Work

Research on a CNC machine tool machining error analysis and prediction method based on digital twin and ISSLO-CNN

Year: 2026
Type: article
Abstract: To address the insufficient hyperparameter optimization accuracy and low convergence efficiency of the stochastic social learning optimization algorithm (SSLO) in CNC machine tool machining error prediction, caused by the fixed distribution of learning factors, lack of elite guidance in learning targets, and static knowledge exchange rate, an improved stochastic social learning optimization algorithm (ISSLO) and digital twin-driven CNN error prediction model are proposed. First, physical measurement signals and virtual simulation signals are synchronously acquired based on a digital twin quadruple model, and a multisource fused feature matrix is generated using a linear fusion strategy as the input samples of the convolutional neural network (CNN). Subsequently, a new improved SSLO algorithm is designed. On the one hand, an elite-guided adaptive learning factor mechanism based on iterative cosine decay and fitness ranking is introduced to dynamically regulate the search step size and guide individuals toward elite solutions. On the other hand, an adaptive perturbation strategy based on historical successful-direction memory and a dynamic knowledge exchange rate adjustment mechanism based on the update success rate are constructed to enhance the improvement efficiency of disadvantaged individuals and information exchange within the population. Finally, ISSLO is employed to iteratively optimize key CNN hyperparameters, including convolution kernel size, number of kernels, learning rate, number of neurons in the fully connected layer, and batch size, and the CNN is trained using the optimal configuration to perform error prediction. Experimental results on the NCMT-DT-Error dataset show that ISSLO-CNN achieves a root mean square error of 3.84 ± 0.16 μm, a mean absolute error of 2.72 ± 0.12 μm, and a coefficient of determination of 0.979 ± 0.003, while its physical consistency metrics comprehensively outperform those of comparison models such as SSLO-CNN. The proposed method has 1.97 M parameters and a single-sample inference time of 3.62 ms, and ablation experiments verify the effectiveness of the two improvement strategies. The results demonstrate that ISSLO-CNN improves prediction accuracy while maintaining low model complexity and inference latency, indicating its potential for online machining error monitoring and intelligent quality control. Its practical closed-loop compensation performance requires further validation.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7213926737
FWCI:
Cites: 26
Cited by:
Related to: 10