Precision turning precision error compensation control system for rolling body

Abstract

The invention relates to the technical field of precision machining control, and discloses a rolling body precision turning precision error compensation control system, which comprises a multi-source heterogeneous sensing module used for collecting various signals; the dynamic stiffness modeling module is used for constructing a circumferential dynamic stiffness distribution model distributed in the circumferential direction of the workpiece; the angle domain phase coupling module is used for outputting a comprehensive feature vector containing the rigidity modulation deformation and the equivalent main shaft geometric error; the error prediction module is used for predicting and outputting a roundness error prediction value of the next angle step length; and the compensation execution module is used for dynamically superposing the position compensation instruction into an original interpolation instruction of the numerical control system and driving a servo shaft to execute displacement compensation. According to the method, by constructing the circumferential dynamic stiffness model and applying angle domain phase coupling, precise calculation of sensor data space-time alignment and cutting point elastic deformation is achieved, and the real-time prediction precision and the compensation effect of the rolling body turning roundness error are improved.

Description

Technical Field

[0001] This invention relates to the field of precision machining control technology, specifically to a rolling element precision turning accuracy error compensation control system. Background Technology

[0002] As the core load-bearing component of precision bearings, the roundness accuracy of rolling elements directly affects the overall rotational accuracy, vibration and noise levels, and fatigue life of the bearing. In the hard turning process for large-sized rolling elements (such as rollers in wind turbine bearings and rollers in high-speed rail axle box bearings), ensuring the micron-level roundness accuracy of the finished product is a critical and extremely challenging process requirement.

[0003] Traditional turning error compensation techniques primarily rely on offline detection or open-loop compensation based on static models. These methods typically assume the workpiece is a rigid body during machining or only consider the geometric errors of the machine tool itself, neglecting the complex dynamic physical interactions during cutting. However, in actual precision hard turning, the workpiece is clamped by a powerful three-jaw chuck, resulting in significant non-uniformity in its radial stiffness distribution along the circumference. Specifically, the stiffness is high in the contact area between the jaws, while the stiffness is low in the suspended area between the jaws. When the cutting tool cuts along the circumference, a constant cutting force acts on the workpiece surface with constantly changing stiffness, inevitably leading to periodically fluctuating elastic yield deformation. This replication error caused by the "stiffness-force" coupling effect is one of the important reasons for the roundness deviation of rolling elements.

[0004] Existing error compensation schemes have failed to effectively address the aforementioned time-varying stiffness problem. While some schemes incorporate cutting force monitoring, they often simply use the force signal as a threshold for monitoring or employ fixed stiffness coefficients for linear conversion, failing to reflect the true deformation characteristics of the workpiece at different rotation angles. Furthermore, existing multi-sensor monitoring systems typically acquire data independently, failing to adequately address the phase difference between different sensors (such as the displacement sensor mounted on the spindle end and the force gauge mounted on the base) in their spatial physical positions. This spatiotemporal asynchrony leads to a phase misalignment between the acquired error signal and the actual physical phenomena occurring at the cutting point, resulting in delayed compensation or even reverse overcompensation, making it difficult to meet the stringent sub-micron precision requirements of high-end rolling element manufacturing. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a rolling element precision turning accuracy error compensation control system, which solves the problems of existing rolling element turning error compensation technologies failing to consider the non-uniform distribution characteristics of circumferential stiffness of the workpiece caused by clamping, and the phase asynchrony of multi-source sensors in physical space, which leads to distortion of elastic deformation characterization and compensation lag at the cutting point.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a rolling element precision turning accuracy error compensation control system, the system comprising:

[0008] A multi-source heterogeneous sensing module is used to acquire spindle radial runout signals, three-dimensional cutting force signals, workpiece clamping contour signals, and spindle rotary encoder signals in real time.

[0009] The dynamic stiffness modeling module is used to construct a circumferential dynamic stiffness distribution model distributed along the circumference of the workpiece based on the workpiece clamping contour signal.

[0010] The angle domain phase coupling module is used to map the spindle radial runout signal and the three-dimensional cutting force signal to the workpiece rotation angle domain using the spindle rotary encoder signal, calculate the ratio of the angle domain cutting force to the corresponding phase stiffness value to obtain the stiffness modulation deformation, and output a comprehensive feature vector containing the stiffness modulation deformation and the equivalent spindle geometric error.

[0011] The error prediction module is used to input the comprehensive feature vector into a pre-trained long short-term memory network model, predict and output the roundness error prediction value for the next angle step.

[0012] The compensation execution module is used to generate a position compensation command based on the predicted roundness error value, and dynamically superimpose the position compensation command into the original interpolation command of the CNC system to drive the servo axis to perform displacement compensation.

[0013] Preferably, the multi-source heterogeneous sensing module includes:

[0014] The orthogonal displacement sensing unit includes two sets of capacitive displacement sensors orthogonally arranged on the end face of the spindle, used to collect the radial runout signals of the spindle in the X and Y directions;

[0015] The cutting force sensing unit includes a strain gauge force gauge integrated into the lathe fixture base, used to acquire the three-axis cutting force signals in the X-axis, Y-axis and Z-axis directions;

[0016] The clamping contour scanning unit includes a laser displacement sensor mounted perpendicular to the workpiece clamping surface, used to scan the workpiece surface to obtain the workpiece clamping contour signal while the spindle is rotating at low speed.

[0017] Preferably, the orthogonal displacement sensing unit uses the cutting point position where the cutting tool contacts the workpiece as the phase zero reference, the two sets of capacitive displacement sensors have a first installation phase offset angle relative to the phase zero reference, and the laser displacement sensor has a second installation phase offset angle relative to the phase zero reference.

[0018] Preferably, the dynamic stiffness modeling module includes:

[0019] The data preprocessing unit is used to perform noise reduction and reference circle fitting on the workpiece clamping contour signal and calculate the radial deviation relative to the reference circle.

[0020] The stiffness mapping generation unit is used to calculate the stiffness value of the workpiece at different rotation angles based on the nominal radial stiffness of the workpiece material and the radial deviation, and to generate the circumferential dynamic stiffness distribution model.

[0021] Preferably, the stiffness mapping generation unit calculates the stiffness value of the workpiece at different rotation angles by: mapping the ratio of the radial deviation to the normalization factor using the hyperbolic tangent function, multiplying the mapping result by the stiffness fluctuation coefficient and adding it to the unit value, and multiplying the sum by the nominal radial stiffness to obtain the stiffness value at the corresponding angle.

[0022] Preferably, the angle domain phase coupling module includes:

[0023] An angle conversion unit is used to receive the instantaneous angular velocity fed back by the spindle rotary encoder signal and calculate the current spindle rotation angle in the workpiece coordinate system by integrating over time.

[0024] The phase alignment unit is used to align the spindle radial runout signal and the three-dimensional cutting force signal to the current spindle rotation angle based on the physical angle between the installation position of each sensor and the cutting point.

[0025] The feature synthesis unit is used to calculate the resultant force modulus of the three-dimensional cutting force signal in the cutting plane, index the stiffness value corresponding to the current angle from the circumferential dynamic stiffness distribution model, and divide the resultant force modulus by the stiffness value to obtain the stiffness modulation deformation.

[0026] Preferably, the feature synthesis unit calculates the equivalent spindle geometric error by: acquiring the spindle radial runout signals in the X and Y directions aligned to the current spindle rotation angle; multiplying the X-direction spindle radial runout signal by the cosine of the current spindle rotation angle; multiplying the Y-direction spindle radial runout signal by the sine of the current spindle rotation angle; and adding the products of the two to obtain the equivalent spindle geometric error.

[0027] Preferably, the error prediction module includes:

[0028] The feature sequence construction unit is used to combine the stiffness modulation deformation, the equivalent spindle geometric error, the workpiece material hardness parameter, and the spindle speed parameter at the current moment and several past angular steps to form a time-series feature matrix.

[0029] The network inference unit is used to input the temporal feature matrix into the long short-term memory network model, update the cell state through the gating mechanism of forget gate, input gate and output gate, and map to obtain the roundness error prediction value.

[0030] Preferably, the network inference unit calculates the activation values ​​of the forget gate, the input gate, and the output gate by concatenating the hidden state of the previous time step with the feature vector in the temporal feature matrix of the current time step, multiplying the concatenated vector with the corresponding weight matrix and adding a bias term, and applying the Sigmoid activation function to the calculation result.

[0031] Preferably, the compensation execution module includes:

[0032] A deviation calculation unit is used to calculate the numerical deviation between the zero-value target and the predicted roundness error value;

[0033] The control quantity generation unit is used to perform proportional, integral and derivative operations on the numerical deviation using a PID control algorithm to generate a control compensation quantity to offset the error.

[0034] The instruction correction execution unit is used to read the original interpolation instruction from the CNC system in real time and use the control compensation amount as an offset amount to modify the target position coordinates of the original interpolation instruction.

[0035] This invention provides a rolling element precision turning accuracy error compensation control system. It has the following beneficial effects:

[0036] 1. This invention solves the problem of uneven circumferential stiffness caused by workpiece clamping deformation, which is not considered in traditional compensation methods, by constructing a circumferential dynamic stiffness distribution model and applying an angle-domain phase coupling mechanism. The system uses clamping contour data obtained by laser scanning to calculate stiffness changes under different rotation angles, and couples the real-time cutting force with the stiffness value of the corresponding phase to obtain the stiffness modulation deformation. This mechanism enables the feature data of the input error prediction model to reflect the actual elastic deformation of the cutting point under the current stress state, improving the accuracy of the model's characterization of the physical machining process, thereby improving the accuracy of roundness error prediction.

[0037] 2. This invention utilizes a multi-source heterogeneous sensing module in conjunction with a phase alignment unit to eliminate spatial phase differences in data caused by the different physical installation positions of sensors. Through the spindle rotary encoder signal, the signals from the dispersed capacitive displacement sensors, strain gauge force gauges, and laser displacement sensors are uniformly mapped and aligned to the same rotation angle in the workpiece coordinate system. This process ensures that the multidimensional data input to the long short-term memory network remains strictly synchronized in the spatiotemporal dimensions, avoiding difficulties in prediction model convergence or compensation lag caused by phase misalignment, and guaranteeing signal consistency during high-speed turning.

[0038] 3. This invention employs a long short-term memory network model for error timing prediction and combines it with a PID algorithm to directly correct the interpolation commands of the CNC system, achieving real-time active compensation during the turning process. The system can predict the roundness error trend of the next angular step at the current machining moment, and the generated micro-displacement compensation command is directly superimposed on the original interpolation command, driving the servo axis to generate a reverse displacement at the cutting point. This closed-loop control method can offset machining deviations caused by spindle rotation errors and stiffness fluctuations in real time, ensuring the roundness accuracy of the finished rolling elements. Attached Figure Description

[0039] Figure 1 This is a system architecture diagram of the present invention;

[0040] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0042] Reference Figures 1-2 , Figure 1 This is a schematic diagram of the overall structure of a rolling element precision turning accuracy error compensation control system according to an embodiment of the present invention. The present invention provides a rolling element precision turning accuracy error compensation control system, which is applied in a CNC lathe environment for hard turning rolling elements. The system includes a multi-source heterogeneous sensing module, a dynamic stiffness modeling module, an angle domain phase coupling module, an error prediction module, and a compensation execution module. These modules are connected via a data bus or communication interface to achieve real-time data transmission and interaction.

[0043] In the precision error compensation control system for rolling element precision turning, a workpiece polar coordinate system is defined to unify the spatial reference of data collected by various sensors. This coordinate system takes the rotation center of the lathe spindle as its origin and the cutting point where the cutting tool contacts the workpiece as its phase zero reference. The system uses the spindle rotary encoder signal as the time reference for phase synchronization, ensuring that all subsequent calculations of physical quantities are based on the same workpiece rotation angle.

[0044] A multi-source heterogeneous sensing module is connected to multiple physical sensors mounted on the lathe. This module is used to acquire physical state data in real time during the turning process. The physical state data specifically includes spindle radial runout signals, three-dimensional cutting force signals, workpiece clamping contour signals, and spindle rotary encoder signals. The multi-source heterogeneous sensing module converts the acquired analog signals into digital signals and sends them to the dynamic stiffness modeling module and the angle domain phase coupling module.

[0045] The dynamic stiffness modeling module receives workpiece clamping contour signals transmitted from the multi-source heterogeneous sensing module. During the scanning phase before formal cutting, this module identifies the geometric features and stress-deformation characteristics of the workpiece surface based on the clamping contour signals. Based on these identified features, the module constructs a circumferential dynamic stiffness distribution model along the workpiece's circumference. This model describes the workpiece's stiffness properties against radial deformation at different rotational angles.

[0046] The angle domain phase coupling module is connected to both the multi-source heterogeneous sensing module and the dynamic stiffness modeling module. This module receives signals from the spindle rotary encoder, spindle radial runout, three-dimensional cutting force, and the circumferential dynamic stiffness distribution model. Utilizing the spindle rotary encoder signal, the module maps the time-domain spindle radial runout and three-dimensional cutting force signals to the workpiece rotation angle domain, eliminating time asynchrony issues between different sensors.

[0047] The angle-domain phase coupling module is further used to perform physical mechanism coupling calculations. This module combines the circumferential dynamic stiffness distribution model with the mapped angle-domain cutting force to calculate the ratio of the angle-domain cutting force to the corresponding phase stiffness value, thereby obtaining the stiffness modulation deformation. Simultaneously, this module calculates the equivalent spindle geometric error and outputs a comprehensive feature vector containing both the stiffness modulation deformation and the equivalent spindle geometric error. This comprehensive feature vector characterizes the overall machining error state caused by the combined effects of the cutting force and spindle error at the current angle.

[0048] The error prediction module is connected to the angle domain phase coupling module and receives the synthesized feature vector. Internally, this module deploys a pre-trained Long Short-Term Memory (LSTM) network model. The error prediction module inputs the synthesized feature vector as an input sequence into the LSM network model. Through inference calculations by the network model, the error prediction module predicts and outputs the roundness error prediction value for the next angle step. This roundness error prediction value represents the expected radial deviation at the cutting point position in the future.

[0049] The compensation execution module is connected to the error prediction module and the lathe's CNC system. This module receives the predicted roundness error value and generates a position compensation command based on it. The position compensation command includes the servo axis displacement to offset the predicted error. The compensation execution module dynamically superimposes the position compensation command onto the original interpolation command of the CNC system, driving the feed servo axis to perform minute displacement compensation at the cutting point, thereby achieving real-time correction of machining errors.

[0050] Reference Figures 1-2 The multi-source heterogeneous sensing module specifically includes an orthogonal displacement sensing unit, a cutting force sensing unit, a clamping contour scanning unit, and a spindle encoder unit. Each unit synchronously acquires data based on a unified time reference and calibrates its spatial phase relationship relative to the machining cutting point according to its physical installation position.

[0051] The spindle encoder unit is a high-resolution photoelectric encoder coaxially mounted at the rear end of a lathe spindle. This spindle encoder unit is used to output the spindle's rotational pulse signals and instantaneous angular velocity in real time. The system defines the cutting point position where the cutting tool contacts the workpiece as the phase zero reference. The signal provided by the spindle encoder unit is used to establish the mapping relationship between the time domain and the angle domain, at any given moment... Workpiece rotation angle position Calculate using the following formula:

[0052] ;

[0053] In the formula, Indicates time The instantaneous angular velocity; Indicates to Take a mold to keep the angle between 0 and Within the range; Indicates at any time At that time, the instantaneous rotation angle position of the workpiece coordinate system relative to the phase zero reference; This indicates the current sampling time or calculation time.

[0054] The orthogonal displacement sensing unit uses two sets of non-contact capacitive displacement sensors, each mounted on the end face of the spindle front bearing housing. The two sensors are orthogonally arranged at a 90-degree angle in the circumferential direction, corresponding to the X and Y axes of the spindle coordinate system, respectively. Due to space limitations, the capacitive displacement sensors cannot be directly mounted at the cutting point; therefore, they have a first installation phase offset angle relative to the phase zero reference. The orthogonal displacement sensing unit acquires the micron-level radial runout displacement of the spindle during rotation in real time, and the output signal is recorded as follows: and These two components characterize the geometric deviation of the spindle axis trajectory from the ideal center of rotation.

[0055] The cutting force sensing unit is a piezoelectric or strain gauge force gauge integrated inside the lathe fixture base or turret base. This force gauge has high rigidity and can sense the dynamic reaction force transmitted to the base during cutting. The cutting force sensing unit simultaneously acquires three component force signals along the machine tool coordinate system: the X-axis (radial feed direction), the Y-axis (tangential cutting direction), and the Z-axis (axial feed direction), denoted as […]. , and During data acquisition, the cutting force sensing unit amplifies and low-pass filters the original analog signal through a signal conditioning circuit to remove high-frequency noise interference and retain frequency band information related to the cutting process.

[0056] The clamping contour scanning unit is configured as a high-precision laser displacement sensor, which is fixed inside the machine tool guard or on a separate measuring bracket, with the laser beam emitted perpendicular to the cylindrical surface of the workpiece. This sensor has a second mounting phase offset angle relative to a phase zero-point reference. The clamping contour scanning unit operates in two modes: scanning mode and monitoring mode. In the pre-machining scanning mode, the spindle rotates at a set low speed (e.g., 10-60 rpm), and the laser displacement sensor performs non-contact continuous ranging on the workpiece surface to acquire the original contour data of the workpiece in the clamping state. The raw profile data includes the workpiece's own shape errors (such as ellipticity) as well as the elastic deformation characteristics caused by the clamping force applied by the jaws or fixtures.

[0057] To ensure the spatiotemporal consistency of multi-source data, the multi-source heterogeneous sensing module synchronously triggers each sensor channel at the hardware level. All analog signals are sampled by a multi-channel analog-to-digital converter under the control of the same clock pulse. The sampling frequency is set to satisfy the Nyquist sampling theorem and is compatible with the highest spindle speed (e.g., above 10 kHz), thereby ensuring that the instantaneous values ​​of each physical quantity at the same moment can be accurately recovered in subsequent processing.

[0058] Reference Figures 1-2 The dynamic stiffness modeling module specifically includes a data preprocessing unit and a stiffness mapping generation unit. The main function of this module is to transform the static geometric contour features of the workpiece into physical stiffness distribution features, thereby quantitatively describing the differences in the workpiece's ability to resist cutting deformation at different angular positions around its circumference.

[0059] The data preprocessing unit is used to process the raw contour data acquired by the clamping contour scanning unit. Signal cleaning and feature extraction are performed. First, the data preprocessing unit processes the time-domain data according to the spindle speed. Convert to angle domain data The data preprocessing unit performs a reference circle fitting operation and calculates the average radius of the workpiece profile using the least squares method. It also utilizes a moving average filtering algorithm to remove high-frequency noise caused by surface roughness or coolant splashing. Based on the average radius, this data preprocessing unit calculates the radial deviation of each angular position relative to the reference circle. :

[0060] ;

[0061] In the formula, Indicates the angle of rotation around the workpiece. Radial deviation at the location; This represents the workpiece contour data in the angle domain after data preprocessing. This represents the fitted average radius of the workpiece profile.

[0062] The radial deviation This directly reflects the macroscopic deformation of the workpiece. In a typical scenario using a three-jaw chuck for clamping, It will exhibit third harmonic characteristics, that is, the radial radius decreases at the contact point of the chuck (negative deviation), and the radial radius increases in the suspended area between adjacent chucks (positive deviation).

[0063] Stiffness mapping generation element configured based on radial deviation Constructing a circumferential dynamic stiffness distribution model This embodiment is based on the physical fact that the radial stiffness of the workpiece is highest near the clamping point of the jaws, while the stiffness is relatively low in the suspended region between the two jaws. To accurately describe this nonlinear stiffness variation, the stiffness mapping generation unit adopts a nonlinear mapping method based on the hyperbolic tangent function.

[0064] Specifically, the stiffness mapping generation unit pre-stores the nominal radial stiffness of the workpiece material. (Unit: N / m), this parameter is obtained through finite element analysis or experimental modal testing. When constructing the model, this stiffness mapping generation element calculates the workpiece at different rotation angles according to the following formula. Stiffness values ​​below:

[0065] ;

[0066] In the formula, Indicates the nominal radial stiffness; This represents the radial deviation after phase correction; Indicates the angle of the workpiece The instantaneous radial stiffness value at the location; The stiffness fluctuation coefficient is a dimensionless positive number used to characterize the range of stiffness change caused by clamping. This coefficient is related to the clamping force of the fixture and the wall thickness of the workpiece. As a hyperbolic tangent function, its S-curve characteristics can smoothly simulate the transition process of stiffness from high to low, and limit the upper and lower boundaries of stiffness correction, preventing the stiffness value calculation from diverging due to local measurement noise. This is a normalization factor used to normalize the dimensions of the radial deviation and adjust the sensitivity range of the hyperbolic tangent function.

[0067] Based on the above calculations, the stiffness mapping generation unit generates a stiffness map that varies with angle. Periodically varying stiffness sequence It not only reflects the structural stiffness of the workpiece itself, but also couples the influence of the clamping boundary conditions. It can accurately describe the stiffness difference of the cutting point when passing through the "hard phase" (at the chuck) and the "soft phase" (at the suspension point), providing an accurate physical benchmark for subsequent deformation calculation.

[0068] Reference Figures 1-2 The angle domain phase coupling module specifically includes an angle transformation unit, a phase alignment unit, and a feature synthesis unit. The core function of this module is to utilize physical mechanisms to accurately align multi-source heterogeneous data in the spatial domain and synthesize a comprehensive feature vector that characterizes the actual processing error state.

[0069] The angle conversion unit is used to receive the pulse signal and instantaneous angular velocity output from the spindle encoder unit. This angle conversion unit calculates the current spindle rotation angle in the workpiece coordinate system by performing real-time integration of the instantaneous angular velocity. To meet the requirements of discrete control systems, this angle conversion unit divides continuous rotation angles into fixed angle steps, generating a discrete angle index sequence. This index sequence serves as the unique primary key for internal data alignment within the system.

[0070] The phase alignment unit is used to solve the data phase misalignment problem caused by the dispersed physical installation positions of sensors. This phase alignment unit pre-stores the physical installation angle parameters of each sensor relative to the cutting point (phase zero point), including the first installation phase offset angle. (Corresponding to capacitive displacement sensor) and second mounting phase offset angle (Corresponding laser displacement sensor). For time... The phase alignment unit maps and aligns the acquired sensor data to the current angle of the cutting point according to the following logic. :

[0071] For the spindle radial runout signal, read the corresponding angle. The sampled values;

[0072] For cutting force signals, considering the extremely short mechanical hysteresis in force signal transmission, they directly correspond to the current angle. The sampled values;

[0073] For the stiffness model, read the corresponding angle. The estimated stiffness value.

[0074] The feature synthesis unit is used to perform key physical quantity synthesis calculations and outputs a comprehensive feature vector that includes stiffness modulation deformation and equivalent principal axis geometric error.

[0075] The feature synthesis unit calculates the equivalent principal axis geometric error. This error characterizes the projected component of the spindle axis offset in the radial direction at the cutting point. The calculation formula is as follows:

[0076] ;

[0077] In the formula, and These are the X-axis and Y-axis principal axis runout values ​​after phase alignment; and The projection coefficients enable the projection of the jump vector in the orthogonal coordinate system onto the direction of the rotating cutting vector. Indicates the first Each angle step When, the equivalent principal axis geometric error is calculated; Indicates the index of the current discrete spindle rotation angle; This indicates the first installation phase offset angle of the capacitive displacement sensor relative to the phase zero reference.

[0078] This feature synthesis unit calculates the stiffness modulation deformation. This is the core physical feature of this embodiment, reflecting the actual elastic deflection caused by the cutting force acting on a non-uniform stiffness workpiece. This unit first calculates the resultant force modulus in the cutting plane. Subsequently, the current angle is indexed from the circular dynamic stiffness distribution model. Corresponding instantaneous stiffness value Finally, the ratio is calculated using a modified form of Hooke's Law:

[0079] ;

[0080] In the formula, This represents the calculated stiffness modulation deformation. Indicates the angle The resultant force modulus of the cutting plane at the location; This represents the instantaneous radial stiffness value indexed from the circumferential dynamic stiffness distribution model.

[0081] This formula allows the system to distinguish between two distinct physical conditions: "large cutting force acting on a high-rigidity region" and "small cutting force acting on a low-rigidity region," thus describing the causes of machining errors more accurately than simply using force or displacement signals.

[0082] Finally, the feature synthesis unit packages the calculated physical quantities with other process parameters to construct the output. The comprehensive feature vector of each angle step length :

[0083] ;

[0084] In the formula, Main spindle speed; For feed rate; The hardness parameter of the workpiece material; Represents the comprehensive feature vector; This represents the calculated stiffness modulation deformation. Indicates the first Each angle step When, the equivalent principal axis geometric error is calculated. This vector It is then transmitted to the error prediction module as input features for the deep learning model.

[0085] Reference Figures 1-2 The error prediction module specifically includes a feature sequence construction unit and a network inference unit. This error prediction module adopts a deep learning architecture based on Long Short-Term Memory (LSTM) networks, aiming to capture the time-dependent and nonlinear dynamic features of the cutting process, thereby achieving advanced prediction of roundness error.

[0086] The feature sequence construction unit processes the real-time data stream input from the angle domain phase coupling module. Considering the "memory effect" of the turning process (i.e., the current machining state is influenced by the state at past moments), this feature sequence construction unit employs a sliding window mechanism to construct the input data. Specifically, this feature sequence construction unit will use the current moment... and before The comprehensive feature vector of each historical moment Combined, forming a dimension Temporal feature matrix ,in This represents the dimension of the feature vector. This is the temporal feature matrix. It comprehensively includes historical information such as the trend of cutting force changes, stiffness fluctuation history, and spindle runout trajectory.

[0087] The network inference unit is used to load and run a pre-trained LSTM network model. The core component of this LSTM network model is the LSTM memory cell, which regulates information storage and forgetting through a sophisticated gating mechanism. For each time step in the input sequence... The LSTM unit performs the following logical operations internally:

[0088] Forget Gate Calculation: First, calculate the activation value of the forget gate. The calculation method is as follows: The hidden state from the previous time step... With the input feature vector at the current time Perform concatenation, then combine the concatenated vector with the forget gate weight matrix. Multiply and add the bias term Finally, the Sigmoid activation function is applied to the calculation results. .

[0089] ;

[0090] In the formula, Indicates the current time step The calculated forget gate activation vector; This represents the Sigmoid activation function; Represents the forget gate weight matrix; This represents the hidden state vector from the previous time step. This represents the input feature vector at the current moment; This represents the forget gate bias vector.

[0091] The forgetting gate determines the cell's state at the previous moment. How much information will be discarded?

[0092] Input Gate and Candidate State Calculation: Calculating the activation value of the input gate. and candidate cell status Input gate Similarly, the proportion of the current input information retained is determined by the Sigmoid function; candidate states Then through the hyperbolic tangent function Generate new candidate information.

[0093] ;

[0094] ;

[0095] In the formula, Indicates the current time step The calculated input gate activation vector; This represents the Sigmoid activation function; This represents the hidden state vector from the previous time step. This represents the input feature vector at the current moment; Represents the input gate bias vector; Indicates the current time step The generated candidate cell state vector; Represents the hyperbolic tangent activation function; Represents the candidate state weight matrix; This represents the candidate state bias vector.

[0096] Cell State Update: Update the current cell state based on the outputs of the forget gate and the input gate. This is the core memory carrier of LSTM.

[0097] ;

[0098] In the formula, Indicates the current time step Updated cell state vector; Indicates the current time step The calculated forget gate activation vector; This represents the cell state vector at the previous time step; Indicates the current time step The calculated input gate activation vector; Indicates the current time step The generated candidate cell state vector; This represents element-wise multiplication. It enables the long-term preservation of important historical information and the integration of current, new information.

[0099] Output Gate and Hidden State Output: Calculating the Activation Value of the Output Gate And calculate the current hidden state based on the updated cell state. .

[0100] ;

[0101] ;

[0102] In the formula, Indicates the current time step The calculated output gate activation vector; This represents the Sigmoid activation function; This represents the output gate weight matrix; This represents the hidden state vector from the previous time step. This represents the input feature vector at the current moment; This represents the output gate bias vector; Indicates the current time step The calculated hidden state vector; This represents element-wise multiplication. It enables the long-term preservation of important historical information and the integration of current, new information. Represents the hyperbolic tangent activation function; Indicates the current time step Updated cell state vector.

[0103] After processing by multiple LSTM units, the network inference unit stores the hidden state of the last time step. The data is fed into a fully connected layer (Dense Layer). The fully connected layer uses a linear transformation to map the high-dimensional hidden state features into a one-dimensional scalar output, namely the predicted roundness error for the next angular step. This prediction quantifies the expected radial deviation of the workpiece surface from the ideal circle at the future cutting point location.

[0104] Furthermore, the network inference unit employs Dropout technology (random deactivation), randomly freezing some neurons during inference to enhance the model's generalization ability and prevent overfitting to processing conditions. Through the above calculation process, the error prediction module can output accurate error prediction results with extremely high response speed (milliseconds), providing a forward-looking decision-making basis for subsequent compensation control.

[0105] Reference Figures 1-2 The compensation execution module specifically includes a deviation calculation unit, a control quantity generation unit, and an instruction correction execution unit. As the system's execution terminal, this module is responsible for converting the data-level prediction results into physical-level mechanical actions, thereby achieving closed-loop error compensation.

[0106] The deviation calculation unit is used to establish the objective function for compensation control. The system's preset ideal machining target is zero roundness error, i.e. This unit receives the roundness error prediction value from the error prediction module. And calculate the numerical deviation between it and the ideal target. :

[0107] ;

[0108] In the formula, Indicates the first The displacement compensation deviation value calculated for each control cycle; This represents the predicted roundness error value output by the error prediction module.

[0109] This deviation value The physical meaning is: in order to offset the predicted error, the servo axis needs to generate a reverse displacement.

[0110] The control quantity generation unit is used to run a digital PID control algorithm to ensure that the servo axis's movements can quickly and stably track changing deviation signals. This control quantity generation unit processes the deviation signal... Perform proportional (P), integral (I), and derivative (D) operations to generate discrete control compensation quantities. .

[0111] The specific calculation logic is as follows:

[0112] Proportional term: This is used to quickly respond to the current error magnitude;

[0113] Integral term: This is used to eliminate the steady-state cumulative error of the system;

[0114] Differential term: It is used to predict error change trends and suppress overshoot oscillations.

[0115] The final control compensation is the sum of the three: .

[0116] In the formula, , , The preset control gain parameters are adjusted based on the dynamic response characteristics of the machine tool feed axes. Indicates the proportional control component; Indicates the first The displacement compensation deviation value calculated for each control cycle; Indicates the integral control component; Represents the differential control component; Indicates the first The final discrete control compensation quantity generated in each control cycle; This represents the historical cumulative integral value maintained in the previous control cycle; It represents the amount of change in error.

[0117] The instruction correction execution unit is used for real-time communication with the machine tool CNC system. This unit reads the original interpolation instructions generated by the CNC system in real time via a high-speed bus. The raw interpolation command contains the target position coordinates of the tool in the next control cycle.

[0118] The instruction modifies the control compensation amount calculated by the execution unit. As a dynamic offset, it is directly superimposed and modified onto the target position coordinates of the original interpolation command to generate the corrected final position command. :

[0119] ;

[0120] In the formula, This indicates the final position instruction after correction; Indicates the original interpolation command; Indicates the first The final discrete control compensation quantity generated in each control cycle.

[0121] The instruction correction execution unit will The data is sent back to the servo driver, which drives the X-axis servo motor to perform micro-feed or retraction movements. Since the compensation action occurs before the actual cutting point is reached (based on the predicted value), the tool can adjust its position instantly when the error occurs, thereby physically removing excess material or compensating for undercut material, effectively improving the roundness accuracy after machining. Furthermore, this instruction correction execution unit also has saturation limiting logic. When the calculated compensation amount exceeds a preset safety threshold (e.g., ±20μm), the output is forcibly limited to the threshold range to prevent the risk of tool collision or overcutting due to prediction anomalies.

[0122] Reference Figures 1-2 , Figure 2 This is a schematic flowchart of a rolling element precision turning accuracy error compensation control method according to an embodiment of the present invention. This method achieves full closed-loop control from static sensing to dynamic compensation through the coordinated operation of the various modules of the above system. The specific implementation process includes the following four main stages:

[0123] Step S100, Static Scanning and Modeling Stage: Before the workpiece is clamped and formal cutting begins, the system first enters static scanning mode. The spindle is controlled to rotate at a low speed (e.g., 30 rpm), while the laser displacement sensor is activated. The laser displacement sensor performs a full circumferential scan of the workpiece's circumference, acquiring raw contour data containing clamping deformation features. Subsequently, the dynamic stiffness modeling module denoises and fits the raw data to a reference circle, extracting the radial deviation distribution. Based on preset workpiece material parameters and the extracted radial deviation, the system uses a nonlinear mapping function to calculate the stiffness values ​​of the workpiece at various angular positions on the circumference, ultimately generating and storing a circumferential dynamic stiffness distribution model. This model serves as the basic benchmark for evaluating the workpiece's stress and deformation during subsequent cutting processes.

[0124] Step S200, Dynamic Machining and Multi-Source Sensing Stage: After scanning, the machine tool enters the formal turning machining state, and the spindle accelerates to the rated cutting speed. At this time, the multi-source heterogeneous sensing module synchronously activates all sensor channels. The spindle rotary encoder outputs the current spindle rotation angle and angular velocity information in real time, serving as the system's time synchronization spindle. Simultaneously, the orthogonal capacitive displacement sensor continuously acquires the radial runout trajectory of the spindle end face, and the force gauge integrated into the fixture base monitors the three-dimensional cutting force components during the cutting process in real time. All sensor data are accurately timestamped and transmitted to the subsequent processing unit via a high-speed bus.

[0125] Step S300, Coupling Calculation and Error Prediction Stage: Upon receiving real-time data, the angle domain phase coupling module first utilizes the geometric relationship of the sensor mounting phase to align the dispersed time-domain signals to the current cutting point angle. Subsequently, this module indexes the stiffness value of the current angle from a pre-stored circular dynamic stiffness distribution model and divides the measured cutting force by this stiffness value to calculate the current stiffness modulation deformation. Simultaneously, the system combines the projected spindle runout data to synthesize a comprehensive feature vector containing deformation, spindle error, and process parameters. The error prediction module immediately reads this feature vector sequence and inputs it into a pre-trained long short-term memory (LSTM) network model. The neural network model processes the temporal dependencies through internal gating units, quickly deducing and calculating the predicted roundness error value for the next angle step.

[0126] Step S400, Command Correction and Execution Compensation Stage: After obtaining the error prediction value, the compensation execution module immediately calculates the deviation between the prediction value and the zero-error target. The PID control algorithm calculates the required servo axis displacement compensation based on the deviation. The system then reads the current original interpolation command of the CNC system and directly superimposes the calculated compensation amount onto the target position coordinates to generate the corrected final motion command. The servo driver responds to this correction command, driving the X-axis motor to generate a small additional feed or retraction action when the cutting tool reaches the angular position. This action physically directly cancels out the machining error caused by insufficient stiffness or spindle runout. As the spindle continues to rotate, the above steps S200 to S400 are executed cyclically in each control cycle until the entire turning process is completed, thereby achieving real-time, continuous, and active suppression of rolling element roundness error.

Claims

1. A rolling element precision turning accuracy error compensation control system, characterized in that, include: A multi-source heterogeneous sensing module is used to acquire spindle radial runout signals, three-dimensional cutting force signals, workpiece clamping contour signals, and spindle rotary encoder signals in real time. The dynamic stiffness modeling module is used to construct a circumferential dynamic stiffness distribution model distributed along the circumference of the workpiece based on the workpiece clamping contour signal. The angle domain phase coupling module is used to map the spindle radial runout signal and the three-dimensional cutting force signal to the workpiece rotation angle domain using the spindle rotary encoder signal, calculate the ratio of the angle domain cutting force to the corresponding phase stiffness value to obtain the stiffness modulation deformation, and output a comprehensive feature vector containing the stiffness modulation deformation and the equivalent spindle geometric error. The error prediction module is used to input the comprehensive feature vector into a pre-trained long short-term memory network model, predict and output the roundness error prediction value for the next angle step. The compensation execution module is used to generate a position compensation command based on the predicted roundness error value, and dynamically superimpose the position compensation command into the original interpolation command of the CNC system to drive the servo axis to perform displacement compensation.

2. The rolling element precision turning accuracy error compensation control system according to claim 1, characterized in that, The multi-source heterogeneous sensing module includes: The orthogonal displacement sensing unit includes two sets of capacitive displacement sensors orthogonally arranged on the end face of the spindle, used to collect the radial runout signals of the spindle in the X and Y directions; The cutting force sensing unit includes a strain gauge force gauge integrated into the lathe fixture base, used to acquire the three-axis cutting force signals in the X-axis, Y-axis and Z-axis directions; The clamping contour scanning unit includes a laser displacement sensor mounted perpendicular to the workpiece clamping surface, used to scan the workpiece surface to obtain the workpiece clamping contour signal while the spindle is rotating at low speed.

3. The rolling element precision turning accuracy error compensation control system according to claim 2, characterized in that, The orthogonal displacement sensing unit uses the cutting point position where the cutting tool contacts the workpiece as the phase zero reference. The two sets of capacitive displacement sensors have a first installation phase offset angle relative to the phase zero reference, and the laser displacement sensor has a second installation phase offset angle relative to the phase zero reference.

4. The rolling element precision turning accuracy error compensation control system according to claim 1, characterized in that, The dynamic stiffness modeling module includes: The data preprocessing unit is used to perform noise reduction and reference circle fitting on the workpiece clamping contour signal and calculate the radial deviation relative to the reference circle. The stiffness mapping generation unit is used to calculate the stiffness value of the workpiece at different rotation angles based on the nominal radial stiffness of the workpiece material and the radial deviation, and to generate the circumferential dynamic stiffness distribution model.

5. The rolling element precision turning accuracy error compensation control system according to claim 4, characterized in that, The stiffness mapping generation unit calculates the stiffness value of the workpiece at different rotation angles by using the hyperbolic tangent function to map the ratio of the radial deviation to the normalization factor, multiplying the mapping result by the stiffness fluctuation coefficient and adding it to the unit value, and then multiplying the result by the nominal radial stiffness to obtain the stiffness value at the corresponding angle.

6. The rolling element precision turning accuracy error compensation control system according to claim 1, characterized in that, The angle domain phase coupling module includes: An angle conversion unit is used to receive the instantaneous angular velocity fed back by the spindle rotary encoder signal and calculate the current spindle rotation angle in the workpiece coordinate system by integrating over time. The phase alignment unit is used to align the spindle radial runout signal and the three-dimensional cutting force signal to the current spindle rotation angle based on the physical angle between the installation position of each sensor and the cutting point. The feature synthesis unit is used to calculate the resultant force modulus of the three-dimensional cutting force signal in the cutting plane, index the stiffness value corresponding to the current angle from the circumferential dynamic stiffness distribution model, and divide the resultant force modulus by the stiffness value to obtain the stiffness modulation deformation.

7. A rolling element precision turning accuracy error compensation control system according to claim 6, characterized in that, The feature synthesis unit calculates the equivalent spindle geometric error by: acquiring the spindle radial runout signals in the X and Y directions aligned to the current spindle rotation angle; multiplying the X-direction spindle radial runout signal by the cosine of the current spindle rotation angle; multiplying the Y-direction spindle radial runout signal by the sine of the current spindle rotation angle; and adding the products of the two to obtain the equivalent spindle geometric error.

8. The rolling element precision turning accuracy error compensation control system according to claim 1, characterized in that, The error prediction module includes: The feature sequence construction unit is used to combine the stiffness modulation deformation, the equivalent spindle geometric error, the workpiece material hardness parameter, and the spindle speed parameter at the current moment and several past angular steps to form a time-series feature matrix. The network inference unit is used to input the temporal feature matrix into the long short-term memory network model, update the cell state through the gating mechanism of forget gate, input gate and output gate, and map to obtain the roundness error prediction value.

9. A rolling element precision turning accuracy error compensation control system according to claim 8, characterized in that, The network inference unit calculates the activation values ​​of the forget gate, the input gate, and the output gate by concatenating the hidden state of the previous time step with the feature vector in the temporal feature matrix of the current time step, multiplying the concatenated vector with the corresponding weight matrix and adding a bias term, and applying the Sigmoid activation function to the calculation result.

10. A rolling element precision turning accuracy compensation control system according to claim 1, characterized in that, The compensation execution module includes: A deviation calculation unit is used to calculate the numerical deviation between the zero-value target and the predicted roundness error value; The control quantity generation unit is used to perform proportional, integral and derivative operations on the numerical deviation using a PID control algorithm to generate a control compensation quantity to offset the error. The instruction correction execution unit is used to read the original interpolation instruction from the CNC system in real time and use the control compensation amount as an offset amount to modify the target position coordinates of the original interpolation instruction.

Images