Prediction of creel change delays in carpet production via artificial neural network approach

In this study, it is aimed to predict the carpet production process and to provide an important source of information for the production planning and sales unit by predicting creel change times. A decision model is proposed for the weaving unit involved in the carpet production to predict the creel change process; thereby, prevent possible delays and deliver the customer demands on time. This process is selected because production stalls in the changes made entirely, and that process disruptions will be experienced in delays on due dates. Various artificial neural network methods have been tried to predict the creel change delays, and a new hybrid model has been obtained to find the best prediction method and to provide much more rapid and high accuracy production optimization compared to classical methods. The dataset having 308 test samples constitutes five input parameters: comb number, color, bench width, loom coil number, and changing coil number, and one output parameter as the creel change delays. These four models are compared to each other using the mean absolute percentage error (MAPE) value to find the most accurate prediction model. It is aimed to select the best structures by evaluating the different structures of each model using the grid-search approach. According to best found results for all models, the MAPE values of the test dataset were calculated as 24%, 30%, 33%, and 18%, in MLP, GRNN, linear regression and hybrid model respectively. Consequently, the hybrid model is determined as the most accurate method when compared to other models. Keywords: Artificial neural network, Hybrid neural networks, Multi layer perceptron, Generalized regression neural networks, Creel change prediction