The interpolator is one of the critical components of industrial robots control, significantly affecting their accuracy. In such technological tasks as welding, laser cutting, coating, and surfacing, in addition to the spatial accuracy of the robot’s endeffector, the accuracy of its velocity during motion along complex trajectories plays an important role. In this paper, we propose a new approach for solving the interpolation problem of a multi-axis industrial robot based on the B-splines. The proposed algorithms can be easily adapted for robots with any kinematics, generating the current, velocity, and position setpoints for the control loops of each of its actuators. A software implementation of the offline interpolator based on the proposed algorithms was developed and executed on B&R industrial controllers. During the experimental studies performed on a SCARA robot, it was demonstrated that the developed algorithmic solutions outperform the standard interpolator of B&R control systems, exceeding it up to 2 times in terms of spatial accuracy and up to 4 times in terms of root mean square velocity deviation. The maximum deviation of the tool’s velocity using the developed algorithms did not exceed 2.4 mm/s, comparable to the results of the most modern planar solutions based on NURBS curves. At the same time, unlike their planar analogs, the solutions proposed in this paper are suitable for multidimensional interpolation. This part is devoted to simulation and field experimental studies of the algorithms described in Part I of this paper, as well as a summary of the the research results

The article is devoted to the issue of constructing the proximity estimation of a pair of autonomous underwater vehicles moving on parallel courses. Two approaches are considered depending on the density of the vehicles in the group: when moving in narrow spaces and when moving on parallel courses for a relatively long time. Exact and approximate methods of constructing such estimates depending on the group density are proposed, since noise manifests itself in fundamentally different ways in these situations. In the case of a dense group, a collision event cannot be considered rare and exact state estimation methods are applied. In the exact method, the equations of dynamics are treated as a conditionally Gaussian system to compute the necessary estimates. For this purpose, the Lipzer-Shiryaev method is used to account for nonlinear dependencies in the equations of the observed variables. In the case of rarefied density (long apparatus movements), coarser approaches in the estimations are allowed. The fact of collision is considered as a rare event, which is led to by the coincidence of factors that form a quite definite sequence in time — the extremal of the optimal control problem. The monitoring problem is formulated as a large deviation control problem. Applying the principle of large deviations, the stochastic problem of collision probability estimation is reduced to a deterministic optimal control problem. A rough estimate of the collision probability for two vehicles is obtained for the limit solution of the averaged system in the paper. As an application of the proposed approach, the problem of motion control of autonomous underwater vehicles moving in the horizontal plane with constant longitudinal velocity at a given depth is considered. Under the condition of nondegeneracy of the diffusion matrix in the equation of observable variables, an algorithm for recovering the transverse coordinates and velocity and calculating the collision risk on this basis is obtained.The paper uses the approach of A. Puchalskii, which requires only controllability of the system by input noise. Knowledge of the extremal allows predicting the collision event, which is used in the paper to estimate the collision risk

The principle of creation of neural network simulator of gas turbine engines in the form of recurrent neural networks and their application in the hardware-in-the-loop simulation for testing and debugging automatic control and condition-monitoring systems is considered. A comparison of NARX and GRU architectures of simulators is carried out. A technique for constructing a gas turbine neural network simulator (model) and its implementation on a hardware-in-the loop simulation testbed is described. Hardware-in-the loop simulation is used for prototyping, testing and certification of developed products (physical objects) with a complex of mathematical models, where real systems are associated with virtual models in which they are located. The results of hardware-in-the loop simulation of the parameters of a gas turbine engine with a real full authority digital engine control system for startup mode, ground mode and flight modes are presented. An analysis of the accuracy and adequacy of the considered models is carried out. The accuracy required to solve control problems and form requirements for electronic control systems units has been confirmed. The intelligent modeling approach can be used to create full-fledged (complete) digital twins, where a model of physical processes and object behavior based on recurrent neural networks can be connected to 3D solid-state modeling to solve problems of object analysis and synthesis, its optimization and reliability increase. The development of such technologies makes it possible to create intelligent models that can be used in digital twins of complex technical systems

This study comprehensively evaluates several Machine Learning (ML) techniques to address the challenge of predicting output currents for demand response management and energy distribution optimization in low-voltage Direct Current (DC) microgrids. The study utilizes an extensive dataset of around 33,334 data sets with diverse electrical characteristics. Several prediction algorithms are used and evaluated in a planned way during this process. These include Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR), Multilayer Perceptron (MLP), Lasso, and Linear Regression (LR). The Random Forest (RF) model outperforms the other models, with a high R^2score of 0.994, indicating a very accurate fit to the observed data. In contrast, the Lasso model has a R^2 score of 0.883, suggesting a somewhat lower effectiveness due to its simplicity. The findings provide a comprehensive assessment of the predictive capabilities of each model, which is further corroborated by other research utilizing measures such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). For instance, the Random Forest (RF) model showcased its robustness in accurately predicting output currents by attaining the lowest Mean Absolute Error (MAE) of 0.289, Mean Squared Error (MSE) of 0.140, and Root Mean Squared Error (RMSE) of 0.374. This comprehensive evaluation enhances the advancement of sustainable and efficient energy distribution networks by emphasizing the potential of Machine Learning (ML) to improve Direct Current (DC) microgrids’ operational efficiency. It also establishes the foundation for future research on integrating these algorithms into real energy systems

The paper presents the synthesis of a tracking system for a quadrocopter considered as a solid body with six degrees of freedom and four control actions (rotor lift forces), considering design constraints on velocities and controls. The plant operates under conditions of parametric and external disturbances, as well as incomplete measurements. The trac king loop is designed in a typical way and consists of translational and rotational motion subsystems with three inputs and three outputs each. Reference trajectories are independently specified for the spatial position of the quadrocopter’s center of mass and yaw angle. The pitch and roll angles have a dual function: in the translational motion subsystem they, together with the total lift force, act as controls, which are considered as reference actions in the rotational motion subsystem. Scientific novelty is related to the developed method of dynamic feedback design using piecewise linear feedback with saturation in regulators, state and disturbance observers, as well as dynamic differentiators of reference actions. Application of the block control principle with combined piecewise linear feedbacks with saturation for design of tracking subsystems of spatial and angular positions allowed to provide stabilization of tracking errors at imposed constraints on velocities and controls. Reduced dynamic observers with piecewise linear correction reduce the computational load. Based on tracking error measurements, they recover composite signals including unmeasured velocities, uncertain parameters, and external disturbances to a specified accuracy without the need for individual identification of uncertain parameters. The feedback formation on the variables of such observers ensures robustness of the tracking system. Instead of numerical differentiation operations, which are problematic to implement, dynamic differentiators with piecewise linear correction are used to recover derivatives of reference actions, which are capable of processing piecewise differentiable signals and do not generate surges of estimation signals at special points. The presented results of numerical modeling confirm the effectiveness of the developed algorithms