A multi-source data-based die-bushing machine control method and numerical control system

Abstract

The application discloses a multi-source data-based walking core machine control method and numerical control system, aiming to solve the problems of difficulty in monitoring high-frequency whirling of a workpiece in a guide sleeve, lack of dynamic sensing capability and insufficient compensation accuracy in the prior art. The method comprises the following steps: synchronously collecting a spindle motor current signal and a back shaft micro-displacement signal; extracting current harmonic amplitude and phase characteristics and axial displacement time sequence characteristics and fusing them; solving a two-dimensional eccentricity vector of a front end of the workpiece in real time through a double-input Kalman filter; generating a half-cycle advance compensation amount in combination with a system delay model, and superimposing the half-cycle advance compensation amount on a trajectory instruction to drive a feed shaft to offset the whirling. The application further discloses a walking core machine numerical control system, which comprises a high-speed analog-to-digital conversion module, an eddy current displacement sensor, a feature reconstruction subsystem, a state estimation kernel and a trajectory feedforward compensation module. Through multi-source data fusion and active feedforward compensation, the application can realize online reconstruction and suppression of the whirling without modifying the guide sleeve.

Description

Technical Field

[0001] This invention belongs to the field of machine tool control, specifically relating to a control method and CNC system for Swiss-type lathes based on multi-source data. Background Technology

[0002] With the rapid development of precision manufacturing technology, the processing efficiency and quality of high-precision parts have become core indicators for measuring the performance of CNC machine tools. In the machining of slender shaft parts with a large length-to-diameter ratio, the Swiss-type CNC lathe, with its unique guide sleeve support structure, can effectively improve the radial stiffness during the cutting process, playing an irreplaceable role in the production of medical implants, precision electronic and optical shaft parts.

[0003] The Swiss-type lathe's CNC system, as the core control hub of the equipment, is responsible for coordinating spindle rotation, tool feed, and the real-time movement trajectory of the workpiece within the sleeve. To ensure dimensional consistency in ultra-precision machining, the CNC system not only needs to implement high-precision interpolation algorithms and trajectory planning, but also must be able to effectively identify and suppress the dynamic coupling vibrations between the complex workpiece and the support structure during machining.

[0004] Traditional Swiss-type lathe control technology suffers from several drawbacks: Firstly, to achieve co-feeding, a micrometer-level assembly gap inevitably exists between the workpiece and the guide sleeve. This causes the workpiece to easily experience high-frequency whirling under cutting forces, manifesting as invisible periodic eccentric rotation. Traditional CNC systems rely solely on preset trajectories for open-loop control, failing to detect the actual dynamic positional shift of the workpiece. Secondly, existing technologies for suppressing this type of vibration are too simplistic, typically relying on conservative parameters such as increasing sleeve manufacturing precision or reducing cutting speed. This significantly increases hardware manufacturing costs and drastically sacrifices machining efficiency and process flexibility. Thirdly, while some improvement schemes attempt to introduce external vibration sensors, the signal extraction location is far from the cutting zone, resulting in effective features being submerged in multi-source environmental noise from the spindle and bed. This makes it difficult to accurately calculate the coupling modes of the workpiece and sleeve, leading to extremely limited compensation accuracy. Fourthly, the CNC system lacks real-time fusion and feedforward compensation capabilities for multi-source state data, resulting in regular vibration marks and dimensional deviations on the machined surface, making it difficult to meet the extreme surface roughness requirements of ultra-precision machining scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a control method and CNC system for Swiss-type lathes based on multi-source data, which can effectively solve the problems in the background technology mentioned above, namely, the difficulty in directly monitoring the periodic high-frequency eddying of the workpiece in the guide sleeve during the Swiss-type lathe machining process, the lack of dynamic sensing capability of traditional CNC systems, and the technical dilemma of insufficient compensation accuracy caused by severe noise interference of external vibration sensors.

[0006] To achieve the above objectives, this invention proposes a control method for a sliding head machine based on multi-source data, comprising the following steps:

[0007] S1. The electrical parameter signals of the main spindle motor and the micro-displacement signals of the back shaft are synchronously acquired through the analog-to-digital conversion module and the displacement sensing module.

[0008] S2. Perform frequency domain feature extraction and time-series preprocessing on the acquired signals to identify the feature vectors characterizing the vortex state of the workpiece.

[0009] S3. Input the feature vector into the dual-input filtering estimation model to solve the two-dimensional eccentricity vector of the workpiece front end relative to the ideal machining axis in real time;

[0010] S4. Based on the two-dimensional eccentric vector, a compensation command is generated through a feedforward compensation algorithm and superimposed on the trajectory planning path of the CNC system to drive the feed axis to perform eddy co-current cancellation motion.

[0011] Preferably, step S2 specifically includes the following steps:

[0012] S21. Perform a fast Fourier transform on the current signal of the spindle motor to extract the harmonic amplitude and phase information of a specific multiple of the spindle rotation frequency.

[0013] S22. Perform detrending and smoothing filtering on the axial micro-displacement data of the back shaft to extract the axial displacement time series features.

[0014] S23. Synchronize and align the harmonic features and axial displacement features according to the preset timestamps to construct a multi-source fusion feature matrix.

[0015] Preferably, step S3 specifically includes the following steps:

[0016] S31. Establish a dynamic coupling model of the workpiece and the guide sleeve, and transform it into a state-space equation, where the state variables include the eccentric displacement and eccentric velocity of the front end of the workpiece in the transverse and longitudinal directions.

[0017] S32. Using harmonic amplitude and phase information as the first observation input and axial displacement timing characteristics as the second observation input, configure the gain matrix of the dual-input Kalman filter.

[0018] S33. Minimize the estimated covariance through iterative calculation and output the optimal estimated two-dimensional eccentric vector.

[0019] Preferably, step S4 specifically includes the following steps:

[0020] S41. Construct a total system delay model that includes servo response lag and signal processing time, and determine the preset lead time constant;

[0021] S42. Calculate the half-cycle advance compensation amount based on the amplitude and phase of the two-dimensional eccentric vector and the advance time constant.

[0022] S43. Convert the half-cycle advance compensation amount into a coordinate correction value and add it to the feed axis command pulse of the current interpolation cycle in real time.

[0023] Preferably, the analog-to-digital conversion module is configured to digitally acquire the power supply circuit current of the spindle motor at a preset high-frequency sampling rate. The selection of the sampling rate must satisfy the sampling theorem to ensure that higher-order harmonic components can be captured. When the spindle motor drives the workpiece to rotate, if the workpiece experiences eccentric eddying within the guide sleeve, the periodic change in cutting force will be fed back to the motor's load torque, thereby causing a slight distortion in the current waveform. By performing high-precision quantization processing on the current signal, raw data support can be provided for subsequent extraction of electrical parameter features reflecting changes in workpiece pose. Anti-aliasing filtering technology is used during the analog-to-digital conversion process to eliminate high-frequency random noise generated by power supply fluctuations and driver carrier waves, ensuring signal purity.

[0024] Preferably, the displacement sensing module employs the eddy current sensing principle and is installed on the outside of the back shaft thrust bearing for non-contact detection of axial micro-displacement. When a Swiss-type lathe processes slender workpieces, the radial eddy current of the workpiece causes minute elastic deformation of the support structure. This deformation manifests as extremely small displacement fluctuations at the axial thrust bearing. The displacement sensing module has a preset high resolution, capable of sensing nanometer-level displacement changes. By monitoring the axial response of the back shaft, this invention can indirectly obtain the coupled vibration modes of the workpiece deep within the guide sleeve. This complements the current signal in a physical mechanism, effectively avoiding the susceptibility to interference inherent in single sensors.

[0025] Preferably, the Fast Fourier Transform employs a sliding window mechanism to perform real-time spectral analysis on the continuous current sampling sequence. By calculating the power spectral density, harmonic components with a fixed multiple relationship to the spindle rotation frequency are identified. Experiments have shown that the workpiece's eccentric eddying is highly correlated with harmonics at two to five times the spindle rotation frequency. The feature extraction logic not only calculates the energy distribution of the harmonics but also accurately extracts their phase deviation relative to the spindle zero-position pulse. This phase information is crucial for subsequently determining the specific spatial direction of the eccentric vector and is a logical prerequisite for achieving precise phase offset compensation.

[0026] Preferably, the dynamic coupling model characterizes the motion law of the workpiece under high-speed rotation and cutting force excitation, constrained by the nonlinear clearance of the guide sleeve. The model introduces equivalent stiffness and equivalent damping parameters, which are initialized based on the elastic modulus of the processed material and the preset clearance value of the sleeve.

[0027] The dynamic coupling model simplifies the workpiece to a concentrated mass point, and models the guide sleeve clearance as a piecewise linear spring-damper; the state-space equation is expressed as: ,in, The process noise vector (usually assumed to be Gaussian white noise, representing model error or unmodeled dynamic disturbances, with dimensions equal to...) same); The observed output vector (usually a sensor measurement, such as a current harmonic signal); The observation matrix (which maps state variables to measurements, its elements) (Obtained through experimental calibration, representing the sensitivity of each state variable to the output). The measurement noise vector is typically assumed to be Gaussian white noise, representing sensor error.

[0028] State vector , representing lateral eccentric displacement and velocity, and longitudinal eccentric displacement and velocity, respectively. System matrix From equivalent stiffness and damping Confirmed, in the following specific form:

[0029]

[0030] in For the workpiece mass, the equivalent stiffness is determined by the material's elastic modulus. Single-sided gap of the sleeve and workpiece overhang length Decide:

[0031]

[0032] Observation matrix It is obtained by calibrating the sensitivity of current harmonics and displacement signals to various state variables, and its general form is as follows:

[0033]

[0034] If the output is a scalar, then It is a row vector; if it is a multi-output, it is a matrix of the corresponding dimension, the elements of which are obtained by calibration.

[0035] During the operation of the dual-input Kalman filter, the system dynamically adjusts the estimated weights based on the observed residuals.

[0036] The dual-input Kalman filter is configured with two independent observation channels: the first channel observation vector. Includes harmonic amplitude and phase, second channel It includes axial displacement and its first-order difference; where For harmonic amplitude, For phase, This is axial displacement. This represents the first-order difference of the axial displacement. Real-time calculation of the residual covariance for each channel. Define the signal-to-noise ratio metric That is, the signal-to-noise ratio of each channel is proportional to the reciprocal of the residual covariance trace. The observation noise covariance matrix is ​​dynamically adjusted. ,in The preset baseline noise level is used. Based on the updated observation noise covariance matrix... Recalculate the Kalman gain:

[0037]

[0038] in The covariance matrix is ​​estimated for the state. The observation matrix (composed of the observation matrices of two channels) is weighted adaptively.

[0039] When the signal-to-noise ratio of the current signal decreases due to heavy-load cutting, the filter automatically increases the weight of the axial displacement observation; conversely, when the displacement signal is weak during the finishing stage, it mainly relies on current harmonics for state reconstruction. This adaptive fusion mechanism ensures that the calculation accuracy of the eccentric vector remains within a preset robust range throughout the entire machining process.

[0040] Preferably, the feedforward compensation algorithm employs predictive control principles. Since there is a physical response time delay between the servo system receiving a command and the motor executing the action, directly using the current eccentricity vector for correction would cause the compensation action to lag behind the actual whirl, and might even exacerbate the vibration. This invention calculates the estimated position of the workpiece whirl trajectory at a specific future time using a total system delay model. The half-cycle advance compensation utilizes the periodicity of whirl, and by pre-setting an angle in the phase, ensures that the tool's corrective motion precisely offsets the workpiece's eccentricity. This active phase-counterbalancing logic significantly improves the relative stability of the contact point between the tool tip and the workpiece.

[0041] On the other hand, the present invention also provides a CNC system for Swiss-type lathes based on multi-source data, for implementing the above-mentioned method, including:

[0042] The data acquisition subsystem includes a high-speed analog-to-digital conversion unit and a high-precision displacement detection unit, which are used to acquire the electrical signal flow of the motor and the displacement timing of the mechanical structure, respectively.

[0043] The feature reconstruction subsystem is connected to the data acquisition subsystem and is equipped with signal processing logic to implement fast Fourier transform and multi-source feature alignment.

[0044] The state estimation kernel, based on a preset Kalman filter algorithm, receives the output of the feature reconstruction subsystem and is used to estimate the spatial eccentricity vector of the workpiece whirl in real time.

[0045] The trajectory controller integrates a feedforward compensation module and a path interpolator to convert the eccentric vector into an axial compensation pulse.

[0046] The drive control module, connected to the trajectory controller, is used to drive the actuator to complete the final eddy suppression action.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention constructs a non-invasive monitoring mechanism within the guide sleeve, utilizing multi-source fusion data of motor current harmonics and back shaft micro-displacement to achieve online reconstruction of the workpiece whirl state within the guide sleeve. First, this non-invasive detection method is fully compatible with existing Swiss-type lathe structures, eliminating the need for complex hardware modifications to the high-precision guide sleeve, significantly reducing the application threshold and maintenance costs. Second, a dual-input Kalman filter algorithm achieves deep fusion of electrical and mechanical signals, effectively separating environmental noise such as spindle rotation and bed vibration, resulting in workpiece whirl vector identification accuracy reaching a preset micron-level standard. In summary, this invention solves the quantitative challenge of invisible vibrations in Swiss-type lathe machining from a sensory perspective, providing a data foundation for subsequent active suppression.

[0049] Furthermore, this invention achieves active intervention in eddy current through a feedforward compensation algorithm, rather than traditional passive damping suppression. Utilizing the half-cycle lead compensation amount generated by the system's total delay model, the turret can synchronously track and cancel out eddy currents based on the workpiece's real-time eccentricity, fundamentally altering the relative motion trajectory between the tool tip and the workpiece. Experimental data shows that after applying the compensation logic of this system, the arithmetic mean roughness of the machined surface can be stably controlled within a preset extremely low range, significantly eliminating regular vibration marks and greatly improving the machining quality and dimensional consistency of slender shaft parts.

[0050] Furthermore, by extracting specific multiples of frequency harmonics through Fast Fourier Transform, electromagnetic load fluctuations caused by mechanical eccentricity can be accurately captured. This sensing method based on internal electrical parameters has extremely high dynamic response speed, can reflect microscopic changes in the cutting zone in real time, and enhances the system's adaptability to complex cutting conditions.

[0051] Furthermore, the introduction of an eddy current displacement sensor to monitor the axial micro-displacement of the back shaft provides another independent physical dimension for eddy current identification. The axial displacement signal has extremely high sensitivity to the modal transformation of the workpiece within the sleeve, and its fusion with the current signal effectively improves the fault tolerance of the state estimation kernel. Even when a single signal source is subjected to local electromagnetic interference or mechanical impact, the continuity and accuracy of the eccentric vector solution can still be maintained.

[0052] Furthermore, the real-time superposition of compensation commands and original trajectory commands in the trajectory controller employs a smoothing algorithm to ensure that the servo motor does not generate additional mechanical shock or overshoot when performing high-frequency correction actions. This deep integration at the interpolation level ensures a dynamic balance between high-precision feed and high-frequency compensation in the CNC system, achieving improved machining accuracy without sacrificing machining efficiency.

[0053] Furthermore, this system has been optimized for ultra-precision machining scenarios such as medical guidewires and optical shafts. By improving process capability indicators, the yield rate of machined parts has been significantly increased. This data-driven intelligent control method eliminates reliance on operator experience and provides core technical support for the intelligent upgrade of Swiss-type CNC systems.

[0054] Furthermore, by establishing a dynamic coupling model between the workpiece and the guide sleeve, the system can predict the vibration tendency of workpieces with different aspect ratios and adaptively adjust the filter parameters accordingly. This model-based design approach enhances the system's versatility and flexibility in processing parts of different specifications and shortens the debugging cycle for new processes.

[0055] Furthermore, the analog-to-digital conversion module employs anti-aliasing filtering and quantization noise suppression techniques, ensuring that even in the complex electromagnetic environment of a workshop, subtle harmonic components of the current signal can still be accurately reconstructed. This provides high-quality input to the state estimation kernel, guaranteeing consistent eddy current suppression across the entire speed range.

[0056] Furthermore, by employing an active compensation strategy, this invention allows users to achieve higher surface quality without reducing cutting parameters. This means that while ensuring machining accuracy, production efficiency is significantly improved, resolving the long-standing contradiction between quality and efficiency in traditional technologies. In summary, this invention, through real-time fusion and feedforward compensation of multi-source data, achieves precise sensing and efficient suppression of whirl in Swiss-type lathe machining, significantly improving the overall performance of precision manufacturing. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the overall technical solution architecture of a Swiss-type lathe CNC system based on multi-source data proposed in this invention;

[0058] Figure 2 This is a schematic diagram of the core principle framework of the workpiece eccentricity vector solution and feedforward compensation based on multi-source data fusion in this invention; Detailed Implementation

[0059] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0061] Example 1

[0062] This embodiment uses the processing of a stainless steel medical guidewire with a diameter of 1 mm and a length of 50 mm as a specific application scenario. When machining such slender parts on a Swiss-type lathe, the stability of the workpiece within the guide sleeve directly determines the surface roughness and dimensional accuracy of the final product.

[0063] In the specific hardware implementation, the Swiss-type lathe CNC system based on multi-source data provided by this invention first connects a high-precision current sensor in series in the power supply circuit of the spindle motor, and connects its output to a high-speed analog-to-digital converter (ADC). The ADC is configured to digitally acquire the three-phase current signal of the spindle motor at a high-frequency sampling rate of 10kHz. This high-frequency sampling not only meets basic control requirements, but more importantly, it can completely preserve the weak high-order harmonic components caused by load fluctuations during spindle rotation. When the spindle motor drives the workpiece to rotate, if the front end of the workpiece generates whirl due to the gap in the guide sleeve, its cutting force will produce periodic and severe fluctuations. These fluctuations will be transmitted to the spindle through the mechanical structure, thereby causing a shift in the frequency and amplitude of the motor stator current. The ADC module integrates a multi-stage anti-aliasing filter, which can effectively filter out the electromagnetic noise generated by the inverter's switching frequency.

[0064] Meanwhile, a displacement sensing module based on the eddy current sensing principle is mounted on the outer housing of the back shaft thrust bearing via a precision bracket. This sensor remains in non-contact with the back shaft structure, with its sensing end face directly opposite the axial vibration monitoring point of the bearing housing. Due to the specific stiffness characteristics of the Swiss-type lathe's load-bearing chain, the radial eddy current of the workpiece is inevitably accompanied by minute elastic deformation of the supporting structure. This deformation generates nanometer-level displacement pulsations in the axial dimension. The eddy current sensor selected in this embodiment has a static resolution better than 10 nanometers, enabling it to detect microscopic dynamic responses imperceptible to the machine tool bed.

[0065] The method for suppressing eddy cog in a Swiss-type machine implemented in this embodiment includes the following detailed steps:

[0066] In step S1, after the system starts, the synchronous clock of the CNC system triggers the analog-to-digital converter module and the displacement sensing module. Through a dedicated hardware timing task, the current data stream of the spindle motor and the micro-displacement data stream of the back shaft are read into the system memory in real time. In order to ensure the phase consistency of the two signals with different physical properties, the system adopts a unified sampling trigger signal to ensure that the electrical parameters and mechanical displacement parameters are aligned at the microsecond level on the time axis.

[0067] In step S2, deep frequency domain feature extraction is performed on the acquired signal. Step S21 uses a Fast Fourier Transform operator to perform real-time spectrum analysis on the current signal with a sliding window of a preset length. The system focuses on scanning harmonics within the range of 2 to 5 times the spindle rotation frequency. This is because dynamic analysis revealed that the eddy trajectory within the guide sleeve typically exhibits asymmetrical elliptical characteristics, with its energy concentrated at specific harmonic points. The system not only records the amplitude of these harmonics but also accurately extracts their phase angle information relative to the spindle zero-position encoder pulse. Step S22 then performs detrending processing on the axial micro-displacement data of the back shaft, eliminating slow drift caused by thermal expansion or static load, retaining only the oscillating components reflecting eddy. Subsequently, a low-pass filtering algorithm is used to remove irrelevant high-frequency mechanical impacts. Step S23 encapsulates the extracted electrical parameter harmonic features and mechanical displacement features into a matrix according to the sampling time sequence, thereby constructing a multi-source fusion feature matrix that can comprehensively characterize the eddy state of the workpiece.

[0068] In step S3, the feature matrix is ​​submitted as the observation input to the built-in dual-input filter estimation model. This model is based on the dynamic coupling relationship between the workpiece and the guide sleeve. In step S31, the system initializes a set of state-space variables based on the elastic modulus of the processed material, the workpiece diameter, and the sleeve's fit clearance. These variables include the eccentricity of the workpiece front end in the lateral direction, the eccentricity in the longitudinal direction, and the corresponding motion velocity. Step S32 configures the core parameters of the dual-input Kalman filter. The harmonic information of the current signal serves as the first observation input reflecting the load change in the cutting zone, and the axial displacement information serves as the second observation input reflecting the structural response. The system automatically monitors the signal-to-noise ratio of the two signals. When the cutting load is large, causing severe current fluctuations, the system dynamically reduces the weight of the current observation and increases the confidence level of the displacement signal. In step S33, through recursive prediction and correction logic, the covariance of the estimation error is continuously reduced, ultimately outputting a two-dimensional eccentricity vector reflecting the position of the workpiece front end relative to the ideal geometric center. This vector points in real time to the actual eccentricity direction and offset of the workpiece at the current moment.

[0069] In step S4, the system converts the calculated two-dimensional eccentricity vector into control commands. First, in step S41, a total system delay model is constructed. The total system delay model is the sum of the delays of each component: ,in: (Calibrated by step response test) (Signal processing algorithm has a fixed processing time) (Current interpolation period). Eddy frequency. Phase of eccentric vector Real-time estimation of rate of change: ,in, This is the rate of change of the phase angle with respect to time, i.e., the instantaneous angular velocity (unit: rad / s). Divide by Then converted to frequency The lead time constant is calculated as follows: ,in, This is the lead time constant (in seconds). It determines how much earlier the compensation action should be issued before the current moment to offset the phase lag caused by the total system delay. Subtracting this total delay from half a eddy cycle gives the amount of time required to issue the command in advance. This design ensures that the phase of the compensation action leads the eddy motion by half a cycle.

[0070] In step S42, based on the periodic rotational characteristics of the whirl, the real-time phase of the two-dimensional eccentric vector is used, combined with the lead time constant, to calculate the estimated position after half a cycle through predictive logic. The purpose of this feedforward logic is to allow the tool tip correction action to be "in place" in advance. Step S43 converts the calculated half-cycle lead compensation amount into coordinate correction values ​​for the X and Z axes, and directly superimposes them into the original feed path of the current interpolation cycle. Finally, the drive control module commands the feed axis motor to perform a canceling motion opposite to the whirl direction and corresponding in amplitude.

[0071] In this embodiment, using the above method, the arithmetic mean surface roughness of the processed medical guidewire was significantly reduced from 1.2 micrometers to 0.35 micrometers. The previously clearly visible periodic ripples completely disappeared, and the dimensional consistency of the workpiece was significantly improved. This demonstrates that the system achieves precise perception and active suppression of workpiece eddying through data-driven means without altering the mechanical structure.

[0072] Example 2

[0073] This embodiment describes the machining scenario of slender titanium alloy shafts for aerospace precision. Titanium alloys have high strength and low thermal conductivity, resulting in cutting forces that are much greater than those of stainless steel during machining, and are accompanied by more complex dynamic mechanical characteristics.

[0074] Unlike Example 1, in this example, the feature extraction logic is specifically optimized for the high hardness characteristics of titanium alloys. During the Fast Fourier Transform (FFT) in step S21, the system expands the scanning range to a 10-fold harmonic range of the spindle frequency to capture high-order harmonic fluctuations caused by intermittent cutting or tool sticking. Because the spindle load is extremely high during titanium alloy machining, the background noise of the current signal increases significantly. Therefore, in step S22, a more refined wavelet denoising algorithm is used to process the back-axis displacement signal, aiming to extract weak eddy mode information from strong environmental vibrations.

[0075] In the state-space equation construction of step S3, a dynamic stiffness compensation factor is introduced because the stiffness and damping characteristics of titanium alloy parts differ from those of general metals. During the operation of the Kalman filter, the update logic of the prediction covariance matrix is ​​set to a nonlinear adaptive mode. When a sudden change in the cutting speed is detected within a preset range, the system quickly adjusts the gain matrix to prevent estimation distortion caused by the switching of cutting states.

[0076] In the feedforward compensation stage of step S4, due to the extremely high precision requirements of the toolpath in titanium alloy machining, a smoother second-derivative continuous control strategy is adopted for the superposition of compensation commands. This means that when the compensation amount is superimposed on the feed axis pulse, not only is the position offset taken into account, but also the sudden acceleration caused by the compensation action is strictly limited, thereby avoiding the tool-biting phenomenon that may occur in the machining of hard materials.

[0077] Experimental data shows that when processing titanium alloy workpieces with an aspect ratio of 40:1, this system, through deep integration of current harmonics and micro-displacements, can accurately predict sudden eddy current changes caused by local material inhomogeneities. Compared with traditional CNC systems that do not have eddy current compensation capabilities, the system used in this embodiment increases the yield rate of parts from 80% to over 95%.

[0078] Example 3

[0079] This embodiment describes an extremely fine shaft with a diameter of less than 0.5 mm used in optical instruments. The physical properties of such parts are highly susceptible to the influence of minute pressure on the guide sleeve, and they exhibit extremely high vortex frequencies, which are often difficult for conventional sensors to detect.

[0080] In the hardware configuration of this embodiment, the installation position of the eddy current sensor is optimized to the back shaft support point closest to the guide sleeve outlet, in order to maximize the acquisition of the vibration projection of the workpiece front end. Because the workpiece diameter is extremely small, the change in the spindle motor current is extremely small, almost at the edge of the sensitivity of conventional sensors. Therefore, in step S1, a programmable gain amplifier circuit is added to the front end of the analog-to-digital conversion module, specifically for linearly amplifying the AC micro-motion component in the spindle motor current by a specific factor, thereby improving the quantization resolution of the effective signal.

[0081] In the feature alignment logic of step S2, considering the phase lag effect of the ultra-thin shaft at high speeds, the system introduces a phase correction value based on the rotational speed function into the multi-source fusion feature matrix using preset experimental benchmark data. This means that when constructing the feature vector, the synchronization of electrical parameters and displacement parameters is no longer a simple physical time alignment, but includes dynamic compensation for the torsional deformation of the workpiece.

[0082] In the state estimation kernel of step S3, considering that the dynamic behavior of the ultrathin shaft is closer to that of a flexible beam, the system expands the state space variables into a composite vector containing multiple vibration modes. The dual-input Kalman filter then acts as a multi-objective observer, not only estimating the eccentricity vector but also monitoring in real time whether the workpiece has a tendency to buckle or become unstable.

[0083] In the feedforward compensation step S4, since the machining of extremely fine shafts is usually accompanied by extremely high revolutions per minute, the lead time constant in the total system delay model is refined to the microsecond level and linearly interpolated to adjust according to the real-time fluctuations of the spindle speed. The execution of the compensation command is directly driven by a high-priority interrupt of the CNC system, ensuring that the total physical lag from signal sensing to action execution is strictly controlled within a preset safety range.

[0084] After applying the technical solution of this embodiment, the cylindricity error of the optical micro-shaft is greatly improved. The fine spiral marks in the middle of the workpiece caused by high-frequency eddying are completely eliminated.

[0085] Example 4

[0086] This embodiment further discloses the specific software module organization structure and logic flow details of the numerical control system in this invention.

[0087] As the perception layer of the entire solution, the data acquisition subsystem's core logic lies in maintaining high-frequency and deterministic data throughput. The high-speed analog-to-digital converter (ADC) uses direct memory access (DMI) technology to transmit the acquired current sequence in blocks to the system cache. Simultaneously, the displacement detection unit uses a high-speed bus interface to convert displacement pulse counts into floating-point sequences with fully synchronized timing. The system's internal clock manager marks both data streams with hardware timestamps to prevent data stream misalignment in a multi-tasking environment.

[0088] The feature reconstruction subsystem, serving as a preprocessing layer, integrates a dedicated digital signal processing instruction set. During Fast Fourier Transform (FFT) execution, the system employs a circular buffer management mode, ensuring that each frame's spectral analysis is completed within a preset numerically controlled cycle. The multi-source feature alignment logic reconstructs the sampling points by comparing the timestamps of two signals and using linear interpolation, thereby ensuring that the data input to the filtering model has strict instantaneous correlation in a physical sense.

[0089] The state estimation kernel is the computational hub of the entire system. Its internal Kalman filter algorithm is broken down into several purely textual logical steps, including prediction, gain calculation, state update, and covariance update. Within each control cycle, the kernel first predicts the current position of the geocentric point based on the state variables and dynamic equations from the previous time step. Then, it reads the observed values ​​output by the feature reconstruction subsystem and calculates the residual between the observed and predicted values. Next, based on preset measurement noise levels and system evolution noise levels, it determines the gain level through a series of matrix element-level operations. Finally, it uses the gain to correct the predicted position and synchronously updates the error covariance.

[0090] The trajectory controller is responsible for converting the calculated eccentricity into actual machine tool motion. Its internal feedforward compensation module monitors the curvature changes of the machining trajectory in real time. When in a straight cutting segment, the superposition weight of the compensation amount is at a standard level; when in a circular transition segment, the system dynamically increases the compensation amount according to the change in centripetal force. When generating the pulse sequence corresponding to each line of G-code, the path interpolator incorporates the coordinate correction value into the differential operation in an incremental form, thereby ensuring the smoothness of the actuator's motion.

[0091] The drive control module, acting as the execution layer, is responsible for converting the logic commands from the trajectory controller into electrical drive signals for the servo motor. This module features ultra-high-speed response characteristics, with the bandwidths of its current and speed loops configured to match the workpiece whirling frequency, thereby ensuring the precise execution of high-frequency trajectory correction actions.

[0092] In summary, this embodiment demonstrates how the various modules of the system, through rigorous logical collaboration, transform the intangible workpiece vortex state into controllable machine tool compensation actions. This data-driven control architecture, supported by a physical model, endows the Swiss-type lathe CNC system with unprecedented dynamic accuracy assurance capabilities.

[0093] Example 5

[0094] This embodiment details the robustness of the system under unsteady conditions. In actual machining processes, tool wear, coolant flow fluctuations, and uneven material hardness can all cause random disturbances to the eddy frequency and amplitude.

[0095] An anomaly detection mechanism has been added to the state estimation kernel in step S3 of this system. When the observation residual of the dual-input Kalman filter exceeds a preset tolerance threshold for multiple consecutive cycles, the system determines that it is currently in a disturbance environment or a period of process abrupt change. At this time, the system will not blindly perform compensation based on erroneous estimation, but will automatically switch to a conservative mode based on energy suppression. In this mode, the system will increase the damping weight of the filter and reduce the proportional gain of the feedforward compensation until the observation residual returns to a stable range.

[0096] Meanwhile, to address potential grid harmonic interference in the spindle motor current signal, a reference channel is added to the feature extraction logic in step S2. The system monitors bus voltage fluctuations and removes them as background noise from the spindle current signal. Through this purely textual signal cancellation logic, the system can achieve stable operation in harsh electromagnetic environments at the factory site, ensuring that the extracted 2nd to 5th harmonics are entirely induced by the eddy currents of the mechanical structure.

[0097] In the feedforward compensation logic of step S4, the system also integrates a predictive verification algorithm. Before being sent to the drive control module, the compensation command generated at each moment is compared in memory with the actual motion feedback from the previous cycle. If the estimated compensation action causes the feed axis's following error to exceed the machine tool's physical limits, the system automatically limits the compensation amount. This multi-level safety protection logic ensures that the system always operates within the machine tool hardware's carrying capacity during high-frequency dynamic compensation.

[0098] Furthermore, this invention provides a self-learning mechanism. During the initial processing of each new workpiece, the system automatically records the whirl characteristic parameters of the workpiece at different rotational speeds and feed rates. These parameters are stored in the system's process database. In subsequent processing of similar parts, the state estimation kernel directly loads this optimized prior knowledge, thereby achieving high-precision initial compensation at the moment processing begins. This intelligent approach based on data accumulation significantly shortens the trial production cycle for high-precision slender shaft parts.

[0099] The logic of this embodiment demonstrates that the system not only possesses extremely high control precision but also demonstrates stability and intelligence in handling complex industrial environments. Through deep fusion and adaptive adjustment of multi-source data, this invention successfully solves the long-standing problem of dynamic error compensation in the ultra-precision machining field of Swiss-type lathes.

[0100] Through the detailed descriptions of the above embodiments, it can be seen that the present invention achieves closed-loop control of the workpiece vortex state within the guide sleeve of a Swiss-type lathe by combining a hardware combination of an analog-to-digital conversion module and a displacement sensing module, along with the methodological logic of feature extraction, Kalman filtering estimation, and feedforward compensation. This scheme abandons the traditional formulaic fixed compensation and instead adopts a dynamically sensing, adaptive data-driven strategy. Throughout the entire implementation process, it does not rely on intrusive sensor installation, ensuring the integrity of the original machine tool structure and machining efficiency, and providing a practical and feasible path to improve the machining quality of slender shaft parts.

[0101] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A control method for a sliding head machine based on multi-source data, characterized in that, Includes the following steps: Step S1: The power supply circuit current signal of the spindle motor reflecting the load fluctuation of the motor is acquired by the analog-to-digital conversion module at a preset high-frequency sampling rate. The axial micro-displacement signal of the back shaft structure is acquired non-contactly and synchronously by the displacement sensing module based on the eddy current sensing principle. The two signals are aligned in phase on the time axis through a unified hardware sampling trigger signal. Step S2: Perform real-time spectrum analysis on the current signal. Within a sliding window of a preset length, extract the harmonic amplitude information of a specific multiple of the spindle rotation frequency and the phase deviation information of the harmonic relative to the spindle encoder pulse through fast Fourier transform. Perform detrending processing and smoothing filtering on the axial micro-displacement signal. After hardware timestamp matching, encapsulate the harmonic features and axial displacement features into a multi-source fusion feature matrix that characterizes the whirl state of the workpiece. Step S3: Construct a dynamic coupling model of the workpiece and the guide sleeve, and transform the dynamic coupling model into a state-space equation that includes the eccentric displacement and eccentric velocity of the workpiece front end in the transverse and longitudinal directions. Use the multi-source fusion feature matrix as the observation input, and dynamically adjust the gain matrix weights of the current observation and displacement observation according to the signal-to-noise ratio through a dual-input Kalman filter. Use recursive prediction and correction logic to minimize the covariance of the estimation error, and calculate the two-dimensional eccentric vector of the workpiece front end relative to the ideal machining axis. The operation of dynamically adjusting the gain matrix weights of the current observation and displacement observation according to the signal-to-noise ratio through a dual-input Kalman filter includes: real-time monitoring of the residual fluctuation intensity of the current signal and axial displacement signal. When the change of cutting load causes the background noise of the current signal to increase, the trust weight of the first observation input is reduced and the trust weight of the second observation input based on the mechanical structure response is increased accordingly to achieve adaptive reconstruction of the workpiece whirl state, ensuring that the calculation accuracy of the eccentric vector is maintained at a preset robust level throughout the entire machining process. Step S4: Construct a system total delay model including servo response, signal processing, and path interpolation delay. Based on the real-time phase and amplitude of the two-dimensional eccentric vector, combined with the lead time constant determined by the system total delay model, calculate the half-cycle lead compensation amount to offset the whirl trajectory through phase prediction logic, convert it into coordinate correction values, and superimpose them in real time into the feed axis command pulse sequence of the current interpolation cycle to drive the feed axis to offset workpiece whirl. The operation of generating compensation commands and superimposing them includes: real-time monitoring of the curvature change of the machining trajectory, dynamically adjusting the superposition gain of the compensation amount between the straight cutting segment and the arc transition segment, and using a second-derivative continuous control strategy to smooth the coordinate correction values ​​to limit the sudden acceleration of the feed axis caused by high-frequency compensation actions, and ensure the smoothness of the actuator's motion when implementing whirl suppression.

2. The method for controlling a sliding head machine based on multi-source data according to claim 1, characterized in that, The operation of extracting harmonic amplitude information of a specific multiple of the spindle rotation frequency in step S2 includes: storing a continuous current sampling sequence using a circular buffer management mode, determining the electromagnetic load fluctuation characteristics induced by mechanical eccentricity by locking the energy center in the power spectral density within the preset multiple of the spindle rotation frequency, and accurately calculating the spatial direction of the energy center relative to the initial phase of the spindle.

3. The method for controlling a sliding head machine based on multi-source data according to claim 1, characterized in that, The method also includes anomaly detection and robustness control steps: when the observation residual of the dual-input Kalman filter exceeds the preset allowable threshold for multiple consecutive control cycles, it is determined that the current environment is in interference or a period of process change. The system automatically switches to energy suppression mode, and increases the damping weight of the filter and reduces the proportional gain of the feedforward compensation until the observation residual returns to the preset stable range.

4. A CNC system for a Swiss-type lathe based on multi-source data, used to implement the method according to any one of claims 1 to 3, characterized in that, The system includes: The data acquisition subsystem integrates a high-speed analog-to-digital converter and a high-precision displacement detection unit. The high-speed analog-to-digital converter is connected to the power supply circuit of the spindle motor through an anti-aliasing filter circuit to acquire a current signal with high dynamic response. The high-precision displacement detection unit is connected to an eddy current sensor installed on the back shaft support structure to acquire the axial displacement timing of the mechanical structure. The feature reconstruction subsystem is connected to the data acquisition subsystem and is configured with dedicated signal processing logic. The dedicated signal processing logic extracts the current harmonic features of a specific frequency by performing a fast Fourier transform and reconstructs the displacement features using a linear interpolation method to ensure that the multi-source features have instantaneous correlation in a physical sense. The state estimation kernel is connected to the feature reconstruction subsystem. It runs an adaptive Kalman filter algorithm based on the dynamic coupling relationship between the workpiece and the guide sleeve. This algorithm is used to receive the multi-source features and estimate the spatial eccentricity vector of the workpiece in real time during the rotation process. The trajectory controller is connected to the state estimation kernel and integrates a feedforward compensation module, a system total delay model and a path interpolator. The feedforward compensation module generates coordinate correction pulses based on the spatial eccentricity vector and the lead time constant. The path interpolator incorporates the coordinate correction pulses into the differential operation of the original machining trajectory in an incremental form. The drive control module, connected to the trajectory controller, is used to convert the superimposed and compensated logic instructions into drive electrical signals for the servo motor, thereby directing the actuator to complete the eddy current suppression action.

5. The Swiss-type lathe CNC system based on multi-source data according to claim 4, characterized in that, The high-speed analog-to-digital conversion unit uses direct memory access technology to transmit the acquired current sequence in blocks to the system cache. The front end of the high-speed analog-to-digital conversion unit is equipped with a programmable gain amplifier circuit, which is used to linearly amplify the AC micro-motion component in the spindle motor current by a specific factor, so as to improve the sensing resolution of the vortex state of small diameter workpieces.

6. The Swiss-type lathe CNC system based on multi-source data according to claim 4, characterized in that, The system also includes a process database and a self-learning module. The self-learning module is configured to automatically record eddy characteristic parameters at different rotational speeds and feed rates during the initial processing of the workpiece, and store the characteristic parameters as prior knowledge in the process database. When processing similar parts in subsequent processing, the state estimation kernel directly loads the corresponding characteristic parameters to optimize the initialization state of the Kalman filter.

7. The Swiss-type lathe CNC system based on multi-source data according to claim 4, characterized in that, The feature reconstruction subsystem also includes reference channel monitoring logic, which is used to acquire bus voltage fluctuation data in real time and remove the fluctuation data as background noise from the spindle current signal to ensure that the extracted harmonic components are entirely derived from the mechanical interaction between the workpiece and the guide sleeve.

8. The Swiss-type lathe CNC system based on multi-source data according to claim 4, characterized in that, The feedforward compensation module in the trajectory controller also integrates prediction verification logic. Before the compensation command generated in each control cycle is sent to the drive control module, the prediction verification logic compares the estimated compensation action with the actual motion feedback of the previous cycle. If the feed axis following error caused by the estimated compensation action exceeds the preset machine tool physical limit, the compensation amount is limited.

Images