Integration of digital twins and statistical process control systems for engineering support of quality management in high-tech mechanical engineering

Abstract

The article examines the integration of digital twins and statistical process control systems (SPC) for engineering support of quality management in high-tech mechanical engineering, ensuring the transition from retrospective discrete control to proactive predictive management of cyber-physical production systems. The purpose of the work is to develop a methodology for coupling deterministic engineering models of cutting and probabilistic models of process variability within a single digital twin framework capable of predicting, in real time, the deviation of accuracy parameters beyond tolerances and forming adaptive control limits. The empirical basis consists of long-term monitoring data of the machining of complex-profile parts made of titanium and heat-resistant nickel alloys on five-axis machining centers, including more than 150,000 measurements of vibration, acoustic emission, drive loads, temperature fields, and geometric deviations recorded by a distributed network of IIoT sensors, edge computing devices, and coordinate measuring machines. The methodology relies on ARIMA models and recurrent neural networks (LSTM) for time series analysis, multivariate correlation-regression statistics, factor analysis and principal component methods, as well as finite element modeling of thermomechanical deformations within the digital twin. It is shown that spindle unit vibration acceleration and predicted tool wear are dominant predictors of machining error, explaining up to 79% of its variance, while the distributions of diagnostic parameters are characterized by pronounced non-normality and leptokurtosis, requiring the use of robust criteria and dynamic control charts. Experimental comparison of traditional SPC and the integrated DT+SPC system revealed an increase in Cp and Cpk indices from 1.124 and 0.982 to 1.678 and 1.543 respectively, a 33% reduction in dimensional standard deviation, and a decrease in defect level from 3456 to 125 ppm, confirming the high efficiency of the proposed approach. It is substantiated that the integration of digital twins makes it possible to transform part of the random process variation into deterministic, to form «corridors of permissible process states» and creates a foundation for implementing virtual metrology and autonomous quality management systems for aerospace and space products.