STATISTICAL QUALITY CONTROL: A MULTIVARIATE APPROACH

Abstract

Many quality control operations in mining deal in controlling more than one variable for meeting the quality specifications. In such cases, an application of the Shewhart's univariate control chart for each of the variables is unsatisfactory as it fails to consider the problem in a multivariate situation ignoring correlation structures amongst the variables. In this paper, an application of the Hotelling's multivariate control chart is demonstrated in an iron ore mine through a case study. The study revealed that the sensitivity of detecting an out-of-control condition is increased using a multivariate control chart. It is suggested that once an out-of-control is detected in multivariate chart, the corresponding univariate control charts should be investigated to identify which variable(s) causes the out-of-control condition. While investigating multivariate and univariate control charts for the case study mine, two possible out of control conditions were encountered: one due to the out-of-control condition of the individual variables and the other relates to the correlation structure of the variables. For the case study mine, it was also revealed that even if the application of control charts will improve the quality of ore, it is difficult to meet all the quality specifications, especially Fe% and Al2O3% on a regular basis. To overcome this problem, it is suggested that the system may require a fundamental change in the mining process or the specification limits may be reviewed. The fundamental change may be the review of cut-off grade, changing of blending ratio of material from different faces and further blending of material.