Abnormal Condition Detection Integrated with Kullback Leibler Divergence and Relative Importance Function for Cement Raw Meal Calcination Process

This paper focus on abnormal condition detection by using Kullback Leibler (KL) divergence with relative importance function. There exist multimodal working conditions, such as normal condition, abnormal condition. KL method was proved to be more sensitive to initial faults than the Hotelling's T-squared statistic. Relative importance function estimation for condition detection has been demonstrated, and relative importance function is always smoother than corresponding ordinary density-ratios. In cement raw meal calcination process, we sampled some important variables, such as calciner temperature, preheater C1 outlet temperature, raw meal flow, and C1 and C5 cone pressure. In actual process, the product quality index is low and it is easy to cause the preheater C5 feeding tube to be blocked. To detect abnormal condition, an abnormal condition detection based on Kullback Leibler divergence with relative importance function was proposed. The actual application results shows that the model proposed can detect abnormal condition by current operating data, and far from fault condition by the practical application results.