Data-Driven Digitalisation of Batch Operations through Automated Reconstruction and Clustering of Process Flow Diagrams

• Automated reconstruction of batch process flow diagrams from historical data • Unsupervised clustering separates formulations by operational process structure • Master-level PFDs reveal redundant steps and quantify batch cycle time variability • Industrial case study demonstrates improved digital maturity and process insight The formulated products industry continues to face challenges stemming from limited digital maturity, such as inconsistent batch-to-batch operational procedures, redundant processing steps, and large variations in batch-cycle times across production campaigns. To address these challenges, this work proposes a domain knowledge guided and fully automated framework that reconstructs and digitalises batch process flow diagrams (PFDs) directly from historical data. The proposed methodology operates in three stages. First, batchwise PFD generation identifies and validates process actions from raw data by combining domain-specific knowledge with data-driven pattern recognition to produce a distinct PFD for each batch. Second, an unsupervised clustering algorithm groups batches with similar operational structure, separating batches of distinct formulations into different clusters. Third, a master-level PFD construction algorithm aggregates the clustered batchwise PFDs to recover the underlying process structure to identify the average and optimum PFD structures. The methodology was demonstrated using an industrial case study involving two formulated products. The results prove that the proposed framework is able to successfully construct accurate batchwise and master-level PFDs from raw data. This study therefore showcases how to automatically digitise vast quantities of historic batch data into representative and optimal PFD structures, paving the way for digital maturity in the formulated products industry.