PSI - Issue 5

Alexander Serov et al. / Procedia Structural Integrity 5 (2017) 1160–1167 Alexander Serov / Structural Integrity Procedia 00 (2017) 000 – 000

1161

2

constructed manually on the basis of results of numerical simulation or experimental results of using similar systems. This way doesn’t guarantee good accuracy of damage identification in the case of high complexity of engineering system. Often prediction of damage or of degradation of properties of engineering structure is complicated by poorly known conditions of its operation. These problems may be solved on the next stage of development of SHM systems. Next generation of technologies developed for monitoring of health of technical systems will include the set of means for the simulation of intelligent processing of data. The need to use intelligent techniques is dictated by continuous increasing of complexity of created technical systems and complexity of technologies that human society has. Phenomena and events in these systems may be essentially different on the scale, dynamics and character from those which are ordinary for us as human beings. Currently analysis of events in these systems is based on the transformation of patterns that characterize these events (and often are not directly interpretable) into the set of patterns which we can interpret. This way may be applied if there are no hard limitations on time boundaries of this analysis. It’s hard to use this way if the dynamics of source events is higher than that in our world. Development of technical systems which are able to cognitive activities will make possible to realize control and management of events in technological field providing the degree of detail which is inaccessible for human consciousness. Development of cognitive technical systems would make possible to realize ideas of Ubiquitous Computing, Weiser (1994) in the field of Structural Health Monitoring. Automated knowledge extraction is one of basic requirements for the construction of Integrated Vehicle Health Monitoring (IVHM) systems, Marzat et al, (2012), Price et al. (2003). IVHM architecture may be used as a basic architecture for the construction of a new type control systems of aircrafts, submarines, and spaceships, Price et al. (2003). We can suppose that IVHM-related direction of research and development in the field of SHM will lead in the future to creation of intelligent technical systems able to self-healing. In this paper we consider the problem of development cognitive architecture and application of this architecture for the monitoring of health of engineering systems. Workflow which traditionally is implemented in SHM systems includes the following set of stages: periodical measurements by the array of sensors; processing results of measurements aiming to extract damage sensitive features; analysis of extracted features to identify current state of the object, Chang at el. (2011), Kim et al. (2007). Engineering structures which are subject to condition monitoring may be characterized as dynamical systems. Structural properties of these systems change with time. Frequently this change has several different time scales. And dynamics of change of structural properties especially for complex technical systems is unknown. All this gives rise to a number of problems which concern adaptation of data processing methods to particular type of monitored object or even to particular instance of the same type. One of possible ways of solution is connected with application of adaptive methods of monitoring. According this way SHM system must adapt itself to particular instance of monitored object. We believe this adaptation may be done by application of principles of Cognitive Science. SHM system by periodical measurements and processing results of these measurements gradually constructs knowledge base. This knowledge base characterizes just one particular instance of engineering system. But we can develop knowledge repository which will be universal. For the solution of this problem we can use Artificial Neural Networks (ANN). Artificial Neural Networks are considered today as a most perspective way of development Artificial Intelligence (AI) based systems. The number of various types of architecture of ANN existing now is quite big (see, for example, Abdelwahab (2016), Schmidhuber (2015)). From this set of architectures we would like to highlight those neural networks which support Machine Learning Algorithms from a Deep Learning class: Deep Learning Architectures, Bengio et al. (2009), Bengio et al. (2015). The most important advantage of Deep Architectures is their ability to learn multiple levels of representation that correspond to different levels of abstraction of outer world. These levels realize the hierarchy of concepts which AI system has due to previous cognitive activity. Automatic construction of hierarchical description of outer world is one of most important problems in the field of AI based SHM systems. Patterns which are ordinary for human consciousness are very rough for representation of world which is observed by cognitive technical systems. Concepts which we learn from our experience (experience of human beings) are not fitted well for the description of events which arise in technical systems. For estimation of health of engineering structures and for prediction of dynamics of health it is necessary to use SHM systems that are 2. Dynamic Artificial Neural Networks