A visual camera system combined with the unmanned aerial vehicle (UAV) onboard edge computer should deploy an efficient object detection ability, increase the frame per second rate of the object of interest, and the wide searching ability of the gimbal camera for finding the emergent landing platform and for future reconnaissance area missions. This paper proposes an approach to enhance the visual capabilities of this system by using the You Only Look Once (YOLO)-based object detection (OD) with Tensor RT^TM acceleration technique, an automated visual tracking gimbal camera control system, and multithread programing for image transmission to the ground station. With lightweight edge computing (EC), the mean average precision (mAP) was satisfied and we achieved a higher frame per second (FPS) rate via YOLO accelerated with TensorRT for an onboard UAV. The OD compares four YOLO models to recognize objects of interest for landing spots at the home university first. Then, the trained dataset with YOLOv4-tiny was successfully applied to another field with a distance of more than 100 km. The system’s capability to accurately recognize a different landing point in new and unknown environments is demonstrated successfully. The proposed approach substantially reduces the data transmission and processing time to ground stations with automated visual tracking gimbal control, and results in rapid OD and the feasibility of using NVIDIA Jetson^TM Xavier NX by deploying YOLOs with more than 35 FPS for the UAV. The enhanced visual landing and future reconnaissance mission capabilities of real-time UAVs were demonstrated.

Metal workpieces are indispensable in the manufacturing industry. Surface defects affect the appearance and efficiency of a workpiece and reduce the safety of manufactured products. Therefore, products must be inspected for surface defects, such as scratches, dirt, and chips. The traditional manual inspection method is time-consuming and labor-intensive, and human error is unavoidable when thousands of products require inspection. Therefore, an automated optical inspection method is often adopted. Traditional automated optical inspection algorithms are insufficient in the detection of defects on metal surfaces, but a convolutional neural network (CNN) may aid in the inspection. However, considerable time is required to select the optimal hyperparameters for a CNN through training and testing. First, we compared the ability of three CNNs, namely VGG-16, ResNet-50, and MobileNet v1, to detect defects on metal surfaces. These models were hypothetically implemented for transfer learning (TL). However, in deploying TL, the phenomenon of apparent convergence in prediction accuracy, followed by divergence in validation accuracy, may create a problem when the image pattern is not known in advance. Second, our developed automated machine-learning (AutoML) model was trained through a random search with the core layers of the network architecture of the three TL models. We developed a retraining criterion for scenarios in which the model exhibited poor training results such that a new neural network architecture and new hyperparameters could be selected for retraining when the defect accuracy criterion in the first TL was not met. Third, we used AutoKeras to execute AutoML and identify a model suitable for a metal-surface-defect dataset. The performance of TL, AutoKeras, and our designed AutoML model was compared. The results of this study were obtained using a small number of metal defect samples. Based on TL, the detection accuracy of VGG-16, ResNet-50, and MobileNet v1 was 91%, 59.00%, and 50%, respectively. Moreover, the AutoKeras model exhibited the highest accuracy of 99.83%. The accuracy of the self-designed AutoML model reached 95.50% when using a core layer module, obtained by combining the modules of VGG-16, ResNet-50, and MobileNet v1. The designed AutoML model effectively and accurately recognized defective and low-quality samples despite low training costs. The defect accuracy of the developed model was close to that of the existing AutoKeras model and thus can contribute to the development of new diagnostic technologies for smart manufacturing.

The prognostic diagnosis of machine-health status is an emerging research topic. In this study, the diagnostic results of hollow ball screws with various ball-nut preloads were obtained using a machine-learning approach. In this method, ball-screw pretension, oil circulation, and ball-nut preload were considered. A feature extraction was used to determine the hollow ball-screw preload status on the basis of vibration signals, servo-motor speed, servo-motor current signals, and linear scale counts. Preloads with 2%, 4%, and 6% ball screws were predesigned, manufactured, and operated. Signal patterns with various preload features, servo-motor speeds, servo-motor current signals, and linear scale counts were classified using the support vector machine (SVM) algorithm. The features of the vibration signal were classified using the genetic algorithm/k-nearest neighbor (GA/KNN) method. The complex and irregular model of the ball-screw-nut preload could be learned and supervised using the driving motion current, ball-screw speed, linear scale positioning, and vibration signals of the ball screw. The experimental results indicate that the prognostic status of the ball-nut preload can be determined using the proposed methodology. The proposed diagnostic method can be used to prognosticate the health status of the machine tool.

This paper attempts to detect and predict the preload loss and life time of the ball screw demands in the machine tool industry. Loss of preload leads to a lower natural frequency and the onset of the nonlinear spring characteristics of the ball screw. Such problems result in rapid downtime in the manufacturing process and gradually undetermined positioning errors. In this paper the preload features are extracted and discriminated using the Hilbert-Huang Transform (HHT) and Multiscale entropy (MSE) method. The starting torque and consecutive quasi steady state constant currents of the machine tool feeding table are conducted experimentally. The test results show the prognostic status of ball screw preload loss can be envisaged by the proposed methodology.