Kyushu University Academic Staff Educational and Research Activities Database
Researcher information (To researchers) Need Help? How to update
Kimura Hajime Last modified date:2018.06.11

Professor / Naval Architecture and Marine Systems Engineering Human Resource Development Course
Faculty of Engineering


Graduate School


Homepage
http://sysplan.nams.kyushu-u.ac.jp/gen/index.html
Hajime Kimura
Dept. of Marine Engineering, Graduate School of Engineering .
Academic Degree
Ph.D
Country of degree conferring institution (Overseas)
No
Field of Specialization
Machine Learning, Artificial Intelligence, Robotics, automatic pipe routing, Reinforcement learning, Statistics, ship building
Total Priod of education and research career in the foreign country
00years00months
Outline Activities
Research activities:

(1) Automatic pipe arrangement for ships and its evaluation
(2) Optimization in Logistics of Manufacturing making use of IT devices or Robots
(3) Performance estimation of ship hulls using statistics

Education activities:
(1) Systems design engineering
(2) Advanced systems design
(3) Marine statistics
(4) Programming exercise
Research
Research Interests
  • A development of automatic piping design system
    keyword : piping, pipe routing, automatic design, equipment arrangement, Dijkstra method, Touch and Cross method
    2010.04~2019.03.
  • Automatic painting robot in shipbuilding
    keyword : shipbuilding, painting, automatic paint system, paint robot
    2018.04~2020.03.
  • Measurement of shapes making use of 3D scannar in shipbuilding
    keyword : 3D-scannar, point clouds, shape recognition, shape measurement
    2014.04~2018.03.
  • Optimization in shipbulding
    keyword : optimization scheduling
    2004.04~2019.03Investigation of optimization techniques in logistics.
  • Assist system for decision of main hull parameters on ground design for shipbuilding
    keyword : Shipbuilding, ground design, main hull parameters
    2008.04~2010.12.
  • Behavior learning of underwater robots using Reinforcement learning
    keyword : Reinforcement learning, underwater robots
    2005.04~2010.03.
  • Mine search robot using electrical sensor
    keyword : robotics, mine search, electrical sensor
    2004.07~2006.03mine detection using an electrical method.
  • underwater sensor using electrical search
    keyword : electrical search, underwater sensor
    2004.07~2006.03.
  • Development of Reinforcement learning
    keyword : reinforcement learning, Robotics
    1995.03~2015.03Reinforcement Learning: Finding control rules through trial and error interactions in the unknown environments.
  • Decision making under uncertainty
    keyword : decision making, Markov decision process
    1995.03Decision making under uncertainty.
Current and Past Project
  • Development of decision support system for early stage of ship hull design
Academic Activities
Reports
1. (Relay Review: Recent Developments in Reinforcement Learning) A Basis of Reinforcement Learning.
Papers
1. Automatic Routing System for Multiple Pipe-lines.
2. An Automatic Piping Algorithm including Elbows and Bends, [URL].
3. Automatic Designing System for Piping and Instruments Arrangement including Branches of Pipes: Multi-objective Optimization of Piping Material Costs and Valve Operability

Automatic designing of piping layout is challenging since it is composed of several numerical and/or combinational optimization problems, e.g., routing problems of pipes including branches, and arrangement problems of equipments. This paper presents a new approach based on a simple idea that the branches of pipes are considered to be a variety of equipment. Accordingly, the pipe routing problems are fairly simplified by removing the branches, and it derives a lot of efficient algorithms to solve the pipe arrangement problems. One is a multi-objective genetic algorithm (MOGA) in which the gene represents both the locations of the equipments and the arrangement of the pipes. And a new simple and efficient crossover operation which appropriately merges two different piping layouts (but of course the PID is the same) is proposed. In order to provide a fairly good initial population for the MOGA, a new heuristics making use of self-organization techniques to arrange equipments is proposed. The efficiency of the proposed approach is demonstrated through two experiments, one is a designing problem including five valves, one pump, and five branches, and the other includes seven valves, one pump, and six branches. The objective of the optimization in the experiments is to minimize the length of the pipes, the number of elbows, and the valve operability cost. The algorithms are programmed using Java language. Although the automatic arrangement system used in the experiments is academic, the concept of the proposed approach will be accepted in practical systems. , [URL].
4. Reinforcement Learning in multi-dimensional state-action space using random tiling and Gibbs sampling.
Works, Software and Database
1. .
Presentations
1. Hajime Kimura, Automatic Piping Arrangement Design Considering Piping Supports and Curved Surfaces of Building Blocks, International Conference on Computer Applications in Shipbuilding, 2017.09, In piping design, consideration must be given to the position and direction in which pipes are passed, in order to properly support pipes from pipe racks and structural members with support. In this paper, a new piping path planning system is proposed in order to automate piping design corresponding to pipe supports and curved hulls. In the proposed system, candidates for positions and directions to which pipes should be passed are given in advance as 'candidate points' from the circumstances of pipe racks and support. Then, the system selects the appropriate candidate points automatically to generate piping paths keeping constraint of many factors, e.g., gravitational flow, or geometrical limitation of the pipe-bending machine, etc. Therefore, it is quite practical. The performance of the proposed system is demonstrated through several simulations.
.
2. Yuto Ando, Hajime Kimura, Automatic Pipe Routing to Avoid Air Pockets, International Conference on Computer Applications in Shipbuilding, 2013.09, [URL], Pipe arrangement is one of the most time-consuming works in ship production because the process requires designers to decide the optimum pipe routes. Previous works focused on finding preferable routes by applying optimization methods, but these methods have not considered the effect of gravity in obtained pipe routes. This paper presents an automatic pipe routing method that avoids air pockets. We call vertical U-shaped pipes “air pockets”. In this paper, the pipe routing problem is considered as a routing problem in a directed and weighted graph. Dijkstra’s method is used in the routing process for generating candidates of optimum routes. In order to avoid making air pockets in the obtained routes, we try to use a new cost function. The performance of this method is shown in several demonstrations..
3. Kimura Hajime, An Automatic Pipe Arrangement Algorithm Considering Elbows and Bends, International Conference on Computer Applications in Shipbuilding, 2012.06, [URL], Nowadays, piping arrangement has been enabled to be more efficient by development and spread of CAD (Computer-Aided Design). However, it is difficult to design piping layout automatically because there are many regulations and functional design rules which must be satisfied. We propose an automatic routing method for simple pipes considering elbows and bends. In practical design of piping layout, there are many bends connecting straight eccentric pipes which have gaps within the pipes' diameter. However, no precedence automatic piping algorithm has been taken into account pipelines with such bends. The proposed method finds piping routes making use of not only elbows but the bends minimizing costs of the path connecting start point to goal point, while avoiding obstacles such as structures, equipments and the other circuits. In our approach, we consider the piping route design problem to a routing problem in a directed and weighted graph. Note that the nodes in the proposed graph have state variables not only locations but directions of the pipes. Consequently, this graph can easily express the bends as simple edges, and then the routing algorithm can easily handle the bends. In addition, the presented method has specifications that the sizes of each cell, which is generated by decomposing of a free space, are not restricted within the diameter of the pipe. The routing algorithm uses Dijkstra's method to provide candidate paths. For practical use, the system adopts XML-file-based interface. This paper presents a new idea to express specific arrangement rules or policies using XML. The efficiency of the proposed method is demonstrated through several experiments..
4. Reinforcement Learning in Multi-Dimensional State-Action Space Using Random Rectangular Coarse Coding and Gibbs Sampling, [URL].
Membership in Academic Society
  • Japan Society of Naval Architects and Ocean Engineers
  • The Society of Instrument and Control Enginerrs
  • Robotics Society of Japan
Educational
Educational Activities
Systems Design Engineering
Marine Statistics
Advanced Systems Design
Computer Programming Exercise