Fault detection and diagnosis (FDD) are very important for engineering systems in industrial applications. One of the most popular approaches is model-based fault detection. Recently, many techniques have been proposed in the FDD area. However, there are still very few reported applications or real-time implementations of the schemes. This paper presents online sensor FDD based on the model-based approach using a Luenberger observer and experimental application on a permanent magnet DC motor. Different kinds of faults are simulated on the motor and experiments are performed to detect the faults. The experimental results demonstrate that this approach could significantly detect the time and size of the faults.