Scalable Deep Learning for Industry 4.0: Speedup with Distributed Deep Learning and Environmental Sustainability Considerations

This paper presents a solution for leveraging High-Performance Computing (HPC) infrastructures, and investigates the integration of distributed deep learning (DDL) techniques to address Industry 4.0 challenges across three distinct applications: intrusion detection with multi-layer perceptron, defect identification with convolutional neural networks, and predictive maintenance with recurrent neural network. Experimental results, underscore the scalability and efficiency of the proposed DDL approach. Notably, computations are accelerated by up to 46 times. In addition to performance metrics, this research places significant emphasis on environmental sustainability. Detailed examination of energy consumption patterns on the HPC infrastructure aims to minimize the carbon footprint associated with deep learning processes. This dual focus on efficiency and sustainability positions the approach as a holistic and responsible solution for Industry 4.0 applications. The practical insights enhance the efficiency of deploying DDL in HPC infrastructure. Additionally, they highlight the significance of eco-friendly AI practices for ethical and environmentally sustainable technological progress.