Deep Learning–Driven Anomaly Detection for IoT-Enabled Smart Engineering Systems

Abstract

The rapid adoption of Internet of Things (IoT) technologies in smart engineering systems has increased the need for reliable anomaly detection mechanisms capable of identifying cyberattacks, operational faults, and abnormal system behaviors in complex cyber–physical environments. Existing rule-based and conventional machine learning approaches often struggle to effectively model the non-linear, high-dimensional, and highly imbalanced nature of IoT-generated multivariate time-series data, thereby limiting their capability to detect subtle and previously unseen anomalies. To address these challenges, this study proposes a deep learning–driven anomaly detection framework based on a hybrid CNN–LSTM autoencoder architecture for modeling spatiotemporal system behavior in IoT-enabled engineering environments. The proposed framework integrates convolutional neural networks for spatial feature extraction with long short-term memory networks for temporal dependency learning, while anomaly detection is performed using reconstruction error analysis and adaptive thresholding under unsupervised learning conditions. Experimental evaluation was conducted using the BATADAL-A dataset, which represents a realistic cyber–physical water distribution system. The results demonstrate stable convergence and strong generalization performance, with closely aligned training and validation losses throughout the learning process. The proposed framework achieved 90% overall accuracy, anomaly precision of 0.83, anomaly recall of 0.22, and an AUC of 0.677, indicating effective modeling of normal operational behavior but limited sensitivity to rare anomalous events. These findings demonstrate that the proposed CNN–LSTM autoencoder provides reliable low–false alarm monitoring for IoT-enabled smart engineering systems while highlighting the need for future improvements to enhance anomaly sensitivity and robustness in safety-critical applications.