Monitoring the ratio of population means of a bivariate normal distribution using CUSUM type control charts (Q1706479)

In this paper, the authors define two CUSUM-type control charts for monitoring the ratio of the means of two (correlated) normal distributions. Given a bivariate \((X,Y)\) normal distribution, with means \(\mu_X\) and \(\mu_Y\), with variances \(\sigma^2_X\) and \(\sigma^2_Y\), and with correlation \(\rho\), the ratio \(\mu_X / \mu_Y\) is considered. All the theory here assumes that the coefficients of variations \(\sigma_X/\mu_X\) and \(\sigma_Y/\mu_Y\) are in the range \((0,0.2]\) so that a suitable approximation of the distribution of \(Z=X/Y\) can be used. For each bivariate sample \(((X_{i,1},Y_{i,1}), \ldots , (X_{i,n},Y_{i,n}))\) the ratio of the means is estimated through \(\hat Z_i = \overline X_i / \overline Y_i\) and two CUSUM charts, namely an upward CUSUM and a downward CUSUM chart, are defined. The Average Run Length (ARL) optimization of such control charts is performed via a Markov chain technique, by partitioning the control interval into a finite set of sub-intervals and then finding the optimal parameters for a give ARL under the in-control hypothesis. On the other hand, an extensive simulation study is presented to show the performances of the two charts under the out-of-control hypothesis. Finally, an application to a real-world problem coming from food industry is illustrated.