<p>The operational safety of underactuated tower cranes is critically dependent on reliable sensor feedback. Sensor failures, which can range from measurement bias to a complete loss of signal, pose a significant risk of performance degradation and instability. This challenge remains inadequately addressed in current control schemes. This paper proposes a novel integrated fault-tolerant control architecture to overcome this limitation. The strategy seamlessly integrates an adaptive sliding mode control (SMC) feedback loop with a neural network-optimized zero-vibration-derivative (NN-ZVD) input shaper in a feedforward-feedback structure. The core innovation lies in a dedicated fault-handling mechanism supported by a fault detection and isolation (FDI) strategy. Under partial sensor faults (e.g., measurement bias), an adaptive observer estimates and compensates for the fault in real-time, preserving high-performance control. Once a complete failure of the sensor is detected, the sliding surface automatically reduces to a lower-order form that depends only on actuated states, while the model-independent NN-ZVD shaper assumes primary anti-swing duties. Rigorous Lyapunov-based analyses are provided to establish the finite-time convergence of the sliding surface and the semi-global uniform ultimate boundedness of the closed-loop states. Extensive physical experiments demonstrate that the proposed method achieves precise positioning and suppresses payload swing effectively, significantly outperforming existing methods under both partial and complete sensor failure scenarios.</p>