The series of studies are carried out on the related problems of bucket tooth fault detection of excavator bucket, and a feasible technical scheme is put forward. On this basis, a fault detection system is designed to realize the fault detection and fault alarm of bucket tooth of excavator, and has a certain effect. However, due to the limited personal ability and the limitation of experimental conditions, the method has not been verified under the actual working conditions, and there are still some problems, such as high false positive rate, long time-consuming, large consumption of hardware resources and so on. Therefore, from the perspective of practical application, many aspects still need continuous research and improvement:
(1) Optimization of data sets: one important reason why deep learning methods can develop rapidly is that they are supported by massive high-quality data sets. There is no public bucket data set on the Internet. The data set used is partially self-made. Combined with the data collected on the Internet, the number is relatively small, and there is still a certain gap compared with the actual working conditions. My ability and energy are limited, and the workload of data set sorting and marking is also very huge. Therefore, in order to improve the recognition effect, more energy needs to be invested in the preparation of data sets. At the same time, it is also hoped that relevant departments and professionals will provide data sources so that better algorithms can be applied to the bucket tooth fault detection of excavator.
(2) Improved model: the basic network of target detection and image segmentation method based on deep learning is vgg-16. Although the model has high feature extraction ability, it has many parameters, general calculation speed and consumes hardware resources. Therefore, reducing model parameters, improving calculation speed and ensuring the recognition performance of the model are the focus of the next research.
(3) Improve the detection accuracy: the proposed method can not rely on a large number of bucket tooth data of the fault excavator to a certain extent, so as to determine the position of the bucket tooth of the fault excavator and judge the wear degree of the bucket tooth of the fault excavator. However, it can be seen from the system experiment that there are still a large number of false detections in the detection process. Therefore, improving the segmentation performance of image segmentation model and using better contour features to calibrate the order of excavator bucket teeth are the focus of future research.