Lin Jingdong

Forming defects prediction for sheet metal forming using Gaussian process regression

Forming defects like fracture and wrinkle are key factors to influence the forming quality of sheet metal. Accurate prediction of forming defects is essential for the sheet metal forming process. In this paper, an approximation model technique based on Gaussian process regression(GPR) is proposed to predict the forming defects in sheet metal forming process. Finite element analysis is applied to simulate the drawing process. Design variables include drawbead resistance coefficient and blank-holding force. Forming limit diagram is used to calculate the values of defects. A dash board drawing case demonstrates that the proposed approach is more accurate and effective than support vector machines method and conventional response surface method.

Research on recognition algorithm of seal fitting point based on compressibility control

According to the particularity of solid rocket motor tightening process seal compressibility which is difficult to measure directly, this article established mechanism model between the compressibility and the key parameters of screw based on analysis of the seal deformation mechanism. Using model prediction and expert control, we created expert rules of seal fitting point torque variation characteristics, and framed the expert estimation model of seal fitting point recognition. Therefore, the recognition accuracy of fitting point is improved. The experiment on hardware platform verified that fitting point recognition algorithm meet the seal compressibility 13% precise control requirements and achieve better effect.

Soft sensor for the compaction density of powders in the elongated metal tube based on Gaussian process regression

The Online measurement of compaction density of powders in the elongated metal tube is typically unavailable due to the limited conditions. To solve this problem, a soft sensor model based on Gaussian process regression method is applied, analyzing the factors that influence the powder density in the compaction process. Compared with Bayesian linear regression and SVM methods, the predicted results show that the proposed soft sensor based on Gaussian process regression model has advantage in predicting the compaction density of powders in the elongated metal tube. With this model, the real-time monitoring and control of compaction density of powders could be satisfied, which could guarantee the final explosive quality of powders in the metal tube.