The operational reliability of shaft systems in hydropower units has become increasingly critical as these units are frequently engaged in grid regulation under new power systems. This paper presents a sensor-driven method for operational status assessment and trend prediction of Francis turbine generator unit shaft systems using vibration and swing signals. Time domain features are extracted from the sensor-acquired signals to construct a multi-dimensional quantitative index system for characterizing the operational state, and a combined Entropy Weight–Coefficient of Variation–TOPSIS model with dynamic health thresholds is established for adaptive condition assessment. To address the nonlinear and non-stationary characteristics inherent in such signals, a decomposition–prediction–reconstruction fusion framework is developed, incorporating Variational Mode Decomposition (VMD) for signal decomposition and noise reduction, iTransformer for capturing global multi-variable interactions, and Bidirectional Long Short-Term Memory (BiLSTM) for bidirectional temporal feature extraction. The hybrid model achieves a coefficient of determination R² of 0.9845 on complex vibration and swing signals, demonstrating its superior prediction capability. Based on the prediction results, health scores and dynamic thresholds are calculated to perform trend analysis and health early warning. A case study is conducted using real-world monitoring data from a 306 MW Francis turbine unit. The results demonstrate that the proposed method effectively characterizes the shaft system operational state, achieving a closed-loop integration from condition monitoring to fault diagnosis and predictive maintenance. The operational status assessment and trend prediction analyses are in good agreement with actual operating conditions, providing reliable technical support for the intelligent health management of hydropower units.