This project addresses the need for fast and reliable defect detection in pharmaceutical manufacturing, where traditional inspection methods often rely on human oversight or cloud based processing that introduces latency and inconsistency. Defects such as micro cracks, improper sealing, or temperature anomalies in vials can compromise drug safety, leading to costly recalls and potential risks to patient health. By leveraging edge computing, the system performs real time, on device analysis that reduces latency while also lowering data transmission requirements and overall carbon footprint compared to cloud dependent approaches. This work directly impacts pharmaceutical manufacturers, quality assurance engineers, and ultimately patients who depend on safe and properly handled medications.
The system solves this problem by using a combination of cameras and sensors to monitor pharmaceutical vials in real time, analyzing the data directly on local edge devices instead of sending everything to the cloud. Computer vision models process images to detect visual defects such as cracks or misalignment, while additional sensors capture temperature and acoustic signals to identify hidden anomalies. These inputs are fused together and analyzed using lightweight machine learning algorithms running on embedded hardware, enabling fast and reliable decision making at the source. By keeping computation local, the system reduces delay, improves efficiency, and minimizes unnecessary data transfer.
By the end of the project, a fully functional edge based inspection system was developed that integrates hardware, software, and machine learning into a unified pipeline for real time pharmaceutical vial monitoring. The deliverables include a working prototype with embedded devices, camera and sensor modules, a trained computer vision model for defect detection, and a control system that processes and logs data locally. In addition, a user interface was created to visualize inspection results, along with documented system architecture, test results, and performance evaluations demonstrating accuracy and low latency. Together, these components form a complete, deployable solution for intelligent quality assurance on the manufacturing floor.