Fourth-order partial differential equations form the basis of models representing many physical problems with bending and curvature, such as those concerned with the dynamics of structural beams. Traditional numerical methods, such as the Finite Difference Method (FDM) and Finite Element Method (FEM), are widely adopted for such problems; however, they face challenges in handling complex geometries, inverse problems, and often incur high computational costs. In the current work, a Physic-Informed Neural Networks (PINNs) framework is proposed as an effective means for the solution of fourth-order PDEs governing the dynamics of Euler-Bernoulli beams. The PINN seamlessly integrates the governing equations with initial and boundary conditions directly in its loss function such that all the physical laws are strictly satisfied. Through automatic differentiation, the model efficiently computes the high-order derivatives involved, a feature that effectively circumvents discretization errors. We demonstrate the effectiveness of the framework in solving various forward and inverse problems, including free vibration, impulsive loading, and identifying system parameters. We have achieved exceptionally good accuracy of 0.00000000000013 in mean squared error for the vibration of a cantilever beam problem. Also, the model identified the unknown physical parameters, namely the flexural rigidity (EI) and mass per unit length (ρA), with absolute errors as low as 0.00017 and 0.00045, respectively. We may infer from our findings that the PINN is endowed with superior precision and computational efficiency in comparison with classical methods. The proposed framework provides an effective, scalable tool for diversified complex engineering applications, such as aircraft wing, turbine blade analysis, MEMS devices modeling, and so on, even with noisy or incomplete data.