Condition monitoring and fault-tolerance agents for grid-tied inverters

This paper proposes innovative techniques for extending the lifespan of grid-tied inverter after detecting an overstress event for inverter components or diagnosing an incipient internal failure. This is performed using a reconfigurable circuit topology working in tandem with two adaptive agents: a fault diagnostic agent and fault-tolerance agent. The fault diagnostic agent uses the trajectory of a condition monitoring vector and its momentum and applies a support vector machine approach. Thus, it is fast and reliable to be used under transient conditions, while the existing techniques either require a long processing time or perform well only under steady-state conditions. Using Q-learning, the fault-tolerance agent will find an optimum action to reduce the amount of detected stress or to isolate and replace the faulty component by an auxiliary component. This agent finds the optimum action using an effective reinforcement learning methodology.