Explosion-induced domino accidents in the chemical industry, such as the 2005 Buncefield and 2019 Xiangshui accidents, can lead to catastrophic losses. Recent studies commonly use probit models (simplified linear regression models) to predict the probability of accident escalation caused by equipment failure due to overpressure conditions but necessitate distinct equations for different equipment types. In order to simplify the number of models and improve their accuracy, this study introduced three machine learning models (random forest model, convolutional neural network model, and deep neural network model), addressing complex nonlinear relationships that conventional regression models may not fully capture. By model training, the DNN model has the highest accuracy (99 %), followed by CNN (94 %) and random RF (95 %). The DNN model was selected as the optimal data-driven model for equipment vulnerability assessment due to their feedforward mechanism's capability to dynamically align parameters with evolving data distributions. The approach developed can not only predict the probability of equipment damage by integrating values related to peak overpressure and equipment type but also effectively address the accuracy validation issues associated with traditional regression models. Besides, this approach can be considered open source model and more explosion data may be used in the future to further improve the model.

A new type of radially polarized (RP) cosine-Gaussian (CG) field is proposed. Through the analytical model, it is found that such RP CG beam exhibits completely different focusing properties from the reported RP plane waves. More importantly, a stable three-dimensional trap of Rayleigh particle accompanied by a subwavelength spin motion can be easily achieved using this RP CG beam.