In the field of data analysis two terms frequently encountered are supervised and unsupervised methods of data classification and clustering methodologies. While supervised methods mostly deal with training classifiers for known symptoms, unsupervised clustering provides exploratory techniques for finding hidden patterns in the data. With huge volumes of data being generated from different systems everyday, what makes a system intelligent is its ability to analyze the data for efficient decision-making based on known or new cluster discovery. The three-fold contribution of this paper can be summarized as the role of unsupervised clustering for intelligent decision-making process, review of existing unsupervised models, including self-organizing maps (SOM), hierarchical tree (HT) model and quality adaptive threshold (QT) model, and lastly a new hybrid model for unsupervised clustering is proposed. For case study, we have taken the example of an intelligent decision making process in the field of fault diagnosis of industrial robots. The unsupervised models were tested on data obtained from an industrial robot used in the semiconductor industry. This paper presents the first set of results obtained from these four methods and discusses further applications of these methods.

This content is only available via PDF.
You do not currently have access to this content.