controlled within ±10r / min, and the single correction amount for the spinning force is controlled within ±100N.

[0035] On the other hand, based on the comparison between the comprehensive deviation vector and the classification threshold, the system selects to maintain the path, insert compensation tracks and densify the path, or terminate and reconstruct the entire trajectory, generating track jump instructions and trajectory reconstruction instructions. The axial sampling interval of the densified path is reduced from 2.0 mm to 0.5~1.0 mm, and the local rotation pressure of the compensation track is reduced by 5%~15% compared with the original track, in order to reduce the risk of further local thinning.

[0036] The real-time controller sends correction values ​​for the rotary wheel radial feed, spindle speed, and spinning force, as well as pass jump commands and trajectory reconstruction commands, to the CNC spinning system, forming a multi-closed-loop collaborative control system consisting of a rotary wheel trajectory closed loop, a process parameter closed loop, a process flow closed loop, and a wall thickness deviation closed loop. Specifically, the rotary wheel trajectory closed loop corrects the rotary wheel motion trajectory based on feedback from the grating encoder and laser 2D displacement sensor; the process parameter closed loop corrects the spindle speed, rotary wheel radial feed, and spinning force based on feedback from the spinning force sensor and spindle torque sensor; the process flow closed loop determines whether to maintain the current pass, add a compensation pass, or reconstruct the trajectory based on the overall deviation level; and the wall thickness deviation closed loop continuously corrects the machining parameters based on the difference between the fused wall thickness and the target wall thickness.

[0037] After each pass, new stable acquisition sequences are obtained by collecting and preprocessing data from each sensor. Fusion calculations are then performed sequentially at multiple measurement locations on the crucible to obtain the fused wall thickness for each part and form a wall thickness distribution. Simultaneously, wall thickness deviation, eccentricity, and roundness error are calculated. If the wall thickness deviation, roundness, and coaxiality all meet the finished product accuracy requirements, processing is complete. If the wall thickness deviation or roundness error in any section exceeds the limit, an additional compensation pass is added, and the process returns to the parameter correction and trajectory decision-making process. If all deviations exceed the limit, the trajectory is reconstructed, full parameter correction is performed, and the process returns to the parameter correction and trajectory decision-making process. Through iterative iteration, the wall thickness deviation and roundness are made to meet the crucible accuracy requirements.

[0038] The working principle of this embodiment is as follows: During crucible spinning, the crucible wall thickness variation is influenced by a combination of factors, including the spinning wheel trajectory, radial feed of the spinning wheel, spindle speed, spinning force, material plastic flow, spindle load, and clamping eccentricity. Laser thickness measurement directly reflects the wall thickness, but it is susceptible to the effects of a highly reflective platinum surface, spinning vibration, and localized runout. Spinning force and spindle load torque reflect material deformation resistance and load variations. Two-dimensional displacement data reflects eccentricity and roundness changes. Therefore, adaptive fusion of laser thickness measurement values, wall thickness values ​​mapped by spinning force, wall thickness values ​​mapped by torque, and eccentricity compensation values ​​can improve the stability and accuracy of wall thickness determination.

[0039] After obtaining the fused wall thickness, the system further combines wall thickness deviation, local wall thinning rate, eccentricity deviation, and roundness error to form a comprehensive deviation vector, expanding the controlled object from a single wall thickness error to a comprehensive error of the forming state. Fuzzy adaptive PID adjusts the radial feed of the spinning wheel, spindle speed, and spinning force in real time based on the comprehensive deviation vector to keep the material deformation within the target range. Hierarchical threshold decision-making selects the maintenance path, adds compensation passes, or reconstructs the trajectory based on the degree of deviation, preventing the continued accumulation of local errors. Through multi-closed-loop collaborative control of the spinning wheel trajectory, process parameters, processing flow, and wall thickness deviation, areas prone to local thinning, eccentricity, and insufficient roundness, such as the crucible mouth, corner transition zone, and bottom center area, can receive targeted compensation, thereby improving the overall wall thickness uniformity, roundness accuracy, and processing consistency of the crucible.

[0040] After 3-5 iterations of optimization according to the method of this embodiment, the final platinum crucible for piezoelectric single crystal growth has a wall thickness deviation of ≤ ±0.018 mm, roundness of ≤ 0.025 mm, and coaxiality of ≤ 0.025 mm, meeting the requirements for high-end piezoelectric single crystal growth. This embodiment can reduce the impact of spinning vibration, surface reflection changes, and single sensor errors on wall thickness detection, improve the real-time performance of wall thickness control, reduce defects such as local thinning, eccentricity, and insufficient roundness, and reduce reliance on manual adjustment and loss of precious metal materials.

[0041] Example 2, based on Example 1, further explains the multi-source data synchronous acquisition and preprocessing process in step S1, and the adaptive weighted fusion and Kalman filtering process in step S2. This example focuses on how to form a stable acquisition sequence during the platinum crucible spinning process, and obtain the fused wall thickness based on the stable acquisition sequence.

[0042] Throughout the spinning process, multiple sensor groups synchronously acquire multi-dimensional status information in a time-scaled, non-contact manner. Laser thickness sensors are symmetrically positioned on the inner and outer walls of the crucible to acquire the original wall thickness signal in real time, with a sampling frequency of no less than 1 kHz and a measurement accuracy of no less than ±0.005 mm. Laser two-dimensional displacement sensors acquire the radial profile, eccentricity, and roundness error of the crucible in real time, with a sampling frequency of no less than 2 kHz. Spinning force sensors are positioned at the spinning wheel mounting base to acquire radial and axial spinning force signals in real time, with a response frequency of no less than 1 kHz. A spindle torque sensor acquires the spindle load torque in real time to characterize the material's plastic deformation strength. An optical encoder acquires the spinning wheel's spatial coordinates, spindle rotation angle, and position information in real time, forming a closed-loop position reference.

[0043] After the acquired data enters the preprocessing unit, a combined filter is first performed. This combined filter includes a first-order hysteresis filter and a median filter with adaptively adjusted window length. The first-order hysteresis filter is used to suppress high-frequency random noise, while the median filter is used to remove impulse-like abrupt interference caused by stick-slip vibration. For any sensing channel, the output value after combined filtering is calculated according to the following formula: ; ; ; Where i is the sensing channel number, i=1, 2, 3, 4, 5 correspond to the laser thickness sensor, laser two-dimensional displacement sensor, rotary pressure sensor, spindle torque sensor and grating encoder respectively; k is the sampling time number; This represents the original acquisition signal of the i-th sensing channel at the k-th sampling time. Let be the normalized scale for the i-th sensing channel, with units equal to 1. same; This is a dimensionless, original, normalized signal; The output signal is a dimensionless filtered signal. This is the dimensionless filtered output signal from the previous sampling time. The output signal is the filtered result of the restored engineering quantity. Let be the dynamic filtering coefficient of the i-th sensing channel at the k-th sampling time, 0 < <1, preferably 0.15~0.35. Laser thickness measurement channel and laser two-dimensional displacement channel. The unit is mm, for the spinning pressure channel. The unit is N, representing the spindle load torque channel. The unit is N·m, which represents the number of channels in a grating encoder. The reference value is selected as mm or rad based on the position or angle.

[0044] The above normalization process ensures that the input parameters of the tanh and asinh functions are dimensionless, thus preventing signals of different units from directly entering the nonlinear functions and causing dimensional mismatch.

[0045] The formula uses the asinh function to perform nonlinear compression on the original signal and the filtered signal from the previous time step, and then uses the tanh function to impose boundary constraints on the output result, making it difficult for spinning vibration and stick-slip impact to directly enter the stable acquisition sequence, thereby improving the reliability of the stable acquisition sequence in step S1.

[0046] Dynamic filter coefficients It adaptively adjusts based on the current vibration intensity and signal stability. When the spinning force and spindle load torque fluctuations increase... Decrease the amount of historical filtered signal to increase its proportion; when the acquired signal is stable... The length of the median filter window is increased to make the current signal response more timely. The length of the median filter window is automatically adjusted according to the fluctuation amplitude of radial spinning force, axial spinning force and spindle load torque. When the machining state is stable, 3 sampling points are taken; when there is slight fluctuation, 5 sampling points are taken; and when there is obvious stick-slip vibration, 7 to 9 sampling points are taken.

[0047] The multi-channel signals after combined filtering are aligned to a unified timescale using a synchronous triggering mechanism. Laser thickness measurements, two-dimensional displacement values, radial spinning force, axial spinning force, spindle load torque, spindle rotation angle, and wheel spatial coordinates are all mapped to the same sampling time, with an alignment error of no more than 0.5ms. This unified timescale alignment ensures that multi-source data at the same sampling time correspond to the same spinning deformation state, avoiding thickness judgment errors caused by different sensor response times.

[0048] The time-scaled data was further analyzed for anomaly identification based on a plastic rheological stress-strain constitutive model of platinum spinning. The reasonable deformation range thresholds were set as follows: radial spinning force 0.8~4.5kN, axial spinning force 0.3~3.0kN, spindle load torque 5~150N·m, wall thickness jump within a single sampling period not exceeding 0.020mm, and local wall thickness thinning rate per pass not exceeding 3.5%. Anomaly identification indicators were calculated using the following formula: ; in, Let be the anomaly detection index at the k-th sampling time; ln is the natural logarithm function; exp is the natural exponential function; The actual equivalent deformation stress is calculated from the radial rotation force, axial rotation force, and spindle load torque. The theoretical equivalent deformation stress is calculated based on the plastic rheological stress-strain constitutive model of platinum material spin forming; Equivalent plastic strain; This is the equivalent plastic strain rate; The reference yield stress for platinum material; It is the absolute value symbol; The wall thickness jump variable within the k-th sampling period; 0.020 is the threshold value for the spindle load torque jump variable in the kth sampling period; 0.020 is the threshold value for the wall thickness jump variable in a single sampling period, in mm; 3 is the threshold value for the spindle load torque jump variable in a single sampling period, in N·m.

[0049] when If the value is greater than 1.20, or if any of the following parameters—radial rotational force, axial rotational force, spindle load torque, wall thickness jump within a single sampling period, or local wall thickness thinning rate in a single pass—exceeds the threshold of the corresponding reasonable deformation range, the preprocessing unit marks the sampled data as an abnormal signal and removes it. The valid data after anomaly removal forms a stable acquisition sequence. This formula utilizes the ln and exp functions to enhance the identification sensitivity when the actual deformation stress deviates from the theoretical deformation stress, and uses the square root and tanh functions to comprehensively constrain wall thickness jumps and torque jumps, thus preventing unreasonable plastic deformation data, stick-slip impact data, and instantaneous measurement anomaly data from entering the subsequent fusion model.

[0050] Based on a stable acquisition sequence, the system employs an adaptive weighted fusion and Kalman filtering approach to construct a wall thickness observation model. This model uses laser-measured thickness values, wall thickness values ​​mapped from spindle pressure, wall thickness values ​​mapped from spindle load torque, and compensation values ​​mapped from eccentricity as fusion inputs. Weighted fusion yields a preliminary fused wall thickness value. Each weight is determined using a normalized weighting function. ; in, is the adaptive weight of the j-th type of fusion input at the k-th sampling time; j is the current fusion input number, j=1, 2, 3, 4 correspond to the laser thickness measurement value, the wall thickness value mapped by the rotation pressure, the wall thickness value mapped by the spindle load torque, and the compensation value mapped by the eccentricity, respectively; m is the summation index, m=1, 2, 3, 4; Let be the dimensionless signal stability factor of the j-th type of fused input at the k-th sampling time; Let be the dimensionless working condition adaptation factor of the j-th type of fused input at the k-th sampling time; ∑ is the summation symbol; the denominator is used to normalize the four types of weights, making + =1.

[0051] Signal stability factor Used to characterize the fluctuation level of the corresponding sensing signal or mapped signal; the more stable the signal, the better. The larger the value; the better the operating condition adaptation factor. Used to characterize the compatibility between the current spinning condition and the corresponding signal source. Within the steady-state spinning range, the laser thickness measurement corresponds to... Increase; within the range of stronger vibration, the wall thickness value mapped by the spinning force and the wall thickness value mapped by the spindle load torque correspond to... , Increase; within the region prone to eccentricity, the compensation value corresponding to the eccentricity mapping. The formula increases the weight distribution. It uses the tanh function to suppress the sudden weight changes caused by an excessively large single stability factor, and uses the ln and exp functions to improve the distinguishability of weight distribution under different working conditions, so that the weight can be dynamically distributed according to vibration, steady state and eccentricity.

[0052] The initial fusion wall thickness is calculated using the following formula: ; in, This is the initial fused wall thickness value at the k-th sampling time. The superscript 0 indicates that this value has not yet been corrected by Kalman filtering. For the wall thickness dimensional parameter, a target wall thickness of 2.0 mm is preferred; For the adaptive weights of the j-th type of fused input; For the j-th type of fusion input, there are four types. All units are in mm. For laser thickness measurement, The wall thickness value mapped by the spinning pressure. The wall thickness value is mapped from the spindle load torque. This is the estimated wall thickness after eccentricity compensation; It can be represented as Where f(e) is the compensation value mapped from the eccentricity, in mm. Therefore, All are dimensionless quantities. j is the fusion input number; ∑ is the weighted summation of the four types of fusion inputs.

[0053] The formula maps wall thickness correlation quantities from different sources and with different fluctuation characteristics to a unified scale using the asinh function, and then constrains the fusion output using the tanh function to avoid abrupt changes in the initial fusion wall thickness caused by a single sensing anomaly or mapping anomaly.

[0054] The wall thickness value mapped by the swirl pressure is obtained through a swirl pressure-wall thickness mapping model. This model takes radial and axial swirl pressure as inputs and wall thickness change as output, and is used to characterize the nonlinear analytical relationship between swirl pressure and wall thickness reduction caused by material flow in the plastic deformation zone. ; in, The wall thickness value is obtained by mapping the spin pressure. Radial rotational force; It is the axial rotational force; This is the estimated wall thickness from the previous state; The basic thinning correction factor; The radial swirl pressure is a nonlinear thinning coefficient; This is the saturation correction factor for axial rotational force; The coefficient representing the combined effect of radial and axial swirl forces; This is the normalization coefficient for radial swirl force; This is the normalization coefficient for axial rotational force. , and The unit is mm. The unit is mm / N. and The unit is ,make , This is a dimensionless quantity, and the unit for the entire expression within the parentheses is mm. Therefore, The unit is mm, which is consistent with the unit of wall thickness.

[0055] This formula uses the ln and exp functions to describe the nonlinear increasing trend of wall thickness reduction as radial spinning pressure increases, the tanh function to limit the estimation abrupt change when axial spinning pressure is too high, and the square root function to characterize the combined plastic deformation intensity under the combined action of radial and axial spinning pressure. When the laser thickness measurement value is affected by vibration or reflection from the platinum surface, this model can infer the trend of wall thickness change through the change in spinning pressure.

[0056] The wall thickness value mapped from the spindle load torque is obtained through a torque-wall thickness mapping model. This model takes the spindle load torque as input and the wall thickness change as output, and is used to characterize the relationship between the material's resistance to plastic deformation and the amount of wall thickness reduction. ; in, The wall thickness value is obtained by mapping the spindle load torque. Main spindle load torque; This refers to the no-load torque or the reference load torque. This is the estimated wall thickness from the previous state; The basic torque correction factor; This is the torque saturation correction coefficient; This is a nonlinear correction coefficient for torque deviation from the reference state; The normalization coefficient for spindle load torque; This is the torque deviation sensitivity coefficient. , and The unit is mm. and The unit is ,make and All are dimensionless quantities, and the unit for the entire expression within parentheses is mm. Therefore, the unit of f(T) is mm.

[0057] This formula uses the tanh function to suppress the estimation abrupt change caused by excessive torque, and uses the ln and exp functions to enhance the ability to identify the wall thickness reduction trend when the spindle load torque deviates from the reference state, so that the spindle load change can be used as an indirect observation of the material's resistance to plastic deformation.

[0058] The compensation value mapped from the eccentricity is obtained through an eccentricity-wall thickness deviation compensation model. This model is constructed based on the geometric relationship between eccentricity, spindle rotation angle, and wall thickness deviation. ; in, For the k-th sampling time, the principal axis rotation angle The wall thickness deviation before eccentricity correction at the location is calculated from the laser thickness measurement value, in mm; For the wall thickness dimensional parameter, in this embodiment, the target wall thickness is taken as 2.0 mm; This is the thickness value measured by laser measurement, in mm; The wall thickness deviation compensation amount is obtained by mapping the eccentricity and the wall thickness deviation before eccentricity correction together, and the unit is mm; This refers to the eccentricity, expressed in mm. The main axis rotation angle is calculated in radians and treated as a dimensionless angular variable. The eccentricity normalization parameter is preferably 0.010 mm; The basic compensation coefficient is expressed in mm. is the eccentric projection compensation coefficient, and is a dimensionless coefficient; This is the nonlinear eccentricity compensation coefficient, in mm; is the wall thickness deviation modulation coefficient, and is a dimensionless coefficient; This is the fourth type of fusion input, which is the estimated wall thickness after mapping the eccentricity and compensating for the wall thickness deviation, in mm.

[0059] This model, based on the original eccentricity and spindle rotation angle compensation, introduces a wall thickness deviation modulation term before eccentricity correction. This allows the geometric projection of the eccentricity in the current circumferential angle direction to participate in the compensation calculation along with the wall thickness deviation. Therefore, the model not only reflects the circumferential measurement offset caused by crucible eccentricity but also dynamically adjusts the compensation intensity according to the actual wall thickness deviation, thereby correcting potential spurious deviations in laser thickness measurements under eccentric conditions and improving the accuracy of calculating the fused wall thickness in the eccentric region.

[0060] This model avoids overcompensation or undercompensation caused by fixed compensation based solely on eccentricity, making the fourth type of fusion input obtained by subsequent adaptive weighted fusion closer to the true wall thickness state, thereby improving the reliability of wall thickness deviation closed-loop control, compensation track insertion, and trajectory reconstruction decisions.

[0061] In this embodiment, the preset vibration intensity thresholds include a spinning force fluctuation intensity threshold and a spindle load torque fluctuation intensity threshold. The spinning force fluctuation intensity threshold is set to 80 N, and the spindle load torque fluctuation intensity threshold is set to 3 N·m. When the fluctuation amplitude of the radial or axial spinning force is greater than 80 N, or the fluctuation amplitude of the spindle load torque is greater than 3 N·m, the current spinning process is determined to be in a high vibration range.

[0062] The preset steady-state determination thresholds include the spinning force steady-state threshold and the spindle load torque steady-state threshold. The spinning force steady-state threshold is set to 30 N, and the spindle load torque steady-state threshold is set to 1 N·m. When the fluctuation amplitudes of both radial and axial spinning force are less than 30 N, and the fluctuation amplitude of the spindle load torque is less than 1 N·m, and this fluctuation lasts for at least 20 ms, the current spinning process is determined to be in the steady-state spinning range.

[0063] The preset eccentricity threshold is set to 0.010 mm. When the eccentricity detected by the laser 2D displacement sensor is greater than 0.010 mm, the current processing area is determined to be an area prone to eccentricity, and the weight ratio of the compensation value mapped from the eccentricity is increased.

[0064] Weight allocation is based on vibration, steady-state, and eccentric dynamic distribution. The fluctuation amplitude of signals collected by the spindle pressure sensor and spindle torque sensor is monitored in real time. When the fluctuation amplitude of radial or axial spindle pressure exceeds 80 N, or the fluctuation amplitude of spindle load torque exceeds 3 N·m, it is determined that the current vibration range is strong. The weight proportions corresponding to the wall thickness values ​​mapped by spindle pressure and spindle load torque are increased, while the weight proportion corresponding to the laser thickness measurement value is correspondingly decreased. At this time, Take a value of 0.25~0.35. Take a value of 0.25~0.35. Take a value of 0.25~0.30. Take a value of 0.10 to 0.15.

[0065] When the fluctuation amplitudes of both radial and axial spinning pressure are less than 30 N, and the fluctuation amplitude of the spindle load torque is less than 1 N·m and lasts for at least 20 ms, it is determined that the current state is in the steady-state spinning range, and the dominant weight ratio corresponding to the laser thickness measurement value is increased. At this time, Take a value of 0.60~0.75. Take a value of 0.10~0.20. Take a value of 0.05~0.15. Take a value of 0.05 to 0.10.

[0066] When the eccentricity detected by the laser 2D displacement sensor exceeds 0.010mm, it is determined that the area is prone to eccentricity, and the weighting of the compensation value mapped from the eccentricity is automatically increased. At this time, Increase the weight to 0.20~0.30, and adjust the remaining weights proportionally. In areas prone to eccentricity, such as the crucible edge and corner transition zone, increase the eccentricity compensation weight even if the vibration intensity does not exceed the vibration intensity threshold, in order to enhance the ability to correct local geometric deviations.

[0067] The preliminary fused wall thickness value obtained by weighted fusion is used as the measurement value of the Kalman filter. The Kalman filter uses time update and measurement update mechanisms, combined with the system state equation, to recursively estimate the crucible wall thickness state: ; in, The wall thickness state vector at the k-th sampling time preferably includes the current wall thickness and the wall thickness change rate; A is the wall thickness state vector at the previous sampling time; B is the state transition matrix, used to describe the natural recursive relationship of the wall thickness state over time; C is the input control matrix, used to describe the influence of process input on the wall thickness state; D is the process input vector at the kth sampling time, preferably including the radial feed of the spinning wheel, the spindle speed, the radial spinning force, the axial spinning force, and the spindle load torque. Radial spinning force; T is the spindle load torque; n is the feed rate; n is the spindle speed; , , , These are the state nonlinearity correction coefficients; This is the process noise vector. The unit is , The unit is , The unit is the reciprocal of the feed rate unit. The unit of is the reciprocal of the unit of spindle speed, so that , , and All are dimensionless quantities; W(k) is the process noise vector, with the same unit as X(k).

[0068] The first row of the column vector in this formula characterizes the influence of radial spinning force and spindle load torque on the wall thickness itself, while the second row characterizes the influence of feed rate and spindle speed on the rate of change of wall thickness. The tanh function is used to limit abrupt changes in state when the spinning force and feed rate inputs are too large, and the ln and exp functions are used to improve the sensitivity of the identification of plastic deformation state by changes in spindle load torque and spindle speed.

[0069] This is the state nonlinear correction scaling matrix, used to convert the dimensionless nonlinear correction quantity into the wall thickness state quantity, preferably: ; in, The unit is mm, used to convert the dimensionless nonlinear correction amount in the first row into the wall thickness correction amount; The unit is mm / sampling period, used to convert the dimensionless nonlinear correction in the second row into a wall thickness change rate correction. By increasing... , , , and All have units consistent with X(k).

[0070] The measurement updates for the Kalman filter are performed according to the following formula: ; in, This is the wall thickness state estimation vector updated by measurement at the k-th sampling time; The predicted state vector obtained by updating over time at the k-th sampling time is labeled "". "Indicates a predicted value that has not been corrected for by current measurements; The Kalman gain at the k-th sampling time; For the initial fusion of wall thickness values; H is the wall thickness scale parameter; H is the measurement matrix used to map the wall thickness state vector into a measurement prediction value that can be compared with the preliminary fused wall thickness value. This is the disturbance correction factor, in mm; Let be the dimensionless combined disturbance quantity at the k-th sampling time, consisting of vibration intensity, eccentricity, and torque fluctuation; therefore... The unit is mm, which is consistent with the unit of the preliminary fusion wall thickness value and the measured predicted value. The unit is determined based on the ratio of the state vector X(k) to the unit of the measurement residual.

[0071] Comprehensive disturbance Calculate according to the following formula: ; in, is the cyclone pressure fluctuation amplitude at the k-th sampling time; 80 is the cyclone pressure fluctuation threshold in the stronger vibration range, in N; 3 represents the amplitude of the spindle load torque fluctuation at the kth sampling time; 3 represents the threshold value of the spindle load torque fluctuation in the area with strong vibration, in N·m. Let be the eccentricity at the k-th sampling time; 0.010 is the eccentricity threshold in mm. This formula uses the square root function to form the comprehensive vibration intensity, the tanh function to limit the vibration disturbance output, and the asinh function to smooth the effect of eccentricity on disturbance compensation.

[0072] After Kalman filtering recursion, the fused wall thickness at the current measurement position is output. Based on the fused wall thickness, the local wall thickness thinning rate, wall thickness deviation, eccentricity deviation, roundness error, and comprehensive deviation vector are calculated. The local wall thickness thinning rate is calculated according to the following formula: ; ; ; in, This is a dimensionless wall thickness variation ratio; The wall thickness of the previous pass or previous state is in mm. The current fused wall thickness is output after Kalman filtering, in mm; The value is a dimensionless sensitivity coefficient. λ(k) is a dimensionless intermediate variable; λ(k) is the local wall thickness thinning rate. The formula outputs 0 when the current fused wall thickness is not less than the wall thickness of the previous state, and outputs a positive value when wall thickness thinning occurs.

[0073] The overall deviation vector is constructed according to the following formula: ; in, This is the comprehensive deviation vector at the k-th sampling time. λ represents the wall thickness deviation in mm; 0.020 is the normalized threshold for wall thickness deviation in mm; λ is the local wall thickness thinning rate; 3.0% is the normalized threshold for local wall thickness thinning rate. The value is the eccentricity deviation in mm; 0.030 is the normalized threshold for eccentricity deviation in mm. 0.030 is the roundness error, in mm; 0.030 is the roundness error normalization threshold, in mm. All function parameters in the formula are dimensionless; when the local wall thickness thinning rate is 0, the corresponding component is 0, to avoid generating non-physical base values ​​in the comprehensive deviation vector.

[0074] This formula unifies wall thickness deviation, local wall thickness thinning rate, eccentricity deviation, and roundness error into dimensionless deviation quantities, allowing different types of deviations to enter the same hierarchical decision model, providing a unified judgment basis for maintaining the path, compensating for track insertion, and trajectory reconstruction in step S3.

[0075] The working principle of this embodiment is as follows: During the spinning process, wall thickness detection is easily affected by platinum surface reflection, equipment vibration, stick-slip impact, material plastic flow, and clamping eccentricity. In the preprocessing stage, noise and abrupt interference are removed by combined filtering, and the homogeneity of multi-source data is ensured by uniform time scale alignment. Then, abnormal signals exceeding the threshold of reasonable deformation range are eliminated based on the plastic rheological stress-strain constitutive model, thereby forming a stable acquisition sequence.

[0076] In the fusion calculation stage, laser thickness measurement provides direct wall thickness observation, wall thickness values ​​mapped by spindle pressure and spindle load torque provide indirect observation of material plastic deformation strength, and compensation values ​​mapped by eccentricity provide geometric deviation correction. Adaptive weights are dynamically adjusted according to vibration, steady state, and eccentricity states, enabling automatic switching of the dominant information source under different processing conditions; Kalman filtering further utilizes the temporal continuity of wall thickness states for recursive estimation. Therefore, this embodiment improves the accuracy, continuity, and anti-interference capability of fused wall thickness, providing reliable state input for subsequent parameter correction, pass jumps, and trajectory reconstruction, thereby solving the problems of unstable single detection signals, delayed wall thickness judgment, and misjudgment of eccentric regions. Through the coordination of normalized scale, state nonlinearity correction scale matrix, and perturbation correction scale coefficient, the input parameters of the nonlinear functions in each formula are dimensionless, and all output results are restored to the corresponding engineering units.

[0077] Example 3, based on Examples 1 and 2, further explains the dual-mode intelligent decision-making in step S3, the multi-closed-loop collaborative control in step S4, and the multi-pass iterative optimization in step S5. This example focuses on how to achieve real-time process parameter correction, intelligent pass switching, self-reconstruction of processing paths, and global iterative optimization based on the comprehensive deviation vector.

[0078] In step S3, the system performs two levels of intelligent decision-making in parallel based on the comprehensive deviation vector, realizing coordinated control of parameter correction and process reconfiguration. The comprehensive deviation vector includes wall thickness deviation, local wall thickness thinning rate, eccentricity deviation, and roundness error.

[0079] To ensure that the magnitude of the composite deviation vector is 0 when there is no deviation, and to ensure that all function input parameters are dimensionless, the magnitude of the composite deviation vector is calculated according to the following formula: ; in, The magnitude of the composite deviation vector; This refers to the wall thickness deviation, in mm. This refers to the local wall thickness thinning rate; Eccentricity deviation, in mm; The value represents the roundness error, in mm; 0.020 mm, 3.0%, 0.030 mm, and 0.030 mm are the normalized reference thresholds for wall thickness deviation, local wall thickness thinning rate, eccentricity deviation, and roundness error, respectively. When... When the value is 0, the wall thickness deviation component is zero, avoiding non-physical initial deviations in M. This formula normalizes and nonlinearly compresses wall thickness deviation, local wall thinning rate, eccentricity deviation, and roundness error, enabling the unified conversion of forming deviations of different dimensions into a comprehensive deviation level index. This improves the accuracy of deviation level judgment, providing a reliable basis for maintaining the current path, inserting compensation passes, or reconstructing the entire trajectory, avoiding insufficient or excessive compensation due to misjudgment of a single deviation index.

[0080] In this embodiment, the first deviation threshold M1 is set to 1.0, and the second deviation threshold M2 is set to 2.0, with the first deviation threshold M1 being less than the second deviation threshold M2. When the magnitude M of the comprehensive deviation vector is less than the first deviation threshold M1, the comprehensive deviation is determined to be in the normal range, and processing continues with the current pass and processing path. When the magnitude M of the comprehensive deviation vector is between the first deviation threshold M1 and the second deviation threshold M2, the comprehensive deviation is determined to be in the medium deviation range, and a local compensation pass is automatically inserted, generating a denser compensation path in the thinner wall and eccentric regions. When the magnitude M of the comprehensive deviation vector is greater than the second deviation threshold M2, the comprehensive deviation is determined to be in the severe deviation range, the current pass is immediately terminated, the process reverts to the designated correction section, and the entire spinning trajectory is reconstructed, simultaneously adjusting the spinneret infeed angle, feed rate, and spinneret-crucible contact area.

[0081] Real-time intelligent correction of process parameters adopts a fuzzy adaptive PID control model. The fuzzy controller takes wall thickness deviation and local wall thickness thinning rate as inputs, and after fuzzification, fuzzy inference, and defuzzification, outputs the online adjustment of the PID control parameters. The fuzzy universe of discourse for wall thickness deviation is set to [-0.04mm, 0.04mm], and the fuzzy universe of discourse for local wall thickness thinning rate is set to [0, 5%]. Fuzzy linguistic variables are divided into seven levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The membership degree of the r-th fuzzy rule is calculated according to the following formula: ; in, denoted as , where r is the activation intensity of the r-th fuzzy rule; r is the fuzzy rule number; Δt is the wall thickness deviation; and λ is the local wall thickness thinning rate. Let be the center value of the membership function of the wall thickness deviation in the r-th fuzzy rule, in mm; is the width parameter of the wall thickness deviation membership function in the r-th fuzzy rule, in mm; Let r be the center value of the membership function of the local wall thickness thinning rate in the r-th fuzzy rule; Let be the width parameter of the membership function of the local wall thickness thinning rate in the r-th fuzzy rule; It is a dimensionless sensitivity coefficient.

[0082] In this formula, and All values ​​are dimensionless. This formula can simultaneously reflect the impact of both the magnitude of wall thickness deviation and the local thinning trend on the control rules, allowing the fuzzy controller to intervene and adjust the system in advance when the wall thickness deviation has not yet seriously exceeded the limit but the local thinning trend has already intensified. This improves the foresight of parameter correction and reduces the risk of further local thinning of the crucible.

[0083] To ensure that the online adjustment of PID parameters has a definite unit, the online adjustment of PID control parameters is calculated according to the following formula: ; in, This refers to the online adjustment amount of the proportional coefficient; This refers to the online adjustment amount of the integral coefficient; R represents the online adjustment amount of the differential coefficients; R represents the total number of fuzzy rules; r represents the fuzzy rule number. Let be the activation strength of the r-th fuzzy rule; , , The unit is , For wall thickness deviation, , and All are dimensionless quantities; , , The unit of is the reciprocal of the unit of local wall thickness thinning rate, where λ is the local wall thickness thinning rate, making λ、 λ and λ are all dimensionless quantities; "-ln2" is used to make the nonlinear term corresponding to λ=0 zero.

[0084] The preferred scaling matrix for adjusting PID parameters is: ; in, , and These coefficients, each with the same units as the proportional, integral, and derivative coefficients, are used to convert dimensionless fuzzy inference results into PID parameter adjustments with practical control meaning. If a normalized PID controller is used, The identity matrix can be used, in which case... , and This is the normalized PID parameter adjustment. Based on the activation intensity of different fuzzy rules, the formula performs online self-tuning of the proportional, integral, and derivative coefficients, enabling the PID controller to dynamically change the control intensity according to wall thickness deviations and local thinning states during the spinning process. This improves the adaptability of the rotary wheel radial feed, spindle speed, and spinning pressure adjustment, and reduces control lag and regulation oscillations caused by fixed PID parameters.

[0085] The PID controller, after online self-tuning, outputs the following correction values: radial feed correction value for the spinning wheel, spindle speed correction value, and spinning force correction value. ; in, This is the radial feed correction value for the spinning wheel, in mm / rev. This is the spindle speed correction value, in r / min; is the correction value for the spinning pressure, in N; G is the normalization matrix corresponding to the wall thickness deviation and the local wall thickness thinning rate, used to convert Δt and λ into dimensionless inputs; This is the normalization matrix corresponding to the eccentricity deviation and roundness error, used to convert Δe and ΔR into dimensionless inputs. This is not the same as the roundness error ΔR; D and Q are unit-corrected gain matrices, and each row of D and Q corresponds to... , and The units are given. Therefore, the input parameters of the tanh and asinh functions are dimensionless, and the outputs, after D and Q conversions, yield process parameter corrections in mm / rev, r / min, and N units, respectively.

[0086] This formula maps wall thickness deviation, local wall thinning rate, eccentricity deviation, and roundness error together into correction values ​​for the radial feed of the spinning wheel, the spindle speed, and the spinning force, allowing wall thickness control and form and position error control to participate in parameter correction simultaneously. This improves eccentricity and roundness errors while controlling wall thickness uniformity, thereby enhancing the overall forming accuracy of the crucible.

[0087] When the fused wall thickness is less than the target wall thickness and the local thinning rate increases, the system reduces the radial feed of the spinning wheel, decreases the spindle speed, and reduces the spinning pressure correction value to reduce the risk of further local thinning. When the fused wall thickness is greater than the target wall thickness and the forming is insufficient, the system appropriately increases the radial feed of the spinning wheel and the feed rate, and increases the spinning pressure correction value to bring the local material flow closer to the target wall thickness. All correction parameters support online self-tuning to adapt to the plastic rheological properties of platinum.

[0088] During the intelligent switching of passes and self-reconstruction of the processing path, the system executes corresponding strategies based on the deviation classification threshold. When the overall deviation is less than the minimum threshold, the current pass and path are maintained and processing continues; when the overall deviation is in the middle range, a local compensation pass is automatically inserted, and a densified compensation path is generated in the defect area; when the overall deviation exceeds the maximum threshold, the current pass is immediately terminated, and the system reverts to the designated correction section, reconstructs the entire spinning trajectory, and simultaneously adjusts the spindle entry angle, feed rate, and contact area.

[0089] The axial step size of the encryption compensation path is determined according to the following formula: ; in, The axial step size for the encryption compensation path, in mm; The minimum axial step distance is preferably 0.5 mm; The standard axial step distance is 2.0 mm; This is a dimensionless comprehensive deviation level adjustment coefficient; The magnitude of the composite deviation vector; The first deviation threshold is set to 1.0; This is a dimensionless eccentricity deviation adjustment coefficient; 0.030 is the eccentricity deviation; 0.030 is the normalized threshold for eccentricity deviation, in mm.

[0090] In this formula, the input parameters of the exponential function and the hyperbolic tangent function are both dimensionless quantities. The unit is composed of and The value is determined to be in mm. This formula automatically adjusts the axial step size of the compensation path based on the overall deviation level and eccentricity deviation, ensuring that areas with more severe deviations and more pronounced eccentricity receive denser compensation paths. This improves the local correction capability for areas with thinner walls and eccentricity, reduces compensation blind spots, and avoids over-machining across the entire area.

[0091] The path self-reconfiguration follows these rules: in areas with thinner walls, the radial feed is reduced, the spindle speed is decreased, and the effective contact area of ​​the spinning wheel is increased; in areas with thicker walls, the radial feed and feed rate are appropriately increased; in eccentric areas, angular partitioning is performed according to the spindle rotation angle, and unequal thickness compensation is implemented to achieve simultaneous correction of roundness and coaxiality. When retracing to the designated correction segment, the designated correction segment is determined based on the axial boundary and circumferential angular boundary of the over-limit area. Local compensation passes cover the over-limit area and the preceding and following transition areas, with the length of the transition area being 10% to 20% of the axial length of the over-limit area.

[0092] In step S4, the real-time controller synchronously sends parameter correction values, trajectory correction commands, and pass jump commands to the CNC spinning system to execute multi-closed-loop collaborative control. The multi-closed loop includes a spinning wheel position trajectory closed loop, a process parameter closed loop, a flow closed loop, and a wall thickness deviation closed loop. The spinning wheel position trajectory closed loop achieves millisecond-level real-time correction; the spindle speed, feed rate, and spinning force constitute the process parameter closed loop; pass jumps, path replacements, and area backtracking constitute the flow closed loop; and the wall thickness deviation constitutes the target closed loop, achieving rapid convergence and stable control.

[0093] Each closed loop is assigned a control priority. The control priority of the process closed loop is higher than that of the wall thickness deviation closed loop, and the control priority of the wall thickness deviation closed loop is higher than that of the rotary wheel position trajectory closed loop and the process parameter closed loop. When a pass change command or trajectory reconstruction command is triggered, the process closed loop responds first, pausing the execution of the rotary wheel position trajectory closed loop and the process parameter closed loop; after the pass change or trajectory reconstruction is completed, the system reactivates the rotary wheel position trajectory closed loop and the process parameter closed loop with the updated pass parameters and trajectory.

[0094] During normal spinning operation, the wall thickness deviation closed loop uses a control period no greater than a preset wall thickness control cycle threshold as the outer loop, while the spinning wheel position trajectory closed loop and the process parameter closed loop use control periods no greater than a preset trajectory control cycle threshold as the inner loop, forming a cascade-parallel hybrid control architecture. In this embodiment, the preset wall thickness control cycle threshold is 10ms, the preset trajectory control cycle threshold is 1ms, and the system control cycle is no greater than 1ms. The correction value output by the wall thickness deviation closed loop serves as the given input for the spinning wheel position trajectory closed loop and the process parameter closed loop, enabling the wall thickness deviation to converge quickly.

[0095] In step S5, after each spinning pass is completed, the system automatically enters a multi-pass iterative optimization process. The laser thickness sensor and the laser two-dimensional displacement sensor move at equal intervals along the crucible's axial direction, driving the crucible to rotate at least one revolution at each axial position, simultaneously acquiring multi-source data at various circumferential angles to form a measurement path covered by a circumferential-axial spiral. The standard axial step spacing is 2.0 mm; in areas with thinner walls, eccentric regions, and corner transition areas, the axial step spacing is reduced to 0.5~1.0 mm.

[0096] For each measurement point along the measurement path, adaptive weighted fusion and Kalman filtering are performed to obtain the fused wall thickness, wall thickness deviation, eccentricity, and roundness error for that measurement point. The system compares the fused wall thickness values ​​of all measurement points with the target wall thickness, maps them to a wall thickness distribution according to spatial coordinates, and identifies the boundaries and deviation amplitudes of areas with insufficient wall thickness and areas with excessive wall thickness. After each pass, the wall thickness deviation and roundness error of each measurement segment are recorded, forming a deviation-pass history dataset.

[0097] When determining whether to add compensation passes or reconstruct the trajectory, the system uses quantification criteria. The preset allowable threshold for local deviation is 0.020 mm, and the preset allowable threshold for local roundness is 0.030 mm. If the absolute value of the wall thickness deviation in any segment is greater than 0.020 mm, or the roundness error in any segment is greater than 0.030 mm, it is determined to be a local over-limit, triggering the addition of local compensation passes.

[0098] When adding additional local compensation passes, the system first calculates the dimensionless predictive compensation coefficient for the q-th defect region, and then converts this coefficient into radial feed rate prediction correction values ​​and feed rate prediction correction values ​​in engineering units. The dimensionless predictive compensation coefficient is calculated according to the following formula: ; ; ; in, is the dimensionless prediction compensation coefficient for the q-th defect region; q is the defect region number; m is the historical pass number; L is the number of historical passes involved in the learning process, preferably the 3 most recent passes; This represents the time decay weight corresponding to the m-th historical trace; the closer it is to the current trace, the greater the decay. The larger, and all The sum is 1; , where is the wall thickness deviation of the q-th defect region after the m-th pass, in mm; 0.020 is the allowable threshold for local wall thickness deviation, in mm; is a dimensionless deviation convergence hysteresis coefficient, used to characterize whether the rate of decrease in wall thickness deviation has slowed down in recent multiple passes; 0.08 is the convergence hysteresis normalization threshold, corresponding to an 8% deviation improvement ratio. The radial feed rate prediction correction value for the q-th defect region, in mm / rev; The radial feed compensation scale is expressed in mm / rev, preferably 0.005 mm / rev. This is the predicted correction value for the feed rate of the q-th defect region, in mm / min; This is the feed rate compensation scale, in mm / min.

[0099] Through the above transformation, the dimensionless prediction compensation coefficient can be mapped to the process parameter correction amount that the CNC spinning system can execute. This formula calculates the prediction compensation coefficient using the continuous deviation variation trend in the deviation-pass historical dataset and converts it into radial feed prediction correction values ​​and feed rate prediction correction values. This allows the compensation pass to no longer rely solely on the current single detection result, but rather to combine historical convergence trends to predict the compensation amount, thereby reducing overcompensation and undercompensation and improving the convergence efficiency of multi-pass iterations.

[0100] The system simultaneously calculates the root mean square (RMS) values ​​of wall thickness deviation and roundness error for all measurement sections, using them as overall deviation indices. The RMS value of wall thickness deviation is calculated using the following formula: ; in, is the root mean square value of the wall thickness deviation, in mm; i is the measurement section number; N is the total number of measurement sections; is the wall thickness deviation of the i-th measurement segment, in mm; 0.020 is the allowable threshold for local wall thickness deviation, in mm.

[0101] In this formula, the weight term is a dimensionless quantity. The unit is mm 2 After square root operation The unit is mm. This formula evaluates the overall wall thickness deviation of all measurement sections using a weighted root mean square method, and increases the influence weight of sections with large deviations in the overall evaluation. This allows for more sensitive identification of whether there is a systematic exceedance of the overall wall thickness deviation of the crucible, providing a quantitative basis for triggering trajectory reconstruction and full parameter correction.

[0102] The root mean square value of roundness error is calculated according to the following formula: ; in, is the root mean square value of roundness error, in mm; i is the measurement section number; N is the total number of measurement sections; Let be the roundness error of the i-th measurement segment, in mm; 0.030 is the local roundness allowable threshold, in mm. In this formula, the weight term is a dimensionless quantity. The unit is mm 2 After square root operation The unit is mm. This formula evaluates the overall roundness error of all measurement segments using a weighted root mean square method, and enhances the influence of areas with large roundness errors on the overall judgment. This avoids local roundness degradation being masked by averaging, improves the ability to identify overall roundness anomalies, and facilitates timely initiation of trajectory reconstruction or roundness compensation.

[0103] In this embodiment, the preset overall wall thickness deviation reconstruction threshold is 0.015mm, and the preset overall roundness reconstruction threshold is 0.020mm. If the root mean square value of the wall thickness deviation... Greater than 0.015 mm, or the root mean square value of roundness error If the value is greater than 0.020 mm, it is determined to be an overall overshoot, triggering trajectory reconstruction and full parameter correction.

[0104] When trajectory reconstruction is triggered, the system extracts the deviation improvement rate from the historical deviation-track dataset for multiple consecutive tracks across the entire region. The deviation improvement rate determination function is calculated according to the following formula: ; in, This is the deviation improvement rate determination value after the m-th pass; m is the pass number. This is the overall deviation index after the m-th pass; This is the overall deviation index after the (m-1)th pass; To avoid extremely small positive numbers with a denominator of zero, it is preferable to select... . and All are dimensionless indicators, therefore It is a dimensionless quantity.

[0105] Overall Deviation Index Calculate according to the following formula: ; in, This is the overall deviation index after the m-th pass; This is the root mean square value of the wall thickness deviation after the m-th pass, in mm; is the root mean square value of the roundness error after the m-th pass, in mm; 0.015 is the overall wall thickness deviation reconstruction threshold, in mm; 0.020 is the overall roundness reconstruction threshold, in mm. In this formula, and All of them are dimensionless quantities, therefore It is a dimensionless overall deviation index.

[0106] This formula unifies the root mean square values ​​of wall thickness deviation and roundness error into an overall deviation index, enabling a comprehensive assessment of wall thickness uniformity and roundness accuracy within the same evaluation system. This improves the comprehensiveness of overall deviation assessment and avoids optimizing only wall thickness while neglecting roundness, or correcting only roundness leading to the accumulation of wall thickness deviations.

[0107] In this embodiment, a preset convergence threshold of 8% is used to determine whether the deviation improvement between consecutive passes is sufficient. The deviation improvement ratio is calculated as the ratio of the decrease in the overall deviation index of two adjacent passes to the overall deviation index of the previous pass, i.e.: ; in, This represents the percentage improvement in deviation after the m-th pass. This is the overall deviation index after the m-th pass; This is the overall deviation index after the (m-1)th pass; To avoid extremely small positive numbers with a denominator of zero, it is preferable to select... . It is a dimensionless quantity. When two consecutive cycles of... When all values ​​are less than 0.08, the rate of improvement of the judgment deviation is lower than the preset convergence threshold.

[0108] Deviation Improvement Rate Judgment Function Used to measure the percentage of deviation improvement The expression for nonlinear smoothing is as follows: ; in, This is the deviation improvement rate determination value after the m-th pass. A value of 0 indicates that the overall deviation has not improved. Equals 0; A value greater than 0 indicates an improvement in overall deviation. Greater than 0; A value less than 0 indicates that the overall deviation has worsened. Less than 0.

[0109] This judgment function avoids false triggering of trajectory reconstruction due to fluctuations in individual trace measurements, and makes trajectory reconstruction judgment more stable. The deviation improvement ratio and deviation improvement rate judgment functions are used to evaluate the convergence speed of the overall deviation between adjacent traces, and reduce the impact of single-detection fluctuations on the judgment results through nonlinear smoothing. When the deviation improvement ratio of two consecutive traces is lower than a preset convergence threshold, the system determines that the current compensation efficiency is insufficient and initiates a larger-scale trajectory reconstruction and parameter correction. This avoids invalid repetitive processing and accelerates the convergence of wall thickness deviation and roundness error.

[0110] When the deviation improvement rate is less than 8% for two consecutive passes, the deviation convergence is considered slow. During trajectory reconstruction, the adjustment range of the rotary wheel entry angle and the variation range of the feed rate are increased simultaneously. The adjustment range of the rotary wheel entry angle is increased from ±1° to ±3°, and the variation range of the feed rate is increased from ±5% to ±15% to accelerate the convergence of wall thickness deviation and roundness error.

[0111] The working principle of this embodiment is as follows: The fusion of wall thickness and overall deviation vectors reflects the current forming state of the crucible, but different types of deviations correspond to different control actions. Wall thickness deviation mainly affects radial feed and spindle pressure compensation, local wall thinning rate mainly affects spindle speed and material flow velocity, and eccentricity deviation and roundness error mainly affect angle zone compensation and trajectory reconstruction. Therefore, the system simultaneously sets two decision modes: real-time intelligent correction of process parameters and intelligent skipping of passes with self-reconstruction of processing paths.

[0112] When the deviation is small, the fuzzy adaptive PID controller continuously adjusts the PID control parameters online to continuously correct the radial feed of the spinning wheel, the spindle speed, and the spinning force correction value according to the forming state, thereby suppressing the accumulation of wall thickness errors. When the deviation enters the medium deviation range, continuous parameter correction alone is insufficient to eliminate local defects. The system automatically inserts a local compensation pass and generates a denser compensation path in areas with thin wall thickness and eccentricity. When the deviation enters the severe deviation range, continuing to process along the current path may amplify the defects. The system immediately terminates the current pass, reverts to the designated correction section, and reconstructs the entire spinning trajectory.

[0113] Multi-loop collaborative control avoids conflicts between different control objectives through a priority mechanism. The process loop prioritizes handling pass jumps and trajectory reconstruction, the wall thickness deviation loop is responsible for target wall thickness convergence, and the turbine position trajectory loop and process parameter loop are responsible for millisecond-level response at the execution level. The outer-loop wall thickness control, together with the inner-loop trajectory and process parameter control, constitutes a cascade-parallel hybrid control architecture, enabling wall thickness deviations to be quickly converted into specific corrections for turbine trajectory and process parameters.

[0114] Multi-pass iterative optimization obtains the crucible's global wall thickness distribution by covering the measurement path with a circumferential-axial helix, and uses the deviation-pass historical dataset to record the trend of defect area changes with each pass. When adding compensation passes or reconstructing the trajectory, the system predicts the compensation amount not only based on the current deviation but also by combining the deviation change trend of consecutive passes, thereby improving the compensation's targeting and reducing over-compensation or under-compensation. By converting the dimensionless predicted compensation coefficients into execution parameters with mm / rev and mm / min units, this embodiment can ensure the dimensional consistency between the algorithm's decision results and the CNC spinning system's execution parameters, improving the implementability and stability of control commands.

[0115] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent control of crucible spin forming wall thickness for piezoelectric single crystal growth, characterized in that, include: Step S1: During the spinning process, data is collected synchronously by a laser thickness sensor, a two-dimensional displacement sensor, a spinning force sensor, a torque sensor, and a grating encoder, and a stable acquisition sequence is obtained after preprocessing. Step S2: Adaptive weighted fusion and Kalman filtering are applied to the stable acquisition sequence to fuse the laser thickness measurement value, the wall thickness value mapped by the rotational pressure, the wall thickness value mapped by the torque, and the compensation value mapped by the eccentricity to obtain the fused wall thickness. The weights are dynamically allocated based on vibration, steady state, and eccentricity. The wall thickness deviation is calculated based on the fused wall thickness and the target wall thickness, and the local wall thickness thinning rate, eccentricity deviation, and roundness error are calculated to generate a comprehensive deviation vector. Step S3, based on the comprehensive deviation vector and the wall thickness deviation, execute in parallel: use fuzzy adaptive PID to output the correction values ​​of the rotary wheel radial feed, spindle speed and spinning force; and according to the comparison between the comprehensive deviation vector and the graded threshold, select to maintain the path, or insert compensation passes and densification paths, or terminate and reconstruct the entire trajectory, and generate pass jump instructions and trajectory reconstruction instructions; Step S4: Send the correction value and instruction to the CNC spinning system to form a multi-closed-loop collaborative control system consisting of a closed loop for the spinning wheel trajectory, a closed loop for process parameters, a closed loop for the process flow, and a closed loop for wall thickness deviation. Step S5: After each pass is completed, a new stable acquisition sequence is obtained by collecting and preprocessing data through each sensor. The fusion calculation as in step S2 is performed sequentially on multiple measurement positions of the crucible to obtain the fused wall thickness of each part to form the wall thickness distribution, and the wall thickness deviation, eccentricity, and roundness error are calculated. If the finished product accuracy is met, the process is complete. If the wall thickness deviation or roundness error of any section exceeds the limit, an additional compensation pass is added and the process returns to step S3. If all deviations exceed the standard, the trajectory is reconstructed and all parameters are corrected before returning to step S3. The process continues until the wall thickness deviation and roundness meet the crucible accuracy requirements.

2. The intelligent control method for crucible spinning wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, Step S1 involves preprocessing the collected data to obtain a stable acquisition sequence, specifically including: The acquired raw signal is subjected to combined filtering, which includes first performing first-order hysteresis filtering to suppress high-frequency random noise, and then performing median filtering with adaptive window length adjustment to remove pulsed abrupt interference caused by stick-slip vibration. The multi-channel signals after the combined filtering are aligned with the same time scale according to the equal time scale synchronization triggering mechanism to ensure the homogeneity of multi-source data at the same sampling time. Based on the plastic rheological stress-strain constitutive model of platinum material spin forming, a reasonable deformation range threshold is set, and abnormal signals in the aligned data that exceed the reasonable deformation range threshold are marked and removed to form the stable acquisition sequence.

3. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, The weights mentioned in step S2 are dynamically allocated based on vibration, steady state, and eccentricity, specifically including: The fluctuation amplitude of the signals collected by the rotary force sensor and the torque sensor is detected in real time. When the fluctuation amplitude exceeds the preset vibration intensity threshold, it is determined that the current vibration is in a strong range. The weight ratio of the wall thickness value mapped by the rotary force and the wall thickness value mapped by the torque is increased, and the weight ratio of the laser thickness measurement value is decreased accordingly. When the fluctuation amplitude is lower than the preset steady-state determination threshold, it is determined that the current state is in the steady-state spinning range, and the dominant weight ratio corresponding to the laser thickness measurement value is increased. When the eccentricity detected by the two-dimensional displacement sensor exceeds the preset eccentricity judgment threshold, it is determined that the current region is prone to eccentricity, and the weight ratio corresponding to the compensation value mapped by the eccentricity is automatically increased.

4. The intelligent control method for crucible spinning wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, The wall thickness values ​​mapped by spinning force and by torque mentioned in step S2 are obtained through the following mapping models: Based on the spinning forming mechanism, a spinning force-wall thickness mapping model is constructed with radial spinning force and axial spinning force as inputs and wall thickness change as output. The spinning force-wall thickness mapping model characterizes the nonlinear analytical relationship between spinning force and wall thickness reduction caused by material flow in the plastic deformation zone. Based on the spinning mechanism, a torque-wall thickness mapping model is constructed with the spindle load torque as input and the wall thickness change as output. The torque-wall thickness mapping model characterizes the correspondence between the material's resistance to plastic deformation and the wall thickness reduction. The compensation value mapped by the eccentricity is obtained through an eccentricity-wall thickness deviation compensation model constructed based on the geometric relationship between eccentricity, spindle rotation angle and wall thickness deviation.

5. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, Step S2, which involves adaptive weighted fusion and Kalman filtering to obtain the fused wall thickness, specifically includes: The laser thickness measurement value, the wall thickness value mapped by the rotational force, the wall thickness value mapped by the torque, and the compensation value mapped by the eccentricity are used as inputs for adaptive weighted fusion, and a preliminary fused wall thickness value is obtained through weighted fusion. The initial fused wall thickness value is used as the measurement value of the Kalman filter. The time update and measurement update mechanism of the Kalman filter are used to perform optimal recursive estimation of the crucible wall thickness state in combination with the system state equation, and the fused wall thickness is output. The system state equation is constructed based on the plastic deformation characteristics of the material during spinning.

6. The intelligent control method for crucible spinning wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, Step S3, which utilizes fuzzy adaptive PID to output correction values ​​for the radial feed of the spinning wheel, the spindle speed, and the spinning force, specifically includes: The wall thickness deviation and the local wall thickness thinning rate are used as inputs to the fuzzy controller. After fuzzification, fuzzy inference and defuzzification, the online adjustment amount of the PID control parameters is output. The proportional, integral, and derivative coefficients of the PID controller are self-tuned in real time using the online adjustment parameters. The PID controller, after online self-tuning, outputs correction values ​​for the radial feed of the spinning wheel, the spindle speed, and the spinning force based on the wall thickness deviation.

7. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, Step S3, which involves comparing the comprehensive deviation vector with the grading threshold to select a maintained path, inserting compensation paths and densified paths, or terminating and reconstructing the entire trajectory, specifically includes: A first deviation threshold and a second deviation threshold are set, wherein the first deviation threshold is less than the second deviation threshold; When the magnitude of the comprehensive deviation vector is less than the first deviation threshold, it is determined that the comprehensive deviation is in the normal range, and the current pass and processing path are maintained to continue processing; When the magnitude of the comprehensive deviation vector is between the first deviation threshold and the second deviation threshold, it is determined that the comprehensive deviation is in the medium deviation range, local compensation paths are automatically inserted, and encrypted compensation paths are generated in the thin wall thickness region and the eccentric region. When the magnitude of the comprehensive deviation vector is greater than the second deviation threshold, it is determined that the comprehensive deviation is in the severe deviation range. The current pass is immediately terminated, and the process is reverted to the designated correction section. The entire spinning trajectory is reconstructed, and the spinneret infeed angle, feed rate, and spinneret-crucible contact area are adjusted synchronously.

8. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, The multi-closed-loop collaborative control described in step S4, which forms a closed loop for the wheel trajectory, a closed loop for process parameters, a closed loop for the flow, and a closed loop for wall thickness deviation, specifically includes: Control priorities are assigned to the rotary wheel trajectory closed loop, process parameter closed loop, process flow closed loop, and wall thickness deviation closed loop, respectively, wherein the control priority of the process flow closed loop is higher than that of the wall thickness deviation closed loop, and the control priority of the wall thickness deviation closed loop is higher than that of the rotary wheel trajectory closed loop and the process parameter closed loop. When the pass change command or trajectory reconstruction command is triggered, the process closed loop responds first, suspending the execution of the rotary wheel trajectory closed loop and the process parameter closed loop. After the pass change or trajectory reconstruction is completed, the rotary wheel trajectory closed loop and the process parameter closed loop are reactivated with the updated pass parameters and trajectory. During normal spinning, the wall thickness deviation closed loop uses a control period no greater than a preset wall thickness control period threshold as the outer loop, and the spinning wheel trajectory closed loop and the process parameter closed loop use a control period no greater than a preset trajectory control period threshold as the inner loop, forming a cascade-parallel hybrid control architecture. The correction amount output by the wall thickness deviation closed loop is used as the given input of the spinning wheel trajectory closed loop and the process parameter closed loop to achieve rapid convergence of the wall thickness deviation.

9. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 1, characterized in that, Step S5, which involves sequentially performing the fusion calculation as described in step S2 at multiple measurement locations of the crucible to obtain the fusion wall thickness at each location to form a wall thickness distribution, specifically includes: The laser thickness sensor and the two-dimensional displacement sensor are controlled to make equal-interval steps along the axial direction of the crucible, and the crucible is driven to rotate at least one revolution at each axial position. Multi-source data at various circumferential angle positions are collected simultaneously to form a measurement path covered by a circumferential-axial spiral. For each measurement point on the measurement path, the adaptive weighted fusion and Kalman filtering are performed to obtain the fused wall thickness value, wall thickness deviation, eccentricity and roundness error of the measurement point. The fused wall thickness values ​​of all measurement points are compared with the target wall thickness, mapped to the wall thickness distribution according to spatial coordinates, and the boundaries and deviation amplitudes of the thinner and thicker wall thickness regions are identified. Record the wall thickness deviation and roundness error of each measurement section after each processing pass, and establish a deviation-processing pass historical dataset.

10. The intelligent control method for crucible spin forming wall thickness for piezoelectric single crystal growth according to claim 9, characterized in that, In step S5, the determination of whether to add compensation tracks or reconstruct the trajectory is based on the following quantitative criteria, and when adding compensation tracks or reconstructing the trajectory is triggered, learning and adjustment are performed based on the aforementioned deviation-track historical dataset: The absolute value of the wall thickness deviation of each measurement section is compared with the preset local deviation allowable threshold, and the absolute value of the roundness error of each measurement section is compared with the preset local roundness allowable threshold. If the wall thickness deviation or roundness error of any section exceeds the corresponding preset local deviation allowable threshold or preset local roundness allowable threshold, it is determined as a local over-limit, and additional local compensation passes are triggered. When adding local compensation passes, the deviation change trend of multiple consecutive passes in the current defect area is extracted from the deviation-pass history dataset. The required compensation amount is predicted based on the deviation change trend, and the radial feed amount and feed rate of the rotary wheel of the densified compensation path of the local compensation pass are adjusted accordingly. Calculate the root mean square value of the wall thickness deviation and the root mean square value of the roundness error of all measurement sections, and use them as the overall deviation index respectively; if the root mean square value of the wall thickness deviation exceeds the preset overall wall thickness deviation reconstruction threshold, or the root mean square value of the roundness error exceeds the preset overall roundness reconstruction threshold, it is determined that the overall deviation exceeds the standard, and the trajectory reconstruction and full parameter correction are triggered. When trajectory reconstruction is triggered, the deviation improvement rate of multiple consecutive tracks in the entire region is extracted from the deviation-track history dataset. If the deviation improvement rate is lower than the preset convergence threshold, the range of the rotary wheel entry angle adjustment and the magnitude of the feed rate change are increased simultaneously during trajectory reconstruction to accelerate deviation convergence.

Images