Crested porcupine optimizer inspired moving deep neural network model for structural response and reliability analyses

Purpose This study aims to develop a novel method for quantifying the failure probability of engineering systems under uncertainty, thereby enhancing the efficiency and accuracy of reliability analysis and ensuring safety and stability throughout the service life. Design/methodology/approach A crested porcupine optimizer inspired moving deep neural network (CPO-MDNN) model is proposed by integrating moving least squares, deep neural network (DNN), meta-heuristic algorithm and Gaussian process. In this method, the moving least squares is adopted to define the radius of compact support region and select effective modeling samples; the DNN is utilized to establish the correlation model between input variables and output response; the crested porcupine optimizer (CPO) is used to find the optimal weights and thresholds of DNN; the Gaussian process is applied to optimize the hyperparameters of the correlation model. Findings A numerical example (the nonlinear function approximation and probability analysis) and an engineering case (the exhaust gas temperature prediction and failure analysis of ship gas engine) are employed to verify the modeling/simulation accuracy and efficiency of the presented method. Results show that CPO-MDNN exhibits superior modeling and simulation performance compared to alternative approaches. Originality/value The effort offers strong support for the health management and optimal design of ship gas engines, while also contributing to the advancement of engineering system reliability theory.