Search

Recent studies have shown great performance of Transformer-based models in long-term time series forecasting due to their ability in capturing long-term dependencies. However, Transformers have their limitations when training on small datasets because of their lack in necessary inductive bias for time series forecasting, and do not show significant benefits in short-time step forecasting as well as that in long-time step as the continuity of sequence is not focused on. In this paper, efficient designs in Transformers are reviewed and a design of decomposing residual convolution neural networks or DRCNN is proposed. The DRCNN method allows to utilize the continuity between data by decomposing data into residual and trend terms which are processed by a designed convolution block or DR-Block. DR-Block has its strength in extracting features by following the structural design of Transformers. In addition, by imitating the multi-head in Transformers, a Multi-head Sequence method is proposed such that the network is enabled to receive longer inputs and more accurate forecasts are obtained. The state-of-the-art performance of the presented model are demonstrated on several datasets. [ABSTRACT FROM AUTHOR]

To improve the adaptability of quadruped robot in multiple scenarios, an adaptive locomotive system based on the double-layered central pattern generator (CPG) is proposed. The novel CPG network consists of double master units and subsets of slave units based on gyroscope signals including yaw and pitch angle. The response of master units provides the ability to control the 1st joins of quadruped robot, while slave units can generate the symmetry signals to control the 2nd and 3rd joints. The CPG network enables the seamless switching of locomotion gaits to stops and starts by using an ultrasonic sensor. Through adjusting the mutually dependent parameters, joints can generate the joint angles to achieve steering behavior. For adaptive movement on an irregular surface, stable ranges of the robot body yaw and pitch angles are proposed by using gyroscope signals. The experimental results verify that the quadruped robot with the proposed double-layered CPG network can perform stable trot pattern in a complex environment. [ABSTRACT FROM AUTHOR]

Arrester is an important lightning protection device in the electrical field. The parameters of arrester such as creepage distance, umbrella distance and diameter are important for product quality, but they are difficult to measure because the shape of arrester is irregular. However, the three-dimensional (3D) reconstruction technique is efficient in measuring arrester parameters. The uniform distributed structure of umbrella skirt on the arrester surface restrict the registration of point cloud. In this paper, a scale-invariant points features histogram (SIPFH) descriptor is proposed; the descriptor combines the characteristics of Scale-invariant Feature Transform (SIFT) and fast point feature histogram (FPFH). Moreover, the improved Levenberg-Marquardt (LM) algorithm is presented, the maximum distance of corresponding points in the iterative process is adjusted to realize the local optimization. The point cloud registration method consists of two parts: primary registration method based on SIPFH, and secondary registration method based on improved LM algorithm. Point clouds of different arresters are collected to establish datasets, some of which have interference. Experimental results indicate that the root mean square error of the method is less than 0.02 m; the average running time is 2.7 s, which is 41.4 % of the conventional method based on FPFH. [ABSTRACT FROM AUTHOR]

Statistical methods, such as Principal component thermography (PCT) and Sparse Principal component thermography (SPCT) have been widely used for signal enhancement of subsurface defects in pulsed thermographic (PT) detection of composite materials. However, PCT and SPCT mainly focus on the temporal variation of thermographic data while leaving the structural variation un-modeled. In this paper, a method of sparse structural principal component thermography (S2PCT) is proposed. In S2PCT, the operation of shift-sampling is first conducted to augment the original thermographic matrix and capture the structural relationships inside the original thermal images. After that, the sparse trick is applied to extract features for defects and reduce signals of noise and non-uniform background. In the case study, two carbon fiber reinforced polymer (CFRP) specimens are detected with PT and the proposed S2PCT is evaluated for visualization enhancing purpose. The results of the experiments have revealed the proposed method helps to highlight the defect signals during the augmentation process, thus showing higher flexibility in reducing interference from background signals. As a conclusion, compared to the original statistical methods, S2PCT has better performance in visualization enhancing of defects. [ABSTRACT FROM AUTHOR]