Manufacturing Big Data Analysis And Intelligent Decision-making In Discrete Manufacturing Workshop

In the intelligent manufacturing era,manufacturing big data(MBD),as the core resource of intelligent manufacture,can be regarded as the key element that empower the manufacture with "intelligence".With the development of information technology,manufacturing Internet of things(Io T)equipment have been largely deployed in the discrete manufacturing workshop,which realize the active perception of manufacturing objects status and comprehensive acquisition of manufacturing data.How to excavate the inherent evolution of the manufacturing system,analyze the complex characteristics of the system’s internal operation,track the production bottleneck,and then perform decisions to modify the production process and improve the vibration and instability during production runs,are the goals need to be reached by utilizing MBD technologies in the discrete manufacturing workshop.Therefore,research on big data analysis and intelligent decision-making is of great significance to improve the capabilities of precise production management and control in the discrete workshop.This research takes the production management and control as the target.Depending on the analysis of the application scenario,a MBD driven analysis and intelligent decision-making architecture for the control of discrete manufacturing workshop is constructed.In this architecture,the integration and business logic of MBD are formulated,and three key technologies are proposed for online production progress prediction,production bottleneck shifting discovery,and intelligent production decision-making.More specifically,the main research contents include the following five aspects:(1)According to the generation and accumulation of MBD in the discrete manufacturing workshop,the concepts,characteristics,and sources of MBD are firstly illustrated in this section.Then,in order to fulfill the needs of MBD analysis and decision-making,the methods regarding spatiotemporal data modeling and ontology-based data integration are proposed for MBD management.On this basis,the main flow of this research is illustrated as “data acquisition—integration modeling—data analytics—application services”,and a MBD driven analysis and intelligent decision-making architecture is constructed accordingly.Furthermore,three relevant enabling technologies are specially proposed for production management and control in the proposed architecture.(2)According to the order progress prediction,the qualitative factors including order composition,real-time WIPs status,and real-time workstation status are firstly analyzed to cover the influencing factors as much as possible.On this basis,a formal descriptive model for order progress prediction is proposed.Hereby,the MBD collected from historical production runs are collected as the training samples for predicting model.In order to quantify the continuous impact of changes in the production status and tasks for production progress,a Stacked sparse autoencoder(S-SAE)model is designed to make online prediction of production progress.Through the feature extraction and dimensionality reduction capabilities of the autoencoder,data representative learning is performed on the high-dimensional manufacturing data.Based on the pre-training of autoencoder,the hidden layers from multiple autoencoders are stacked to form the S-SAE structure.Therefore,the precise prediction for order progress is achieved based on the generalized capability of S-SAE,which can be used to determine whether the future production schedule can meet the requirements of the production plan.Consequently,the online prediction of order progress provides a prerequisite judgment for intelligent decision-making during the production runs.(3)According to the predicted production schedule cannot meet the requirements of the original production plan,it is essential to find the bottleneck manufacturing units in the current system that hinder the original production schedule.Aiming at the production bottleneck and bottleneck shifting problem,the causes of bottleneck phenomenon in the production process are analyzed firstly,and on this basis,the concepts of production bottleneck and bottleneck shifting problems are further defined.Based on the quantification of the bottleneck degree,it is proposed to use the main data and auxiliary data to respectively describe the bottleneck degree and the real-time production status.In order to find the shifting trends of the production bottleneck during production runs,a Parallel gated recurrent units(P-GRUs)model is designed to extract the time series characteristics from the main inputs and auxiliary inputs.By setting the different time lags,the P-GRUs model is able to discover the short-term,mid-term and long-term trends of bottleneck shifting,which can be used for precisely locate and track the occurrence of bottlenecks in future production runs.(4)Considering the predicted order progress and the production bottleneck,a multi agents reinforcement learning mechanism is proposed for decision-making in the discrete manufacturing workshop,which can percept the changing production status and modify the production schedule dynamically.Firstly,the disturbances factors which resulting in the bottleneck generation are detected by complex event processing of the Io T collected data.Moreover,the decision triggering condition is dynamically estimated by calculating the deviations of production progress and original production schedule.Furthermore,the holographic virtual shop floor environment is developed to provide the learning environment for reinforcement learning agents,and in the virtual environment,the agents interact with the environment by precepting the environment status and making adaptive decisions.Furthermore,the minimization of the maximum completion time is set as the reward function for agent and the experience sharing mechanism is used to realize the collaborative decision-making among agents.Finally,the well-trained agents are applied in the actual production environment to realize the autonomous decision-making based on the real-time production status of the workshop,and intelligent optimization of the uncertain disturbance factors in the discrete manufacturing workshop.(5)Based on the deployed Io T devices in the shop floor,the collected MBD are stored in a distributed manner by Mongo DB.Furthermore,according to the methods proposed in this research,a MBD driven analysis and intelligent decision-making platform is developed to apply the MBD technologies for production management and control in the discrete manufacturing workshop.Finally,the prototype system is deployed in an aeroengine parts machinery factory to verify the effectiveness of modules including data collection,data management,production status visualization,online prediction of production progress,discovery of production bottleneck,and intelligent decisions-making,etc.