This chapter presents the neural network modelling of a single link flexible manipulator system. The modelling exercise is presented in two parts. The first part of the chapter uses two popular neural network structure, multilayer perceptron (MLP) and radial basis function (RBF). After appropriate training these are used to identify the dominant vibration modes of a flexible manipulator system. The system identification is realised by minimising the prediction error of the actual plant output and the model output. The second part of the chapter deals with the neural network modelling of dynamic systems. This is to perform parametric identification of a physical system and identify structural features and parameter values including the identification of the model structure. The neural network trained through supervised learning is used for both structure identification and parameter estimation. The technique is then used to model a flexible manipulator system using a composite input torque. The models are developed for hub angle, hub velocity and end-point acceleration.

Chapter Contents:

• 7.1 Introduction
• 7.2 Neural networks
• 7.2.1 MLP neural network
• 7.2.1.1 Backpropagation learning for MLP-NN
• 7.2.1.2 Improving backpropagation
• 7.2.2 RBF neural network
• 7.2.2.1 RBF-NN-functional description
• 7.2.2.2 RBF-NN learning algorithms
• 7.3 Modelling with NN
• 7.4 Model validation techniques
• 7.4.1 One-step ahead prediction
• 7.4.2 Model predicted output
• 7.4.3 Estimation set and test set
• 7.4.4 Correlation tests
• 7.5 Data pre-processing
• 7.6 Case studies
• 7.6.1 Case study 7.1: MLP-NN modelling using composite PRBS input
• 7.6.1.1 OSA prediction of hub angle
• 7.6.1.2 OSA prediction of end-point acceleration
• 7.6.1.3 MPO of end-point acceleration
• 7.6.2 Case study 7.2: RBF-NN modelling using composite PRBS input
• 7.6.2.1 OSA prediction of hub angle
• 7.6.2.2 OSA prediction of end-point acceleration
• 7.6.2.3 MPO of end-point acceleration
• 7.6.3 Case study 7.3: MLP-NN modelling using white noise input
• 7.6.3.1 OSA prediction of hub angle
• 7.6.3.2 OSA prediction of end-point acceleration
• 7.6.3.3 MPO of end-point acceleration
• 7.6.4 Case study 7.4: RBF-NN modelling using white noise input
• 7.6.4.1 OSA prediction of hub angle
• 7.6.4.2 OSA prediction of end-point acceleration
• 7.6.4.3 MPO of end-point acceleration
• 7.6.5 Comparative assessment
• 7.7 Summary

Inspec keywords: manipulator dynamics; vibrations; radial basis function networks; parameter estimation; learning (artificial intelligence); multilayer perceptrons; flexible manipulators

Other keywords: physical system; supervised learning; prediction error minimization; vibration modes; radial basis function; model output; hub angle; MLP; system identification; neural network modelling; dynamic systems; end-point acceleration; parametric identification; composite input torque; RBF; parameter estimation; single link flexible manipulator system; neural network structure; multilayer perceptron; structural feature identification; hub velocity; plant output

Subjects: Vibrations and shock waves (mechanical engineering); Neural computing techniques; Knowledge engineering techniques; Manipulators; Simulation, modelling and identification; Robot and manipulator mechanics