Bearing surface defect detection and classification methods based on machine vision have been widely used in bearing quality inspection due to their high-speed, high-precision, and non-contact advantages. However, traditional machine vision algorithms have low reusability and their development processes are often expensive and time consuming. Several deep learning-based bearing surface defect detection methods have been proposed. However, these deep learning models often require a large number of datasets, which is often difficult to achieve in the actual industry. Transfer learning provides a promising solution to the small sample difficulties associated with deep learning. However, the complexity of the illumination conditions and the huge differences between bearing dataset and ImageNet dataset make it impossible to use the current single model-based transfer learning for bearing defect detection. In this study, we propose a novel transitive transfer learning convolutional neural network (CNN) ensemble framework for classifying bearing surface defects. Only small-scale datasets are needed in this framework. A transfer path and transfer method selection strategy for transitive transfer learning is then proposed to train the deep learning models, which enhances the feature extraction ability of the CNN models on the basis of multiple illuminations. Ablation experiments are conducted to verify the effectiveness of the proposed method. Experimental results show that the proposed transitive transfer learning CNN ensemble framework has the accuracy rate of 97.51%. The average time for detecting each bearing is 155 ms, which can meet the requirements of industrial online detection.

Optical tweezers are widely used for noninvasive and precise micromanipulation of living cells to understand biological processes. By focusing laser beams on cells, direct cell manipulation with optical tweezers can achieve high precision and flexibility. However, direct exposure to the laser beam can lead to negative effects on the cells. These phenomena are also known as photobleaching and photodamage. In this study, we proposed a new indirect cell micromanipulation approach combined with a robot-aided holographic optical tweezer system and 3D nano-printed microtool. The microtool was designed with a V-shaped head and an optical handle part. The V-shaped head can push and trap different sizes of cells as the microtool moves forward by optical trapping of the handle part. In this way, cell exposure to the laser beam can be effectively reduced. The microtool was fabricated with a laser direct writing system by two-photon photopolymerization. A control strategy combined with an imaging processing algorithm was introduced for automated manipulation of the microtool and cells. Experiments were performed to verify the effectiveness of our approach. First, automated microtool transportation and rotation were demonstrated with high precision. Second, indirect optical transportations of cells, with and without an obstacle, were performed to demonstrate the effectiveness of the proposed approach. Third, experiments of fluorescent cell manipulation were performed to confirm that, indicated by the photobleaching effect, indirect manipulation with the microtool could induce less laser exposure compared with direct optical manipulation. The proposed method could be useful in complex biomedical applications where precise cell manipulation and less laser exposure are required.

Vision-based weld seam extraction poses a significant challenge for weldments with complex spatial structures in automated welding. Existing research primarily focuses on identifying weld seams from weldments with given positions and postures, while practical weld path planning requires multiple weld seams identified within arbitrarily placed weldments. This paper proposes a methodology that identifies weld seams from arbitrarily placed spatial planar weldments in a single run. First, by introducing a turntable calibrated with respect to a 3D camera, we perform 3D reconstruction on an arbitrarily placed spatial planar weldment. Second, an improved RANSAC algorithm based on Euclidean clustering is proposed to carry out plane segmentation, focusing on segmentation accuracy. Finally, we present a novel weld seam extraction algorithm leveraging the half-edge data structure to efficiently compute weld seams from the segmented planes. The experiments conducted in this study demonstrate that the average segmentation errors (as an indirect indicator of weld seam extraction error) are reduced by 90.3% to 99.8% over conventional segmentation methods, and the standard deviations are reduced by 64.8% to 97.0%.