In this decade, deep convolutional neural network called DCNN has been attracting attention due to its high ability of image recognition and other applications. In this paper, a design application of DCNN is considered and developed for vision-based defect inspection. As a trial test, three kinds of DCNNs are designed, implemented and tested to inspect small defects, such as, crack, burr, protrusion, chipping and spot phenomena seen in the manufacturing process of resin molded articles. An image generator is also implemented to systematically generate range of relevant deformed version of similar images for training. The designed DCNNs are trained using the generated images and then evaluated through classification experiments. The usefulness of the proposed DCNN design application is demonstrated and discussed.

In this paper, a curved surface fitting method using a raster-scanning window is proposed to smooth the original organized point cloud data (PCD) and extract the original features in the presence of noise. The method allows PCD to be smoothed while keeping its own original features. Then, a Stereolithography (STL) generator is proposed to produce triangulated patches from smoothed PCD, and this helps to deal with a triangular representation of a 3D surface geometry. The process allows reconstructing 3D digital data of a real object described using STL format from original PCD in the presence of noise for reverse engineering. The effectiveness and usefulness of the proposed techniques are demonstrated through actual experiments to smooth PCD, reconstruct STL, generate cutter location source (CLS) data and perform machining of the object model described in the original PCD.

A cast metal is generally produced by using a sand mold. A foamed polystyrene mold is used as the master mold for making the corresponding sand mold. Recently, the development of a flexible machining robot for foamed polystyrene materials is expected to be used in the cast metal industries because of the high cost and inefficiency of conventional large-sized CNC machine tools. To cope with the need, we already presented a machining system based on an industrial robot. Industrial robots have been recognized as a typical biomimetic machine pursuing the capability of human arm. The developed robotic CAM system provides a simple interface without using any robot languages between any CAD/CAM system and the machining robot. In this paper, a trajectory following controller with a capability of generating micro vibrational motion is completed for a ball endmill of the machining robot in order to enhance the machining performance of foamed polystyrene materials with curved surface. The micro vibrational motion is online generated along curved surface by post-processing original CL data. The implementation of the proposed controller and removal machining results along curved surface are illustrated.

Swarm robotics has been focused in order to express more advanced performance using lower performance devices. Swarming behaviors of multiple robots are need for working in industries or in places where a human cannot enter. In the natural world, some creatures which have simple biological function can acquire higher intelligence by swarming. Therefore, biomimetic method is effective in order to establish an efficient swarming system. In the swarming robotic system, because cost of the system increases in proportion to number of a robot, devices of robots are required to be simple. Furthermore, an implemented algorithm into robots should be designed considering a simplification of the individual robot with keeping correctness of the model of the robots. In this paper, a simple algorithm for swarming and following behaviors is proposed assuming that robots had only three sensors to detect distance information. The algorithm consists of nine patterns according to information of three PSD sensors. In the proposed algorithm, each robot's movement is determined based on the inputs of three distance sensors. Then, a virtual simulator for multiple mobile robotics is developed to evaluate the performance of the algorithm, in which the movement of a virtual robot is modeled based on an actual three-wheeled mobile robot. It is assumed in the simulator that the robot moves only to six directions and rotates counterclockwise or clockwise on the position; the robots do not communicate with each other; shape and kinematics of the robot are simplified in the two-dimensional plane. Several simulations using the proposed algorithm are conducted to evaluate the performance for the swarming and following behaviors. Results of the simulations show that the proposed algorithm can provide swarming and following behaviors.

Until now, we have proposed a multiple mobile robots system based on wireless communication using simple hardware and software. The controller of the system is developed based on a subsumption architecture to realize a swarm intelligence. The subsumption architecture is constructed in a supervisory PC server. The subsumption architecture generates eight kinds of simple commands. The server transmits a communication packet including the command to each mobile robot by wireless communication. Each mobile robot moves based on the received command while transmitting a response packet including six Position Sensitive Detector (PSD) sensors' information to the server. The sensors' information is used as the inputs for three agents in the subsumption architecture. In this paper, to realize a more biological behavior, a bend sensor is added in the robot through a compact A/D converter and the corresponding agent is newly designed in the controller of the server. It is expected that not only the added agent works as emergency escape function to defend the robot itself but also other analogue sensors can be easily equipped with. Also, to examine the actual delay of the system in using serial communication and wireless communication, dynamic sampling period is experimentally measured by using multi media timer with 2 ms resolution on Windows system. The dynamic sampling period is the communication time required between the PC server and the robot. In addition, the success rate of communication, which is defined as the ratio of reception times to transmission times, is evaluated. The measured results are shown and discussed.

Position and velocity feedback gains should be tuned suitably to realize a stable position controller. We already proposed genetic algorithms with variable search spaces, which successfully address this problem. The effectiveness was proved through fine-gain tuning of a resolved acceleration controller, which is one of the model-based robotic servo controllers. Additionally, the viscous and Coulomb frictions of the joints were considered in the dynamic model of the manipulator, to realize a more realistic robotic simulation. As a result, undesirable oscillations caused by the friction were observed because of the nonlinearity of the friction model. In this paper, we add a low-pass filter to the velocity term in the resolved acceleration controller to reduce the influence of the friction. The improved controller is evaluated through fine-gain tuning for a trajectory-following control problem. Simulations are conducted by using the dynamic model of a PUMA560 manipulator. The results demonstrate the effectiveness of the improved controller.

In this paper, a new desktop orthogonal-type robot, which has abilities of compliant motion and stick-slip motion, is first presented for lapping small metallic molds with curved surface. The robot consists of three single-axis devices with a high position resolution of 1 μm. A thin wood stick tool is attached to the tip of the z-axis. The tool tip has a small ball-end shape. The control system is composed of a force feedback loop, position feedback loop and position feedforward loop. The force feedback loop controls the polishing force consisting of tool contact force and kinetic friction forces. The position feedback loop controls the position in pick feed direction, e.g., z-direction. The position feedforward loop leads the tool tip along a desired trajectory called cutter location data (CL data). The CL data are generated from the main-processor of a CAM system. The proposed robot realizes a compliant motion required for the surface following control along a spiral path. In order to improve the lapping performance, a small stick-slip motion control strategy is further added to the control system. The small stick-slip motion is orthogonally generated to the direction of the tool moving direction. Generally, the stick-slip motion is an undesirable phenomenon and should be eliminated in precision machineries. However, the proposed robot employs a small stick-slip motion to improve the lapping quality. The effectiveness of the robot is examined through an actual lapping test of an LED lens mold with a diameter of 4 mm.

In this paper, a new desktop orthogonal-type robot with a CAD/CAM-based position/force controller is presented for finishing small metallic molds with curved surface. The robot consists of a three-axis robot whose single one has a high position accuracy of 1 mum, which means that it can perform higher resolutions of position and force, compared to general industrial robots. A thin wood stick tool with a ball-end tip is attached to the z-axis through a force sensor. The control system of the robot is comprised of a force feedback loop, position feedback loop and position feedforward loop. The force feedback loop controls the polishing force consisting of tool contact force and kinetic friction forces. The position feedback loop controls the position in pick feed direction. Further, the position feedforward loop leads the tool tip along a spiral path. We first evaluate the backlash that causes inaccuracy in position at the tip of the abrasive tool, by simply measuring the position and force. Next, a surface following control experiment along a lens mold is conducted, in which the mold has axis-symmetric concave areas. Finally, a LED lens mold is further finished by using the proposed system in order to demonstrate the performance and promise.

Cutter location (CL) data with normal vectors can be used for not only a desired trajectory of tool's translational motions but also a desired force direction for molds. In this paper, such normalized tool vectors from 3-axis CL data are generated for a polishing robot. The resultant CL data allow the polishing robot based on an industrial robot to realize a teaching-less operation of position and force. The present robot can also control the polishing force consisting of the contact and kinetic friction forces. During the polishing of a mold, a position feedback loop has a delicate contribution to the force feedback loop in Cartesian space so as not to make the tool deviate from the desired trajectory. The effectiveness and validity of the proposed robot have been proved by actual polishing experiments using an aluminum mold with curved surface.

In this paper, a finishing system with a mounted abrasive tool is proposed for polishing of metallic molds. The shape of the mounted abrasive tool is a ball-end type. When a metallic mold with curved surface is polished, not only the orientation of the mounted abrasive tool is fixed but also its revolution is locked. The motion of the mounted abrasive tool is feedforward controlled based on an initial trajectory calculated in advance. The surface is polished by the polishing force which consists of a contact force, motion and viscous friction forces acting between the mold and tool. In this case, the velocities in the normal and tangent directions are delicately regulated so that the polishing force can track the desired value. The effectiveness of the proposed system is proved by experiments using an industrial robot with a PC based controller.

A furniture polishing robot which uses a trajectory generator based on cutter location data, is proposed. The position and orientation of the polishing tool is suitably controlled by referring to the output from the trajectory generator. The trajectory generator yields a zigzag path, whirl path or their combinational path according to the shape of each workpiece. Using the trajectory generator, the present polishing robot doesn't need the conventional complicated teaching process. The contact force acting between the polishing tool and the workpiece is controlled through a desired impedance model to follow in Cartesian space. The effectiveness and promise of the proposed polishing robot are demonstrated through some experiments concerning polishing tasks using an industrial robot JS-10 with a PC based controller.

Impedance control allows the manipulator to change the mechanical impedance such as inertia, damping and stiffness, acting between the end-effector and its environment. However, to achieve stable force control under unknown stiff environments, complicated tuning of desired impedance parameters is needed. Among the parameters, the desired damping is the most significant to suppress overshoots and oscillations. In the paper generalized fuzzy environment models with anisotropy are proposed to systematically determine the desired damping against unknown environments. The models learned with genetic algorithms, can estimate each directional stiffness of the environment and yield the desired damping, considering the critical damping condition of the control system. Position and force control simulations are shown to demonstrate the effectiveness and promise of the models.

A position based impedance control with a fuzzy compensator is proposed to improve a force control performance, in which the environment has unknown dynamics or curved surface. The main attraction of the fuzzy compensator is that it generates proper position compensation as feedforward command so that overshoots and oscillations in the direction of force control can be suppressed satisfactorily. Profiling control simulations using the dynamics of the PUMA560 manipulator are shown to demonstrate the effectiveness and promise of the fuzzy compensator under the environment having unknown dynamic characteristics.