Cyber-physical industrial systems are internet-enabled physical entities embedded with computers and control components consisting of sensors and actuators. However, interconnecting the cyber and physical spaces led to new security challenges. This paper presents a recovery controller based on a physics-informed neural network (PINN) to enhance the resilience of cyber-physical systems (CPS) against false data injection attacks (FDIA). The PINN-based controller is trained to predict corrective actions that can restore the desired operating conditions of the CPS after an attack. The proposed approach is validated on a quadruple water tank process, a benchmark system for CPS control. Results show that the PINN-based recovery controller can effectively restore the system's desired operating conditions, outperforming conventional recovery controllers that do not incorporate the physical dynamics of the CPS in their design.