The results of different performance measures advocate the following:
In the aymmertic class of distribution like normal,logistic, and t distribution,the proposed Tukey-CUSUM chart offers better preformance as compared to TCC for smaller shifts, while for larger shifts TCC gets superior in general. It is evidenced by different measures such as ARL MDRL EQL and other performance measure
The supporting results provided in tables I-V and Figure2 may be seen inthe form of different individual and overall performance measures
The ACL structuer of the proposed Tukey-CUSUM is more effective espective especially for skwed distributions. Table I-V and fiiguer 1 may be seen in favor of these findings in the form of different measures such as . We see Tukey-CUSUM offers poor ARL other hand, for positive shifts Tukey-CUSUM offers better ARL performance
In asymmetric class of distribution, Tukey-CUSUM offer depending upon the amount of Skewness, as may be seen in Figuer2 and table I-V using different RL based measures
TCC and Tukey-CUSUM offer relatively better ARL performance for negative shifts. At positive shift CUSUM and Tukey-CUSUM compete quite closely. TCC has relatively better ARL1 performance at negative shift while Tukey-CUSUM is dominating at positive shift. It may be observed in tableI-V and figure2
Sample selection criteria
In practical applications, sample selection is of greater concern in process monitoring. We may have less number of observations or a huge set of data. An appropriate sample size may be the one such that the underlying process behavior is identifiend. According to Quesenberry, 300 sample are sufficient for the construction of control limits. They developed the correction terms for the accurate control limits. Albers and Kallenberg also indicated 300 sample that mat be reduced to 40 samples uder corrective actions. Lee applied the asymmetrical control limits using 100 observation to the in control process. For our study proposal we have followed the same inspirations to select such samples so that an efficient performance of proposed design is achieved. We have selectsd different sample of size n for Phase-l estimation and ARL performance of the proposed design is evaluated at pre-specified. The results are reported in table VI. From these results, it is observed that our proposed design requires approximately 100to200 observation to reasonably match the expected outcomes. The larger the ARL0 value, the more observations we need to have smaller duscrepancy from the expected outcomes and ti achieve an efficient performance by the proposed scheme
The specific steps for Phase1 and Phase 2
step1 : select Phase-1 data set, fit an appropriate distribution, and estimate its parameter from the data set
Step2: construct Phase-1 chart using the limits LCL where mu and sigma refer to mean and standard deviation estimated from the data set. These estimatirs may be some efficient and robust estimators like Q2 and IQR.
Step3: screen the unusual samples points by plotting theme against the estimated control limits and discarding any data points falling outside the limits. repeat all these setps until all points show stability
Step4: after screening, obtain the final parameter estimate Q1 and using the clean data set.
Using the final estimaties, construct Phase-2 Tukey-CUSUM chart in the form of Cpi ,Cni and also the decision interval of the chart.
Periodically collect the new data set and use the Phase-2 Tukey-CUSUM chart for online monitoring of process parameter