Big Data-based methods for functional safety case preparation

Abstract

Aim. The paper aims to overview the opportunities, approaches and techniques of studying and ensuring functional safety of transportation systems, including those driverless, with the use of Big Data. It is noted, that the modern technology that underpins next-generation transportation systems that operate in ever-evolving conditions, with significant numbers of passengers, requires modified control systems design. With the growth of agglomerations, many suburban systems merge with urban ones, and their traffic intervals are close in size to those of the metro. Under these conditions, there is a transition from human-machine systems to automatic systems, characterized by varying degrees of automation. Widespread deployment of digital telecommunications, process automation and remote data collection and management technology is under way in railway transportation. Variations in the behaviour of transportation systems as a type of cyberphysical system cause a paradigm shift from line-andstaff to adaptive management with fundamentally non-linear systems with variable structure and parameters. Methods. Control and management systems are conventionally assessed for Lyapunov’s stability. In this case, the behaviour of a stable system can with a 100% probability be predicted in the neighbourhood of the ε-tube. For the examined supervised systems, in which stability is ensured through the introduction of a supervisor algorithm, speaking of a strict Lyapunov’s stability would not be correct. The idea of controlled algorithms can extend not only to ANN, but also to other intelligent algorithms. Thus, a scope of systems and knowledge is identified that is not covered by the relevant regulatory documents and methods of safety case preparation. Identifying and eliminating abnormal signals of such systems would allow defining the boundaries of the set of acceptable processes more clearly, thus, in some cases, increasing the speed of the decision algorithms by disabling an entire branch of unfavourable scenarios. Results. For non-linear transportation systems with variable stricture and parameters, examples are considered of machine learning/Big Data application in analysing the functional safety of complex control/management systems in railway transportation. The paper proposes the concept of application of supervised artificial neural networks combined with model checking. A special attention is given to artificial neural networks with control elements that are considered as a new subclass of neural networks. Conclusion. Updated requirements are defined for transportation systems using artificial intelligence as part of adaptive train schedule management and autonomous train control. That will ultimately allow developing an entire line of research associated with the operation of complex systems with counterintuitive behaviour from AI-based system functional safety estimation and machine learning to safety case preparation of intelligent supervised control/management systems based on formal verification.