Permanent Magnet Synchronous Motors (PMSM) have the advantages of high air gap flux density, small torque ripple, large torque/inertia ratio, and high efficiency, and have been widely used in small and medium-capacity servo systems. Due to the complex operation of the servo system, the PMSM itself is a multi-variable, nonlinear, and strongly coupled system. Therefore, the general controller adopts a PI controller and is easily affected by uncertainties such as motor parameter changes and load disturbances. Dynamic response and immunity to interference cannot be well balanced. In order to overcome the shortcomings of PI controllers, various control strategies to eliminate the influence of uncertainty have been proposed [1–5]. However, these strategies are basically based on a linear design model. Actually, the electromagnetic torque inevitably includes ripple torque such as ripple torque and cogging torque. In addition to the uncertainty of the load torque, there is a large amount of nonlinearity and uncertainty in the torque, sometimes causing The quality of control has been severely degraded. As a result, research has been carried out on reckless people, professors and doctoral supervisors who are engaged in intelligent control, multivariable control and their application in AC drives.
Stickiness is not guaranteed. This paper regards the pulsating torque and the load torque as measurable disturbances, and proposes a pole configuration self-tuning feedforward control (PPST FC) strategy for permanent magnet synchronous motor based on neural network.
The pole configuration self-correcting feedforward control combines feedforward and feedback to implement an adaptive pole configuration of the system, but it alone cannot solve the measurement problem of interference. Neural networks have been widely used as a new means in system identification and control, and have been applied in the field of AC drives [6, 7]. This paper uses a recursive neural network as a load torque observer and literature [8]. The method proposed in the paper provides fast training for recursive neural networks.
2 The mathematical model of PMSM permanent-magnet synchronous motor controlled by vector vector is that in the formula, u and u are d and q-axis stator voltage respectively, i and i are d and q-axis stator currents, and d and q-axis stator magnetic sum are respectively L is the d and q axis inductances of the stator windings, R is the stator resistance, is the electrical angular velocity, p is the differential sign, L is the d-axis mutual inductance between stator and rotor, I is the equivalent d-axis excitation current of the permanent magnet, P For pole pair numbers, T is the electromagnetic torque, T is the load torque, J is the moment of inertia, B is the damping coefficient, and is the rotor angular velocity.
The basic principle of permanent magnet synchronous motor control is vector control. If i = 0, the d-axis stator flux linkage does not change, and permanent magnet synchronous motor neutralization I is a constant, so the electromagnetic torque T is proportional to i, ie, 8) Substituting into (7) the frequency domain model can be added to the zero-order keeper and z-transformation of equation (9) to obtain the discretization equation as the 3-point neural network integrated load torque observer before PMSM pole configuration self-correction The feedforward control 3. 1 pole configuration self-correcting feedforward controller expresses (10) as (k) as input, and the structure of the control system is shown in Fig. 1. The conventional pole configuration design method [9] can be used to control The equation is substituting u(k) in equation (12) into equation (11) to obtain a given stable closed-loop pole polynomial T(z), resulting in a pole configuration equation) coprime, and equation (14) has a unique solution.
The choice of polynomial D(z) is the same as for the conventional linear object design. In order to effectively eliminate the static error, an integrator is added to the controller, so D(z) satisfies the changes of the parameters of the object model control and decision due to changes of the PM SM parameters and load disturbances, and an explicit algorithm is used to first identify the process model. Parameters, then find the control law. Applying a recursive least-squares algorithm with forgetting factor to estimate the parameter in the equation 3.2. The neural load of the integrated load torque observer PM SM actually contains the ripple caused by the back-EMF or stator current harmonics. Torque and cogging torque caused by the interaction of the stator core with the rotor magnetic field (also known as the cogging effect). The ripple torque is related to the stator current and the rotor position. The cogging torque is related to the rotor position, and the relationship among them is very complicated and difficult to accurately represent. In addition, the load torque itself has a lot of non-linearity and uncertainty. So PM SM simplifies the accurate calculation of the model. For this reason, based on the fact that the main load torque in the actual situation is basically known, the load torque that is known to the actual system is directly introduced and set to T(k) and the uncertainties are added to the ripple rotation in the electromagnetic torque. The pulsating torque, such as moment and cogging torque, is trained using a recurrent neural network and is set to f(k). Because the recursive neural network can directly identify the black box system (see Figure 1), there is a comprehensive load torque for the diagonal recursive neural network and fast training algorithm used in [9].
The steps of PMSM pole configuration self-tuning feedforward control with neural network integrated load torque observer are as follows: 1) Determine the desired pole polynomial T(z 2) to measure the output y(k) of the object and determine T(k) , Using neural network to find f 3) Model identification of formula (17) 4) On-line identification of neural network 4 Experimental simulation Using the PMSM vector control system of Kollmorgen, the experiment was conducted, PM SM parameters. Use a DC motor to provide a definite load torque. A variable inertia mechanism is mounted on the permanent magnet synchronous motor, and the rotor inertial moment and the damping coefficient are changed by placing different iron plates thereon.
In the experiment, the speed is given as 700 r min 1, and the external load torque T = 0 is started. At 1 s, the load torque of 3 N m is added. Experiments were conducted in 3 cases: 1) Rated condition 2) Increase J about 5 times, B not change 3) Increase B about 5 times, and J does not change. For the comparison of effects, experiments were performed using a conventional PI speed controller. A comprehensive load torque observer consisting of a recursive neural network was first trained offline for rated no-load conditions. The number of online training steps is 5 and the forgetting factor is 0. 99. Figure 2 shows the result of PI control. Figure 3 (a) and (b) are the control results of the added load torque as T(k), respectively. Comparison shows the effectiveness of the proposed control strategy.
The PPSTFC response of the detector (a) Considers the applied load torque as the result of f(k). (b) Considers the added load torque as the result of T(k). Li Hongru et al: Band neural network observer The permanent magnet synchronous motor pole configuration self-correction feedforward control 5 Conclusion This paper presents a pole-configuration permanent magnet synchronous motor with self-tuning feedforward control strategy with neural network integrated load torque observer. A comprehensive load torque observer is constructed by a recursive neural network, so that the integrated load torque is regarded as measurable disturbance, and a self-correcting feedforward control of the pole configuration of the permanent magnet synchronous motor is realized, and uncertainties such as parameter changes and load disturbances are performed. Effective feed forward compensation. Theoretical analysis and experimental simulation prove that the proposed control strategy has strong robustness and is obviously superior to the traditional control strategy.
Despite the adoption of a fast training method, the on-line training of a comprehensive load torque observer composed of recursive neural networks cannot be optimized due to the limitation of the method and the calculation speed of the microcomputer. In this paper, the known load torque is directly introduced, and the method of training using the recurrent neural network for the uncertain part is very practical and effective. The experimental simulation results fully demonstrate that the direct introduction of the load torque has the best control effect.
Replacement Uv Lamp,Germicidal Uv Lamp,Ultraviolet Uv Light,Uv Lamp R-Can
Foshan Kingrate Optoelectronic Technology Co., Ltd. , https://www.kingrateuv.com