Document Type : Article
Authors
Dep. of Industrial Engineering Shahed University
Abstract
Nowadays, in many production and service environment, the quality of a product or process is represented by two or more quality characteristics. Therefore, multivariate and multi-attribute control charts have been widely developed by many authors, separately. Sometimes, the quality of a product or a process is represented by correlated variables and attribute quality characteristics. For example, in plastic manufacturing companies, the number of defects in one product, as an attribute, has a correlation with the weight of the product, as a variable quality characteristic. To the best of our knowledge, there is no method to monitor this type of quality characteristic. Note that monitoring correlated variables and attribute quality characteristics separately, without considering the correlation structure, leads to increasing the overall probability of Type I error in the control chart. An appropriate approach in designing control charts is defining confidence limits. There are some methods to etermine confidence limits for correlated random data, such as described by Bonferoni (Hayter and Tsui, 1994) and Sidak (1967). But, the first method requires large samples, which lead to less application, and the second has a weakness in neglecting the correlation between quality characteristics. In this paper, we propose two methods to monitor multivariate-attribute processes, based on the bootstrap technique (Jhun et al. 2007), which has none of the drawbacks of the mentioned methods (Bonferoni and Sidak). In the first method, new confidence limits for multivariate-attribute quality characteristics are presented by using the bootstrap technique. In the second method, the bootstrap technique is used to determine confidence limits for EWMA control statistics, which are used for correlated attribute and variable quality characteristics. Finally, these confidence limits are used to monitor the process. Based on the signal rule, whenever each control chart signals, the process will be out-of-control and the corresponding quality characteristic is introduced as the source of variation. This research is performed in Phase II, thus, it is assumed that the distributions of quality characteristics are known based on Phase I analyses. The obtained results confirms the efficiency of the proposed method compared to the traditional method based on the accuracy of probability of Type I error, especially under small values of Type I error probability.
Keywords
)