Design and Development of a Multi-Agent System Environment to Monitor and Analyze Data Originating from Automated Test Equipment

The Industrial Internet of Things, Big Data, and Artificial Intelligence are among the pivotal technologies fueling the Fourth Industrial Revolution. SPEA SpA, an Italian company manufacturing Automatic Test Equipment, aims to develop a series of software tools to better monitor the health and productivity of their machines and implement Predictive Maintenance. This thesis focuses on the project and development of a novel Multi-Agent System designed for such purpose. A Multi-Agent System or MAS tackles complex problems by decomposing them into smaller tasks and assigning them to intelligent units described as agents; in doing so it exploits the emergence of intelligence from the interaction of individuals, and can autonomously devise solutions where a monolithic system would fail or be too difficult to realize. Such an architecture offers expandability, high efficiency, and robustness, parallelizing computations and overcoming individual faults. Moreover, the proposed design allows the possibility of developing a single module and reusing it at different levels (namely, machine, plant, and corporate), creating a system that is easy to configure and maintain, and can be expanded with new features at will. In fact, the subdivision and isolation of tasks lay the ground for streamlined development, testing, and deployment of agents as the system evolves. The key contributions of this thesis include the study of the core components, the Agents Manager Program, and the Database structure and synchronization mechanism, along with a Zero-Configuration Networking protocol to manage the connection of equipment transparently. The MAS proposed in this thesis is based on Simple Reflex Agents operating, for the most part, in a structured MongoDB environment, which acts as a blackboard for agents to communicate the result of their computations. Data integrity and near real-time reactivity are paramount, and so is robustness as the system works in multi-threading and is designed to prevent and withstand agent malfunction. Considerations about agent scheduling are also taken into account as a topic of preliminary experimentation to be expanded upon at a deeper stage of development. The schema flexibility of the document-based technology renders the system adaptable to add-ons making it future-proof, and the horizontal scalability well matches the need for managing the large quantities of data generated by several machines in high throughput industrial settings, which are essential and lay the basis of the soon to be implemented Machine Learning algorithms for predictive maintenance. To evaluate the results, several qualitative and quantitative tests have been performed both in simulated environments and on-premise, meeting and exceeding the requirements, and confirming the applicability of the architecture to high throughput production environments. The system is demonstrated to be reliable and robust, and the synchronization works promptly and transparently. Due to its potential, the system calls for further research and development for deployment readiness, positioning it as a foundational step toward advanced predictive maintenance solutions.