With the rapid development of Industry 4.0 technologies, the application of artificial intelligence is a more critical task than ever to inject intelligence into the smart manufacturing practice. To satisfy the demand of modern production, developing a real-time automated decision-making system with control and optimization is essential. Due to the advantage of reinforcement learning, it becomes a practical tool for finding complicated control policies as well as for solving optimization problems for production system. In this paper, we concentrate on the control and optimization of production system operations to improve the efficiency and productivity of the manufacturing process. Specifically, we consider serial production lines with finite buffers and machines characterized by the Bernoulli reliability model. Additionally, we assume that multiple shared workforce units can be allocated dynamically during production via a real-time production bottleneck identification and mitigation mechanism in order to increase throughput. Firstly, we discuss the solution to this problem in systems with three and four machines, where two shared workforce units are being assigned during the production. The optimal control policy is derived byQ-learningalgorithm. Then, the study is extended to longer lines with multiple shared workforce units. An aggregation-based hierarchical decomposition approach is proposed to search for the optimized control policy through decomposing the serial line and then construct control policies from bottom up. Numerical experiments are carried out to demonstrate the performance of the proposed approach.

Magnetic resonance imaging (MRI) is a powerful noninvasive diagnostic imaging tool that provides unparalleled soft tissue contrast and anatomical detail. Noise contamination, especially in accelerated and/or low-field acquisitions, can significantly degrade image quality and diagnostic accuracy. Supervised learning based denoising approaches have achieved impressive performance but require high signal-to-noise ratio (SNR) labels, which are often unavailable. Self-supervised learning holds promise to address the label scarcity issue, but existing self-supervised denoising methods tend to oversmooth fine spatial features and often yield inferior performance than supervised methods. We introduce Corruption2Self (C2S), a novel score-based self-supervised framework for MRI denoising. At the core of C2S is a generalized denoising score matching (GDSM) loss, which extends denoising score matching to work directly with noisy observations by modeling the conditional expectation of higher-SNR images given further corrupted observations. This allows the model to effectively learn denoising across multiple noise levels directly from noisy data. Additionally, we incorporate a reparameterization of noise levels to stabilize training and enhance convergence, and introduce a detail refinement extension to balance noise reduction with the preservation of fine spatial features. Moreover, C2S can be extended to multi-contrast denoising by leveraging complementary information across different MRI contrasts. We demonstrate that our method achieves state-of-the-art performance among self-supervised methods and competitive results compared to supervised counterparts across varying noise conditions and MRI contrasts on the M4Raw and fastMRI dataset. The project website is available at: https://jiachentu.github.io/Corruption2Self-Self-Supervised-Denoising/.

In production systems, the bottleneck is typically understood as the operation that affects the system performance in the strongest manner. Identification and mitigation of bottleneck are among the most important problems in production systems research and practice. The current literature offers numerous methods for the identification of production bottlenecks. Most of these methods focus on steady-state system behavior under constant parameters, and the corresponding mitigation approach is typically to prioritize the improvement of that particular operation. This applies when the improvement cannot be dynamically reallocated to other operations in the system, and the improvement is to be fixed to an operation for a long period of time. In this article, we attempt to study the bottleneck problem in a dynamic environment. Especially, we consider serial production lines with finite capacity buffers and the Bernoulli reliability machines and use a control-theoretic approach to formulate the bottleneck identification and mitigation problem as a state-based feedback control problem with the objective being the maximization of steady-state throughput. Computation formulas and procedures are developed to calculate the performance metrics of such systems with dynamic bottleneck control. Properties of the optimal control policies are investigated, and a computationally efficient algorithm is developed to obtain an effective real-time bottleneck indicator. Numerical experiments are used to demonstrate the efficacy of the proposed method, and a case study is presented to illustrate its application.Note to Practitioners—It is very common in manufacturing practice that operators, engineers, and managers need to make decisions on-the-fly to ensure efficient production operations. One of the key tasks involved is to identify real-time bottleneck (RTBN). This is typically carried out based on the practitioners’ intuition and experience. With the advances in sensing, computing, and communication technologies emerging in the Industry 4.0 era, manufacturers now have the capability to collect, process, and visualize an enormous amount of data in real time. This provides a foundation for more rigorous real-time production management and control. The research described in this article is intended to provide such a tool for RTBN identification in production lines and, thus, guide timely decision-making on the factory floor for smarter production activities.