Neural Network based ModelReference Adaptive Control Structure for aFlexible Joint withHardNonlinearities
Pierre Sicard · 2004
Thispaperproposes acontrol strategy basedon artificial neural networks (ANN)forapositioning system with aflexible transmission element, taking intoaccount Coulomb friction forbothmotorandload, andusing axariable learning rateforadaptation toparameter changes andtoaccelerate convergence. Theinverse modelofthis system isunrealizable. Thecontrol structure consists ofafeedforiard ANN thatap- proximates theinverse ofthemodel, anANN feedback control law, areference modelandtheadaptation process oftheANNs withvariable learning rate.Inthis structure, thelearning rate ofthefeedback ANN issensitive toloadiner-tia variations. The contribution ofthis paperistoresolve thisweakness bypro- posing asupervisor thatadapts theneural networks learning rate. Simulation results highlight theperformance ofthecon- troller tocompensate thenonlinear friction terms, inparticu- larCoulombfriction, andflexibility, anditsrobustness tothe loadanddrive motorinertia parameter changes. Internal sta- bility, apotential problem withsuchasystem, isalsoverified.