Safety and reliability analysis of the solid propellant casting molding process based on FFTA and PSO-BPNN

This paper proposes a physics-based machine learning model to analyze the safety and reliability of solid propellant casting molding processes. The model identifies the relationship between process variables that may lead to failure events and process safety. The fuzzy fault tree analysis (FFTA), as a typical physical model, can provide reasonable physical criteria and reliable a priori knowledge for back propagation neural network (BPNN). All information mapped into BPNN is used to explore the nonlinear relationships of the data and establish dynamic rules. The particle swarm optimization (PSO) algorithm is used to improve the performance of the BPNN model (PSO-BPNN), and a risk prediction model with a maximum error of 0.0006 is obtained. The results show that the proposed model can provide high precision evaluation results. A sensitivity analysis is also performed based on the mean impact value (MIV) algorithm. The importance of curing temperature, casting vacuum, curing time, casting time, and vacuum degree is determined. The above methods help realize dynamic risk analysis of the solid propellants production process and provide timely warning and feasible reference for unsafe processes.

Automated summary

This paper proposes a physics-based machine learning model to analyze the safety and reliability of solid propellant casting molding processes. The model identifies the relationship between process variables that may lead to failure events and process safety. The fuzzy fault tree analysis (FFTA) provides physical criteria and reliable a priori knowledge for the back propagation neural network (BPNN). The particle swarm optimization (PSO) algorithm is used to improve the performance of the BPNN model, resulting in a risk prediction model with a maximum error of 0.0006. The proposed model provides high precision evaluation results and the importance of various process variables have been determined using sensitivity analysis.