Modern combustion engines from the personnel and commercial vehicle sector require a continuously increasing amount of testing for achieving the targets exhaust emission, drivability and fuel consumption. The reason is the similarly increasing number of actors, which influence the behavior of the engine and which are the basis for the further going progress in the development of combustion engines. The complexity of the necessary measuring devices and the huge number of sensors result in a situation, where the test-bed engineer is not able to check and verify the whole amount of data. At the same time, alternative measurement strategies become more and more common. DoE, the Design of Experiments, effects that measurement points are driven randomly distributed. Even a well versed test-bed engineer has no real chance to prove the data online, while the measurement points are driven in that way. The result of these changes in the engine development process is an increase in faulty measurements, higher test-bed costs and a poor efficiency of the testing facilities. With help of this approach, these difficulties should be minimized and a quality insurance introduced in the testing procedure. A testing program, which is supplied online with the actual measurement data, observes them for eventually existing mismatches. An error message is signalized and plotted in a text file in the case of detected mismatches. The work can be subdivided in two main parts, namely the fault detection (FD) and the fault isolation (FI). The detection in based on an extensive collection of single methods, which describe a certain characteristic of the engine. These methods include limit value observations, thermodynamic equations, comparisons of specified values and measurement based polynomial models, if their generality could be demonstrated on several different engine-types. The function of the fault isolation process is the analysis, which of the measurement channels has been responsible for the eventually detected conspicuousness. Thereby, two completely different approaches come into use. The first is a logical technique based on mathematical correction numbers, the other is a probability based method in combination with expert knowledge. This expert knowledge is fixed by weighting all used channels corresponding to there influence in the single methods. The developed concept differs clearly from comparable works in the field of fault diagnosis. Fault analysis developments are mostly used only for well known systems, which have been analyzed and measured in a detailed way. Examples are the OBD (On Board Diagnose) in the engine control unit and most other applications in the field of monitoring. In contrast, modern test-beds systems make a change from one engine to another in several minutes possible. The engines dwell time on such a test-bed varies from one day to several weeks. With every change of the engine, most of the measuring devices are changed too. The knowledge about the specific engine and the adapted sensors is limited. Therefore, a detailed and time-consuming parameterization can not be the solution to make the models run. The program itself must figure out and conclude from the existing measurement channels, which models will work and are convenient to realize a fault diagnosis. The existing paper includes the description of the necessary aspects which are needed in such a program. Furthermore, some of the used plausibility checks and the two isolation methods are described.