Adaptive neural speed controllers applied for a drive system with an elastic mechanical coupling – A comparative study

• Neurocontrollers with on-line weight adaptation are used in a drive system with elastic coupling.
• The developed adaptive neurocontrollers use only easy measurable motor speed.
• An original fuzzy learning factor of all adaptation algorithms of NN is proposed.
• The ability of neurocontrollers for torsional vibration damping was analyzed and compared.

This paper presents an analysis and comparison of neural-adaptive controllers applied in a control structure of an electrical drive with an elastic mechanical coupling between the driving motor and a load machine, using only one state variable used in the feedback loop (a motor speed). However, the presented considerations can be assumed as a general neural speed control of the drive with a fast enough electromagnetic torque control loop of an electrical machine. This is justified by analogy with a design process independent of the parameters of a specific drive system and its electromagnetic torque control loop. Four types of neuro-controllers and training methods are analyzed: Adaptive Linear Neuron with Delta Rule, Multi-Layer Perceptrons Neural Network with the Backpropagation method, Feedforward Network with Adaptive Interaction adaptation and Radial Basis Function Neural Network with gradient algorithm, applied as speed controllers. Two main problematic issues related to neural controllers trained on-line are discussed: initial parameters selection for a neural network and determination of learning factors used in adaptation algorithms. Simulations are confirmed in experiment tests, using dSPACE1103 card. All the tested neurocontrollers are compared to a classical PI solution with one state variable used in the feedback loop of the analyzed drive system.

Figure optionsDownload as PowerPoint slide