Chaos Engineering for Enhanced Resilience of Cyber-Physical Systems

Abstract

Cyber-physical systems (CPS) incorporate the complex and large-scale engineered systems behind critical infrastructure operations, such as water distribution networks, energy delivery systems, healthcare services, manufacturing systems, and transportation networks. Industrial CPS in particular need to simultaneously satisfy requirements of available, secure, safe and reliable system operation against diverse threats, in an adaptive and sustainable way. These adverse events can be of accidental or malicious nature and may include natural disasters, hardware or software faults, cyberattacks, or even infrastructure design and implementation faults. They may drastically affect the results of CPS algorithms and mechanisms, and subsequently the operations of industrial control systems (ICS) deployed in those critical infrastructures. Such a demanding combination of properties and threats calls for resilience-enhancement methodologies and techniques, working in real-time operation. However, the analysis of CPS resilience is a difficult task as it involves evaluation of various interdependent layers with heterogeneous computing equipment, physical components, network technologies, and data analytics. In this paper, we apply the principles of chaos engineering (CE) to industrial CPS, in order to demonstrate the benefits of such practices on system resilience. The systemic uncertainty of adverse events can be tamed by applying runtime CE-based analyses to CPS in production, in order to predict environment changes and thus apply mitigation measures limiting the range and severity of the event, and minimizing its blast radius.

Enhancing Operational Resilience of Critical Infrastructure Processes Through Chaos Engineering

“…Eq. ( 3) demonstrates that synchronization of output measurements is possible in steady-state conditions, with the deviation being the same at all ICS components [14].…”

“…Our study builds upon previous research on applying CE principles to CPS [14] and aims to demonstrate the benefits of using CE to enhance a CPS's operational resilience. To achieve this objective, we integrate CE experimentation [16] [17] into the simulation phase of DT models using material flow networks (MFNs).…”

“…The use of CE experiments in industrial CPS is new, and to the best of our knowledge, it was first introduced in [14]. In their approach, the authors effectively applied CE principles to an industrial CPS testbed, demonstrating its capability to predict environmental changes and implement mitigation measures that control the severity of adverse events.…”

“…However, significant challenges still need to be addressed when conducting CE experiments on real-world operational CPS without disrupting normal operations and system availability. Similar to the work in [14], our approach employs CE to assess industrial CPS resilience by conducting experiments that introduce random and unpredictable behaviors into the system. We built upon this previous work by incorporating MFN to create a DT model of the production system.…”

“…where the state 𝑥(𝑡) ∈ ℝ 𝑛 with 𝑥(0) = 𝑥 0 , the measurement output deviations 𝛥𝑦(𝑡) ∈ ℝ 𝑚 , the control input values 𝑐(𝑡) ∈ ℝ 𝑚 that represent the set points for state regulation during LPG purification, and 𝑧(𝑡) ∈ ℝ 𝑛 denotes the vector of measurement signals including disturbances, measurement noise, and unknown control variables. The system is assumed to be defined over domains 𝒱 ⊆ ℝ 𝑚 × ℝ 𝑛 and 𝒲 ⊆ ℝ 𝑛 for which [14]:…”