| 摘要: | 近年來，主動式磁浮軸承(active magnetic bearing, AMB)由於具備非接觸(noncontact)與無摩擦(frictionless)特性，已經成功且廣泛地實現在各種應用之中，包括非旋轉裝置與旋轉裝置。而對於非旋轉裝置最重要之控制目標為受控體必須能達到精密定位或精確追隨預先設定之軌跡；另一方面，對於旋轉裝置最重要之控制目標為轉軸必須能夠精準地被調節與維持在狹小孔隙內的中心位置。有鑑於此，本論文之目標即為發展一個包含兩個徑向主動式磁浮軸承(radial AMBs)與一個推力型主動式磁浮軸承(thrust AMB)的全懸浮五軸主動式磁浮軸承控制系統(fully suspended five degree-of-freedom AMB control system)，以同時建立旋轉裝置與非旋轉裝置之控制技術，並滿足實際的應用需求。 論文中首先詳述磁浮軸承的系統架構、動態分析與實驗設計，再針對推力型主動式磁浮軸承提出不同的追隨控制方法包括無模型控制(model free control)法如以赫米特多項式為基礎之遞迴式類神經網路(Hermite polynomial-based recurrent neural network, HPBRNN)控制；結合滑動模式控制(sliding-mode control, SMC)、適應控制(adaptive control)、遞迴式赫米特類神經網路(recurrent Hermite neural network, RHNN)優點之智慧型滑動模式控制(intelligent SMC)法如適應性互補式滑動模式控制(adaptive complementary SMC, ACSMC)與強健性無奇異終端滑動模式控制(robust non-singular terminal SMC, RNTSMC)等，以實現轉軸在軸向之精密軌跡追隨控制。而針對五軸主動式磁浮軸承系統的五軸同時控制，本論文先提出分散式比例-積分-微分類神經網路控制器(decentralized proportional-integral-derivative neural network controller, PIDNN)，使控制器具備線上控制增益調整能力。再提出一解耦動態模型(decoupled dynamic model)，將原本各軸耦合之五軸主動式磁浮軸承系統轉換為五個獨立之子系統，使系統可以分散式控制概念進行控制系統之設計。基於該解耦動態模型，本論文再提出分散式智慧型雙重積分滑動模式控制系統(decentralized intelligent double integral SMC system, IDISMC)以進一步提高控制品質，其中所設計之雙重積分滑動面使控制律具有積分控制特性，且控制律中各項控制增益及系統不確定性可同時藉由本論文所提出之改良型比例積分微分類神經網路估測器(modified PIDNN observer, MPIDNN)分別進行線上調整與估測，可有效減少穩態誤差，適合用於高度非線性、易受外來干擾影響與精度要求高之五軸主動式磁浮軸承系統。 本論文以個人電腦(personal computer, PC)為實驗平台，並設計多種測試條件與性能量測，對使用不同控制方法之推力型主動式磁浮軸承系統與全懸浮五軸主動式磁浮軸承系統分別進行可行性之驗證。由實驗結果可知，使用本論文所提出之控制方法，推力型主動式磁浮軸承控制系統在外加0.38kg負載時，追隨誤差平均值仍可維持於1μm以內；而五軸主動式磁浮軸承系統在轉速達到4800RPM時，各軸偏差量之均方根值仍可穩定調節於0.1mm以內，亦即1/4氣隙中。因此，本論文所發展之推力型主動式磁浮軸承控制系統與全懸浮五軸主動式磁浮軸承控制系統，確實均具備優異之控制特性與強健性。In recent years, active magnetic bearings (AMBs) with noncontact and frictionless characteristics have been successfully and widely implemented in various kinds of applications including the non-rotating and rotating devices. The most important control object for the non-rotating devices is the controlled devices should be positioned or tracked to the pre-defined trajectories precisely. On the other hand, the most important control object for the rotating devices is the controlled rotor should be regulated and stabilized in the centers within the narrow aperture perfectly. For this reason, the purpose of this dissertation is to develop a fully suspended five degree-of-freedom (DOF) AMB control system, which is composed of two radial AMBs (RAMBs) and one thrust AMB (TAMB), to build up the technologies for both non-rotating and rotating devices and fulfill the requirements of the practical applications. In the beginning of this dissertation, the system dissections, dynamic analyses, and experimental designs of the AMB systems are presented in detail. Then, various tracking controllers including model-free control methods such as Hermite polynomial-based recurrent neural network (HPBRNN) control; intelligent sliding-mode control (SMC) methods, which combine the merits of SMC, adaptive control, and recurrent Hermite neural network (RHNN), such as adaptive complementary SMC (ACSMC) and robust non-singular terminal SMC (RNTSMC) and so on are proposed to control the rotor in the axial direction of the TAMB system for the tracking of various reference trajectories. Moreover, to control the five-axes of the five-DOF AMB system simultaneously, a decentralized proportional-integral-derivative neural network (PIDNN) controller with on-line tuning control gains is proposed first. Then, a decoupled dynamic model is proposed to transfer the original coupled five-DOF AMB system into five independent subsystems for the purpose of decentralized control. Based on the decoupled dynamic model, a decentralized intelligent double integral SMC (IDISMC) system is further proposed to improve the control performance of the five-DOF AMB control system. In the control law of the IDISMC, the designed double integral sliding surface reinforces the control law with integral control feature. Furthermore, the control gains of the IDISMC can be adjusted on-line and the system uncertainty can also be observed simultaneously by using of a modified PIDNN (MPIDNN) observer. Thus, the proposed IDISMC system with reduced steady-state error is suitable for the highly nonlinear, highly sensitive to the external disturbance, and high precision requirement five-DOF AMB system. In this dissertation, the validities of the TAMB and five-DOF AMB control systems using different control methods are verified by various testing conditions and performance measures in personal computer-based experimentation. According to the experimental results, the average tracking error of the TAMB system with 0.38kg load can be kept within 1μm and the root mean square of regulating error of the five-DOF AMB system operated at 4800RPM can be maintained within 0.1mm, i.e. 1/4 air gap. Obviously, both the developed TAMB and five-DOF AMB control systems possess good control performances and robustness. |
