Due to the advances in technology, most of the data are being as streams with large volumes and high flow. Different approaches were proposed to deal with such data streams. Clustering is one of the most important algorithms utilized to solve the data stream problems. K-Mean algorithm was considered as one of the best classical clustering algorithms. It is suitable for mining and analyzing data stream, because it depends on the centroid, number of data elements, and distance measurement. K-Mean algorithm needs continuous updates to determine the number and size of clusters. Data stream processing requires an immediate processing at any time of the data stream entrance. In this paper sensors data streams clustering are considered. The proposed system is composed of four stages. Collecting streams of data (based on date and time) from real wireless sensors represent the 1st stage. The 2nd stage is to construct an offline implementation to propose primary number of clusters with variable sizes. The 3rd stage is an online implementation to the arriving data from sensors. In this stage an Adaptive Dynamic Diameter and Border Threshold (ADDBT) is utilized to change the clusters sizes in an adaptive way. ADDBT is modified based on a proposed equation to estimate the acceptable distance from online data stream by variant thresholds to produce additional new clusters. It has an ability to process drift data by calculating the Cohesion Index for all resulting clusters. By these stages, the problem of determining the value of K and determining the size of clusters is solved by making them non-fixed and adaptive so that their value can be estimated according to the quality of the online data stream. In the 4th stage, the resulting clusters were modified, merged, deleted, and evaluated. The final results then be discussed for the candidate clusters with the original clusters in terms of dynamic diameters and the cohesion of the clusters.