Treffer: ANOMALY DETECTION USINGARTIFICIAL NEURAL NETWORKSIN A FEDERATED LEARNING SETUPWITH PYTHON SOCKETINTEGRATION

In the realm of manufacturing, ensuring optimal product quality is of utmost importance. This emphasizes the necessity for efficient anomaly detection, which identifies deviations from established standards to prevent malfunctions, defects, and subpar performance. With machine learning emerging as a quite useful tool in anomaly detection, there is a growing concern over the confidentiality of the substantial data required for machine learning. FederatedLearning offers a promising solution, enabling collaborative learning without the need for direct data sharing, thereby safeguarding data security. The integrity of data and efficiency in anomaly detection not only ensures product quality but also reduces potential costs and protects market reputation. The balance of data security and efficient machine learning is essential for the modern manufacturing industry. This study makes use of a specialized modular manufacturing simulator, a result of a strategic collaboration between MDU and ABB. This simulator, mimicking real-world manufacturing scenarios such as a modular ice-cream factory, presents a complex system filled with numerous Internet of Things devices. The research emphasized the merging of software engineering principles with AI-driven anomaly detection. An important part of this research is the Modular Ice cream factory Dataset on Anomalies in Sensors, which has been obtained from the simulator. Utilizing Python socket-based server-client architecture and federated learning approach, specifically artificial neural networks, and comparing their accuracy and performance with a centralized approach using the modular ice cream factory dataset on anomalies in sensors. Results reveal the comparative effectiveness of the federated artificial neural networks, slightly outperforming the centralized artificial neural networks with a 1.09% higher average accuracy. Federated approach looks promising, but we still need to explore its effects on how long processes take in