Receiver operating characteristic analysis for the selection of threshold values for detection of capping in powder compression

The acoustic emission (AE) energy obtained from compressing lactose powder to form pharmaceutical tablets was chosen for condition monitoring of the tablets. The method used was based on the setting of an AE energy decision threshold such that problems of tablet capping and lamination were successfully identified. Capping and lamination are the most common types of problem that can occur in tablets manufacturing using a powder compression process. To assess the performance of a classifier, use was made of a receiver operating characteristic curve (ROC) obtained by plotting the correct detection probability against the false alarm probability based on AE energy distributions for capped and non-capped tablets. The area under the ROC curve, referred to as the AUC, determines the level of competency of the classifier. A value of 0.5 suggests a mere hazarding of guesses whilst a value of 1 indicates correct classification every time. The AE energy approach for tablet capping monitoring gives an AUC value of 0.96, thereby suggesting the possibility of a highly accurate classifier. With the assumption that penalties for false alarm and missed detection are equally severe, using the graphical method of expected penalty cost (EPC), the optimal AE energy decision threshold was established to be 1.2x10(8) units, at which the maximum correct capping detection rate of 95% was achieved. The paper also explains how a decision threshold can be obtained when the two penalties are not equal.