Genetic neural network and application in welding robot error compensation

Abstract

For the error analysis of a welding robot, based on the Vittorio granularity encoding, this paper proposes an enhanced genetic neural network using binary and real-valued blend encoding method. The neural network topology adopts binary encoding which reserves the virtues of Vittorio granularity encoding, and the connection weights adopt real-valued encoding, the Solis&Wets algorithm brings the virtues of evolutionary programming and evolutionary strategy to the new genetic algorithm. In addition, the combination of genetic algorithm and Solis&Wets algorithm, elitist preserving make the genetic search space more diverse and accelerate the convergence speed of genetic algorithm; dynamic parameter encoding substituting Vittorio granularity encoding not only improves the optimization accuracy of connection weights, but also avoids the fitness violent and discontinuous change due to the Vittorio granularity change. Simulation and experimental results verify this algorithm can overcome premature convergence of genetic algorithm and improve the robot pose accuracy effectively.