Detection and isolation of dos and integrity cyber attacks in cyber-physical systems with a neural network-based architecture

New applications of industrial automation request great flexibility in the systems, supported by the increase in the interconnection between its components, allowing access to all the information of the system and its reconfiguration based on the changes that occur during its operations, with the purpose of reaching optimum points of operation. These aspects promote the Smart Factory paradigm, integrating physical and digital systems to create smarts products and processes capable of transforming conventional value chains, forming the Cyber-Physical Systems (CPSs). This flexibility opens a large gap that affects the security of control systems since the new communication links can be used by people to generate attacks that produce risk in these applications. This is a recent problem in the control systems, which originally were centralized and later were implemented as interconnected systems through isolated networks. To protect these systems, strategies that have presented acceptable results in other environments, such as office environments, have been chosen. However, the characteristics of these applications are not the same, and the results achieved are not as expected. This problem has motivated several efforts in order to contribute from different approaches to increase the security of control systems. Based on the above, this work proposes an architecture based on artificial neural networks for detection and isolation of cyber attacks Denial of Service (DoS) and integrity in CPS. Simulation results of two test benches, the Secure Water Treatment (SWaT) dataset, and a tanks system, show the effectiveness of the proposal. Regarding the SWaT dataset, the scores obtained from the recall and F1 score metrics was 0.95 and was higher than other reported works, while, in terms of precision and accuracy, it obtained a score of 0.95 which is close to other proposed methods. With respect to the interconnected tank system, scores of 0.96, 0.83, 0.81, and 0.83 were obtained for the accuracy, precision, F1 score, and recall metrics, respectively. The high true negatives rate in both cases is noteworthy. In general terms, the proposal has a high effectiveness in detecting and locating the proposed attacks

• Autores: Paredes Valencia, Carlos Mario; Martínez Castro, Diego; Ibarra-Junquera, Vrani; González Potes, Apolinar
• Tipo de recurso: Article of investigation
• Fecha de publicación: 2021
• Institución: Universidad Autónoma de Occidente
• Repositorio: RED: Repositorio Educativo Digital UAO
• Idioma: eng
• OAI Identifier: oai:red.uao.edu.co:10614/13869
• Acceso en línea: https://hdl.handle.net/10614/13869; https://red.uao.edu.co/
• Palabra clave: Detección de anomalías (Seguridad informática); Redes neuronales (Computadores); Anomaly detection (Computer security); Neural networks (Computer science); Anomaly detection; Anomaly isolation; Artificial neural networks; Cyber Physical System
• Rights: openAccess
• License: Derechos reservados - MDPI, 2021