Dynamic Data Linkage Keys Generation Technology for Process Flow Data Analysis using Users Criteria

The digitalization of chemical processes, which use advancement of AI for data management and analysis, offers significant potential to boost operational efficiency and drive innovation in the pharmaceutical, chemical, and materials science industries. Process such as Single-Screw Extruder, Co-rotating extruder and Twin-Screw Extruders process, can benefit from digitalization, which could enhance the operation efficiency of mixing, compounding, processing materials, such as plastics and polymers. However, the digitalization of chemical processes could be challenging, as it requires managing independent diverse and complex process flow data analysis, especially when there is lack of critical Linkage Keys (LKs) for data linkage. In the extrusion process, new processing steps or materials often result in entity name changes. User-defined criteria rely on these changes because they require complete data linkage that matches the updated entity names. Existing data linkage methods like Re-Link[1] match words using lexical and semantic similarity. However, they often fall short when linking relationships based on knowledge, facts, and theories, as they primarily focus on mathematical similarity. Bit Vector [2] further enhances record matching and entity resolution by utilizing patterns. Despite these advancements, these methods struggle to achieve the knowledge-based linkage based on user criteria when entity names change, and when the same equipment name is referred to differently across datasets. To address these challenges, we propose a novel data linkage technique using Dynamic Multiple LKs (DMLKs), driven by user-defined criteria derived from historical data & expert knowledge. Our approach utilizes a two-stage ensemble ML model. In the first stage, the model generates numerous DMLKs by learning from meaning, facts, patterns, theories, principles, and procedures, which helps in recognizing entity name change based on diverse patterns and terminologies. In the second stage, these LKs are optimized using user-defined criteria, such as process name, material name, and sensing parameters, ensuring the LKs align with the most desired criteria, focusing on the most relevant data for effective linkage. Our technique achieved a maximum 83.4% precision score and a 12.6% improvement over the Re-Link. This improvement demonstrates our model's ability to prioritize the most relevant data features based on user criteria. By simplifying the linkage of complex chemical processes, the DMLKs approach enhances the management of independent process flow data, significantly boosting operational efficiency in the chemical industry, particularly in extrusion processes.