Cyber-Physical System Security in Industry 4.0 Manufacturing Using Parallel LSTM-CNN Based Orchestration Framework

Abstract

The rapid evolution of Industry 4.0 has enabled smart factories where
interconnected machines, sensing devices, and automation platforms collaborate through cyberphysical
systems (CPS). While this connectivity enhances productivity and operational
flexibility, it simultaneously increases exposure to sophisticated cyber threats capable of
manipulating control commands, altering operational data streams, or damaging high-precision
industrial equipment. To address these vulnerabilities, this proposal introduces a novel CPS
security framework that integrates a Long Short-Term Memory–Convolutional Neural Network
(LSTM-CNN) hybrid model with a Parallel Orchestration (PLO) mechanism. The LSTM-CNN
architecture combines the strengths of both models: the CNN component automatically extracts
spatial and local feature patterns from structured command data and sensor matrices, while the
LSTM component captures long-term temporal dependencies from sequential machine-state
data. This hybrid design enables comprehensive learning of both spatial correlations and
temporal dynamics, allowing early detection of anomalous behavior triggered by malicious
inputs or unauthorized control modifications. The Parallel Orchestration (PLO) mechanism
coordinates multiple detection streams concurrently, ensuring that textual command sequences,
network traffic logs, and machine-state time-series signals are analysed in parallel.