In order to accurately locate the key processing links affecting the pass rate of the first delivery of the finished gun, the Bayesian network is selected to construct a causal model between the processing parameters and the pass rate. By selecting the long short-term memory (LSTM) neural network model as the time series prediction model for the pass rate of the first delivery of guns, the pass rate of the first delivery of guns in the next batch can be predicted more accurately, and the key processing links can be further located. The results show that the prediction can provide a theoretical reference for the next targeted improvement of the production process.