A Cyber-Physical Architecture for Physics-Based Behavioural Fingerprinting in Industrial Uptime Verification

Abstract

Industrial systems increasingly rely on software-generated operational logs for performance auditing, contractual billing, and regulatory compliance. However, such logs may not reliably represent actual machine activity due to misconfiguration, replay attacks, or deliberate falsification. This creates a fundamental gap between reported and physically executed operations. This study presents a cyber-physical software architecture that enables independent verification of machine uptime using physics-based behavioural fingerprinting. The proposed architecture integrates non-intrusive sensing, signal processing, behavioural modelling, and a verification engine to compare reported operational logs with independently observed physical activity derived from vibration, acoustic, and power signals. A proof-of-concept implementation was developed using a controlled experimental setup based on an industrial electric motor instrumented with external sensors. Behavioural features were extracted and classified using a supervised learning approach to reconstruct observed machine states. The verification engine quantified discrepancies between claimed and observed uptime using a fraud scoring formulation. Evaluation across representative scenarios, including legitimate operation, idle fraud, partial over-reporting, and environmental disturbance, demonstrated strong discrimination capability. The framework achieved classification accuracy in the range of 90% to 97% across scenarios, with precision and recall values consistently above 0.89. The model demonstrated a receiver operating characteristic area under curve value of 0.96, with performance variability across repeated runs remaining low (±1.8%). The system operates in near real-time, with processing latency of approximately 40–60 ms per signal window. These results establish the feasibility of physics-based behavioural fingerprinting as an independent verification mechanism for industrial performance auditing. The proposed architecture provides a foundation for scalable deployment, with future work focusing on large-scale validation across heterogeneous industrial systems, adaptive modelling strategies, and robustness against adversarial signal manipulation.