In order to operate autonomous combine harvesters in rice fields, the harvester needs to detect obstacles on its path. One important such obstacle is a human. This research develops a 3D-LIDAR based method to detect and distinguish humans from other environmental objects, such as rice plants and the ground. Distance and intensity values from the 3D-LIDAR sensor were used for this detection. Once this data was captured, a filter was applied to mask rice plants and the ground, and then the remaining points plotted on a grid map. Evaluation of the resulting obstacle detection and collision avoidance system were undertaken during rice harvesting experiments. It was able to detect and respond to various obstacle (a human standing and sitting outside of the rice field, a human standing inside of the rice field, and a PVC pole) at the rate of 2.0 fps during harvesting.

Because most hazardous accidents involve lateral rollover, it is important to investigate tractor roll motion to predict lateral tractor rollover and improve tractor safety. This study reconstructed the ground in front of a tractor using a stereo-vision camera installed on top of the tractor weight. Then, the obtained road surface height was input into a tractor behavior simulation to compare the actual tractor movement with the simulated movement. The ground was reconstructed according to different camera angles. The root mean square error (RMSE) of the roll angular velocity was determined in terms of ground obstacles at each camera angle. The tractor simulation exhibited roll motion that was very similar to that of the actual tractor on the same ground.

A method is described for numerically analyzing the vibration resistance of diesel engine cooling fans. The coupled vibration of the blade and rotating-shaft systems in the whirl vibration of the cooling fan is formulated using a four-degree-of-freedom model. A detailed simulation model constructed using a multi-body dynamics model is used to express the dynamic behavior of the cooling fan. Comparison of the simulation results with the experimental results demonstrated the effectiveness of the simulation model.

To save seeds over the same planting area as that of conventionally sown broadcast seedlings, arrayed mat seedlings were prepared by sowing seeds in 26-furrow arrays on coco-peat mats, transplanted with the use of a conventional 8-row transplanter and grown along the broadcast (control) seedlings. The arrayed seedlings were prepared with 40 % less seeds than were the control and their foliar ages were 3 to 4 at transplanting. Neither seedling was suitable for a 7 mm-vertical feed in terms of missing hills at transplanting. The arrayed seedlings at transplanting missed 11.3 % and 6.3 % of hills in the 10 and 13 mm-vertical feed, respectively, but they were still larger than those in the control (9.0 % and 3.3 %). With the arrayed seedlings, the hill mortality during the growth period was low (＜5 %) and the grain yield was 500 g/m², which were not different from the control.