Faults in the intake and exhaust path of turbocharged common-rail diesel engines lead to an increase of emissions and to performance losses. Fault detection strategies based on plausibility checks, threshold based trend or limit checking of sensor data are not able to detect and isolate all faults appearing in the intake and exhaust path without increasing of the number of sensors. The need to minimize mass and reduce cost, including the number of sensors, while maintaining robust performance leads to higher application of models for intake and exhaust path components. Therefore a concept of model based fault detection with parity equations is considered. It contains the following parts: modeling, residual generation with parity equations using parallel nonlinear models, fault to symptom transformation with masking of residuals dependent on the operating point and limit violation checking of the residuals.A semi-physical, isentropic efficiency based model of the turbocharger with variable geometry (VGT) can help to detect more faults and to extend the operation region of the fault detection. The developed semi-physical turbocharger model contains a physical core with thermodynamic power calculation based on enthalpy and mass-flow equations and is modified by mathematical approaches in order to expand the operation region of the model and increase the model accuracy. It is used for the calculation of a set of residuals. The models are parameterized with stationary and quasistationary measurement data from an engine testbench. Residuals are calculated first from the difference between the model output and measured value. In addition other quantities for fault-free and actual operation are calculated. Fault detection is based on the deviations if they transgress thresholds.The detection of different faults using model based residuals is validated using dynamic testbench measurements with real faults in the intake and exhaust path and operation over a wide range. For this purpose leakages and restrictions are implemented in different locations of the engine. Measurement data in faulty and fault-free operation is used for the validation of the fault detection methods. Summing up it is shown with real measurements from the engine testbench, that all leakage and restriction faults can be detected. In addition the capability to diagnose the faults is outlined. The presented contribution is developed in cooperation between the Institute of Automatic Control at TU-Darmstadt and GM Powertrain Europe.