Identification of position-related geometric error terms from profile error in four-axis ultra-precision machine tool based on Informer model

Abstract

As modern manufacturing progresses, demands for product quality and precision continuously increase. Ultra-precision machine tools, serving as carriers for high-accuracy workpieces, are widely utilized. The stringent quality control at each stage in these machines, making component accuracy a critical factor influencing workpiece profile precision. Measuring and compensating for geometric errors at specific machining positions constitutes an effective method for enhancing the precision of ultra-precision machine tools. This study aims to employ a data-driven intelligent algorithm to identify the position-related geometric error terms from turned surface profile error, aiming to acquire geometric error data for the ultra-precision machine tool. Initially, an error model between workpiece and tool is established to analyze the effects of position-related geometric error terms, followed by a sensitivity analysis of spatial sub-errors within the total error vector across the X-Y-Z directions. The results indicate that there are no coupling relationships between sensitivity coefficients, enabling the separation of spatial sub-errors and the establishment of corresponding sub-models for position-related geometric error terms, thus generating the theoretical dataset required for model training. Then, using the least square method to fit the turned tool marks and the circle lines, the profile error of the turned surface is obtained for model actual prediction. Subsequently, an identification model for turned surface profile error and position-related geometric error is developed based on the Informer deep learning framework. This model predicts error using a data-driven approach informed by the theoretical sub-models. Finally, experimental data from ultra-precision turned surface profile error are utilized for decoupling and tracing, leading to the predictions of position-related geometric error terms. Validation results demonstrate that the proposed model effectively decouples profile errors and successfully traces the position-related geometric error terms of the ultra-precision machine tool.