|
|
 Выпуск 86
- Burnaev E. Anomaly detection based on surrogate models
Predictive maintenance methods are used to detect as soon as possible significant changes (disorders) in the operation of mechanisms. The main purpose of this approach to maintenance is to continuously monitor and correct the technical condition or completely replace the mechanisms before the detected changes become critical for the operation of individual components or the system as a whole. The diagnostic capabilities of predictive maintenance methods have significantly increased in recent years due to the improvement of sensory observation technologies and the development of new information processing algorithms. The use of predictive maintenance has a number of advantages in comparison with other practiced approaches to maintenance, namely, the ability to conduct more accurate and timely monitoring of the health of individual parts and the entire system as a whole; continuous monitoring and analysis of internal and external conditions improves the safety of operation and allows for a more rapid and in some cases preventive response to possible accidents and failures; significant reduction in maintenance costs, due to the exclusion of planned replacements of technically sound and reliable system components. In this paper, we consider the problem of constructing predictive models (surrogate models) to solve the predictive maintenance problem. The special features of the problems of detecting anomalies and predicting failures are analyzed. An overview of the main needs of industrial applications and a description of the structure of the corresponding predictive maintenance systems is provided. An example of using methods for detecting anomalies based on surrogate models for predictive maintenance of shovel machines is given.
- Furtat I., Nekhoroshikh A., Gushchin P. Robust stabilization of linear plants in presence of disturbances and high-frequency measurement noises
A solution is proposed for the robust stabilization of linear dynamic plants with unknown parameters belonging to a known compact set, bounded external disturbances, and bounded high-frequency measurement noises. The synthesis of the control algorithm is divided into two steps. At the first step a filtering algorithm is synthesized, which makes it possible to reduce the influence of measurement noises on the output variable of the plant. If the measurement noises can be represented as the sum of sinusoidal signals, then constructive conditions for choosing parameters in the filtering algorithm are proposed. At the second step, a control algorithm is synthesized with the attenuation of the influence of parametric uncertainty and external disturbances. This algorithm is based on the use of finite differences in continuous time, which avoids the use of dynamic observers that increase the dimension of a closed-loop system. The simulation results illustrating the effectiveness of the proposed algorithm in comparison with some existing analogues are presented. Thus, a comparative analysis with the results of Astolfi D., Marconi L., Isidori A. etc. has been showed that the proposed control algorithm have less dynamic order, guarantees higher accuracy with respect to the output signal and its derivatives. Moreover, in contrast to the results of Astolfi D., Marconi L., Isidori A. etc., in the proposed algorithm, choosing the algorithm parameters is easier due to independent filter settings and the control law, while choosing parameters in the controller of Astolfi D., Marconi L., Isidori A. etc. performed simultaneously for the whole algorithm.
- Glushchenko A., Petrov V., Lastochkin K. Control quality improvement of dc motor on basis of its linearization and compensation of unmodeled dynamics
The aim of this research is to develop an approach to DC motor control as an alternative to the conventional cascade control. Disadvantages of the mentioned classical method are shown, including inability to compensate effectively the influence of unknown dynamics and disturbances (load torque). The solution to these problems lies in the joint use of: 1) the feedback linearization method to separate the unknown dynamics from the control object description and 2) the second Lyapunov method to compensate it. A DC motor linearization based on the solution of the inverse dynamics problem is proposed in this research. It allows to consider limitations on physical signals of current and voltage of anchor circuit. A linear adapter with real-time adjusted parameters on the basis of formulas obtained with the help of the Lyapunov second method is proposed to compensate the unmodeled drive dynamics. The formulas distinctive feature is that when they are used, it is necessary to know the control object gain sign only. The stability of the system with an adapter is proved using Uniform Ultimate Boundedness. Experimental verification of the proposed approach is conducted using the DC drive model with non-stationary parameters. As an example, it is shown that the adaptive system is able to compensate the plant non-stationarity, when the armature circuit parameters are changed by 1.5 times from their nominal values, the inertia moment is changed by two times from its nominal value, and the load torque is equaled to half of the value of the torque, which corresponds to the motor cutoff current. A discussion of the results and further research aims are shown at the end of the paper.
- Zorkaltsev V., Polkovskaya M. Multiplicative model of trend detection and seasonal fluctuations: application to the dynamics of prices for agricultural products
The problem of a trend detection, and periodic and random fluctuations analysis for time series is considered. The problem of the inconsistency of additivity and multiplicativity as two requirements to the methods of selecting components is discussed. This needs developing of alternative methods of time series decomposition with the distinguished components interacting either additively or multiplicatively. A description of additive and multiplicative models for trend detection, seasonal and random fluctuations analysis in the monthly economic data is given. The prices dynamics analysis for agricultural products is considered as an application of the multiplicative model. It justifies the necessity of isolating the trend and seasonal components from the series of monthly prices in order to ensure more efficient planning of agricultural production, reduce risks, and choose optimal storage and sale periods for products, taking into account the action of random factors in the production and sale of products. The expediency of applying the multiplicative version of the component decomposition model for analyzing and forecasting prices is argued. The results of using the model when analyzing the dynamics of prices for certain types of agricultural products in the Irkutsk region are presented.
- Sochnev A. Production resources planning based on net models
The application of simulation models based on Petri nets for resource planning systems is considered. A formalized procedure for the formation of an inverse Petri net for a class of net-processes (Workflow nets) has been developed. Formal rules for inverse simulation are presented in this paper. A mechanism of material needs planning has been proposed. It’s based on an inverse Petri net and inverse simulation. The issues of using some traditional optimization techniques in the implementation of reverse network simulation, in particular, the application of the rules of the priority of operations are considered. Some of the most typical formulations of the problem of optimal resource allocation have been determined, for which the approach proposed in the article provides a fast and fairly accurate solution for practical use. An example of solving the problem of planning resource requirements for a site of a typical medium-series machine-building enterprise is given and the features of the practical use of the proposed approach are revealed. The structure of the considered production system is quite typical for machine-building production, which makes it possible to predict the positive effect of the approach and is the basis for its replication in such systems.
|
|
