Twin migration modeling and fault diagnosis of traditional Chinese medicine granulation production line under variable working conditions

Traditional Chinese medicine (TCM) pharmaceutical production is characterised by a wide variety of products, small batch sizes, and complex and variable operating conditions. Digital twin models established for specific scenarios lack the ability to adapt to changes in operating conditions, making it difficult to quickly and accurately identify equipment faults in pharmaceutical production. This study proposes a research method for twin migration modelling and fault diagnosis of traditional Chinese medicine granulation production lines under variable operating conditions. By analysing the factors influencing dynamic changes in operating conditions caused by pharmaceutical production, a framework for the adaptive migration of pharmaceutical process twin models is established. This framework analyses and judges the fault states and time-varying characteristics of processing equipment for different product specifications. The study introduces the Swin Transformer, CNN, and GRU to fuse spatial and temporal features, and designs a multi-level adaptive migration strategy including identification, training, updating, and prediction. This approach enables rapid model adaptation and maintains high performance under new operating conditions, addressing the challenges of knowledge transfer under conditions of data scarcity, similar faults within and different faults between equipment, and heterogeneous equipment, thereby effectively improving the prediction accuracy of equipment faults under complex operating conditions. Experimental results show that in typical variable operating condition scenarios such as significant fluctuations in process parameters and product batch switching, the fault prediction accuracy reaches 0.98, validating the effectiveness and practicality of the method. This study achieved adaptive updates of multi-product, variable-condition pharmaceutical process twin models. The fault prediction error of the transferred model is below 0.05. The developed method can be applied to precise fault prediction of equipment under other complex conditions, providing