Industrial Internet of Things (IIoT) technologies have emerged with significant security challenges due to increasing interconnectivity and network complexity. Thus, this study proposes and tests a centralized deep-learning intrusion detection system (IDS) particularly designed for IIoT networks. Convolutional Neural Networks (CNNs) and Multilayer Perceptrons (MLPs) were implemented in PyTorch and TensorFlow, and their performance was assessed on the CIC IoT-DIAD 2024 dataset to simulate real industrial traffic and varied attack vectors. Accuracy, precision, recall, F1-score, and confusion matrices were used for the evaluation metrics. Results showed that the TensorFlow CNN achieved the highest detection rate (80.1%), followed very closely by the PyTorch CNN (77.3%), highlighting the superior performance of CNN architectures, especially on TensorFlow, in recognizing intricate spatial patterns in IIoT traffic. The comparative research enhances IIoT cybersecurity research by identifying efficient deep learning models for intrusion detection and proposing a framework for adoption in industrial systems to improve resilience and mitigate vulnerability to advanced cyberattacks.