Abstract: The impoundment and sediment trapping by the Three Gorges Project (TGP) and upstream reservoirs have significantly altered the flow-sediment regime in the middle and lower Yangtze River (MLYR), leading to intensified channel bed scour and frequent bank failure incidents. These pose severe threats to levee safety and the stability of the fluvial regime. To elucidate the formation mechanisms and evolutionary patterns of bank failures in the MLYR and enhance capabilities for monitoring, early warning, and mitigation, this study employed field investigations and mechanical analyses to clarify the mechanisms underlying strip-type bank failures and arc-shaped bank erosion and to analyze the hydraulic failure modes of riprap revetments. A cross-sectional scale bank failure model, PFEMS (Process-Based Bank Failure Model System), was developed, incorporating key factors such as toe scour, hydraulic failure of riprap revetments, and groundwater fluctuation. This model was further extended into a multi-scale hydro-morphodynamic model coupling one-dimensional to three-dimensional sediment transport and bank retreat processes. An integrated early warning system for bank failure was established by fusing multi-source data (including groundwater monitoring and remote sensing imagery) with dynamic simulations and machine learning techniques. Critical emergency stabilization strategies for significant downstream arc-shaped bank erosion events were identified, and the layout scheme for emergency engineering works was optimized using the Zhinan Village (at Yangzhong City) arc-shaped bank erosion event as a case study. Key findings are as follows: ① Bank failure mechanisms: Strip-type bank failures assume as a cyclical "toe scouring-bank failure-re-scouring" process triggered by flow-induced toe erosion. Arc-shaped bank erosion originates from near-bank deep scour pits wedging into the bank, causing slope instability, with subsequent expansion driven by vortex scouring. Hydraulic failure of riprap revetments follows a progressive three-stage mode: initiation of individual particle movement-sliding of rock clusters-mass collapse. ② Model performance: The PFEMS model significantly outperformed the BSTEM software in identifying failure timing and predicting failure morphology. The multi-scale coupled model successfully simulated the entire bank failure process.The 1D model achieved a 64% accuracy rate in predicting bank failure zones within the Jingjiang Reach in 2024. The 3D model simulated the width of the collapse in Zhinan Village, Yangzhong, with an error of only 8.4%. ③ Technical application: The system was successfully applied for risk classification of bank failures in the Jingjiang Reach in 2024, significantly enhancing the scientific rigour and accuracy of monitoring and early warning. The optimized mitigation scheme for arc-shaped bank erosion contributed to an average 14.5% reduction in flow velocity within the scour pit, a 17.4% decrease in collapse area, and a reduction of the bank slope angle from 10.5° to 7.7°. These research outcomes provide systematic scientific and technological support for bank failure risk management in the MLYR, holding pivotal theoretical and engineering significance for ensuring river safety and promoting sustainable regional development.