AL-SNN: an active learning based spline neural network for risk modelling and reliability design

A novel AL-SNN framework is proposed to address three key tasks in the design of reliability test plans and the evaluation of associated risks. Specifically, the framework integrates active learning with a spline neural network to support risk modeling and reliability design within the field of reliability engineering. A new risk space definition is introduced to facilitate t he visualization of risk trends under varying test parameters. Cubic spline functions are utilized for surface fitting to enhance model accuracy. The framework effectively supports risk computation, surface fitting, boundary identification, and risk prediction. Its performance is validated through simulation studies and a case analysis involving an exponentia l product with prior information. The proposed method demonstrates significant improvements in both computational efficiency and predictive accuracy for reliability test planning and risk assessment.