The work describes swarm artificial intelligence (SAI) systems, designed to predict the technical state of engineering, and also for predictive medicine systems. The cluster of computing cores of an SAI system is a set of interacting neurosingular machines (iNESIMA), consisting of many interacting ganglia. Each ganglion is a modified Turing machine. An inverse neurosingular machine is built on the basis of failure physics, models of continuum mechanics with a microstructure and simulates evolution processes in multiphase materials based on their topological and structural characteristics. On the basis of the neurosingular model, a characteristic such as RUL is redefined, which is a multidimensional numerical array. Using the elements of this array, the functional classes of the incoming time series are indexed. The task of optimizing predictive maintenance strategies is solved in each class of failure progression timeline (FPTL), and the optimizing strategy is to vary the control parameters of the mechanisms in real time on the basis of RUL estimates using iNESIMA and the formation of optimizing control loops of inverse iNESIMA. The self-maintenance solution is presented in terms of neurosingular machine and inverse neurosingular machine. This work is the culmination of recent studies and previously published results.