In this paper, we introduce a comprehensive framework for extracting and mining sensor data. This framework consists of a new formulation for the association rules, distributed extraction mechanism, and a compressed structure for the data along with the mining algorithm that is able to extract the knowledge out of it. The new formulation define the temporal relations between sensors and map them to the association rules, a well know data mining technique, a direct application of the extracted relations is predicting the sources of future events, estimating the value of missed events, or identifying faulty nodes. The proposed distributed extraction is designed to improve the network life time by reducing number of messages needed to generate the required data for the mining process, experiments have shown that our distributed extraction solution is able to reduce number of exchanged messages by 50% compared to a centralized solution. The compressed representation structure, which we call it Positional Lexicographic Tree (PLT), is able to partition and compressed the data and provides an easy access mechanism for manipulating the data, we successfully compared the mining process of the PLT with the FP-Growth, a well know mining algorithm, results have shown that PLT outperform FP-Growth in both CPU time and memory usage.