Identifying rock mass joint sets and systems are important tasks in oil exploration as well as mining and civil engineering projects. Up to 10 characteristic of joints are normally gathered by field survey, among them only two features (dip and dip direction) are basically used for identifying joint sets. Present study applies clustering techniques on 7 characteristic of Asmari formation joints, in Iranian oil fields, and identifies joint sets. Results showed that more than two features of joints should be considered to reach a firm result. Where rose diagram and stereonet could only show 2 or 3 joint sets, clustering of joints with 7 features and K- means clustering method, confirmed that 6 joint sets occur in the area of study.
Dip and dip direction are two common features for joint clustering. Rose diagram and stereonet are commonly used for representing the result of joint clustering. Earth science experts often utilize just one joint feature (dip direction or strike in rose diagram), or two features (dip and dip direction in stereonet) for joint clustering. However it has been shown that using only 1 or 2 features is not adequate for joint clustering (Memarian and Fergusson, 2003). For example, it is impossible to recognize two joints with similar strike, but different dip direction with strike rose diagram; however their recognition with dip direction rose diagram is possible. Meanwhile it is impossible to recognize two joints with different dip but similar strike and dip direction, even with a dip direction rose diagram. Another example is two joint sets with similar dip and dip direction but different roughness, causing different effects in wall stability. Such joint sets cannot be separated with rose diagram or stereonet either, because the rose diagram or stereographic projection of them will overlay each other. On a similar note, if variance of some similar joint set dips is great, separation of them with stereonet is impossible. Therefore rose diagram and stereonet seem to have a lot of shortcomings in joint clustering. To solve these kinds of problems, new clustering methods are necessary. This is particularly important in joint studies, where various features of joints like dip, dip direction, continuities, spacing, roughness, type and amount of infilling are available. In the new approach of joint clustering, the whole features of joints should be taken into account. In this way, the space dimension of joint clustering will increase up to n which is the number of joint features. Recently, there have been some studies to solve the problems associated with conventional 1304 a b methods of joint clustering. However those studies also suffer from the limitation of small feature space and fixed clustering method (up to 3 joint features and 1 clustering or classification method). So ranking the effectiveness of joint features, and capability of clustering (or classification) methods in joint clustering, has not been investigated in previous researches. In the following a brief summary of the main jointclustering researches is presented.