In this thesis, an intelligent controller for gun and/or sight stabilization of turret subsystems is developed using artificial neural networks. A classical proportional, integral and derivative (PID) controller equipped with a non-linear unbalance compensation algorithm is used as the low-level controller. The gains of this PID controller are tuned using a multilayered back-propagation neural network. These gains are modeled as a function of the error between the command and feedback signals and this model is generated by the function fitting property of neural networks as an estimate. The network is called as the &ldquo / Neural PID Tuner&rdquo / and it takes the current and previous errors as inputs and outputs the PID gains of the controller.
Columb friction is the most important non-linearity in turret subsystems that heavily lower the efficiency of the controller. Another multilayered back-propagation neural network is used in order to increase the performance of the PID controller by identifying and compensating this Columb friction. This network utilizes the error between the output of the PID controller driving the physical system with Columb friction and the output of the identical PID controller driving a virtual equivalent linear system without Columb friction. The linear dynamics of the physical system is identified using a single layer linear neural network with pure linear activation function and the equivalent virtual linear system is emulated using this identification.
The proposed methods are applied to both computer simulations and hardware experimental setup. In addition, sensitivity and performance analysis are performed both by using the mathematical model and hardware experimental setup.
| Identifer | oai:union.ndltd.org:METU/oai:etd.lib.metu.edu.tr:http://etd.lib.metu.edu.tr/upload/12607852/index.pdf |
| Date | 01 November 2006 |
| Creators | Gumusay, Ozdemir |
| Contributors | Erkmen, Aydan M. |
| Publisher | METU |
| Source Sets | Middle East Technical Univ. |
| Language | English |
| Detected Language | English |
| Type | M.S. Thesis |
| Format | text/pdf |
| Rights | To liberate the content for public access |
Page generated in 0.0175 seconds