This paper introduces a novel physics-informed neural network (PINN) approach to solve nonsmooth dynamic problems involving friction-induced vibration. The authors address the challenge of accurately simulating dynamic behaviors due to friction, which is common in various engineering fields such as aerospace, high-speed railways, robotics, and human/artificial joints. The proposed framework integrates the theoretical formulations of nonsmooth multibody dynamics into the training process of the neural network, improving both accuracy and efficiency compared to conventional time-stepping methods. Four high-accuracy PINN strategies are proposed: single PINN, dual PINN, advanced single PINN, and advanced dual PINN. These methods are tested on two typical dynamics problems: a 1-dimensional contact problem with stick-slip and a 2-dimensional contact problem considering separation-reattachment and stick-slip. The results show that both single and dual PINN methods outperform conventional methods in the 1-dimensional stick-slip problem, while the advanced single and advanced dual PINN methods provide better accuracy in the 2-dimensional problem, even in cases where conventional methods fail. The paper highlights the importance of understanding friction-induced vibration (FIV) and its complex dynamics, particularly in systems with multiple contact points. It discusses the challenges and limitations of traditional methods and emphasizes the potential of the proposed PINN approach in simulating these complex behaviors more accurately and efficiently.This paper introduces a novel physics-informed neural network (PINN) approach to solve nonsmooth dynamic problems involving friction-induced vibration. The authors address the challenge of accurately simulating dynamic behaviors due to friction, which is common in various engineering fields such as aerospace, high-speed railways, robotics, and human/artificial joints. The proposed framework integrates the theoretical formulations of nonsmooth multibody dynamics into the training process of the neural network, improving both accuracy and efficiency compared to conventional time-stepping methods. Four high-accuracy PINN strategies are proposed: single PINN, dual PINN, advanced single PINN, and advanced dual PINN. These methods are tested on two typical dynamics problems: a 1-dimensional contact problem with stick-slip and a 2-dimensional contact problem considering separation-reattachment and stick-slip. The results show that both single and dual PINN methods outperform conventional methods in the 1-dimensional stick-slip problem, while the advanced single and advanced dual PINN methods provide better accuracy in the 2-dimensional problem, even in cases where conventional methods fail. The paper highlights the importance of understanding friction-induced vibration (FIV) and its complex dynamics, particularly in systems with multiple contact points. It discusses the challenges and limitations of traditional methods and emphasizes the potential of the proposed PINN approach in simulating these complex behaviors more accurately and efficiently.