Concepts for Processing Non-stationary Loading for Creating Digital Twins of Full-Scale Objects from the Point of View of Durability

Modern methods for assessing the durability of structures under random loading have significant limitations, namely, they consider only stationary ergodic Gaussian random processes. At the same time, in engineering practice it is necessary to work with real non-stationary loading processes everywhere, be it stationary units or moving vehicles. At the same time, their processing requires large time expenditures and strongly depends on the specialist’s qualification. Moreover, the existing methods of processing of random loading processes are unsuitable for automation and, as a consequence, cannot be used as algorithmic support for digital doubles of natural products for estimation of their residual resource. Recently more and more researches affecting the processing of non-stationary signals with the use of empirical signal decomposition and machine learning methods new for the scientific community appear. In the article the existing innovative approaches in estimation of durability with application of modern methods of signal processing and methods of artificial neural networks are considered. In addition, two possible concepts for processing non-stationary loading processes with the possibility of further application of the developed methods to create digital doubles are proposed. The first concept implies the decomposition of a non-stationary random process with the subsequent estimation of maximum distribution densities for each obtained process and, summarizing the damage from the action of each process, the calculation of the final durability. The second concept is the application of frequency-time analysis methods and convolutional neural network algorithms to predict the material fatigue curve for a specific nonstationary loading process.