||To be able to analyze large datasets, one has to take a more general view on the data. Clustering methods are able to detect regions of similarity in the data. Crisp clusterings have great disadvantages when used for controlling and predictive tasks. In this paper we will describe a multi stage approach to find also global minima. We will first describe a method to clean the data and reduce the number of datapoints. Then several methods to choose the initial number of clusters and cluster centers are described. Afterwards a modified fuzzy clustering method is used, to gain the final clustering.