Search

The monitoring of small magnitude faults is a specification that should be accomplished by data-driven fault diagnosis techniques in multimode industrial processes. This is a challenging problem because it is difficult to identifythedata of each stable operation mode considering the presence the transitions between them and the noise of the measurements. A method for the identification of stable operation modes is proposed in this paper. This is based in a modification of the clustering algorithm known as k-medians and is named incremental k-medians. His main advantages are: is robust against the presence of transitions, outliers and noise in the dataset, it is not required to know the number of modes previously to parameterize it and it is not required to make multiple runs of it. The monitoring scheme is built upon the identified modes. The proposed strategy is tested using a continuous stirred tank heater (CSTH). Results obtained for a fault with different magnitude in each mode demonstrate the effectiveness of the proposal. [ABSTRACT FROM AUTHOR]

Independent Component Analysis (ICA) is a statistical technique applied through the last ten years for dimensionality reduction in fault diagnosis on industrial systems. FastICA is the most popular algorithm in order to find the independent components. The independence measure and the feature selection criteria are two of the most important parameters to set on this algorithm. We propose in this paper a procedure to find a good combination of them based on achieve a good classification rate reducing the dimension of data the most possible. The proposed procedure is applied on the Tennessee Eastman Process plantas a model study. [ABSTRACT FROM AUTHOR]