Model-based fault diagnosis methodologies rely on an accurate mathematical representation of a system's dynamics to effectively detect and localize faults. However, creating such models can be challenging, particularly for complex systems operating under diverse conditions. Furthermore, faults affecting the system can also modify its dynamics.
Given the limitations of model-based fault diagnosis, this study introduces a data-driven approach within the Blind System Identification framework. This approach can identify both the fault and the linear-time invariant model simultaneously. The mathematical formulation of this problem is expressed as a constrained least squares problem involving rank and sparsity constraints. To illustrate the application of this methodology, we demonstrate its effectiveness in diagnosing faults in Air Data Sensors using actual flight data obtained from the Cessna Citation II aircraft.