Deep Neural Network for Operational Abnormality Detection for a Central Fill Pharmacy System

Traditional manufacturing processes are evolving into intelligent processes that aim to keep up with Industry 4.0. The main objective of any manufacturing system is to meet customers’ needs by delivering low-cost and high-quality products. Advanced technology, such as digital twins, complex sensing systems, and real-time analysis, can help identify operational errors that impact system productivity and performance. However, operational errors are often unavoidable despite advanced upgrades in automated or semi-automated manufacturing systems. Therefore, fault detection and diagnosis are essential steps that can help achieve the desired process performance. Early-stage fault detection can empower the system to take timely action and reduce the likelihood of failures or unfavorable outcomes. Neural networks and machine learning algorithms have become powerful tools for fault detection and abnormal process prediction. This research employs several preprocessing methods to clean, handle the missing values, and prepare the data for classification models while focusing on the top ten medication brands repeatedly ordered by customers. Different Neural Networks architectures were developed to classify operational faults for a Central Fill Pharmacy (CFP) system. A two-input Deep Neural Network (DNN) model with a concatenation layer was proposed, and the performance of the proposed model is compared to Logistic Regression (LR) and Gaussian Naïve Bayes (GNB). The results indicate that the proposed DNN outperforms LR and GNB and achieves a classification accuracy of 99%. The proposed model will help detect abnormal operations and potentially prevent system performance degradation.