Designing the dq-frame current regulator for single-phase voltage-source inverters is a very challenging task. Since only one real current signal exists in the circuit, an orthogonal signal generation (OSG) block is required to generate the virtual orthogonal signal. Thus, ac variables can be turned into equivalent dc quantities through an αβ/dq transformation. However, the OSG block makes the control system complex and introduces an extra transient disturbance. Consequently, the dynamic performance is deteriorated. In this study, the reference-current-based OSG method is analysed thoroughly. Based on this structure, the dq-axes decoupling control, which is widely discussed for three-phase systems and usually neglected for singlephase systems, is studied. Two decoupling techniques, i.e. the reference-current feed-forward control and the quasi-complex vector proportional-integrator control, are implemented and analysed. The proposed theories and control schemes are evaluated by experimental results.

A full digital deadbeat controller is proposed for the speed loop to obtain better dynamic response performance in permanent magnet synchronous motor servo system. Given an expected target error function and speed command in Z-domain form, according to the deadbeat zero-pole configuration rules, the digital controller is derived to make the current command output stable in several sampling periods and the speed response also becomes stable without ripples. To reject disturbance from the load variation, an estimation method, based on model reference adaptive system through Popov's hyper stability criteria, is presented to identify online the moment of inertia and load torque. The two quantities are essential for the speed loop to achieve adaptive control by load compensation. The simulation and experimental results reveal the effectiveness of the proposed speed controller.

No abstract

No abstract

In order to identify parameters of permanent magnet synchronous motor(PMSM) on-line, single-layer neural networks (SLNN) with gradient descent is proposed. SLNN can study and adapt itself by change its weigh values while PMSM is running. The output of PMSM's status variants is a function about the estimated parameters, including stator resistance, d-q axial inductance, rotor flux and moment of inertia, which are included in SLNN's weight vector, so the estimated parameters can be iterated in SLNN after computing their gradients. Changing the learning rate of SLNN makes it available to choose the emphasis on estimated accuracy or on convergence rate. The servo PI parameters are adjusted according to the identified values .The experimental results and simulations have illustrated its simplicity, validity and efficiency.