Improvement of an Adaptive Threshold Technique for Fault Detection in a Cement Rotary Kiln

Abstract

In this paper, we suggest to use a two-dimensional plot characterizing the statistical variability of a large-sized multivariate data. This graphical representation is based on the mean and variance values. The two statistical parameters plot is used to assess fault detection performances in a cement rotary kiln system. The adaptive threshold technique is shown to result in more accurate and reliable detection. The threshold is established through several repeated experiments under the healthy mode with the same operating conditions. An adequate statistical test is used to examine the validity of the adaptive threshold estimation approach. At each mean’s subinterval for all experiments, a confidence interval closely linked to the distribution frequencies of the variance as a random variable is obtained. In addition, several significance levels are considered to show the performances of the proposed adaptive thresholding technique compared to the limitations of the fixed threshold through the rate of false alarms. Two different experimental faults are considered to demonstrate the effectiveness and accuracy of the adaptive threshold in terms of no false alarms and negligibly small missed alarms in comparison to the fixed threshold technique.