Key technologies towards smart manufacturing based on swarm intelligence and edge computing

Abstract

In recent years, the manufacturing industry has faced various global challenges. One of these challenges is the increasing frequent adjustment and reconfiguration of production lines, necessitated by the diversification of customer demands. To improve the processing efficiency after production line reconfiguration, this paper puts forward a group learning architecture towards intelligent equipment. With this architecture, swarm intelligence can be accomplished through group learning. From the perspective of edge intelligence, this paper addresses key technology issues within the areas of data acquisition and preprocessing, cyber-physical fusion, knowledge extraction and sharing, and equipment performance self-optimization. The proposed approaches are much more useful for improving the processing efficiency of the reconfigured production line.

Introduction

Due to the ever increasing interconnectivity in the digital age, the number of terminal devices and the data they generate has grown dramatically [1]. According to a study on the Internet of Things (IoT), released by Transforma Insights, the number of activated IoT devices in the world reached 7.6 billion by the end of 2019. This number is expected to further grow to 24.1 billion in a decade [2]. Facing this huge number of terminal devices, a data processing mode that depends heavily on cloud computing cannot meet suitable efficiency needs, as it cannot guarantee real-time computing performance, requires large resources and cannot ensure data privacy in the process of data transmission [3].

To tackle these problems, the concept of edge computing has been put forward for processing data within the large multiplicity of terminals [4, 5]. Condry et al. [6] proposed a security model that combines terminal devices and control system gateways, to enhance the security of the system through multi-factor authentication via encrypted communication channels. Wan et al. [7] proposed a scheme for optimizing the edge computing network – by allowing the interfaces of the network management devices to be defined by the software. Then, through constrained grouping of requirements, Li et al. [8] used a variety of schemes to set up the transmission path, and realized the optimization of transmission delay and throughput. Further, to solve the problem caused by the rigidity of traditional IoT systems, Suganuma et al. [9] designed a new batch system model that has good environmental adaptability and is highly responsive to customer demands, which can be used in multi-agent edge computing architectures. Byers et al. [10] summarized the requirements of IoT network architecture by analyzing cases of edge computing. As the fusion of emerging information technologies (such as IoT [11], [12], [13], big data [14,15], cloud computing [16,17], artificial intelligence (AI) [18], and 5 G [19], [20], [21]) and manufacturing processes deepens, a manufacturing system based on cyber-physical fusion [22,23] can help in enabling the rapid development of manufacturing processes. Thus, to better meet the requirements of customized/personal production, advances in theory and methods for optimizing the machine group operation based on edge computing can significantly improve the level of intelligent collaboration between machines within the production line.

This increased collaboration can reduce performance requirements on the individual machines, enabling them to perform more complex processes [24]. Moreover, the reconfiguration of the intelligent machine group can expand the function of the entire production system, endowing the production system with higher flexibility and adaptability. Therefore, equipping intelligent machines with swarm intelligence not only enables them to perform complex manufacturing tasks [25] but also mitigates the deficiencies of the individual equipment. At present, there exist enormous caveats and challenges in the applicability of swarm intelligence in intelligent production machine groups. These are summarized as follows:

• The software/hardware inconsistency between different machines implies that a heterogeneous network needs to be constructed to facilitate information exchange between the machines, which involves multi-protocol, multi-language, and cross-platform data fusion.

• The machine group needs to acquire through group learning the ability to recognize the changes in the manufacturing environment, and to adapt to new types of task; so that tunable experience parameters can be reutilized and experience knowledge is shared among the machines.

• The machine group needs to ensure that the production tasks can be executed according to the functional requirements of the manufacturing system and the external disturbances. It also should be robust, extensible, and adaptive.

For manufacturing environments where a group of intelligent manufacturing machines collaborate, the following research objectives can be formulated to address the problems related to real-time sensing; multi-protocol cross-platform dynamic interaction; experience parameter reuse and knowledge sharing; scalability, adaptability, and robustness of the manufacturing machine group:

• To establish the cyber-physical fusion system architecture based on edge computing, a cross-platform information interaction mechanism for intelligent machines with heterogeneous software /hardware should be established. Further, it should be designed a data fusion method that allows for state-behavior spatiotemporal sensing – within the industrial heterogeneous network environment.

• A formal semantic model for complex interactive behavior of machines should be designed. Also, it is required a cloud-assisted knowledge base for group learning, using the information extracted from the environmental sensor data, task planning data, and decision-making data. So that behavior learning and experience sharing among the manufacturing machines is possible.

• Lastly, a task-oriented equipment self-optimization model based on group learning should be established. This self-optimization can be designed using AI through self-learning / transfer learning, and can endow scalability, adaptability, and increased robustness to the manufacturing machine group.