Anomaly detection is critical for maintaining operational stability and preventing unplanned downtimes in industrial systems. This paper explores and compares supervised and semi-supervised deep learning based approaches within industrial systems, leveraging the OPC-UA protocol for data communication and execution of the evaluated algorithms within a finite state machine. This study presents a case study of anomaly detection in a tinplate lids system, demonstrating the performance of both approaches. Integrating with OPC-UA ensures near real-time data access, interoperability, and scalability across diverse industrial environments. Experimental results highlight the strengths and limitations of each method, providing insights into their applicability for modern industrial anomaly detection challenges. Average Accuracy, Execution Time (e.g., CPU and GPU), Round Trip Time, and End-to-End Delay are used to evaluate the performance of the proposed approaches. The code will be available on GitHub, facilitating further research.