| Industrial Internet of Things is the fundamental of smart manufacturing and an important support condition to realize the transformation of manufacturing industry into digitalization,intelligence and networking.With the rapid development of IIoT,the number of sensors and terminal devices deployed at production sites is increasing,and the computing tasks generated by them are also increasing,which makes it difficult for the computing resources in the network to meet the delay requirements for task completion.In addition,the conventional cloud computing model is also unable to meet the requirements due to the long communication latency over long transmission distances.To solve this problem,edge computing is introduced to the IIoT system to sink the computing resources of the network and provide offloading services for the nearby computing of devices in the system.However,the communication and computation capacities of edge servers are limited,and how to design an effective task offloading and communication resource allocation strategy to improve the resource utilization of edge servers in order to reduce the task completion delay and improve the productivity has become one of the hot research direction in the industry.To address the above problems,this thesis develops the theoretical research on latency optimization for two scales of edge computing-enabled IIoT at the workshop level and plant level according to the properties of IIoT systems at different scales,as follows:1)A task offloading strategy based on improved deep reinforcement learning is proposed for the workshop-level task offloading.This study takes the signal processing task as an example,and splits this task into two subtasks with timing dependencies so as to achieve parallel processing of the tasks.Based on this,the system state transfer is modeled as a Markov decision process,and the system state is precisely described from the perspective of the amount of data to be processed,and the optimization goal of minimizing the total delay in completing tasks in the system is equivalently translated into minimizing the amount of data to be processed in the system.To solve this problem,a task offloading strategy based on deep reinforcement learning is proposed,and the network structure is improved to enhance the convergence of the algorithm.Simulation experiments demonstrate that the proposed strategy can significantly reduce the task completion delay.2)An expert system-based optimal delay search strategy is proposed for the plant-level joint task offloading and resource allocation.In this study,the impact of task offloading and communication resource allocation on task completion delay is considered,and the goal is to minimize the completion delay of the last finished task in the system.Firstly,the decision space of task offloading is reduced by using expert knowledge,and then the optimal task offloading decision and the corresponding optimal communication resource allocation decision are found by heuristic search.Simulation experiments demonstrate that the proposed strategy can effectively improve the average rate of task execution and reduce the completion time of the last finished task in the system.Based on the above theoretical research,A tool fault diagnosis system for machining industry is designed in the thesis.3D vibration signal collectors are used as the terminal devices and several Raspberry Pis and an IPC to conpose a two-layer edge computing platform to realize real-time acquisition,filtering and analysis of machine tool vibration signals.The system is settled in the field to be tested.The actual test results prove that the platform can run smoothly in the field,and the execution rate of tasks can meet the requirements in actual production. |
