中村 優介

Labor shortages and long working hours have long been problems at construction sites. In order to solve this problem and improve the productivity of the sites, automation is progressing on various occasions according to the process, and AGV is used especially when transporting materials. There are various driving methods for AGV, but in this research, we let AGV drive automatically after learning using data obtained by detecting obstacles with a drone so that it can travel safety with any loads.

In this paper, image features are used to realize the repositioning of logs in between processing. The position-posture estimation by image features has some problems such as missing materials and textures. To solve the problem of position-posture estimation, we employed Affine-SIFT and segmentation matching. By adopting these two methods, it is possible to estimate the position and orientation regardless of some missing materials and textures. In order to verify the effectiveness of our method, a wall and a corner of a traditional wooden building were created. By creating them, the effectiveness of our method is confirmed.

Under a static machining path of numerical controlled machining, when the machining path is completed successfully, the machining of the member is as accurate as the trajectory of the tool. However, it is difficult to confirm that the machining was performed accurately for a machining path that changes dynamically through feedback control. In addition, as the machined surface may have a texture peculiar to the blade, the machined surface cannot simply be compared with the designed shape. In this study, we will examine an achievement evaluation method for evaluating the accuracy of dynamic machining.

The conventional workflow for producing wooden parts requires the wood workpiece to be firmly secured in the position predetermined by a processing path generation step. As a result, it is difficult to process curved raw wood pieces using machines. To overcome this difficulty, we developed a single-shot posture estimation technique using pre-reconstructed 3D points and features, as well as a processing path transformation technique that adjusts the path according to the actual posture of the fixed workpiece. A curved beam named “taiko bari” was processed and subsequently tested. The results demonstrated that the proposed techniques are suitable for practical applications.

Following a static path, pre-calculated as the trajectory of the tool, cannot reflect potential changing states during the machining process. Therefore, adjusting to differences occurring in the environment between the assumed and actual conditions, such as variations in the physical properties of the material, is not possible. In this research, we dynamically change the tool path by monitoring the load applied to the tool during processing. Furthermore, applying our results to processing with a chisel demonstrates that parts can be machined appropriately by dynamic paths.

At precut fabrication sites for high-mix low-volume production, multi-axis machines are operated using a dedicated CAD/CAM that derives an operating path concurrently with detailed modeling. Because a CAD/CAM based workflow has complicated frontloading and data management issues, the authors developed a generic CAM that creates an operation path from a 3D geometry of a part. This paper reports a process path deriving method for a circular saw using 3D shape recognition based on PointNet. The proposed method classifies shape types and segments cut patterns for joints of parts represented as a point cloud. A validation experiment showed the joints are classified and segmented properly.

There is a problem that thinned woods are overly small diameter and un-uniform shape to saw up into lumbers as building elements. This research attempts to solve this problem by developing thinned woods processing system that consists of a thinned wood management database and robotic fabrication. In the database, thinned woods are managed as features based on photogrammetry. Appropriate materials are suggested when intended shapes are input into an inquiry function. In the robotic fabrication, a lathe-like work bench synchronously rotates the material with the robot’s movement. Validation by building log walls shows that variational shapes of thinned woods can be processed efficiently.

Five-axis machines are capable of complicated movement; however, their operation requires both experience and knowledge. For applying these machines to diversified small-quantity production, a sophisticated operating system that assists the operation in deriving the toolpath and avoiding errors is indispensable. From this perspective, the authors attempted to develop an automatic toolpath deriving method that detects processable polygons with a circular saw by analyzing a target geometry. A movement simulation method was also developed to ensure safety in operation. This paper reports the implementation of the system, and its validation through building a catenary dome that has miter joints.

5-axis machines are capable of complicated movement. However, their processing shapes are highly limited by the production systems optimized for particular building systems. This study aims to realize a 5-axis machine that can machine various shapes by the maximal activation through the development of the self-built machine and its operating software. Shapes that are processed with a circular saw are classified into three types. New log designs are machined by analyzing existing logs according to these types. In addition, algorithmically twisted logs are machined accurately. The results show that these approaches have the possibility of creating new building systems.