A Neural Network Model for Modeling and Analysis of Ink Printing Systems

This paper proposes a method for constructing a dynamic recursive neural network model for research and analysis the processes of ink circular and axial distribution and transfer in ink printing systems. The structural and mathematical model of the neural network is developed, which takes into account the parameters of the ink printing system elements and the operating modes of the ink feeder and ink distributing subsystems; the oscillator, plate and offset cylinders. The informational model of the training process the recursive neural network of the ink print system is presented, which is based on the correction of synaptic weights by the method of backpropagation through time. Training of the neural network is based on the parameters of the input zonal distribution and the ink thickness in the corresponding zones of the imprints obtained as a result of the physical experiment and the form data. Trained network is used for modeling and researching the processes of the ink distribution and transfer in the ink printing system. In addition, this network is suitable for determining the parameters of the input distribution for printing forms with different structure and density of filling their printing elements. The approbation of the neural network based on the nine-zone test form. The efficiency of the neural network is confirmed by the convergence of the simulation results and the physical experiment.