Industrial System Research Group


 To promote DX in the shipbuilding production process, we have been advancing the "Digital Shipyard" concept. The Digital Shipyard concept aims to create a shipyard where everything is expressed numerically, everything is planned numerically, and everything is completed according to plan, thus eliminating or digitalizing ambiguity that arises in various aspects of shipbuilding process.  Our group is engaged in research and development related to the Digital Shipyard, and as part of this effort, we are developing "Shipbuilding Simulator" which can precisely duplicate workers' flexible movements in shipbuilding work, including supplemental jobs. We are also committed to developing data linkage and standardization across shipbuilding design and production. In addition, AI utilization in shipbuilding design and planning is the target. Through these R&D efforts, we aim to establish a framework that enables construction of advanced ships with outstanding production efficiency.

Overview of our research

1. Study on construction of "Digital Shipyard"

 Research and development of a "digital shipyard" aiming at "a shipyard where everything is expressed numerically, everything is numerically planned, and everything is completed as planned" toward the realization of a short-term delivery of ships in a shipyard going. In particular, we are paying attention to information on the design and construction of shipbuilding, and are engaged in research to create information related to design and construction, and research to effectively transmit and display the created information to designers and workers.

  1. R & D of shipbuilding simulation technology that reproduces detailed movements of workers
  2. R & D of work support system for various shipbuilding works (bending, painting, etc.)
  3. R & D of AR technology and VR technology to shipbuilding
  4. R & D on shipbuilding robots (assistant work support robots, block surface plate management by drone, CFRP design CAD/CAM for shipbuilding, etc.)

 We are conducting research and development on shipbuilding simulation technology to reproduce the detailed movements of workers. Fig. 1 shows a shipbuilding simulation that reproduces the detailed movements of the workers at a small assembly process. The practical application of such shipbuilding simulation technology enables the shipbuilding process to be precisely reproduced and helps planning worker allocation, calculating costs, and planning facility maintenance.

Fig.1 Example of shipbuilding simulation

 Shipbuilding involves various on-site operations such as bending steel plates, assembly, and painting. Workers on-site work while looking at the drawings, and we are engaged in research and development to support site operations using AR (Augmented Reality) technology as shown in Fig. 2. The aim is to shorten the construction period by making the work more efficient.

Fig. 2 AR application to support bending process

 Ships are constructed by assembling parts of the hull structure, called blocks, on a surface plate. We have been developing the block surface management system shown in Fig. 3 to check whether the blocks under construction are manufactured as planned.

Fig.3 Block surface plate management system

2. Research on design and planning using AI (noise, nesting)

 In today’s world of the third AI boom, research and development using deep learning and deep reinforcement learning, which have evolved neural network technology, is progressing. In the current shipbuilding industry, it is required to connect these AI technologies to social implementation. Therefore, research is being conducted to utilize AI technology in the early stages of shipyard design (the stage of basic design and basic planning).

  1. Onboard noise prediction using neural networks
  2. Nesting by reinforcement learning (steel placement planning)

 In the design stage, predicting the noise level inside a ship has been determined empirically and numerically. Simple and flexible empirical methods have been required to select and place components through trial and error in the design process. Neural network predictions are close to empirical judgments, so we have applied them to shipboard noise predictions. As shown in Figure 4, a web application is created and all the sample data that will improve prediction accuracy is kept and controlled.

Fig. 4 Neural network based noise prediction web application and its operation image

 Nesting is the process of arranging parts so that the amount of waste materials is reduced as much as possible, as shown in Fig.5. We have tried to increase the yield rate at the design stage, and we have applied deep reinforcement learning, which is one of the learning methods of AI, to this task.

Fig.5 Nesting

 Fig.6 shows an example of AI nesting results. Satisfactory results were obtained in self-learning without using results placed by humans, while by performing imitation learning using results placed by humans, we have confirmed the possibility that imitation learning can be used at a level that does not require manual correction.

Fig. 6 Nesting results by reinforcement learning

3. Research on Advancement of Inspection Technology

 Visual inspection is the main form to detect structural defects and damage. Recently, drones have been used as one means to get easy access to inspection points. Our group has been working on the following research and development to apply them to the inspection of ship cargo tanks and offshore wind turbine blades.

  1. Image recognition of damage inside cargo tanks using deep learning AI
  2. An efficient method of inspecting offshore wind turbine blades

 Assuming that drones are used to inspect the ship's internal structure, we studied utilizing image recognition using deep learning (Faster R-CNN) to find cracks. Fig. 7 shows the result of damage recognition based on the results of deep learning on the damage images taken by an inspector with a digital camera. It is not useful enough to employ the current deep learning method alone for the field. It is necessary to advance this research by utilizing knowledge related to important inspection points.

Fig. 7 An example of damage image recognition results by deep learning

 We are conducting research on efficient inspection technology for wind turbine blades in order to prevent lowering of the operating rate of offshore wind power generation, which is expected to spread in the future. One of the inspection technologies is image inspection technology by drone. Offshore wind turbines are located offshore, making them more difficult to access than onshore wind turbines. Also, because the wind turbine itself is larger than on land, the blades of the wind turbine are also in a higher position, making visual inspection more difficult. It is hoped that image observation by drone will solve these problems. As an example, at the time of inspection, the wind turbine is stopped rotating and the drone shoots with the blade stationary, but the drone flight path for inspecting the front and back and front and rear edges of the blade in a short time (Fig. 8) was examined.

Fig.8 Inspection method for offshore wind turbine blades