Purpose <p>This study aims to develop an Industrial Internet of Things (IIoT)-based framework that integrates deep learning and cloud analytics for the predictive maintenance of CNC lathe spindle units. It addresses the challenge of vibration signature feature extraction and selection for the early detection of spindle degradation to optimize maintenance scheduling, improve product quality, and minimize downtime in industrial environments.</p> Methods <p>An accelerated run-to-failure experimental testbed was developed to acquire spindle degradation data using vibration sensors. Neighborhood Component Analysis (NCA) was employed for feature selection. Long Short-Term Memory (LSTM) and bi-directional LSTM (bi-LSTM) deep learning models were trained using the processed time-series data for Remaining Useful Life (RUL) estimation. The framework was implemented on the MATLAB-powered ThingSpeak™ IoT platform, enabling real-time data acquisition, cloud analytics, and remote visualization.</p> Results <p>The proposed system successfully predicted the spindle unit’s RUL with high accuracy and robustness. Among the tested architectures, the LSTM model demonstrated the most balanced and consistent performance, achieving an RMSE of 40.01 time-steps and high A20-index values of 97.30% and 92.57% across the two test datasets. While the hybrid LSTM + bi-LSTM model achieved the lowest RMSE of 31.65 time-steps and an A20-index of 99.72% on one test set, its lower performance on the other set highlighted limitations in generalization. The cloud-based decision support system generated timely failure alerts, while the user dashboard provided continuous health status updates.</p> Conclusion <p>The study demonstrates a viable IIoT-based predictive maintenance decision-making framework CNC lathe spindle unit. By integrating advanced sensor-driven data acquisition, deep learning algorithms, and cloud-based analytics, this framework supports proactive maintenance strategies through predictive analytics.</p>