Fault Detection For Sulphur Flotation Process Based On Froth Texture Unit Distribution

The sulfur flotation process is a complex physical process without any reagent owing to the natural high hydrophobicity of sulfide mineral, where the high grade of sulfur flotation can be realized in a stable condition through reasonably regulating the corresponding operation (air flow, valve opening and so on) Any error operation in the flotation process or equipment failure, will cause the fault condition, and affect the flotation effect. At present, it is difficult to detect the fault condition in real-time and accurately. Therefore, developing an effective fault detection method based on machine vision to improve the sulfur flotation grade is of great significance.There is a close relationship between texture features and flotation conditions. Many researchers calculated the second order statistics based on traditional gray level co-occurrence matrix(GLCM) such as angular second moment, entropy, moment of inertia to recognize the texture features. However, the actual flotation foam texture is very complex. The second order statistics based on GLCM is difficult to accurately describe the texture feature, which will affect the detection of fault conditions. Texture unit is the basic unit of the texture spectrum, can more precisely describe the texture feature. This paper designs a new nonparametric kernel density estimation algorithm with fixed kernel functions to approximating the foam texture unit distribution. It solves the problem that the traditional kernel density estimation can not compare the two different texture unit distribution. At the same time, the infinite dimensional texture unit distribution function is turned into the weight vector with finite dimension. The principal component analysis model under normal conditions can be established based on the main information extracted from weight vector by means of PC A method. The T2statistics reflect the distance between the sample and PCA model and the control limits of T2statistic can be used as the threshold to detect the fault condition. The effectiveness of this method has been proved by simulation experiment and industrial application.