Reinforcement learning methods based on GPU accelerated industrial control hardware

Abstract

Reinforcement learning is a promising approach for manufacturing processes. Process knowledge can be gained automatically, and autonomous tuning of control is possible. However, the use of reinforcement learning in a production environment imposes specific requirements that must be met for a successful application. This article defines those requirements and evaluates three reinforcement learning methods to explore their applicability. The results show that convolutional neural networks are computationally heavy and violate the real-time execution requirements. A new architecture is presented and validated that allows using GPU-based hardware acceleration while meeting the real-time execution requirements.

New Progress in Artificial Intelligence Algorithm Research Based on Big Data Processing of IOT Systems on Intelligent Production Lines

“…It not only makes the way of communication between people leap forward but also makes it possible to communicate between people and things and between things and things. In a word, the Internet of Things technology has turned the whole world into a whole [ 18 , 19 ]. Networking, IOT, interconnection, automation, perception, and intelligence are the basic characteristics of the Internet of Things.…”

Real-time high-performance computing for embedded control systems

“…Other authors address the compliance of GPUs regarding functional safety certification [47][48] [49] by proposing the use of language subsets or the adaptation of safety standards. Regarding industrial applications, some works analyze the challenges of using GPUs in embedded systems [50][51] [52], while others analyze the exploitation of GPUs parallelism when executing common control workloads [53][54] or advanced control techniques like predictive control [55] and reinforcement learning-based control [56].…”