In view of the time-varying complexity of the heat source for the ball screw feed system, this paper proposes an adaptive inverse problem-solving method to estimate the time-varying heat source and temperature field of the feed system under working conditions. The feed system includes multiple heat sources, and the rapid change of the moving heat source increases the difficulty of its identification. This paper attempts to develop a numerical calculation method for identifying the heat source by combining the experiment with the optimization algorithm. Firstly, based on the theory of heat transfer, a new dynamic thermal network model was proposed. The temperature data signal and the position signal of the moving nut captured by the sensors are used as input to optimize the solution of the time-varying heat source. Then, based on the data obtained from the experiment, finite element software parametric programming was used to optimize the estimate of the heat source, and the results of the two heat source prediction methods are compared and verified. The other measured temperature points obtained by the experiment were used to compare and verify the inverse method of this numerical calculation, which illustrates the reliability and advantages of the dynamic thermal network combined with the genetic algorithm for the inverse method. The method based on the on-line monitoring of temperature sensors proposed in this paper has a strong application value for heat source and temperature field estimation of complex mechanical structures.
High-speed machining is a key technology for greatly improving productivity and reducing production costs. Angular contact ball bearing is the key component of ball screw feed system. Predictive analysis of bearings thermal behavior is significant to ensure the working accuracy of machine tools. Due to the continuous dynamic changes under actual working conditions, it is difficult to predict and model of the bearings heating characteristics. This paper develops a comprehensive calculation method for predicting the thermodynamic characteristics on high speed feed shaft bearings. Based on quasi-static model and thermal dynamic model, Monte Carlo optimization algorithm is used to inverse the temperature field experimental temperature data. The real heating characteristics of heat source are comprehensively predicted and analyzed, so as to accurately predict the bearing thermal dynamic performances of feed shaft bearings. The research work in this paper lays a strong foundation for the thermal error modeling and thermodynamic analysis of the ball screw feed system of machine tools.
In order to solve the problem that the flatness error contour curve of the guideway mounting surface on machine tools can not be intuitive description, a novel method is proposed in this paper. The principle is to divide the measured surface with a certain length into several short surfaces with overlapping region, and the segmented surfaces are measured respectively, then the flatness error contours of the segmented surfaces can be obtained, finally, the segmented flatness error contours are spliced together to reconstruct the flatness error contour of the entire surface. On this basis, the detection instrument is designed, the detection principle is given, the benchmark plate is customized to identify the starting position and ending position of measurement, the reference surface and the measured surface are measured respectively in combination with the detection instrument and the benchmark plate. Then, an algorithm is developed to convert the distance difference between the reference surface and the measured surface into the flatness error contours of segmented surfaces. Next, using the geometric relationship between overlapping region of segmented flatness error contours, an algorithm is studied which can splice these segmented flatness error contours and reconstruct the flatness error contour of the entire surface. Meanwhile, mathematical models of the algorithms are established, and the measurement experiment is carried out. The results demonstrate that this method can reconstruct the flatness error contour of long surface accurately.
The thermal error of the ball screw feed system will reduce the positioning accuracy of the machine tool, and the main reason for the thermal error is the friction heat generated by bearings and screw nut in the feed system. Therefore, a time-varying thermal error prediction method of the ball screw feed system based on the inverse method is proposed in this paper. Based on data-driven and dynamic thermal network model, the heat source estimation method of ball screw feed system under unsteady thermal effect is established, and the thermodynamic equilibrium equation is deduced into explicit form for heat source load identification. Aiming at the common matrix ill-conditioned problem of load identification, the regularization algorithm is used to identify the heat source load, and the optimal selection method of regularization parameters is proposed based on Monte Carlo algorithm. Using the collected temperature experiment data and the position data of the moving nut, the dynamic heat source load, temperature field and thermal error of the feed system under the actual working condition are predicted and analyzed by using the inverse method. Finally, the accuracy and effectiveness of the prediction method are verified by experiment. The inverse method proposed in this paper has great application potential for the prediction and estimation of heat source and temperature field of machine tools and various mechanical structures.
Graph Neural Networks (GNNs) have become a pivotal framework for modeling graph-structured data, enabling a wide range of applications from social network analysis to molecular chemistry. By integrating large language models (LLMs), text-attributed graphs (TAGs) enhance node representations with rich textual semantics, significantly boosting the expressive power of graph-based learning. However, this sophisticated synergy introduces critical vulnerabilities, as Graph-LLMs are susceptible to adversarial attacks on both their structural topology and textual attributes. Although specialized attack methods have been designed for each of these aspects, no work has yet unified them into a comprehensive approach. In this work, we propose the Interpretable Multi-Dimensional Graph Attack (IMDGA), a novel human-centric adversarial attack framework designed to orchestrate multi-level perturbations across both graph structure and textual features. IMDGA utilizes three tightly integrated modules to craft attacks that balance interpretability and impact, enabling a deeper understanding of Graph-LLM vulnerabilities. Through rigorous theoretical analysis and comprehensive empirical evaluations on diverse datasets and architectures, IMDGA demonstrates superior interpretability, attack effectiveness, stealthiness, and robustness compared to existing methods. By exposing critical weaknesses in TAG representation learning, this work uncovers a previously underexplored semantic dimension of vulnerability in Graph-LLMs, offering valuable insights for improving their resilience. Our code and resources are publicly available at https://anonymous.4open.science/r/IMDGA-7289.