In this work we address the design and analysis challenge of a laboratory benchmark system, suited for testing and comparing model-based fault diagnosis algorithms. The motivation is the democratization of research in fault diagnosis, which is hindered by the lack of accessible data from real-world systems. A liquid-level control system with three interconnected storage tanks was selected for the physical process. The fault modes under consideration — liquid leak at the tanks — were planted in a manner that their impact can be measured and quantified. Commercially available sensing, actuating, and data acquisition electronic components were used for interfacing the monitoring unit (cyber part) with the process (physical part). A detailed description of the first-principles mathematical modeling is provided for deriving the state-space equations of the physical process. System identification, using elementary least-squares estimation, was performed to estimate the parameters of the parametric model using input/output data. The validation of the identified dynamic model and its agreement with the collected data showcase the capabilities of the proposed system for testing and comparing model-based fault diagnosis algorithms.