CN-121979103-A - Ultra-precise machining error compensation method, medium and device based on multi-mode information fusion

CN121979103ACN 121979103 ACN121979103 ACN 121979103ACN-121979103-A

Abstract

The invention relates to the technical field of ultra-precise intelligent manufacturing, in particular to an ultra-precise machining error compensation method, medium and equipment based on multi-mode information fusion. According to the method, three-dimensional shape data of a workpiece are obtained through an in-situ measurement device, temperature field and vibration data are synchronously collected, a multi-mode space-time tensor is constructed, a physical information neural network is input to decouple static geometric errors, time-varying thermal drift errors and dynamic vibration errors, and a four-dimensional correction tool path is generated to realize reverse compensation. According to the invention, physical constraint is embedded through PINN models, so that the problem of false thermal drift misjudgment in the traditional method is solved, experiments show that the RMS error in free-form surface processing is reduced to below 5nm, the precision is improved by more than 70%, meanwhile, error decoupling physical interpretability is given, the generalization capability of a small sample is enhanced, and a technical basis is provided for process optimization.

Inventors

XU YING
ZHU BEIBEI
LUO CUI
QIN LIN
ZHANG QISHENG
LI BO
HUANG WEN
LI YUAN

Assignees

上海航天控制技术研究所

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (10)

1. The ultra-precise machining error compensation method based on multi-mode information fusion is characterized by comprising the following steps of: in the initial processing process, three-dimensional shape data of the surface of a workpiece are obtained by utilizing an in-situ measurement device integrated in the working space of the ultra-precise machine tool; The method comprises the steps that temperature field data and vibration data in an initial processing process are synchronously collected through distributed temperature sensors and dynamic vibration sensors which are arranged on a heat source part and a moving part of a machine tool; Aligning the three-dimensional morphology data, the temperature field data and the vibration data according to time stamps to construct a multi-mode space-time tensor; Inputting the multi-modal space-time tensor into a physical information neural network to decouple and separate the total processing error into a static geometric error, a time-varying thermal drift error and a dynamic vibration error, wherein the physical information neural network has a composite loss function The following conditions are satisfied: ; Wherein: outputting the mean square error of the total error and the measured data for the network; The violation degree of the thermal error component output by the network to the thermal conduction equation is calculated by utilizing automatic differentiation; the violation degree of dynamic error components output by the network to the forced vibration equation is calculated by utilizing automatic differentiation; weight coefficients that are physical constraints; based on the static geometry errors, a four-dimensional modified tool path is generated comprising spatial coordinates (x, y, z) and machining time t to spatially reverse compensate for the static geometry errors.
2. The method of claim 1, wherein after generating a four-dimensional modified tool path comprising spatial coordinates (x, y, z) and a machining time t, the method further comprises: adopting a power spectrum density analysis method to verify whether the corrected tool path is within the dynamic capacity range of the corresponding machine tool shaft; if the corrected tool path has the condition that the machine tool axis is not feasible dynamically, correcting the corrected tool path based on the optimizer of the jerk constraint; the acceleration constraint optimizer satisfies the following conditions: ; Constraint conditions: ; Wherein, the Correcting the cutter path after optimization; Correcting the tool path before optimizing; servo bandwidth limits for corresponding machine tool axes in a machine tool system; The third derivative of the tool path in the displacement direction of the machine tool shaft after optimization, namely jerk, is used for correcting.
3. The method of claim 1, wherein the physical information neural network employs a dual stream feature extraction architecture: space flow branching, namely processing geometrical coordinate data by adopting a graph convolution network or a point cloud network, and extracting local curvature space characteristics; A time flow branch, namely processing a temperature and vibration sequence by adopting a long-term memory network or a one-dimensional convolution network, and extracting thermal inertia and vibration mode characteristics; and a fusion layer for fusing the spatial features and the time features by adopting an attention mechanism and outputting the decoupled error components.
4. The method of claim 1, wherein the four-dimensional modified tool path The following conditions are satisfied: ; Wherein, the In order to design a curved surface, The static geometrical errors are obtained for decoupling.
5. The method according to claim 1, characterized in that the ultra-precise machine tool is equipped with a position servo-controlled spindle and at least two linear motion axes to perform slow tool servo machining; The in-situ measurement device comprises an in-situ measurement probe integrated into the working space of the machine tool for measuring the surface topography of the workpiece without unloading the workpiece from the machine tool fixture, thereby maintaining a uniform coordinate system between machining and measurement operations.
6. The method of claim 5, wherein the temperature sensor is a fiber bragg grating sensor, the machine tool heat source part comprises a spindle front and rear bearing seat, a linear motor coil winding and a machine tool body structure point, the vibration sensor is a high-frequency response piezoelectric accelerometer or a MEMS accelerometer, and the moving part comprises a tool clamping end and a spindle rotor.
7. The method of claim 5, wherein the initial machining employs a spindle-only tool radius compensation strategy in which all tool radius compensation movements are strictly limited to be performed on the spindle to avoid introducing dynamic tracking errors on the linear motion axis.
8. The method of claim 1, wherein the in situ measurement device employs a continuous helical scan path to acquire three-dimensional topography data of the workpiece surface.
9. A non-transitory computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements a multi-modal information fusion-based ultra-precision machining error compensation method according to any one of claims 1 to 8.
10. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a multi-modal information fusion-based ultra-precision machining error compensation method according to any one of claims 1 to 8 when the computer program is executed by the processor.

Description

Ultra-precise machining error compensation method, medium and device based on multi-mode information fusion Technical Field The invention relates to the technical field of ultra-precise intelligent manufacturing, in particular to an ultra-precise machining error compensation method, medium and equipment based on multi-mode information fusion. Background With the rapid development of aerospace, next generation lithography, biomedical imaging and other fields, the processing precision requirements on free-form surface optical elements (such as sine wave grids and microlens arrays) have been raised to the nanometer level. Single Point Diamond Turning (SPDT) and Slow Tool Servo (STS) techniques are the dominant processes for manufacturing such surfaces. However, inherent geometrical errors of machine tools, thermal errors caused by environmental temperature fluctuations, and dynamic errors caused by high-frequency cutting forces become bottlenecks that limit further improvement of machining accuracy. The existing error compensation technology is mainly divided into off-line measurement compensation and in-situ measurement compensation. Offline measurements suffer from severe "repositioning errors" (loss of reference) and are difficult to meet on a nanometer scale alignment requirements. Although the existing in-situ measurement technology solves the reference problem, the following key technical defects exist: 1. Ignoring the time-variability of the measurement process, high resolution in situ scanning typically takes longer (tens of minutes). During this time, the heat generated by the rotation of the machine spindle can cause thermal drift of the structure on the order of microns. The prior art generally assumes that the measurement data is "static" and compensates directly for the inverse of the measurement. This can lead to false thermal drift generated during measurement being misinterpreted as a geometry error on the workpiece surface, thereby introducing a new inverse error in the correction process. 2. The traditional error compensation model is mostly in linear superposition, and the physical constraint on error generation mechanisms (thermal deformation and vibration) is lacked, so that the model generalization capability is poor under small sample data, and the complex dynamic processing environment is difficult to deal with. 3. The dynamic feasibility verification is insufficient in that the directly generated correction path may contain high-frequency components exceeding the servo bandwidth of the machine tool, so that errors cannot be compensated, and machine tool tremors are induced. Therefore, a new method is needed to distinguish between "true geometric errors" and "time-varying thermal/dynamic errors" and to compensate based on a mechanism model. Disclosure of Invention Aiming at one of the technical problems, the invention adopts the following technical scheme: according to one aspect of the present invention, there is provided an ultra-precise machining error compensation method based on multi-modal information fusion, the method comprising the steps of: In the initial processing process, three-dimensional shape data of the surface of a workpiece are acquired by utilizing an in-situ measurement device integrated in the working space of the ultra-precise machine tool. Temperature field data and vibration data in the initial processing process are synchronously acquired through distributed temperature sensors and dynamic vibration sensors which are arranged on a heat source part and a moving part of the machine tool. And aligning the three-dimensional morphology data, the temperature field data and the vibration data according to the time stamp to construct the multi-mode space-time tensor. The multi-modal spatiotemporal tensor is input to a physical information neural network to decouple the total machining error into a static geometry error, a time-varying thermal drift error, and a dynamic vibration error. Composite loss function of physical information neural networkThe following conditions are satisfied:。 Wherein: and outputting the total error and the mean square error of the measured data for the network. An automatic differential calculation is utilized for the degree of violation of the thermal conduction equation by the thermal error component of the network output.And (3) utilizing automatic differential calculation for the violation degree of the dynamic error component output by the network to the forced vibration equation.Is a weight coefficient of the physical constraint term. Based on the static geometry errors, a four-dimensional modified tool path is generated that includes spatial coordinates (x, y, z) and machining time t to spatially counter-compensate the static geometry errors. According to a second aspect of the present invention, there is provided a non-transitory computer readable storage medium storing a computer program which when executed by a processor implement