Sleeve machining error dynamic feedback regulation method based on fuzzy logic operation

Abstract

The present application relates to the technical field of industrial equipment control, in particular to a sleeve machining error dynamic feedback adjustment method based on fuzzy logic operation, comprising: acquiring pusher transient current, workbench positioning angle deviation, main shaft vibration acceleration, real-time temperature and cylindricality deviation data of offline sleeve in the machining process; extracting pusher current out-of-tolerance characteristic quantity, generating initial feed ratio offset instruction of the first main shaft for feedforward control; extracting main shaft vibration effective value and temperature rise gradient, synchronously inputting them into a comprehensive state fuzzy evaluation model with the angle deviation, outputting a comprehensive error severity scalar, and using a conversion matrix to decouple the tool compensation and rotational speed offset of the remaining main shafts to execute cutting feedback adjustment; using continuous batch cylindricality difference differential slope to generate boundary offset, and globally translating and calibrating the fuzzy model reference. The present application realizes closed-loop adjustment of machining error through fuzzy evaluation of multi-source data.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment control technology, specifically to a method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operations. Background Technology

[0002] Sleeve-type parts, as core components in mechanical transmission and high-pressure fluid systems, have their inner and outer cylindrical surfaces whose geometric accuracy directly determines the overall operational performance and lifespan of the equipment. To meet the demands of modern industry for the mass production and high efficiency of these parts, multi-station rotary indexing CNC machine tools are commonly used for continuous collaborative machining.

[0003] In conventional CNC sleeve machining, to ensure machining accuracy, the industry typically relies on the closed-loop control capability of the CNC system and periodic sampling compensation. Existing error control methods are mostly based on single-dimensional delay feedback or static model compensation, such as fixed-step tool compensation after offline measurement of the part, or simple linear thermal displacement compensation based on a single temperature sensor.

[0004] However, under the high-frequency continuous operation of multi-station machine tools, machining errors are actually a nonlinear result caused by a combination of various disturbances, such as front-end material fluctuations, high-frequency dynamic loads, and low-frequency thermal drift of the system. Because these factors exhibit highly dynamic evolution characteristics in continuous machining, existing independent variable or static compensation mechanisms are difficult to accurately map this systematic error drift trend, and are prone to undercompensation or over-adjustment under complex operating conditions.

[0005] In summary, how to effectively suppress the accumulation of machining errors caused by complex disturbance factors under the highly dynamic working conditions of multi-station continuous machining, so as to ensure the long-term consistency of the machining accuracy of parts, is a technical problem that urgently needs to be solved in this field.

[0006] To address this, a dynamic feedback adjustment method for sleeve machining error based on fuzzy logic operations is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a dynamic feedback adjustment method for sleeve machining errors based on fuzzy logic operations. Through multi-source data fuzzy evaluation, closed-loop adjustment of machining errors is achieved. This includes acquiring data on the transient current during material feeding, table positioning angle deviation, spindle vibration acceleration, real-time temperature, and the cylindricity deviation of the finished sleeve; extracting the out-of-tolerance characteristic quantity of the material feeding current to generate the initial feed rate offset command for the first spindle for feedforward control; extracting the effective value of spindle vibration and temperature rise gradient, and synchronously inputting them and the angle deviation into a comprehensive state fuzzy evaluation model to output a comprehensive error severity scalar; and using a transformation matrix to decouple the tool compensation and speed offset of the remaining spindles to perform cutting feedback adjustment; and using the cylindricity difference slope of consecutive batches to generate boundary offsets for global translation calibration of the fuzzy model reference.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A dynamic feedback adjustment method for sleeve machining error based on fuzzy logic operations includes: Acquire transient current data during the material feeding process, table positioning angle deviation data, spindle vibration acceleration data, real-time temperature data, and cylindricity deviation data of the unloading sleeve; Extract the peak current value of the transient current data during the feeding cycle, calculate the difference between it and the preset reference value to obtain the current deviation feature, and generate the first initial feed rate offset command of the spindle based on the current deviation feature for feedforward control. The effective value of the spindle vibration acceleration data is extracted, and the time derivative of the real-time temperature data is calculated to generate the temperature rise gradient value. This value is then synchronously input into the comprehensive state fuzzy evaluation model along with the table positioning angle deviation data. The comprehensive state fuzzy evaluation model infers and outputs the comprehensive error severity scalar, which is then multiplied by the transformation matrix to parse out the radial tool compensation fine adjustment amount and speed offset amount of the remaining spindles and send them down to execute the cutting feedback adjustment. Extract the cylindricity deviation data from consecutive batches to construct a time series and calculate the first-order difference slope. Generate the boundary offset based on the first-order difference slope and superimpose it onto the benchmark value of the comprehensive state fuzzy evaluation model for global calibration.

[0009] Preferably, the process of acquiring various data during the processing includes: acquiring transient current data of the sleeve blank during feeding by sampling at a preset sampling period using a current sensor connected to the DC bus of the servo driver of the linear servo pusher mechanism; acquiring table positioning angle deviation data by reading the angle difference between the actual angle and the theoretical position during each indexing action locking using a high-precision circular grating installed on the rotating shaft end of the multi-station rotary indexing table; acquiring spindle vibration acceleration data during continuous cutting by using a triaxial piezoelectric accelerometer bolted to the surface of the bearing seat at the front end of the vertical spindle; acquiring real-time temperature data by using a thermal resistance sensor attached to the surface of the housing of the vertical spindle; and extracting spatial point cloud data of the outer cylindrical surface of the finished sleeve sampled by the coordinate measuring machine based on the contact scanning probe along the lower line, and calculating the cylindricity deviation data using a least squares cylindrical fitting algorithm.

[0010] Preferably, the process of generating the initial feed rate offset command for the first spindle based on the current deviation feature for feedforward control includes: acquiring a pre-calibrated reference current envelope of the linear servo pusher mechanism under no-load operation, and dynamically time-aligning the acquired pusher transient current data with the reference current envelope based on waveform fluctuation characteristics; subtracting the aligned pusher transient current data from the reference current envelope to extract the current residual sequence; extracting the maximum amplitude of the current residual sequence as the current peak value within the effective clamping time window when the sleeve blank is pushed into the fixture, and subtracting it from the preset standard clamping resistance reference value of the blank material to obtain the current deviation feature; searching and matching the primary feed degradation ratio corresponding to the current deviation feature based on a pre-established mapping table of current deviation and tool cutting load; limiting the primary feed degradation ratio based on the preset fracture safety threshold of the first spindle tool assembly, and generating the initial feed rate offset command containing the limited degradation ratio for feedforward control.

[0011] Preferably, the process of extracting the effective value of the spindle vibration acceleration data and generating the temperature rise gradient value by calculating the time derivative of the real-time temperature data includes: using a bandpass filter to isolate the acquired spindle vibration acceleration data by frequency band, wherein the lower cutoff frequency of the bandpass filter is set to the spindle tooth frequency and the upper cutoff frequency is set to the switching carrier frequency of the machine tool servo drive; within the range where the other spindles are in steady-state cutting feed, the effective frequency band data is dynamically truncated based on a sliding time window with a preset step size, and the root mean square of the vibration amplitude within each sliding time window is calculated as the effective value of the spindle vibration acceleration data; the acquired real-time temperature data is smoothed and reconstructed using a moving average filtering algorithm to obtain a smoothed temperature curve of the spindle; according to a preset macroscopic time interval, the smoothed temperature curve of the spindle is subjected to discrete difference operation along the time axis, and the temperature rise increment within a unit time interval is extracted as the temperature rise gradient value.

[0012] Preferably, the process of outputting the comprehensive error severity scalar includes: using the table positioning angle deviation data as a spatial geometric deviation factor, using the extracted effective value as a transient alternating load factor, and using the generated temperature rise gradient value as a thermal inertia drift factor; using a membership function based on the machine tool factory precision acceptance specification and the physical limit calibration of tool wear, mapping the spatial geometric deviation factor, transient alternating load factor, and thermal inertia drift factor to their respective fuzzy domains; establishing a fuzzy rule base based on the multi-field coupled physical failure mechanism, which contains a three-dimensional basic mapping matrix, and its benchmark mapping criterion is that when the values ​​of the transient alternating load factor and the thermal inertia drift factor tend to When the output of the comprehensive error severity scalar is directed toward the positive maximum domain, it shifts toward the positive maximum domain. Based on the benchmark mapping criterion, the fuzzy rule base is configured with nonlinear penalty rules: when it is determined that the thermal inertia drift factor is in a preset thermal expansion amplification range, and the spatial geometric deviation factor deviates from the benchmark zero range, a thermal deformation amplification effect determination is triggered, and a cross-domain coupling penalty mechanism is executed to nonlinearly increase the penalty weight of the corresponding rule. The area centroid method is used to perform defuzzification feature dimensionality reduction calculation on the reasoning results of all triggered rules in the fuzzy rule base, generating a normalized value as the comprehensive error severity scalar for the current indexing station.

[0013] Preferably, the process of issuing and executing cutting feedback adjustment includes: extracting a multi-dimensional control transformation matrix pre-stored in the machine tool CNC system, wherein the multi-dimensional control transformation matrix is ​​constructed from radial dimension sensitivity weights and cutting heat sensitivity weights for different machining processes of each of the other spindles; performing a dimension-expanding mapping calculation on the comprehensive error severity scalar and the multi-dimensional control transformation matrix to generate a comprehensive control increment vector containing independent correction coefficients for each of the other spindles; and performing dimensional decoupling on the comprehensive control increment vector to extract the original values ​​corresponding to each of the other spindles. Radial tool compensation value and original speed offset value; the original radial tool compensation value and the original speed offset value are compared with the preset maximum tool wear compensation extreme value and the spindle safe speed boundary, respectively, and anti-collision soft limiting truncation processing is performed to generate the radial tool compensation fine adjustment amount and the speed offset amount after safety verification; the radial tool compensation fine adjustment amount and the speed offset amount after safety verification are encapsulated into a control message conforming to the machine tool underlying communication bus protocol, and sent to the corresponding servo driver according to the preset smooth interpolation cycle to control the other spindles to perform progressive compensation during continuous cutting.

[0014] Preferably, the process of obtaining the weights in the multidimensional control conversion matrix includes: during the trial cutting stage of the machine tool system, keeping other machining parameters unchanged, sequentially applying a preset limit Z-axis feed speed excitation to each of the remaining spindles individually, measuring the physical change in the actual radial expansion deviation of the inner hole of the test piece, and directly labeling the physical change with positive and negative signs as the feed amount size sensitivity weight of the corresponding spindle; sequentially applying a preset limit spindle speed excitation to each of the remaining spindles individually, while simultaneously increasing the Z-axis feed speed of the corresponding spindle proportionally to lock the feed per tooth, measuring the steady-state temperature rise increment when the corresponding spindle reaches thermal equilibrium steady state, and calculating the maximum speed offset required by the corresponding spindle to offset thermal distortion based on the temperature rise increment and the spindle safety speed boundary, retaining its positive and negative signs, and labeling it as the cutting thermal sensitivity weight of the corresponding spindle; and performing matrix splicing of the feed amount size sensitivity weight and the cutting thermal sensitivity weight obtained from the calibration of each of the remaining spindles to construct the multidimensional control conversion matrix.

[0015] Preferably, the process of global calibration in the benchmark value superimposed on the comprehensive state fuzzy evaluation model includes: extracting the cylindricity deviation data of a preset number of consecutive batches before the current processing is completed, and constructing the time series based on a sliding data window; calculating the arithmetic mean of the difference between the cylindricity deviation data of adjacent batches within the sliding data window divided by the adjacent processing time interval, as the first-order difference slope; comparing the absolute value of the first-order difference slope with a preset drift tolerance threshold: if the absolute value of the first-order difference slope is less than or equal to the drift tolerance threshold, it is determined that the current processing error is in the normal fluctuation range, and the boundary offset is zero; if the absolute value of the first-order difference slope is greater than the drift tolerance threshold, the first-order difference slope is multiplied by a preset macroscopic compensation gain and the average time span constant of the current sliding data window to calculate the boundary offset; and the boundary offset is directly superimposed on the coordinates of the center point of the membership function of the spatial geometric deviation factor in the input layer of the comprehensive state fuzzy evaluation model to complete the translation calibration of the model judgment benchmark.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By dynamically normalizing and subtracting the reference current envelope, the physical impedance of pure material feeding is physically decoupled from the background current caused by friction of the machine tool's inherent guideways. This decoupling mechanism avoids the misinterpretation of static frictional force changes caused by mechanical aging as cutting resistance. Combined with a fracture safety threshold for amplitude limiting and feedforward command generation, the first spindle can perform speed reduction based on the actual material resistance, effectively mitigating the transmission of transient mechanical shocks to subsequent multi-station oscillations in the initial machining stage.

[0017] 2. Sensor data is mapped to spatial geometric deviation, transient alternating load, and thermal inertia drift factor, and a cross-domain coupling penalty mechanism for thermal deformation amplification effect is introduced. When machine tool thermal expansion and indexing deviation occur simultaneously, the geometric amplification law of thermodynamic distortion is reflected by nonlinearly increasing the weight. This compensates for the undercompensation defect that is prone to occur in conventional independent linear evaluation models when multiple physics fields are coupled, making the evaluation results truly reflect the dynamic degradation trend under high-cycle operation.

[0018] 3. By utilizing a multi-dimensional transformation matrix that incorporates dimensional and cutting thermal sensitivity, a single error scalar is decoupled into independent tool compensation and spindle speed correction values ​​that conform to the machining characteristics of each process. Combined with anti-collision limiting verification of wear extreme values ​​and a smooth interpolation distribution mechanism, this ensures that control increments are applied to each spindle progressively during continuous tool feed. This process curbs systematic error drift while avoiding the risk of surface step marks induced by directly injecting step compensation commands into the servo system. Attached Figure Description

[0019] Figure 1A schematic diagram of the dynamic feedback adjustment method for sleeve machining error based on fuzzy logic operation provided by the present invention; Figure 2 A schematic diagram illustrating the process of obtaining the weights in the multidimensional control transformation matrix provided by this invention; Figure 3 This is a schematic diagram of the global calibration process for the benchmark values ​​superimposed on the comprehensive state fuzzy evaluation model provided by the present invention. Detailed Implementation

[0020] The technical solutions of 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.

[0021] Please see Figures 1 to 3 This invention provides a dynamic feedback adjustment method for sleeve machining error based on fuzzy logic operations, the technical solution of which is as follows: A dynamic feedback adjustment method for sleeve machining error based on fuzzy logic operation is described in the following process: Figure 1 As shown, it includes: Acquire transient current data during the material feeding process, table positioning angle deviation data, spindle vibration acceleration data, real-time temperature data, and cylindricity deviation data of the unloading sleeve; Extract the peak current value of the transient current data during the feeding cycle, calculate the difference between it and the preset reference value to obtain the current deviation feature, and generate the first initial feed rate offset command of the spindle based on the current deviation feature for feedforward control. The effective value of the spindle vibration acceleration data is extracted, and the time derivative of the real-time temperature data is calculated to generate the temperature rise gradient value. This value is then synchronously input into the comprehensive state fuzzy evaluation model along with the table positioning angle deviation data. The comprehensive state fuzzy evaluation model infers and outputs the comprehensive error severity scalar, which is then multiplied by the transformation matrix to parse out the radial tool compensation fine adjustment amount and speed offset amount of the remaining spindles and send them down to execute the cutting feedback adjustment. Extract the cylindricity deviation data from consecutive batches to construct a time series and calculate the first-order difference slope. Generate the boundary offset based on the first-order difference slope and superimpose it onto the benchmark value of the comprehensive state fuzzy evaluation model for global calibration.

[0022] Furthermore, the process of acquiring various data during the processing includes: obtaining transient current data of the sleeve blank during feeding by sampling at a preset sampling period using a current sensor connected to the DC bus of the servo driver of the linear servo pusher mechanism; obtaining the table positioning angle deviation data by reading the angle difference between the actual angle and the theoretical position during each indexing action locking using a high-precision circular grating installed on the rotating shaft end of the multi-station rotary indexing table; obtaining the spindle vibration acceleration data during continuous cutting by using a triaxial piezoelectric accelerometer bolted to the surface of the bearing seat at the front end of the vertical spindle; obtaining the real-time temperature data by using a thermal resistance sensor attached to the surface of the housing of the vertical spindle; and extracting the spatial point cloud data of the outer cylindrical surface of the finished sleeve sampled by the coordinate measuring machine based on the contact scanning probe along the lower line, and calculating the cylindricity deviation data using a least-squares cylindrical fitting algorithm.

[0023] When acquiring transient current data for the pusher, a closed-loop Hall current sensor is connected in series to the DC bus circuit of the servo driver of the linear servo pusher mechanism. When the machine tool control system issues a pusher start command, the external data acquisition channel is synchronously triggered to continuously sample and convert the analog signal of the bus current to digital. The sampling process continues until the servo motor reaches the preset torque threshold based on the standard clamping resistance of the blank and the clamping fixture provides a feedback clamping signal. All discrete current amplitude points recorded in chronological order within this effective pusher time window are constructed into a time series to obtain the transient current data of the sleeve blank during the feeding process.

[0024] When acquiring the table positioning angle deviation data, during the machine tool control system initialization phase, the zero-position pulse mark of the high-precision absolute circular grating installed at the end of the central rotating shaft of the multi-station rotary indexing table is pre-read and aligned with the mechanical zero-point coordinates of the machine tool CNC system. During continuous machining operation, when the indexing table completes a single rotation indexing and the underlying mechanical or hydraulic locking mechanism is in place and triggers the rising edge of the locking confirmation signal, the control system immediately latches the actual absolute angle position output by the current circular grating. Subsequently, the actual absolute angle position is algebraically subtracted from the theoretical indexing angle coordinates preset in the machine tool machining program for that station, and the output difference is the table positioning angle deviation data.

[0025] When acquiring spindle vibration acceleration data and real-time temperature data, mounting holes are pre-machined on the surface of the bearing housing at the front end of the vertical spindle to rigidly fix the triaxial piezoelectric accelerometer to this surface using bolts, specifically for vibration signals. Based on the maximum spindle speed set in the machine tool program and the number of cutting edges of the assembled tool, the upper limit of the fundamental frequency of cutting chatter is calculated. The specific calculation formula is as follows: ,in This is the upper limit of the fundamental frequency, measured in Hertz. This refers to the maximum spindle speed, measured in revolutions per minute (rpm). The number of cutting edges of the assembled cutting tool is expressed in revolutions per second.

[0026] Based on the Nyquist sampling theorem, the preset high-frequency sampling rate threshold of the acquisition channel is strictly set to more than twice the upper limit of the fundamental frequency. During the feed interval where the machine tool executes the interpolation command and drives the tool to perform continuous cutting, the weak charge signal output by the piezoelectric sensor is conditioned into a voltage signal by a charge amplifier, and the vibration waveform in the three-dimensional orthogonal direction is recorded in real time according to the above-set high-frequency sampling rate as the spindle vibration acceleration data. For the temperature signal, a PT100 type thermal resistance sensor is tightly attached to the surface of the vertical spindle housing. Its resistance signal that changes with temperature is converted into a standard electrical signal by a temperature transmitter, and continuous recording is performed throughout the entire machining cycle with a low-frequency sampling period of seconds to generate real-time temperature data.

[0027] When acquiring the cylindricity deviation data of the finished sleeve, after the finished sleeve is machined and transferred to the coordinate measuring machine, the contact scanning probe is controlled to perform a multi-section circumferential or helical mechanical contact sliding scan along the outer cylindrical surface of the finished sleeve. This collects the three-dimensional coordinates of each contact point in the machine tool's measurement coordinate system, forming spatial point cloud data. After extracting this spatial point cloud data, the system first uses the initial least squares method to roughly fit all spatial point clouds to generate an initial centerline, and calculates the radial distance from each measured spatial point to this initial centerline. Subsequently, based on the radial distance data of each measured spatial point, the average and standard deviation are calculated. and apply 3 The criteria are used to screen outliers in one-dimensional features, removing those whose radial distance deviates from the mean by more than 3. The measurement points are used to eliminate abrupt noise coordinates caused by unremoved iron filings or residual cutting fluid on the surface. The cleaned spatial point cloud data is then substituted into the mathematical model of the least squares cylinder fitting algorithm. Through iterative calculation, the ideal cylinder axis that minimizes the sum of the squares of the normal distances from all measured spatial points to the fitted cylinder surface is found. The maximum and minimum radial distances from each measured point in the cleaned point cloud to this ideal cylinder axis are calculated, and the difference between the two is calculated to output the specific cylindricity deviation data.

[0028] A physically specific electromechanical sensing link is configured for each error characteristic, directly binding the actions of processes such as feeding, indexing, cutting, and inspection to a defined signal measurement loop. This mechanism eliminates the risk of distortion from purely theoretical algorithm derivations, ensuring that the collected heterogeneous data can accurately reflect the transient mechanical and low-frequency thermal microscopic changes of the machine tool under high-cycle continuous operation, providing reliable underlying data support for error decoupling under complex and highly dynamic working conditions.

[0029] Furthermore, the process of generating the initial feed rate offset command for the first spindle based on the current deviation characteristic for feedforward control includes: acquiring a pre-calibrated reference current envelope of the linear servo pusher mechanism under no-load operation, and dynamically time-aligning the acquired pusher transient current data with the reference current envelope based on waveform fluctuation characteristics; subtracting the aligned pusher transient current data from the reference current envelope to extract the current residual sequence; extracting the maximum amplitude of the current residual sequence as the current peak value within the effective clamping time window when the sleeve blank is pushed into the fixture, and subtracting it from the preset standard clamping resistance reference value of the blank material to obtain the current deviation characteristic; searching and matching the primary feed degradation ratio corresponding to the current deviation characteristic based on a pre-established mapping table of current deviation and tool cutting load; limiting the primary feed degradation ratio based on the preset fracture safety threshold of the first spindle assembled tool, and generating the initial feed rate offset command containing the limited degradation ratio for feedforward control.

[0030] During the calibration phase before machine tool production, the linear servo pusher mechanism is controlled to continuously run a preset number of pusher cycles at the standard pusher speed under no-load conditions without a sleeve blank. The bus current sequence of each cycle is collected synchronously, and the same phase is accumulated synchronously and the average value is calculated. After removing random noise, a reference current envelope reflecting the inherent frictional resistance of the guide rail is generated. Subsequently, a standard sleeve blank whose size and material meet the tolerance requirements is selected, and a preset number of reference pusher clamping measurements are performed. The peak value of the current residual after removing the basic mechanical friction background value in each measurement is extracted and the average value is calculated. This value is solidified and calibrated as the reference value of the clamping resistance of the standard size blank. The value of the safety approximation margin is set based on: extracting the maximum positive axial tolerance value from the blank's incoming material specification, adding the mechanical repeatability positioning error value of the pusher servo axis, and taking 1.2 to 1.5 times the sum of the two as the safety approximation margin; the moment when the indexing table fixture is fully locked and outputs a clamping confirmation signal is set as the end of the time window. In actual processing, the starting pulse for the forward movement of the pusher axis issued by the CNC system of the machine tool is used as the starting point for data acquisition and the signal of the absolute position of the target reached fed back by the pusher axis servo driver is used as the ending point for data acquisition. After acquiring the current pusher transient current data, a dynamic time warping algorithm is used to calculate the minimum cumulative distance matrix between the current pusher transient current data and the reference current envelope, using the Euclidean distance of the current amplitude as the metric standard, and to search for the optimal bending path. Based on the optimal bending path, the time axis of the current pusher transient current data is elastically stretched or compressed so that its peaks and troughs and other undulation features are strictly aligned with the reference current envelope in the time dimension.

[0031] The current transient current data of the pusher, aligned with the time axis, is algebraically subtracted from the reference current envelope at discrete time points to eliminate the background values ​​of basic mechanical friction, outputting a current residual sequence that purely characterizes the dynamic resistance of the pusher. An effective clamping time window is set based on the absolute position feedback coordinates of the machine tool pusher axis. The coordinate point where the pusher rod contacts the end face of the sleeve blank is set as the start point of the time window, and the moment when the indexing table fixture is fully locked and outputs a clamping confirmation signal is set as the end point of the time window. Within this effective clamping time window, the current residual sequence is traversed, and the largest positive amplitude is extracted as the current peak value within the pusher cycle. The current peak value is algebraically subtracted from the pre-calibrated standard-sized blank clamping resistance reference value; the resulting difference is the current deviation characteristic quantity characterizing the blank interference or hardness abrupt change.

[0032] A pre-established mapping table of current deviation and tool cutting load in the machine tool control system is extracted. This mapping table records the correspondence between different current deviation ranges and recommended feed rates calibrated through offline cutting force measurement experiments. The specific steps of the offline cutting force measurement experiment are as follows: sleeve blank samples with different interference dimensions and hardness gradients are used for pushing and trial cutting respectively. The cutting resistance is monitored in real time using a spindle force gauge. Under the premise of keeping the cutting resistance equal to the rated load of the tool, the pushing current deviation value and safe cutting feed rate corresponding to each sample are recorded, and they are compiled into the mapping table one by one. The calculated current deviation characteristic is used as an index and input into the mapping table. The corresponding primary feed degradation ratio is retrieved by interval matching. The fracture safety threshold preset for the tool assembled on the first spindle is extracted from the machine tool program. This threshold is defined as the lower limit of the limit feed rate corresponding to the minimum allowable feed per tooth at the current spindle speed. The retrieved primary feed rate reduction ratio is compared with the lower limit of the ultimate feed rate. If the primary feed rate reduction ratio is lower than the lower limit of the ultimate feed rate, the reduction ratio is forcibly truncated and assigned the value of the lower limit of the ultimate feed rate; if it is higher than or equal to the lower limit, the original value is retained. Finally, the reduction ratio, after numerical limiting verification, is encapsulated into a standard feed rate offset instruction and sent to the machine tool CNC system for feedforward execution before the first spindle begins cutting feed.

[0033] By dynamically normalizing and subtracting the reference current envelope, the background interference caused by the pure material pushing physical impedance and the inherent friction of the machine tool guideways is decoupled, thus restoring the true front-end clamping fluctuations. Combined with the fracture safety threshold limiting to generate feedforward commands, the first spindle adaptively reduces its speed based on the actual material resistance, smoothing out the transient mechanical shocks caused by material inhomogeneity in the initial machining stage and preventing the nonlinear physical transmission of vibration from a single station to subsequent indexing processes.

[0034] Further, the process of extracting the effective value of the spindle vibration acceleration data and generating the temperature rise gradient value by calculating the time derivative of the real-time temperature data includes: using a bandpass filter to isolate the acquired spindle vibration acceleration data by frequency band, wherein the lower cutoff frequency of the bandpass filter is set to the spindle tooth frequency and the upper cutoff frequency is set to the switching carrier frequency of the machine tool servo drive; within the range where the other spindles are in steady-state cutting feed, the effective frequency band data is dynamically truncated based on a sliding time window with a preset step size, and the root mean square of the vibration amplitude within each sliding time window is calculated as the effective value of the spindle vibration acceleration data; the acquired real-time temperature data is smoothed and reconstructed using a moving average filtering algorithm to obtain a smoothed temperature curve of the spindle; according to a preset macroscopic time interval, the smoothed temperature curve of the spindle is discretely differentially processed along the time axis, and the temperature rise increment within a unit time interval is extracted as the temperature rise gradient value.

[0035] Extract the spindle speed value set in the current CNC machining program of the machine tool, divide it by 60 to calculate the number of spindle revolutions per second, and then multiply it by the number of cutting edges of the assembled tool to obtain the spindle tooth pass frequency, which is used as the lower cutoff frequency of the bandpass filter. Read the pulse width modulation parameters of the machine tool servo driver and extract the switching carrier frequency value of the inverter power device, which is used as the upper cutoff frequency of the bandpass filter. Construct a digital bandpass filter based on the lower and upper cutoff frequencies. The digital bandpass filter is specifically a finite impulse response filter designed based on a Hamming window, and its filter order is set to 64th to 128th order according to the preset stopband attenuation requirements. Input the acquired spindle vibration acceleration data sequence into the digital bandpass filter for discrete convolution operation, truncate the signal amplitude outside the frequency range, and output the effective frequency band data sequence that retains only the frequency band between the fundamental frequency and the carrier frequency.

[0036] The interpolation status flag and actual feed rate feedback data of the machine tool CNC system are continuously read. When the interpolation status flag is determined to be in the cutting command segment of linear or circular interpolation, and the actual feed rate reaches the preset tolerance range of ±2% to ±5% of the set feed rate, the corresponding spindle is determined to have entered the steady-state cutting feed range. Within the steady-state cutting feed range, a sliding time window containing the preset sampling point length and step size is set. To ensure that feature extraction covers the complete mechanical excitation cycle, the time span of the sliding time window is forcibly constrained to be greater than or equal to the single rotation cycle of the spindle. That is, the set value of the sampling point length must be greater than or equal to the ratio of the high-frequency vibration sampling rate to the number of spindle rotations per second. At the same time, the step size is forcibly set to one-quarter to one-half of the sampling point length to ensure a data overlap rate of 50% to 75% between adjacent sliding time windows, preventing the omission of high-frequency transient flutter features at the window segmentation edge. The sliding time window is dynamically truncated along the time axis on the effective frequency band data sequence. For the discrete vibration amplitude sequence truncated within each sliding time window, the root mean square statistical algorithm is directly applied for calculation, and the calculated result is used as the effective value of the spindle vibration acceleration data within the current sliding time window.

[0037] The acquired real-time temperature data sequence is extracted, and a first-order moving average filtering algorithm is used. Based on a sliding data window with a preset sampling point length (the sampling point length is the product of a time window spanning 2 to 5 seconds and the low-frequency temperature sampling frequency), the mean value of the continuously backtracked original temperature sampling values ​​is iteratively calculated. The smoothly reconstructed temperature values ​​are output sequentially, and the main axis smooth temperature curve is generated by splicing them together according to the time sequence.

[0038] The complete cycle time of a single sleeve machining indexing action of the machine tool is extracted and fixed as a macroscopic time interval characterizing thermal inertia hysteresis. On the smoothed temperature curve of the spindle, the current smoothed temperature value corresponding to the current time node is extracted, and backtracking along the time axis, the previous smoothed temperature value before the macroscopic time interval is extracted. If the absolute time node calculated by backtracking does not exactly coincide with the discrete sampling timestamp of the smoothed temperature curve of the spindle, a linear interpolation algorithm is used to automatically extract the smoothed temperature values ​​of the two nearest adjacent sampling points to the absolute time node and perform time-weighted summation to obtain the numerically continuous previous smoothed temperature value. The current smoothed temperature value is algebraically subtracted from the previous smoothed temperature value, and the difference is divided by the macroscopic time interval to perform discrete difference operation to obtain the temperature rise increment within a unit time interval, that is, to generate the temperature rise gradient value.

[0039] By defining the filtering boundary using the machine tool spindle rotation fundamental frequency and the driver switching carrier frequency, low-frequency mechanical resonance and high-frequency electrical harmonics are precisely isolated, extracting the effective vibration energy characterizing actual cutting chatter. Combining smooth reconstruction and macroscopic time difference, the amplification and distortion of minute temperature measurement noise during derivative calculations are avoided. This signal cleaning logic removes random noise interference from the highly dynamic machining environment, ensuring that the extracted alternating load and thermal drift characteristics strictly correspond to the actual physical deterioration state of the machine tool.

[0040] Furthermore, the process of outputting the comprehensive error severity scalar includes: using the table positioning angle deviation data as a spatial geometric deviation factor, using the extracted effective value as a transient alternating load factor, and using the generated temperature rise gradient value as a thermal inertia drift factor; using a membership function based on the machine tool factory precision acceptance specification and the physical limit calibration of tool wear, mapping the spatial geometric deviation factor, transient alternating load factor, and thermal inertia drift factor to their respective fuzzy domains; establishing a fuzzy rule base based on the multi-field coupled physical failure mechanism, which contains a three-dimensional basic mapping matrix, and its benchmark mapping criterion is when the values ​​of the transient alternating load factor and the thermal inertia drift factor... When the output of the comprehensive error severity scalar tends towards the positive maximum domain, it shifts towards the positive maximum domain. Based on the benchmark mapping criterion, the fuzzy rule base is configured with nonlinear penalty rules: when it is determined that the thermal inertia drift factor is in a preset thermal expansion amplification range, and the spatial geometric deviation factor deviates from the benchmark zero range, a thermal deformation amplification effect determination is triggered, and a cross-domain coupling penalty mechanism is executed to nonlinearly increase the penalty weight of the corresponding rule. The area centroid method is used to perform defuzzification feature dimensionality reduction calculation on the reasoning results of all triggered rules in the fuzzy rule base, generating a normalized value as the comprehensive error severity scalar for the current indexing station.

[0041] Extract the table positioning angle deviation data, retain its positive and negative signs, and directly assign it as the spatial geometric deviation factor; extract the effective value of the spindle vibration acceleration data and assign it as the transient alternating load factor; extract the generated temperature rise gradient value and assign it as the thermal inertia drift factor. Referencing the machine tool's factory precision acceptance specifications, extract the maximum allowable angular displacement deviation extreme value of the indexing table to construct the first physical boundary; based on the fatigue wear physical curve of the assembled tool material, extract the critical point of the effective vibration value that induces chipping to construct the second physical boundary; extract the maximum allowable steady-state temperature rise slope of the spindle bearing design to construct the third physical boundary. Using the above three physical boundaries as reference scaling factors, perform full-range normalization division processing on the spatial geometric deviation factor, transient alternating load factor, and thermal inertia drift factor respectively, so that their numerical ranges are uniformly mapped to the standard fuzzy domain of the interval [-1, 1]. The specific full-range normalization equation is: ,in, The original values ​​of the input factors. This corresponds to the physical boundary extreme value.

[0042] Within the standard fuzzy domain, triangular membership functions covering discrete linguistic variables (the set of discrete linguistic variables is specifically divided into five fuzzy subsets: negative large NB, negative small NS, zero ZE, positive small PS, and positive large PB) are deployed for each of the three factors. The normalized values ​​of each factor are substituted into the triangular membership function equation to calculate the corresponding membership activation probability value.

[0043] Within the standard fuzzy domain, triangular membership functions are assigned to three factors, each encompassing discrete linguistic variables (specifically divided into five fuzzy subsets: negative large NB, negative small NS, zero ZE, positive small PS, and positive large PB). The general piecewise algebraic equation for the triangular membership function is defined as follows: in, These are the normalized factor values. b These are the coordinates of the left base, vertex, and right base of the triangular membership function, respectively. For the five fuzzy subsets, their corresponding vertex coordinates b are arranged equidistantly along the domain interval [-1, 1] as {-1, -0.5, 0, 0.5, 1}, and the base coordinates of adjacent fuzzy subsets are forcibly set to overlap at the vertices of adjacent subsets. The normalized factor values ​​are substituted into the triangular membership function equation of the corresponding boundary parameters to calculate the corresponding membership activation probability values.

[0044] A fuzzy rule base matrix based on the multi-field coupled physical failure mechanism is established. This matrix contains complete three-dimensional mapping logic. To clarify the benchmark mapping criteria, typical boundary condition mapping rules are listed below: Rule 1: The transient alternating load factor (IF) is PB, the thermal inertia drift factor (PB) is PB, and the spatial geometric deviation factor (PB) is PB; the comprehensive error severity scalar (THEN) is PB. Rule 2: The transient alternating load factor is PB, the thermal inertia drift factor is ZE, and the spatial geometric deviation factor is ZE. The comprehensive error severity scalar is PS.

[0045] When running the above basic mapping criteria, the activated antecedents of the rules in the current period are scanned one by one. For the i-th triggered rule, the minimum value of the membership degree of each factor participating in the rule combination is extracted using the minimum value algorithm, and this value is used as the initial inference antecedent for activation. Its mathematical formula is: in, , , These are the membership values ​​of the current input variable in the corresponding fuzzy subset. Furthermore, the fuzzy rule base is configured with a cross-domain coupling nonlinear penalty rule: when it is determined that the input of the thermal inertia drift factor is in a preset thermal expansion amplification range (i.e., the linguistic variable with the highest membership is positive PB), and simultaneously it is determined that the spatial geometric deviation factor crosses the zero range and is in a large deviation range (i.e., the linguistic variable with the highest membership is positive PB or negative NB), it is confirmed that the working condition combination triggering condition is met. At this time, a preset nonlinear penalty constant is forcibly introduced. (Its set value range is 1.2 to 1.5), for the activation degree of the preliminary inference antecedent. Perform a power function amplification operation based on the exponent to obtain the actual output weight of the consequent of the rule. The specific nonlinear penalty algebraic equation is as follows: The nonlinear penalty constant The setting is based on the physical response boundary of the machine tool servo drive: if <1.2, the penalty amplification weight increment is too weak to overcome the mechanical inertia and static friction of the feed transmission chain, easily leading to undercompensation; if >1.5, excessively large transient step changes in weight increments can easily lead to overload of the underlying servo drive and induce high-frequency resonance of the spindle and surface tool marks. Ultimately, the result after the above nonlinear amplification penalty will be affected. The actual output weights, which are the consequents of this rule, are mapped to the output.

[0046] For all triggered rules in the fuzzy rule base, the maximum-minimum synthesis method is used to maximize the independent output membership function curves obtained from the consequents of each rule, generating a global polygonal envelope contour equation over the output domain. The output variable The components are distributed within a predefined normalized universe of discourse [0, 1]. Within the global integral boundary of the normalized universe of discourse, a numerical integration operation using the area centroid method is performed on the overall polygonal envelope contour equation to obtain the absolute coordinates of the integral centroid. Its discretized integral equation in the continuous domain is: in, The total number of discrete nodes in the output domain [0, 1] divided by a preset sampling step size (the sampling step size is set to 0.005 to 0.01, i.e., the total number of discrete nodes n is forcibly constrained to be between 100 and 200 to match the computation cycle time of the underlying microprocessor of the CNC system), Let be the coordinates of the universe of discourse of the i-th discrete node. The membership value of the global polygon envelope contour corresponding to the discrete node; the absolute coordinates of the centroid obtained by numerical integration. The normalized absolute value generated is truncated to a preset decimal place and then latched and issued as the comprehensive error severity scalar for the current indexing station.

[0047] Multi-source data is mapped to spatial geometric deviation, alternating load, and thermal inertia drift factor, and a cross-domain coupled nonlinear penalty rule is constructed. When thermal expansion intensifies and indexing deviation occurs simultaneously, the system adaptively increases the rule weight, truly mapping the physical amplification law of thermodynamic distortion on spatial geometric errors, and avoiding the evaluation lag and insufficient compensation problems that often occur when multiple physics fields interact under complex conditions.

[0048] Furthermore, the process of issuing and executing cutting feedback adjustment includes: extracting a multi-dimensional control transformation matrix pre-stored in the machine tool CNC system, wherein the multi-dimensional control transformation matrix is ​​constructed from radial dimension sensitivity weights and cutting heat sensitivity weights for different machining processes of each of the other spindles; performing a dimension-expanding mapping calculation on the comprehensive error severity scalar and the multi-dimensional control transformation matrix to generate a comprehensive control increment vector containing independent correction coefficients for each of the other spindles; and performing dimensional decoupling on the comprehensive control increment vector to extract the original values ​​corresponding to each of the other spindles. The initial radial tool compensation value and the original speed offset value are calculated. The initial radial tool compensation value and the original speed offset value are compared with the preset maximum tool wear compensation extreme value and the spindle safe speed boundary, respectively. Anti-collision soft limiting truncation processing is performed to generate the radial tool compensation fine adjustment amount and the speed offset amount after safety verification. The radial tool compensation fine adjustment amount and the speed offset amount after safety verification are encapsulated into a control message conforming to the machine tool's underlying communication bus protocol and sent to the corresponding servo driver according to the preset smooth interpolation cycle to control the other spindles to perform progressive compensation during continuous cutting.

[0049] A multidimensional control transformation matrix, pre-calibrated based on offline process trial cutting and stored in the machine tool CNC system, is extracted. This multidimensional control transformation matrix is ​​a 2×k-dimensional matrix, where k is the total number of other spindles currently participating in the linkage of the machine tool. The first row of the matrix contains the maximum physical compensation reference value (in millimeters or micrometers) of the corresponding radial dimension pre-calibrated for each of the other spindles in different machining processes. The second row contains the maximum offset reference value of the corresponding rotational speed caused by cutting heat (in revolutions per minute). The numerical form of the comprehensive error severity scalar is then subjected to scalar-matrix multiplication and dimension expansion mapping calculation with the 2×k-dimensional multidimensional control transformation matrix. That is, the scalar value is multiplied by each reference value element in the matrix to generate a comprehensive control increment vector matrix with real physical dimensions, which is also 2×k-dimensional.

[0050] The integrated control incremental vector matrix is ​​decoupled by row dimension: each column element in the first row of the matrix is ​​extracted sequentially as the original radial tool compensation value for each of the other spindles; each column element in the second row of the matrix is ​​extracted sequentially as the original speed offset value for each of the other spindles. Extract the maximum tool wear compensation extreme value preset for each station tool in the machine tool tool management system, as well as the upper and lower boundaries of the safe speed corresponding to each spindle motor. For each of the remaining spindles, a conditional branch logic is used to perform a soft-limiting truncation process for the anti-collision mechanism: if the absolute value of the corresponding original radial tool compensation value is greater than the maximum tool wear compensation extreme value, it is forcibly assigned the maximum tool wear compensation extreme value with the original positive and negative signs; for the speed offset, firstly, the current actual operating speed of the corresponding spindle is read in real time, and the actual operating speed is algebraically added to the original speed offset value to obtain the expected absolute speed; if the expected absolute speed exceeds the upper or lower boundary of the safe speed, the expected absolute speed is forcibly truncated to the corresponding safe speed boundary limit value, and the current actual operating speed is subtracted from the limit value to solve the speed offset amount after safety verification in reverse; thus, the radial tool compensation fine-tuning amount and the speed offset amount after safety verification are generated.

[0051] The radial tool compensation fine-tuning amount and the speed offset amount, after safety verification, are mapped and encapsulated according to the message frame format of the machine tool's underlying industrial Ethernet communication bus to generate a control message. The smooth interpolation cycle span set by the machine tool CNC system is obtained and divided by the single-step control cycle time of the underlying servo communication to calculate the total number of interpolation subdivision steps N. The encapsulated radial tool compensation fine-tuning amount and the speed offset amount are then divided by the total number of interpolation subdivision steps N to calculate the minute compensation step amount within a single servo communication cycle. During the continuous cutting process of controlling the remaining spindles, a discrete time sequence is set. Discrete recursive equations are used based on a single servo communication cycle. The accumulated compensation setpoint is periodically sent to the corresponding servo driver (where... The compensation setpoint is the same as the one used in the previous cycle. The step size is the small compensation step size, until the step size of the Nth cycle is issued, thereby controlling the remaining spindles to perform smooth, stepless progressive compensation.

[0052] By utilizing a multi-dimensional transformation matrix incorporating dimensional and cutting thermal sensitivity, a single error scalar is decoupled into independent tool compensation and speed correction quantities that conform to the actual state of each process. Soft-limiting verification is performed by combining wear extreme values ​​and speed boundaries, eliminating the safety hazard of machine tool collisions induced by intervention commands. By sending control messages through a smooth interpolation cycle, the servo drive performs progressive state fine-tuning, avoiding step tool marks defects on the sleeve surface caused by instantaneous compensation commands during cutting.

[0053] Furthermore, the process of obtaining the weights in the multidimensional control transformation matrix includes: during the trial cutting phase of the machine tool system, keeping other machining parameters constant, sequentially applying a unit Z-axis feed rate step excitation to each of the remaining spindles individually, measuring the physical change in the actual radial expansion deviation of the inner hole of the test piece, calculating the algebraic ratio of the physical change to the unit Z-axis feed rate step excitation and retaining its positive and negative signs, and calibrating it as the feed rate size sensitivity weight of the corresponding spindle; sequentially applying a unit spindle feed rate step excitation to each of the remaining spindles individually. A step excitation of spindle speed is applied, and the Z-axis feed rate of the corresponding spindle is increased proportionally to lock the feed per tooth. The steady-state temperature rise increment when the corresponding spindle reaches thermal equilibrium is measured. The algebraic ratio of the steady-state temperature rise increment to the unit spindle speed step excitation is calculated, and its positive or negative sign is retained. This ratio is calibrated as the cutting heat sensitivity weight of the corresponding spindle. The feed rate size sensitivity weights of the other spindles obtained from the calibration are then matrix-concatenated with the cutting heat sensitivity weights to construct the multidimensional control transformation matrix. The specific process is as follows: Figure 2 As shown.

[0054] During the trial cutting phase before the machine tool system is put into actual production, a process reference state is first established. All spindles of the machine tool are set to the reference spindle speed and reference Z-axis feed speed specified in the process documents. A preset number of standard sleeve test pieces are continuously machined. The actual inner diameter of each standard sleeve test piece is measured using a coordinate measuring machine. The actual inner diameter is algebraically subtracted from the theoretical nominal diameter, and the arithmetic mean is obtained. This arithmetic mean is used as the reference inner diameter enlargement deviation with positive and negative spatial directions. At the same time, the machine is continuously run under the reference machining state, and the reference steady-state temperature value when the surface of each spindle bearing seat reaches thermal equilibrium is recorded.

[0055] When obtaining the feed rate dimensional sensitivity weight, the machining parameters of other spindles of the machine tool are kept completely constant at the reference state. A preset amplitude Z-axis feed rate step excitation is applied to the i-th spindle currently participating in the linkage. The amplitude of the Z-axis feed rate step excitation is set to 5% to 15% of the reference Z-axis feed rate. After applying this feed rate step excitation, a preset batch of test sleeve specimens are continuously machined. The actual radial enlargement deviation of the inner hole of each test sleeve specimen is extracted using a coordinate measuring machine and the arithmetic mean is calculated. The average actual radial enlargement deviation of the inner hole is algebraically subtracted from the reference inner hole enlargement deviation to obtain the radial enlargement physical change induced by the change in the Z-axis feed resistance of a single spindle. The physical change is retained in its positive and negative spatial directions and directly used as the original dimensional sensitivity weight of the i-th spindle under the ultimate excitation. Determine whether the absolute value of the original size sensitivity weight is greater than a preset rigid dead zone threshold. The preset rigid dead zone threshold is set as the minimum physical measurement resolution of the corresponding coordinate measuring machine head. This threshold serves as a minimum value truncation boundary to prevent division by zero overflow when the underlying microprocessor performs inversion operations and to physically filter out interference from background measurement noise. If the value is less than the threshold, the weight of the spindle is forcibly set to zero to prevent arithmetic overflow. If the value is greater than the threshold, the original size sensitivity weight is directly fixed and calibrated as the feed size sensitivity weight of the i-th other spindle. This step is performed iteratively to obtain the feed size sensitivity weights of all k other spindles.

[0056] After completing the above independent univariate isolation test, extract the k feed rate size sensitivity weights obtained from calibration, arrange them in the order of the spindle physical channel number to construct a horizontal one-dimensional row vector, which serves as the first row of the matrix; extract the k cutting heat sensitivity weights obtained from calibration, arrange them in the same order of the spindle physical channel number to construct another horizontal one-dimensional row vector, which serves as the second row of the matrix, so that the elements of the second row of the multidimensional control conversion matrix are unified as the maximum offset reference value of the rotational speed. The first row vector and the second row vector are concatenated in the column direction to generate a 2×k-dimensional real number matrix with a definite physical calibration mapping relationship and real control dimensions, that is, to construct and generate the multidimensional control conversion matrix, and store it in the parameter area of ​​the machine tool CNC system.

[0057] A step excitation test with unit tool compensation and unit speed was used, and the sensitivity weight of the transformation matrix was calibrated based on the absolute ratio of the system's physical change to the steady-state temperature rise increment. This calibration process directly measured the true error transmission coefficient of each axis response to external intervention for a specific machine tool, abandoning the static matrix set by manual experience, and ensuring that the decoupling allocation mechanism of multi-axis commands during operation maintains a high degree of consistency with the actual cutting physical transmission characteristics.

[0058] Further, the process of global calibration in the benchmark value superimposed on the comprehensive state fuzzy evaluation model includes: extracting the cylindricity deviation data of a preset number of consecutive batches before the current processing is completed, and constructing the time series based on a sliding data window; calculating the arithmetic mean of the difference between the cylindricity deviation data of adjacent batches within the sliding data window divided by the adjacent processing time interval, as the first-order difference slope; comparing the absolute value of the first-order difference slope with a preset drift tolerance threshold: if the absolute value of the first-order difference slope is less than or equal to the drift tolerance threshold, it is determined that the current processing error is in the normal fluctuation range, and the boundary offset is zero; if the absolute value of the first-order difference slope is greater than the drift tolerance threshold, the first-order difference slope is multiplied by a preset macroscopic compensation gain and the average time span constant of the current sliding data window to calculate the boundary offset; the boundary offset is directly superimposed on the coordinates of the center point of the membership function of the spatial geometric deviation factor in the input layer of the comprehensive state fuzzy evaluation model to complete the translation calibration of the model judgment benchmark, the specific process is as follows. Figure 3 As shown.

[0059] Extract the cylindricity deviation data from a preset number of consecutive batches (with the sliding data window sequence length set to 5 to 10 consecutive machining cycles) before the current machining is completed, and sort them according to their machining time to construct a time series based on the sliding data window. For this time series, a backward difference discretization algorithm is used to calculate the algebraic difference between the cylindricity deviation data of each adjacent batch within the sliding data window, and the actual absolute machining time interval of the corresponding adjacent batches is extracted; the algebraic difference is divided by the actual absolute machining time interval to obtain the corresponding time partial derivative. The calculated time partial derivative sequences are summed and the arithmetic mean is calculated, which is used as the first-order difference slope characterizing the macroscopic deterioration rate of the current machine tool spatial geometric error.

[0060] Extract the pre-set drift tolerance threshold of the machine tool system. This threshold is calibrated based on the steady-state thermal expansion reference rate induced by natural temperature fluctuations in the workshop environment where the machine tool is located. Compare the absolute value of the first-order difference slope with the drift tolerance threshold: if the absolute value of the first-order difference slope is less than or equal to the drift tolerance threshold, it is determined that the current machining error of the machine tool is within the controlled thermal equilibrium normal fluctuation range, and the boundary offset is forcibly generated to be zero.

[0061] If the absolute value of the first-order difference slope is greater than the drift tolerance threshold, the first-order difference slope is multiplied by a preset macroscopic compensation gain and the average time span constant of the current sliding data window to calculate the boundary offset. The specific calculation equation is as follows: in, This is the boundary offset. The first-order difference slope, The average time span constant, The macroscopic compensation gain is a preset value, and the equation for calculating the macroscopic compensation gain is as follows: in, This represents the physical extreme value of the maximum permissible geometric tolerance of the machine tool. A dimensionless calibration constant is introduced to eliminate the time dimension, ensuring that the boundary offset generated by the final product is a dimensionless normalized pure value that can be directly mapped to the domain interval [-1, 1]. The dimensionless calibration constant... Aimed at balancing the tracking response speed of the machine tool system to long-cycle environmental drift and the steady-state anti-vibration capability of the control system, the constant is set to a value ranging from 0.05 to 0.25. During the calibration phase before the machine tool is put into production, this constant... The specific value is determined through a step response test: set an initial small value. The value is tested and increased gradually, and the cylindricity compensation feedback curve of the sleeve is monitored in real time. When the compensation feedback curve first shows over-compensation and overshoot oscillation, the critical constant value at this time is recorded, and 0.5 to 0.7 times the critical constant value is taken as the final solidified dimensionless calibration constant. .

[0062] To prevent the coordinate topology of the fuzzy domain from collapsing, the generated boundary offset is input into a preset saturation limiter: when the absolute value of the boundary offset is greater than the preset maximum allowable offset threshold (which is set to be strictly less than the original distance between the center points of adjacent membership functions), it is forcibly truncated to the maximum allowable offset threshold with the original sign.

[0063] Extract the original center point coordinate sequence of the five discrete membership functions (NB, NS, ZE, PS, PB) corresponding to the spatial geometric deviation factor in the input layer of the comprehensive state fuzzy evaluation model. The boundary offset, after amplitude limiting verification, is algebraically added to the original center point coordinate sequences of the five discrete membership functions and their corresponding left and right bottom point coordinate sequences. This achieves synchronous rigid translation of the geometric envelope of the entire membership function, and the translated new coordinate sequences are all updated and written into the fuzzy rule base, thereby completing the adaptive translation calibration of the model judgment benchmark.

[0064] A sliding data window is used to obtain the cylindricity difference slope for consecutive batches, and a system drift tolerance threshold is introduced as the dead zone determination boundary. This macroscopic calibration logic triggers the reference translation only when the long-term error trend exceeds the physical dead zone, effectively filtering out interference from single-sample outliers caused by probe contamination or accidental chip inclusions in the workshop. This anti-jitter mechanism prevents the control system from experiencing high-frequency oscillations and overshooting under dynamic operating conditions, thereby locking the global machining reference in the long-cycle batch dimension.

[0065] Example 2: In this machining scenario, multiple vertical spindles on a multi-station special machine are equipped with fixed-size reamers, which perform vertical downward feed internal hole reaming on the sleeve blank on the fixture along the Z-axis.

[0066] During continuous cutting, the system synchronously collects multi-source data from the machine tool: it uses a digital bandpass filter to perform discrete convolution filtering on the high-frequency vibration data of the spindle, extracts the effective vibration sequence and calculates the root mean square effective value to generate a transient alternating load factor; it uses a surface temperature sensor to collect the spindle temperature, and through first-order moving average filtering and backward difference operation, it generates the temperature rise gradient value under the current machining cycle, which serves as the thermal inertia drift factor; at the same time, it extracts the actual inner hole radial dimension deviation of the lower sleeve test piece and retains its positive and negative spatial signs as a spatial geometric deviation factor. The system introduces the pre-calibrated physical boundary tolerance of the machine tool as a reference scaling factor, and normalizes and maps the above three factors to the standard fuzzy universe of discourse [-1, 1].

[0067] The normalized factors are input into the initialized comprehensive state fuzzy evaluation model. When the system determines that the membership degree of the current thermal inertia drift factor is in the "positive" range and the spatial geometric deviation factor is in the "large deviation" range, the corresponding rule antecedent in the fuzzy rule base is triggered. The system uses the minimum operation rule to calculate the activation degree and introduces a preset nonlinear penalty constant to perform a power function operation of the sub-exponential to obtain the consequent weight after amplification penalty. Subsequently, the maximum-minimum synthesis method and the area centroid method are used to perform defuzzification numerical integration to output the comprehensive error severity scalar of the current indexing station.

[0068] The system retrieves a multidimensional control transformation matrix stored in the underlying parameter area. This matrix contains the spindle feed size sensitivity weight and cutting heat sensitivity weight derived from single-variable isolated trial cutting calibration. The system performs a scalar multiplication and dimension expansion mapping between the comprehensive error severity scalar and this matrix, decoupling the corresponding original feed rate offset value and original speed offset value. After reading the actual operating speed and performing soft-limiting truncation with a preset safety limit, the system generates the safety-verified feed rate fine-tuning amount and speed offset amount. Finally, the CNC system issues the small compensation step amount according to the interpolation subdivision cycle, smoothly reducing the Z-axis feed rate to decrease axial cutting resistance, and simultaneously adjusting the spindle speed to achieve progressive online compensation that effectively suppresses the internal hole enlargement and spindle thermal distortion.

[0069] To address environmental baseline drift caused by long-term, high-volume production, the system periodically extracts the time series of inner hole diameter deviations from consecutive batches within a sliding data window and calculates their average first-order difference slope. If the absolute value of this slope exceeds the natural drift tolerance threshold calibrated under no-load conditions, the system multiplies it by a macroscopic compensation gain containing the reciprocal of the tolerance extremum to generate a boundary offset. After being truncated and limited by a preset maximum allowable offset threshold, this boundary offset is algebraically added sequentially to the original coordinate sequence of the membership function in the fuzzy model, achieving a rigid spatial translation of the model envelope, thereby completing the adaptive global calibration of the judgment benchmark based on environmental temperature drift.

[0070] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation, characterized in that, include: Acquire transient current data during the material feeding process, table positioning angle deviation data, spindle vibration acceleration data, real-time temperature data, and cylindricity deviation data of the unloading sleeve; Extract the peak current value of the transient current data during the feeding cycle, calculate the difference between it and the preset reference value to obtain the current deviation feature, and generate the first initial feed rate offset command of the spindle based on the current deviation feature for feedforward control. The effective value of the spindle vibration acceleration data is extracted, and the time derivative of the real-time temperature data is calculated to generate the temperature rise gradient value. This value is then synchronously input into the comprehensive state fuzzy evaluation model along with the table positioning angle deviation data. The comprehensive state fuzzy evaluation model infers and outputs the comprehensive error severity scalar, which is then multiplied by the transformation matrix to parse out the radial tool compensation fine adjustment amount and speed offset amount of the remaining spindles and send them down to execute the cutting feedback adjustment. Extract the cylindricity deviation data of consecutive batches to construct a sliding data window time series, calculate the arithmetic mean of the difference between the cylindricity deviation data of adjacent batches divided by the adjacent processing time interval to obtain the first-order difference slope, generate the boundary offset based on the first-order difference slope, and superimpose it on the benchmark value of the comprehensive state fuzzy evaluation model for global calibration.

2. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of acquiring various data during the processing includes: obtaining transient current data of the sleeve blank during feeding by sampling at a preset sampling period using a current sensor connected to the DC bus of the servo driver of the linear servo pusher mechanism; obtaining the table positioning angle deviation data by reading the angle difference between the actual angle and the theoretical position during each indexing action locking using a high-precision circular grating installed on the rotating shaft end of the multi-station rotary indexing table; obtaining the spindle vibration acceleration data during continuous cutting by using a triaxial piezoelectric accelerometer bolted to the surface of the bearing seat at the front end of the vertical spindle; obtaining the real-time temperature data by using a thermal resistance sensor attached to the surface of the vertical spindle housing; and extracting the spatial point cloud data of the outer cylindrical surface of the finished sleeve sampled by the three-coordinate measuring machine based on the contact scanning probe along the lower line, and calculating the cylindricity deviation data using a least-squares cylindrical fitting algorithm.

3. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of generating the initial feed rate offset command for the first spindle based on the current deviation characteristic includes: acquiring a pre-calibrated reference current envelope of the linear servo pusher mechanism under no-load operation, and dynamically time-aligning the acquired pusher transient current data with the reference current envelope based on waveform fluctuation characteristics; subtracting the aligned pusher transient current data from the reference current envelope to extract the current residual sequence; extracting the maximum amplitude of the current residual sequence as the current peak value within the effective clamping time window when the sleeve blank is pushed into the fixture, and subtracting it from the preset standard clamping resistance reference value of the blank material to obtain the current deviation characteristic; searching and matching the primary feed degradation ratio corresponding to the current deviation characteristic based on a pre-established mapping table of current deviation and tool cutting load; limiting the primary feed degradation ratio based on the preset fracture safety threshold of the first spindle tool assembly, and generating the initial feed rate offset command containing the limited degradation ratio for feedforward control.

4. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of extracting the effective value of the spindle vibration acceleration data and generating the temperature rise gradient value by calculating the time derivative of the real-time temperature data includes: using a bandpass filter to isolate the acquired spindle vibration acceleration data by frequency band, wherein the lower cutoff frequency of the bandpass filter is set to the spindle tooth frequency and the upper cutoff frequency is set to the switching carrier frequency of the machine tool servo drive; within the range where the other spindles are in steady-state cutting feed, the effective frequency band data is dynamically truncated based on a sliding time window with a preset step size, and the root mean square of the vibration amplitude within each sliding time window is calculated as the effective value of the spindle vibration acceleration data; the acquired real-time temperature data is smoothed and reconstructed using a moving average filtering algorithm to obtain a smoothed spindle temperature curve; according to a preset macroscopic time interval, the smoothed spindle temperature curve is discretized along the time axis, and the temperature rise increment within a unit time interval is extracted as the temperature rise gradient value.

5. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of outputting the comprehensive error severity scalar includes: using the table positioning angle deviation data as the spatial geometric deviation factor, using the extracted effective value as the transient alternating load factor, and using the generated temperature rise gradient value as the thermal inertia drift factor; using a membership function based on the machine tool factory precision acceptance specification and the physical limit calibration of tool wear, mapping the spatial geometric deviation factor, transient alternating load factor, and thermal inertia drift factor to their respective fuzzy domains; establishing a fuzzy rule base based on the multi-field coupled physical failure mechanism, which contains a three-dimensional basic mapping matrix, and its benchmark mapping criterion is that when the values ​​of the transient alternating load factor and the thermal inertia drift factor tend to be When the positive maximum domain is reached, the output comprehensive error severity scalar is shifted towards the positive maximum domain. Based on the benchmark mapping criterion, the fuzzy rule base is configured with nonlinear penalty rules: when it is determined that the thermal inertia drift factor is in a preset thermal expansion amplification range, and the spatial geometric deviation factor deviates from the benchmark zero range, a thermal deformation amplification effect determination is triggered, and a cross-domain coupling penalty mechanism is executed to nonlinearly increase the penalty weight of the corresponding rule. The area centroid method is used to perform defuzzification feature dimensionality reduction calculation on the inference results of all triggered rules in the fuzzy rule base, generating a normalized value, which serves as the comprehensive error severity scalar for the current indexing station.

6. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of issuing and executing cutting feedback adjustment includes: extracting a multi-dimensional control transformation matrix pre-stored in the machine tool CNC system, wherein the multi-dimensional control transformation matrix is ​​constructed from radial dimension sensitivity weights and cutting heat sensitivity weights for different machining processes of each of the other spindles; performing a dimension-expanding mapping calculation on the comprehensive error severity scalar and the multi-dimensional control transformation matrix to generate a comprehensive control increment vector containing independent correction coefficients for each of the other spindles; and decoupling the comprehensive control increment vector in terms of dimension to extract the original radial dimension sensitivity weights corresponding to each of the other spindles. The original radial tool compensation value and the original speed offset value are compared with the preset maximum tool wear compensation extreme value and the spindle safe speed boundary, respectively. The anti-collision machine soft limiting truncation process is performed to generate the radial tool compensation fine adjustment amount and the speed offset amount after safety verification. The radial tool compensation fine adjustment amount and the speed offset amount after safety verification are encapsulated into a control message conforming to the machine tool's underlying communication bus protocol and sent to the corresponding servo driver according to the preset smooth interpolation cycle to control the other spindles to perform progressive compensation during continuous cutting.

7. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 6, characterized in that, The process of obtaining the weights in the multidimensional control transformation matrix includes: during the trial cutting stage of the machine tool system, keeping other machining parameters unchanged, applying a preset limit Z-axis feed speed excitation to each of the remaining spindles in turn, measuring the physical change in the actual radial expansion deviation of the inner hole of the test piece, and directly labeling the physical change with positive and negative signs as the feed amount size sensitivity weight of the corresponding spindle; applying a preset limit spindle speed excitation to each of the remaining spindles in turn, and simultaneously increasing the Z-axis feed speed of the corresponding spindle proportionally to lock the feed per tooth, measuring the steady-state temperature rise increment when the corresponding spindle reaches thermal equilibrium steady state, and calculating the maximum speed offset required by the corresponding spindle to offset thermal distortion based on the temperature rise increment and the spindle safety speed boundary, retaining its positive and negative signs, and labeling it as the cutting thermal sensitivity weight of the corresponding spindle; and performing matrix splicing of the feed amount size sensitivity weight and the cutting thermal sensitivity weight of the corresponding remaining spindles to construct the multidimensional control transformation matrix.

8. The method for dynamic feedback adjustment of sleeve machining error based on fuzzy logic operation according to claim 1, characterized in that, The process of global calibration of the benchmark value superimposed on the comprehensive state fuzzy evaluation model includes: extracting the cylindricity deviation data of a preset number of consecutive batches before the current processing is completed, and constructing the time series based on a sliding data window; calculating the arithmetic mean of the difference between the cylindricity deviation data of adjacent batches within the sliding data window divided by the adjacent processing time interval, as the first-order difference slope; comparing the absolute value of the first-order difference slope with a preset drift tolerance threshold: if the absolute value of the first-order difference slope is less than or equal to the drift tolerance threshold, it is determined that the current processing error is in the normal fluctuation range, and the boundary offset is zero; if the absolute value of the first-order difference slope is greater than the drift tolerance threshold, the first-order difference slope is multiplied by a preset macroscopic compensation gain and the average time span constant of the current sliding data window to calculate the boundary offset; and directly superimposing the boundary offset onto the coordinates of the center point of the membership function of the spatial geometric deviation factor in the input layer of the comprehensive state fuzzy evaluation model to complete the translation calibration of the model judgment benchmark.

Images