Quantum computing is increasingly explored in artificial intelligence and anomaly detection due to its potential to model complex data structures using high-dimensional feature embeddings. One of the most promising directions in this field involves quantum kernel methods, which allow for the computation of similarities between data samples in the Hilbert space of a quantum system. These kernels can capture nonlinear relationships and hidden patterns that are often inaccessible to traditional classical techniques, making them particularly useful in scenarios where subtle correlations are critical. In this work, we propose a hybrid quantum-classical approach for anomaly detection in industrial control systems. The method combines classical autoencoder for dimensionality reduction with parameterized quantum circuits (PQCs), which are used to compute quantum kernel-based similarity measures. To optimize the structure of the quantum kernel, the Evolutionary Variational Quantum Algorithm (EVOVAQ) is employed, allowing task-specific tuning to maximize performance in terms of classification or detection metrics. The proposed approach is applied to two different datasets: a simulated Hydrogen Transport Network (HTN), which reflects physical sensor data in the gas transport framework and leakage monitoring, and the Secure Water Treatment (SWaT) scenario, which simulates the behavior of a cyber-physical water treatment system. Both datasets present real-world challenges such as noisy measurements, correlated sensor data, and complex temporal dynamics. By embedding data into quantum Hilbert spaces, our method aims to improve the sensitivity to anomalous patterns that may resemble normal operational fluctuations, offering a novel strategy for enhancing the robustness of anomaly detection in critical infrastructure systems.