The paper proposes a computational strategy for detection of changes in the performance of machines and complex systems. Once a distinct change is discovered, the information can be further used for diagnosing the reason for the system malfunction. The proposed computational strategy consists of several steps. First of all, two different radial basis function (RBF) network models are learned based on given operation data from the system during two different time periods. Then these models are used to predict the respective outputs on the even grid of the input space. Finally, the difference between the two models outputs is used as new data set for generation of the fuzzy rules by a specially proposed algorithm. These rules shown in a fuzzy manner how much is the difference of the system performance in each area of the input space. Three original algorithms combined into two learning strategies for the type of normalized RBF network models are first given in the paper and then evaluated on a test example. The proposed computational strategy is applied to evaluation and detection of changes in real operation of a diesel engine for a hydraulic excavator. Three different sets of data have been examined in order to discover the degree of difference between the respective three operation periods.