Fault Injection Framework for Organic Computing Architecture

In the future, autonomous systems such as self-driving cars must robustly and flexibly handle various fault situations. Static models of faults and countermeasures are standard in classical approaches; however, such static models are no longer efficient as the complexity of fault scenarios tremendously increased. The bio-inspired concept of organic computing applies biological concepts to technical systems. Organic computing utilizes an artificial hormone system as a decentralized mechanism, i.e., a middleware that continuously monitors and organizes task allocations to computing nodes in distributed real-time embedded systems. By introducing different types of artificial hormones for the tasks, task allocations are realized by constantly establishing hormone balances via distributed closed control loops. This process handles the increasing complexity of e.g., distributed control systems by enabling self-configuration, -adaptation, -improvement and -healing. Such an organic computing environment inherently overcomes system-level faults, such as computation node failures, since missing (i.e., non-executed) tasks directly lead to hormone imbalances that are compensated for, thereby restoring these tasks. However, faults in the artificial hormone system, e.g., due to incorrect hormone values, are currently not covered by the organic computing environment and can result in adverse and critical system behavior and even complete system failure. To address these types of faults and thus improve the capabilities and safety of current organic computing systems, this paper presents a fault injection framework to analyze the effects of an extended range of fault cases, including faults in the artificial hormone system. Such analyses are important as they mark the foundation for future fault-handling strategies and safety features in next-generation organic computing systems for autonomous systems. The statistical analyses and results based on the fault injection framework reveal the most safety-related fault cases for which the paper outlines respective fault handling strategies.