Horizontal accuracy prediction model of CNC machine tool using regression algorithm and multi-point load feedback
Journal
Journal of Intelligent Manufacturing
ISSN
09565515
Date Issued
2026
Author(s)
Abstract
The horizontal accuracy of computer numerical control (CNC) equipment plays a crucial role in determining machining precision, as it directly influences static errors, component alignment, and positional accuracy. Inadequate horizontal alignment can cause deformation or unevenness of the machine table, reducing machining accuracy and increasing the likelihood of product defects—ultimately compromising overall product quality. Because placing a leveler on the workpiece during CNC operation is impractical, current industrial practices primarily rely on periodic inspections and manual adjustments. However, these approaches do not enable real-time measurement and cannot detect tilt or support anomalies during machining, thereby introducing hidden quality risks. To address these limitations, this study proposes an indirect method for measuring horizontal accuracy based on force-sensor signals. Four force sensors were mounted beneath the supporting feet of a CNC lathe to capture load variations under different tilt conditions. Corresponding tilt angles were labeled using a high-precision electronic leveler, and the resulting dataset was used to train several regression and classification models for predicting horizontal accuracy. Among the evaluated models—Linear Regression Model(LRM), Random Forest Regression Model(RF-RM), and a Backpropagation Neural Network Regression Model (BPNN-RM)—the BPNN-RM demonstrated superior performance owing to its flexible architecture and adaptive learning capability, achieving a mean absolute residual angle of 0.0160°, a root mean square error (RMSE) of 0.0193°, and an R² value of 0.878. These results indicate a high level of predictive accuracy comparable to that of commercial electronic levelers. Compared with classification algorithms, regression-based models offer higher angular resolution, making them more suitable for precise tilt estimation. The proposed method thus enhances the feasibility of real-time horizontal accuracy prediction. When integrated into CNC control systems, the proposed method can enable real-time monitoring and early detection of tilt anomalies during machining, thereby improving both product quality and machine operational stability.
Subjects
Artificial intelligence
Horizontal accuracy
Internet of things
Machine tool
Real-time measuring
Publisher
Springer
Type
journal article