A novel data-driven framework for CNC machine maintenance: integrating deep learning and risk-based prioritization

In advanced industrial environments, predictive maintenance driven by multi-sensor data and intelligent modeling within the Industrial Internet of Things has become a pivotal strategy for mitigating unexpected failures and enhancing the operational efficiency of computer numerical control machines. The timely identification of incipient failure patterns is essential to prevent costly downtime, performance degradation and secondary system damage. This study proposes a novel integrated four-stage data-driven framework grounded in a digital twin architecture. In the first stage, data from eight heterogeneous sensors are collected and systematically preprocessed to ensure analytical reliability. In the second stage, temporal dependencies and latent signal dynamics are captured using long short-term memory and bidirectional long short-term memory networks for high-fidelity behavioral forecasting. The extracted representations are subsequently utilized by a deep neural network to probabilistically estimate component-level failure risks. In the final stage, the predicted failure probabilities are embedded into a failure mode and effects analysis structure and transformed into risk priority numbers, enabling quantitative and intelligent maintenance prioritization across four hierarchical intervention levels. The proposed framework is evaluated within a data-driven simulation environment based on the digital twin concept and benchmarked against conventional reactive maintenance policies. Experimental results indicate a 35% reduction in failure occurrence, a 34% decrease in downtime and an increase in mean time between failures from 7.7 to 11.76. These findings demonstrate substantial improvements in system reliability, cost efficiency and sustainable industrial performance.