Deep learning-based methods have shown excellent potential on object detection and pose estimation with vast amounts of training data to achieve good performance. Obtaining enough comprehensive manual labeling training data is a time-consuming and error-prone task in industrial scenes, and most current time-saving synthetic data generation methods require detailed high-quality CAD models. Instead, we introduce an image-to-image translation based synthetic data generation approach, which requires only untextured CAD models and a small number of real images. The proposed approach starts with edge maps extracted by mesh model projection. And each edge map is provided with realistic appearance learned from real images through image-to-image translation to achieve patch-level realism, and then superimposed on randomly selected backgrounds with a simple Cut-and-Paste strategy. Experiments on our custom dataset and T-LESS dataset show that our synthetic data is competitive with uncomprehensive real data and gives a considerable improvement when compared to the same degree of model rendering images. In addition, the proposed approach allows for extensibility to accommodate new parts with similar texture and structure. And it shows the potential to train image translation model on only one single object and synthesize data for multiples in one take.

To improve virtual human motion simulation and generation efficiency in virtual assembly and maintenance (A&M) environment, based on a detailed analysis of virtual human control methods in A&M simulation, a virtual human hybrid control method is presented in this paper. In the proposed method, virtual human A&M motion is divided into three types: primitive motion (PM), fine motion (FM) and transitional motion (TM). For PM, a real operator can drive a virtual human in real time for A&M operation via virtual reality interactive devices; in this way, the PM can be quickly generated. As for FM and TM generation, their semantic-level parameterized models including engineering attributes are created; then, their generation processes are represented; in the generation processes, a virtual human posture automatic reasoning model is established to support auto-generating A&M motion. The approaches in the present study are validated by generating virtual human motions in partial assembly process of auto engine. Results show that the proposed method can improve A&M motion simulation efficiency and support the fast generation of realistic A&M motion.

Cone crushers are used in the aggregates and mining industries to crush rock material. The pressure on cone crusher liners is the key factor that influences the hydraulic pressure, power draw and liner wear. In order to dynamically analyze and calculate cone crusher performance along with liner wear, a series of experiments are performed to obtain the crushed rock material samples from a crushing plant at different time intervals. In this study, piston die tests are carried out and a model relating compression coefficient, compression ratio and particle size distribution to a corresponding pressure is presented. On this basis, a new wear prediction model is proposed combining the empirical model for predicting liner wear with time parameter. A simple and practical model, based on the wear model and interparticle breakage, is presented for calculating compression ratio of each crushing zone along with liner wear. Furthermore, the size distribution of the product is calculated based on existing size reduction process model. A method of analysis of product size distribution and shape in the crushing process considering liner wear is proposed. Finally, the validity of the wear model is verified via testing. The result shows that there is a significant improvement of the prediction of cone crusher performance considering liner wear as compared to the previous model.