Abstract In order to accurately and quickly predict the anti-impact performance of the plastic shell of a mining sensor, taking the plastic shell of a mine dust concentration sensor as the research object, finite element simulation software is used to simulate and predict its performance in the impact test. Compared with the experimental results, the accuracy of finite element simulation is verified. The batch simulation is carried out by using the finite element method, and the BP neural network and the PSO (particle swarm optimization)-BP neural network are established. The comparison results show that the PSO (particle swarm optimization)-BP neural network has higher accuracy than the BP neural network, and can more accurately predict the mechanical properties of the mine sensor shell under impact conditions. At the same time, compared with the finite element method, it has the advantage of efficiency. The prediction time can be improved to 5-10 milliseconds by using PSO-BP neural network mode, compared to hours of the finite element simulation. This method provides a research method for efficiently and accurately obtaining the impact mechanical performance of mine plastic shells and optimizing structural parameters.