Quantum-Inspired Neural Networks For Real-Time Intrusion Detection In Smart Manufacturing Systems

Abstract

With the continuous development of Smart Manufacturing Systems (SMS) as a result of extensive use of the Industrial Internet of Things (IIoT), the cybersecurity threats are on the rise. Conventional Intrusion Detection Systems (IDS) founded on classical Deep Learning (DL) usually face a major performance bottleneck, namely, high detection latency and high computational overhead, which do not fit the strict real-time demands of the current industrial production lines. To overcome these difficulties, the current paper suggests a new Quantum-Inspired Neural Network (QiNN) architecture that is designed to be used in real-time intrusion detection in the SMS context. The QiNN model, by applying quantum mechanical concepts (superposition and interference) to a classical computational architecture, improves the recognition of patterns without increasing the architecture size. The given framework was strictly tested on the state-of-the-art X-IIoTID dataset that is the industrial traffic of the modern device-independent nature. According to experimental findings, QiNN model has a better detection accuracy of 99.2% and F1-score of 0.98, which is considerably higher than the traditional CNN and LSTM models. Importantly, the model had an impressive latency of 1.2 ms in detection and also reduced trainable parameters by 70 percent relative to classical deep learning models. These results indicate that quantum-inspired computing represents a viable and effective way to secure next-generation industrial infrastructure against advanced and polymorphic cyber-attacks and has the potential to bring a scalable solution to edge-computing devices resource-constrained by their location on the factory floor.

Keywords: Quantum-Inspired Machine Learning, Industry 4.0, Smart Manufacturing, Intrusion Detection System (IDS), Industrial IoT (IIoT), Cybersecurity, Real-Time Analytics, Neural Networks, Edge Computing, X-IIoTID Dataset