A Hierarchical Clustering Federated Learning System Based on Industry 4.0

Introduction

Material requirement planning (MRP) emerged when people began using information computations to assist manufacturing and production operations. M.R.P. is a decision-making model for cost management (i.e., raw material purchase schedules and quantities) after manufacturers receive purchase orders. As computer performance evolves over time, MRP is gradually incorporated into the master production schedule (MPS.), rough-cut capacity planning (RCCP), and capacity requirement planning (CRP). It is combined with raw materials, human resources, equipment, and costs to form a manufacturing resource plan (MRP-II). Because material requirement planning and manufacturing resource planning share the same acronym, they are referred to as “MRP I” and “MRP II,” respectively. Over time, enterprises began to integrate MRP II into accounting, finance, human resources, and other MRP II-related systems to develop contemporary enterprise resource planning (ERP). MRP II is associated with the effective use of resources in operational job scheduling on shop floors to decrease production costs and increase production capacity. Namely, it signifies that production lines on shop floors can comprise different workers and production equipment working together to manufacture products. Over the past few decades, many studies regarding job sequencing theories and other applications have been conducted. (R. Ruiz &T. Stützle.,2007; Luo et al.,2013; Zhao et al.,2021). Commonly seen scheduling models include job sequencing, flow shop, job shop, mixed shop (D-Shop), and open shop (Noor, Sahar, et al., 2009).

The Internet of Things (IoT) has emerged following the confluence of Internet ubiquity and the minimization of the chip manufacturing process. The application of IoT in shop floors enables manufacturing equipment to display operating equipment-related information to on-site personnel in real-time or transmit related parameters to backend servers to perform related artificial intelligence (A.I.) computations that allow the optimal production parameters of various items to be obtained. On a shop floor, each production line is composed of a variety of machines and personnel. Before switching to different machines to produce different things, the machines may need to be turned off to allow parameters to be adjusted before restarting. Thus, parameters may need to adjust in response to various scheduling patterns, necessitating different settings. Industry 4.0 refers to transmitting production equipment parameters to backend computation systems to obtain optimal machine operation parameters under certain scheduling constraints. Many Industry 4.0 devices rely on A.I. models to conduct intelligent automated processes. Servers must constantly train the A.I. models to enhance identification accuracy and improve production yield. The concept of federated learning (F.L.) has been developed to help devices calculate relevant parameters on the client side before encrypting and sending the data to the server. After collecting the uploaded data, the server generates an optimal solution through gradient descent. Finally, the server sends the optimal solution back to the client side for updates, improving the A.I. model accuracy. The FL technique only needs to collect the parameters of the device without needing to collect client-side information, thus avoiding privacy concerns.