Based On The Statistical Distribution Of Bubble Size In Flotation Process Fault Diagnosis

It is approved that the froth appearance characteristics like bubble size and the flotation process performance are closely related. The change of the bubble size reflects the variety of flotation operational condition and the occurrence of process faults, which also gives feedback of the separation performance and in return provides guidance on the adjusting of reagent addition being one of the key operating variables. The research on bubble size is of great importance to the formation of the high-quality froth layer and the optimization of the whole flotation process. The requirement for the safety and reliability is demanded in industry field. In order to guarantee the flotation running reliably and effectively, it is highly desired to detect the fault timely and accurately which has occurred or may occur, which makes the modeling of fault detection and diagnosis significant. Hence, bubble size based fault detection and diagnosis of the flotation reagent addition constitute the main content studied and discussed in this thesis.After the froth images are collected by developed computer vision based monitoring system, it is found that the distribution of surface bubble size is non-Gaussian. Therefore the distribution can not be assumed simply to be normal distribution, and can not be described by single-valued features such as mean and variance, etc. It is put forward in this paper that non-parametric estimation algorithm of the probability density function (PDF) is used to describe the bubble size distribution accurately. In order to achieve the comparison of different PDFs, an improved kernel density estimator is established to describe the PDF distribution of flotation bubble size. A method of fixing basis function is designed to get the weight coefficient vector, which lays the foundation of the basis model for fault detection and diagnosis.In the flotation monitoring system, the feedback information from output is the PDF distribution of bubble size instead of a single variable value. It is different from traditional fault detection and diagnosis methods. It is proposed to establish a model of fault detection and diagnosis based on output PDF in this work. The improved kernel density estimation is used to approximate the output PDF that is transformed into nonlinear dynamic model. Then a new filter based model of fault detection of flotation reagent addition is established. Based on linear matrix inequality (LMI), a feasible method of fault detection is obtained. In the industrial field a large number of fault video samples are collected, the simulation results prove that the method has high accuracy.Through studying profoundly the qualitative relationship between reagent addition and bubble size distribution and its corresponding weight values, a new filter model of fault diagnosis of flotation reagent addition is established to diagnose two types of reagent fault. Based on linear matrix inequality (LMI), a feasible method of fault diagnosis is obtained. Experimental results on industrial fault video samples demonstrate the high accuracy of the proposed diagnosis method.