QSPR Studies For Prediction Of Heat Sensitivity Of Energetic Compounds Based On ANN And Hot-spot Ignition Theory

Energetic compounds are widely used for both military and civil applications. Thesensitivity of energetic materials is an important indicator for the reliability and safety in theapplication. However, the necessary tests are sometimes time-consuming or even impossiblefor newly designed compounds. Consequently, the sensitivity prediction with molecularstructures has become important and urgent, and got considerable attention in the last decades.In this work, a large amount of molecular structure parameters from quantum chemicalcalculation were analysed and used in modelling with mathematical statistics methods.The main contents and results are as follow:(1)68energetic compounds have been carried on the quantitative calculation usingquantum chemical calculation method. A large number of molecular structure parameters thatrepresent the molecular structures were got.(2) After theoretical analysis on the data obtained various parameters,8parameters werechosen for structure describing using statistical method and taking actual physical meaningin thermal explosion into consideration. The model inputs and heat sensitivity had got a clearquantitative physical relationship, rather than simple data analysis. The work, making anadvance on how molecular descriptors influence the sensitivity, provided some instructionfor further designing of new kinds of energetic compounds.(3) After the obtained structure descriptors were normalization processed, appropriatetraining methods was chosen and model parameters were adjusted, model was developed bymeans of artificial neural network (ANN), and its predicted outcomes were compared tomeasured values. It showed good predictive power for the test set: a Pearson’s r of0.884, anadjusted R2of0.771and a Q2of0.801were achieved by the ANN model.(4) Key parameters of the obtained neural network model were read out to produce anindependent operation of the program interface for convenient using. Firstly, we can inputthe parameters of theoretical calculation of the molecular structure into the input box, thenclick and calculate, the predicted value of heat sensivity will be read in the output box. It’seasy to operate. The applicable scope of the model were analyzed with williams method.Limit value h of an energetic compound for the model application could be calculated fromits structure descriptors (0<h<0.5745).