Trustworthy Data Analysis and Sensor Data Protection in Cyber-Physical Systems

Abstract

Cyber-Physical Systems are widely used in critical infrastructures such as the power grids, water purification systems, nuclear plants, oil refinery and compressor plants, food manufacturing, etc. Anomalies in these systems can be a result of cybersecurity attacks, failed sensors or communication channels. Undetected anomalies may lead to process failure, cause financial damage and have significant impact on human lives. Thus, it is important to detect anomalies at early stages and to protect data in Cyber-Physical Systems. In this paper, we propose the novel on-the-fly NIST-compliant key generation scheme for a secure data container used to transfer and store sensor data. The data container delivers data from the low-level field sensors to high-level data analysis servers in a protected form. It provides data confidentiality and integrity, as well as data origin integrity, a fine-grained role-based and attribute-based access control. As a result, the anomaly detector runs on trustworthy data sets, protected from unauthorized adversarial modifications. Our solution can be easily integrated with many existing Cyber-Physical Systems and IT infrastructures since our secure data container supports RESTful API and is implemented in two modifications: (1) signed, watermarked and encrypted spreadsheet file; (2) signed and encrypted JSON file. In addition, we implemented several machine learning models based on a Random Forest, a k-Nearest Neighbors, a Support Vector Machine and a Neural Network algorithms for the detection of various anomalies and attacks in a gas pipeline system. We will demonstrate that our anomaly detection models achieve high detection rate with an average accuracy of 97.7% for two industrial data sets collected by the Mississippi State University's Critical Infrastructure Protection Center and Oak Ridge National Laboratories (ORNL)