Method and arrangement for neural modeling of a paper winding device

A method for generating optimal control quantities for a paper winding device depending on a desired winding quality of the paper coil, further comprising the step of:

• a) supplying influencing quantities and control quantities of the paper winding device to be optimized in a first optimization step to the neural model prepared according to the following steps of:
• measuring and storing influencing and control quantities of a paper winding device in a first step for training of a neural network at least depending on a wound number of layers and an appertaining coil radius of a paper coil,
• determining one of web strength in a paper web and a quantity correlated with the web strength depending on the influencing and control quantities of the paper winding device as target quantity for the neural network from a relationship between the coil radius and the number of layers of the paper coil measured in said measuring and storing step, and
• training the neural network as a model of the paper winding device via a conventional learning method at least with the influencing and control quantities of the paper winding device as input quantities and one of the web strength and the quantity correlated with the web strength as output quantities, and
• determining one of a first web strength and a first quantity correlated with the first web strength from the neural network with the aid of the model
• b) determining a first winding quality in a second step from the known relationship between the one of the web strength and the quantity correlated with the web strength and a winding quality from the first web strength or the correlated first quantity, and
• c) comparing the first winding quality to the desired winding quality and supplying at least second control quantities of the paper winding device to be optimized in a further optimization step to the model, continuing until the winding quality determined from a) and b) sufficiently accurately corresponds to the desired winding quality.

An operating method for a paper winding device, comprising the step of:

• supplying influencing quantities and control quantities of the paper winding device to be optimized in a first optimization step to the neural model prepared according to the following steps of:
• measuring and storing influencing and control quantities of a paper winding device in a first step for training of a neural network at least depending on a wound number of layers and an appertaining coil radius of a paper coil,
• determining one of web strength in a paper web and a quantity correlated with the web strength depending on the influencing and control quantities of the paper winding device as target quantity for the neural network from a relationship between the coil radius and the number of layers of the paper coil measured in said measuring and storing step, and
• training the neural network as a model of the paper winding device via a conventional learning method at least with the influencing and control quantities of the paper winding device as input quantities and one of the web strength and the quantity correlated with the web strength as output quantities, and
• determining one of a first web strength and a first quantity correlated with the first web strength from the neural network with the aid of the model,
• determining a first winding quality in a second step from the known relationship between the one of the web strength and the quantity correlated with the web strength and a winding quality from the first web strength or the correlated first quantity, and
• comparing the first winding quality to the desired winding quality and supplying at least second control quantities of the paper winding device to be optimized in a further optimization step to the model, continuing until the winding quality determined from said determining steps sufficiently accurately corresponds and the desired winding quality,
• supplying control quantities to the paper winding device.