A reinforcement learning with switching controllers for a continuous action space

Reinforcement learning (RL) attracts much attention as a technique for realizing computational intelligence such as adaptive and autonomous decentralized systems. In general, however, it is not easy to put RL to practical use. This difficulty includes the problem of designing a suitable action space for an agent, i.e., satisfying two requirements in trade-off: (i) to keep the characteristics (or structure) of an original search space as much as possible in order to seek strategies that lie close to the optimal, and (ii) to reduce the search space as much as possible in order to expedite the learning process. In order to design a suitable action space adaptively, in this article, we propose a RL model with switching controllers based on Q-learning and an actor-critic to mimic the process of an infant’s motor development in which gross motor skills develop before fine motor skills. Then a method for switching controllers is constructed by introducing and referring to the “entropy.” Further, through computational experiments by using a path-planning problem with continuous action space, the validity and potential of the proposed method have been confirmed.     

A High-Speed Optical Data Highway for Industrial Control Systems

Recent application of distributed control systems to large-and medium-scale industrial plants requires effective and high-speed communication among the control devices each other to ensure high-performance operation.     

On phenomenological failure criteria for composite bolted joint analysis

There has been a recent trend to use phenomenological strength models in combination with point stress or average stress failure criteria to form strength models for the analysis of composite bolted joints. Several papers published using these approaches have demonstrated practical reductions in the size of the empirical data base required to perform bolted joint analysis for specific materials and laminates. Since the inception of phenomenological failure models for bolted joints several refinements have been proposed, but the generality of the method is not clearly defined. This paper addresses the concepts underlying the formulation of these strength models, compares the capabilities of several commonly used failure criteria and discusses the limits of applicability through example analyses and comparison with experimental data. The results indicate that differences in predicted strengths for the models examined were insignificant. Failure mode predictions based on single point failure location information varied from model to model and differed significantly from experimentally observed failure modes for some geometries. The models cannot be applied with confidence for the analysis of a wide class of laminates and geometries without calibrating the model through empirical data.     

The development of simplified ultrasonic CT system and its application to the evaluation of weld metal

In this paper, as a new measurement method to estimate the change of material condition, the simplified ultrasonic CT system, which uses the information of three directions, that is, 90°, + 45° and −45° about inspection plane is proposed. Use of simplified CT system has two merits: Firstly, the measurement time is very short compared with general CT. Secondly, it can detect sensitively small defect in vertical or slant direction about inspection plane because the obtained image is CT image calculated from three directions. From these merits, this method can be considered as an effective method to evaluate material conditions. The basic performance of the proposed method was confirmed through several specimens with several simple defects. In order to confirm the applicability of actual NDT, several kinds of welded specimens are investigated. The result showed that the CT image obtained had good agreement with actual defect of specimens.     

Selective sorption of heavy metal thiocyanate complexes on crown ether resin

The sorption behavior of heavy metal thiocyanate complexes was investigated for dibenzo-18-crown-6 (DB18C6) resin and bis[2-(o-methoxyphenoxy)ethyl]ether (BMPE) resin. The DB18C6 resin showed a high sorption ability and the degree of zinc sorption increased significantly with thiocyanate concentration. This behavior was not observed for BMPE resin. The sorption behavior was influenced by the countercation species, and the degree of sorption of zinc ions showed the maximum when the potassium thiocyanate was used as a complexing salt. The sorbed species appear to be KZn(SCN)₃, K₂Zn(SCN)₄ in the potassium thiocyanate system, and Zn(SCN)₂ in the lithium thiocyanate system, respectively, according to analysis of the sorption equilibrium. Sodium and ammonium thiocyanate systems show an intermediate behavior of the two. The sorption selectivity for DB18C6 resin depended not only on the hydrophobic nature of heavy metal thiocyanate complexes but also on the stabilization of counter cation species with crown ether matrix, and the sorption selectively was found to be effectively controlled by countercation species according to the cation-chelation mechanism.     

Multi-fluid multi-phase subchannel analysis of subassembly accidents of LMFRs

Analyses of experiments simulating hypothetical subassembly accidents such as a large-scale inlet blockage in a Liquid Metal Cooled Fast Reactor (LMFR) have been performed with computer program KAMUI. With the use of relatively simple but reasonable constitutive models, the code has been applied to the SCARABEE experiments BE+1 and APL1 to validate the analytical capability against the accident conditions under the multi-pin geometry. The results show that the key events such as sodium boiling, clad melting, molten clad relocation, molten clad freezing were adequately simulated taking into account the effect of heat loss to the coolant flow in the outside channel of the test section.