Search

Deep learning provides a promising way to extract effective representations from raw data in an end-to-end fashion and has proven its effectiveness in various domains such as computer vision, natural language processing, etc. However, in domains such as content/product recommendation and risk management, where sequence of event data is the most used raw data form and experts derived features are more commonly used, deep learning models struggle to dominate the game. In this paper, we propose a symbolic testing framework that helps to answer the question of what kinds of expert-derived features could be learned by a neural network. Inspired by this testing framework, we introduce an efficient architecture named SHORING, which contains two components: \textit{event network} and \textit{sequence network}. The \textit{event} network learns arbitrarily yet efficiently high-order \textit{event-level} embeddings via a provable reparameterization trick, the \textit{sequence} network aggregates from sequence of \textit{event-level} embeddings. We argue that SHORING is capable of learning certain standard symbolic expressions which the standard multi-head self-attention network fails to learn, and conduct comprehensive experiments and ablation studies on four synthetic datasets and three real-world datasets. The results show that SHORING empirically outperforms the state-of-the-art methods., Comment: 18 pages, 4 figures

NILM is an important technology to realize power consumption behavior perception. In view of the problem that the existing load characteristic model can not match the multi-state load and the huge amount of calculation of identification algorithm can not be applied to smart meter, this paper proposes a method for building a load multi-state model library and a weighted matching algorithm based on Manhattan distance. The effective identification characteristics of each state of electrical equipment are used as characteristic unit to building the multi-state identification model library. Through the multi-time scale window scanning the total power sequence, the load characteristics in the process of load switching and state change are extracted. Using the weighted algorithm based on the improved Manhattan distance, load identification is carried out by calculating the matching distance with the library. The test results show that the proposed method can identify the load category quickly and accurately.

For traditional constant impedance control, the robot suffers from constant stiffness, poor flexibility, large wear and high energy consumption in the process of movement. To address these problems, a variable impedance control method based on reinforcement learning (RL) algorithm Deep Q Network (DQN) is proposed in this paper. Our method can optimize the reference trajectory and gain schedule simultaneously according to the completion of task and the complexity of surroundings. Simulation experiments show that, compared with the constant impedance control, the proposed algorithm can adjust impedance in real time while manipulator is executing the task, which implies a better compliance, less wear and less control energy.

Collision avoidance implies that extra motion in joint space must be taken, which might exert unexpected influences on the execution of the desired end-effector tasks. In this paper, a novel framework for generating collision-free trajectories while respecting task priorities is proposed. Firstly, a data-driven approach is applied to learn an efficient representation of the distance decision function of the system. The function is then working as the collision avoidance constraints in the inverse kinematics (IK) solver, which avoids the collision between manipulators. To eliminate undesired influences of the extra motion for collision avoidance on the execution of tasks, task constraints are proposed to control the task priorities, offering the system with the ability to trade off between collision avoidance and task execution. Furthermore, the overall framework is formulated as a QP (quadratic programming), therein guarantees a real time performance. Numerical simulations are conducted to demonstrate the effectiveness and efficiency of the presented method.

For the purpose of delving deep into the preparation process and early working performance of inorganic cementitious composite material, and analyzing its application prospect in pavement emergency repair, the slag, fly ash and silica fume are used as solid cementitious materials, and sodium hydroxide and sodium silicate are used as alkali-activators to prepare inorganic polymer-based fast-hardening concrete (IPFC). The early working performance of the specimens with different contents of slag, fly ash and silica fume were compared and analyzed. The results show that, all the kinds of concrete mixtures have good fluidity, cohesiveness and water retentiveness. The addition of fly ash and silica fume can prolong the initial setting time of concrete, and the higher the content, the longer the extension time. And the slump of the mixture is inhibited by the addition of fly ash. When the content of fly ash is 15%, the slump of the mixture decreases by 19.2%. However, silica fume can improve the slump of the mixture. When the content of silica fume is 15%, the slump of the mixture increases by 19.1%.

In robotic assembly processes, detecting target pose is indispensable for robot manipulation algorithms. Such a tedious work hinders the further application of robot manipulation algorithms to real industrial processes. In this paper, a vision-based position/impedance control framework is proposed. Therein, the objective pose is firstly obtained by a motion capture system, where a kinematic equation is solved to Figure out the target assembly pose. Next the robot arm motion is approximated by an impedance model, in this way, the position/impedance control law of the robot is derived, which drives the robot to perform a compliant assembly manipulation. Finally, a memory stick assembly experiment is conducted on a 6-DOF robot arm equipped with a motion capture system and a force-torque sensor. Experiment shows the efficiency of the vision-based position/impedance hybrid controller in terms of autonomous positioning and compliant assembly. Due to its convenience and generality, the present controller can be expected to apply to more general industrial scenarios.

In the multi-robot collaborative manufacturing, such as the server motherboard assembly task in the 3C industry, the characteristic of the operating process and the uncertainty of the assembly environment make this problem difficult to be solved in an efficient way. This paper addresses the key points of this problem. Firstly, a series of end effectors for 3C assembly operation is designed. After analyzing precision and scope for different steps in the assembly process, the positioning systems with different precision and scope are designed and are chosen for better performance in different operations. Finally, compliant control is introduced in the inserting process where the positioning system does not perform well to further increase the success rate. Experiments illustrate the superiority of our ideas. Due to its convenience and generality, the present approach can be expected to apply to more general industrial scenarios.

Underwater vehicles are being emphasized as highly integrated and intelligent devices for a significant number of oceanic operations. However, their precise operation is usually hindered by disturbances from a tether or manipulator because their propellers are unable to realize a stable suspension. A dynamic multi-body model-based adaptive controller was designed to allow the controller of the vehicle to observe and compensate for disturbances from a tether or manipulator. Disturbances, including those from a tether or manipulator, are deduced for the observation of the controller. An analysis of a tether disturbance covers the conditions of the surface, the underwater area, and the vehicle end point. Interactions between the vehicle and manipulator are mainly composed of coupling forces and restoring moments. To verify the robustness of the controller, path-following experiments on a streamlined autonomous underwater vehicle experiencing various disturbances were conducted in Song Hua Lake in China. Furthermore, path-following experiments for a tethered open frame remote operated vehicle were verified for accurate cruising with a controller and an observer, and vehicle and manipulator coordinate motion control during the simulation and experiments verified the effectiveness of the controller and observer for underwater operation. This study provides instructions for the control of an underwater vehicle experiencing disturbances from a tether or manipulator.

Solid waste (slag and fly ash) based geopolymer (SWG-SF) was developed on the basis of response surface methodology (RSM), using the fly ash to slag ratio (FSR), alkali content (AC), makeup of alkali activator (n) and binder to water ratio (BWR) as design parameters. Flexural strength (ff-3d) and compressive strength (fc-3d) after 3 days of standard curing were tested, the early strength systems were established, the prediction of strength system was validated by comparison, and the effects of design parameters on ff-3d were studied. The test result shows that the reliability and accuracy of the early strength system of SWG-SF by adopting RSM can be verified in terms of mathematics and application, and the early strength system can be used for theoretical study and practice guidelines; according to analysis on the strength system, ff-3d increases linearly with BWR, and the increasing rate is closely related to FSR; ff-3d increases linearly with n, and the increasing rate is closely related to AC; ff-3d decreases continuously with FSR, but the decreasing rate gets slower and slower, and BWR has certain influence on the change rate of strength; and ff-3d increases at the beginning and then decreases with AC, which suggests that AC has an optimum value (0.146–0.055n). Thus it can be seen that RSM can be widely used in the preparation of new solid waste based geopolymer and has promising applications in the utilization of solid waste.

Underwater autonomous manipulation is a challenging task which not only includes a complicated multibody dynamic and hydro-dynamic process, but also involves the limited observation environment. This paper is proposed to investigate the current state of art about underwater autonomous technology. Although a lot of research aiming at autonomous under-water vehicle manipulator system (UVMS) of large-scale AUV have been carried out by scientists in Italy, Spain etc. However, there has been a long lack of studies on portable AUV. So this paper presented a newly designed streamlined portable AUV and the numerical simulation of horizontal plane motions, which can produce positive outcomes in the further control and accurate manipulation.

In order to study the effect of VAE latex powder on the tensile and shear properties of styrene-acrylic emulsion-based cement compound joint compound (PCJS), the tensile test and shear test of PCJS under different VAE latex powder content (0~5%) were carried out. The test results show that the incorporation of VAE latex powder improves the tensile and shear mechanical properties of PCJS. When the amount of latex powder is 5%, compared with PCJS without latex powder, the shear modulus of PCJS increased by 32% and 35%; with the increase of VAE latex powder, the tensile and shear deformation properties of PCJS increased first and then decreased. When the amount of latex powder was 3%, the shear deformation performance of PCJS is the best; the incorporation of VAE latex powder has no obvious effect on the tensile elongation improvement of PCJS at break. In addition, the mechanism analysis of the effect of VAE latex powder on tensile and shear properties of PCJS was carried out.

Effective thrust allocation for underwater vehicle is very important to realize complex control task. In order to generate optimal thrust allocation according to control command, a thrust optimal allocation scheme has been designed for broad type of underwater vehicles. Corresponding to horizontal and vertical thruster configuration, a force allocation model has been established to realize optimal allocation. Infinity-norm optimization has been combined with 2-norm optimization to construct a bi-criteria primal-dual neural network optimal allocation scheme. In the experiment of open frame remote operated vehicle, bi-criteria primal-dual optimization outperformed 2-norm optimization in the fault tolerance control and accurate path following. In the thrust optimal allocation simulations for underwater vehicle manipulator system, 4 vertical thrust have been optimal allocated to realize diving and attitude maintenance successfully during manipulation process. Thus the thrust optimal allocation has been verified.

NaOH and sodium silicate-activated slag and fly ash based geopolymer concrete (NSSFGC) and NaOH and Na2CO3-activated slag and fly ash based geopolymer concrete (NNSFGC) with strength grades of C30 were prepared, the dynamic compressive deformation behavior under impact loading was studied contrastly. The results indicate that, the compressive deformation of NSSFGC is greater than that of NNSFGC in the stage of compaction, which indicates that the volume of microdefect of NSSFGC is greater than that of NNSFGC; the elasticity modulus of NNSFGC is larger than that of NSSFGC; NSSFGC has excellent deformation performance. Thus it can be seen that, alkali-activator types have a big impact on the dynamic compressive deformation behavior of GC, and compared to the alkali-activator prepared with NaOH and Na2CO3, the alkali-activator prepared with NaOH and sodium silicate can be beneficial to give full play to deformation property; so it is better to prepare GC with NaOH and sodium silicate, when used in the Earthquake Engineering.

Cementitious materials, ceramic aggregates, and fiber are used as basic materials; ceramics cement–based porous material (CCPM) and basalt fiber-reinforced ceramics cement–based porous material (BFRCCPM) are prepared on the basis of Closest Packing Theory. After the tests, the internal relation between the average change rate of incident energy and the physical essence of the ensemble damage modes is discussed, and the change law of energy is probed. BFRCCPM’s compression energy properties during the dynamic deformation damage process are analyzed, and based on that, the effect of basalt fiber on the dynamic damage characteristics are studied. Results show that the damage level and energy absorption of CCPM and BFRCCPM get more serious or increase with the average change rate of incident energy; the damage level of BFRCCPM is obviously lower than that of CCPM. Moreover, BFRCCPM possesses better energy absorbing properties than CCPM. This advantage is more obvious when the average change rate of in...

This article presents a comprehensive review on a non-invasive, high resolution and high contrast imaging modality, called thermoacoustic tomography (TAT). Details about the principles, history and future directions of TAT are described.

A series of ruthenium substituted polyoxotungstates were used to catalyze the cleavage oxidation of styrene. Under all experimental conditions investigated, reaction products were benzaldehyde and benzoic acid. Kinetic studies showed that these reactions were consecutive reactions consisting of first oxidizing styrene into benzaldehyde and then oxidizing the benzaldehyde into benzoic acid. In this work, the first oxidation step was carefully studied. It has been shown to be first order with respect to substrate and catalysts but zero order with respect to NaIO4 under the conditions studied. The activation energies of the first oxidation catalyzed by different polyanions are closely similar to one another, but the value of pre-exponential factors, differed by orders of magnitudes. Such differences could possibly be accounted for by the phase transfer effect as experiments showed that lowering the concentration of phase transfer reagent for low distribution ratio (Caq/Corg) catalysts would enhance the pre-exponential factor value and selectivity of the reaction.