Search

A new principle MOEMS accelerometer based on the magneto-optical rotation effect is developed. The design of the magnetic film integrates with the proof mass, using a magneto-optic crystal to detect the magnetic field changes induced by the external acceleration to produce the magneto-optical rotation effect. By detecting the optical rotation angle, the high precision measurement of the acceleration is realized. The structure of the accelerometer was fabricated through structural design, simulation and micro-nano processing technology. The test results show that the noise spectral density of the MOEMS accelerometer is $10.5\mu \text{g}/\surd $ Hz at 1Hz, the bias instability is $37.6\mu \text{g}$ and the output sensitivity is 330.2mV/g. [ABSTRACT FROM AUTHOR]

Due to the high cost and shortage of rare earth materials, electrically excited synchronous machines (EESMs) become a potential alternative for electric vehicles. With the help of inductive power transfer (IPT), the EESM would easily realize the brushless excitation to improve its reliability. However, the dependence of IPT and EESM leads to significant control challenge, especially considering the inductive load characteristics. When the traditional modeling approaches are applied to the whole system, a complicated 13th-order model is built, which is unfavorable for dynamic analysis. For simplification, this article proposes a systematic and universal reduced-order model approach based on the local linear approximation of resonant voltage/current. A reduced ninth-order model of system is then obtained based on the approach. The frequency domain characteristic of primary-side control is analyzed, and the reduced-order model has high accuracy compared with the full-order one. Finally, the experimental results are given out to verify the effectiveness of the proposed modeling method. [ABSTRACT FROM AUTHOR]

Recent research on nonlinear multimode interference (NL-MMI) based mode-locking method provides a new approach for ultrafast all-fiber lasers. Here, we report on the polarization dynamics of group velocity locked vector solitons (GVLVSs) and noise-like vector pulses (NLVPs) in a graded-index multimode fiber (GIMF) based all-fiber laser. Due to the non-polarization sensitive characteristic of the NL-MMI, emitted pulses automatically evolve into the vector multi-pulse complexes constituting two orthogonally polarized components. Beyond the traditional polarization resolved measurement, dispersive Fourier transform (DFT) and polarization beam splitting techniques are jointly adopted to enable insights into the transient polarization dynamics of the vector solitons. It is found in real time that the two orthogonally polarized components oppositely shift their central wavelengths to compensate the polarization dispersion. In addition, stationary and pulsating NLVPs are respectively unfolded by the DFT based polarization resolved measurement. Especially, the pulsating NLVPs before and after polarization resolved measurement are accompanied with the identical periodic spectral breathing process. All these findings reveal the vectorial nature in NL-MMI based ultrafast fiber lasers, as well as highlight the transient polarization dynamics and the long-term pulsation in dissipative systems. [ABSTRACT FROM AUTHOR]

The application and theoretical analysis of fault tolerant learning are very important for neural networks. Our objective here is to realize fault tolerant sparse multilayer perceptron (MLP) networks. The stochastic gradient descent method has been employed to perform online learning for MLPs. For weight noise injection-based network models, it is a common strategy to add a weight decay regularizer while constructing the objective function for learning. However, this ${l}_{2}$ -norm penalty does not generate sparse optimal solutions. In this paper, a group lasso penalty term is used as a regularizer, where a group is defined by the set of weights connected to a node from nodes in the preceding layer. Group lasso penalty enables us to prune redundant hidden nodes. Due to its nondifferentiability at the origin, a smooth approximation of the group lasso penalty is developed. Then, a rigorous proof for the asymptotic convergence of the learning algorithm is provided. Finally, some simulations have been performed to verify the sparseness of the network and the theoretical results. [ABSTRACT FROM AUTHOR]