 |
Visits: 85174464
AMS now reading: 1
|
|
|
|
|
 |
|
|
| Žilková J., Timko J. |
| NEURAL MODELLING OF A TINNING LINE’S TECHNOLOGICAL PART |
Abstract
The present article deals with a proposal of the neural model of a continuous tinning line’s technological part. The tinning line is intended for production of sheets for packaging purposes.
The conti-line’s technological section is made up of six controlled electrical drives with separately excited motors that allow to separately control individual electrical, mechanical and technological variables so that the specified manufacturing technology would be adhered to. Our intention was to monitor speed of the belt in any technological line sections with the use of the line neural model. Proposed for individual drives were controllers of both the current and speed, when all six sections of the line were modelled within the Matlab-Simulink program environment.
Considered at suggesting neural networks were multi-layer forward and cascade neural networks. Elected for individual neural models’ inputs were values of the belt requisite speed, observed values of the belt speed in (k-1) step and armature currents in k-th, (k-1) and (k-2) steps for individual drives. Used for the output layer neurons were linear and, in the hidden layer, tan-sigmoid activation functions. The neural networks were proposed in the Matlab environment by using the Neural Networks Toolbox, and they were tested by and in the Simulink program. Whereas at final testing more favourable results have been attained with cascade networks any belt speed observers are resolved as cascade networks. Training of individual neural networks was performed off-line, whilst adaptation of the network parameters was ensured by Levenberg-Marquard’s modification of the back-propagation algorithm.
Assessed at the end of the article are the results and the mentioned line speeds’ waveforms observed at testing one of the neural models.
|
| Keywords: estimation|modelling|tinning line|DC motor|artificial neural network|cascade network|feedforward network| |
| No 3 (2006), p. 282-290 |
|
|
|
|
|
|
|
Editors and press
AMS online
