Background: Modern spinning preparatory has undergone drastic technological changes, but still, individual’s expertise-based decisions govern the complex and non-linear multivariant relationships prevailing amongst raw material (cotton) variables, machine variables, process variables (waste), and product (card sliver) quality. The scientifically precise prediction regarding the cleanliness and quality of card sliver and waste control for the given inputted cotton variables processed on the state-of-the-art machinery setup without waiting for the production and testing of card sliver is still impossible. Methods: The present work describes the use of Aritificial Neural Networks (ANN) for ruling out these limitations on scientific grounds. A complex system targeted at ANN was developed using the "newff" function on the mill's five-year database. Single-group ANN was initially designed to determine the influence of inherent variations in raw cotton fibre properties and trash content on blow room and card performance. A multi-group approach of ANN was developed at a later stage to define the influence of complex interactions amongst various fibre properties on three main quality measures considered at blow room and carding, viz., i) influence of blow room and card on fibre length properties, ii) fibre damage at blow room or improvement at card, and iii) degree of cleanliness of the output material. Results: Reverse modelling for both groups was also successfully designed to demarcate feed cotton quality and cleanliness requirements for targeted blow room or card cleaning performance. Conclusion: A high level of positive endurance was observed for all ANNs. Multigroup networking has proven to be more precise than single group networking.
Read full abstract