Efficient Task Management Mechanism Based on Learning to Scheduling in Smart Factory

Abstract

Smart factory also known as smart manufacturing is an emerging field with the revolution of industry 4.0. The smart factory concept is an integration of internet of things technologies, computing platforms, cyber-physical systems, control mechanisms, data modeling and simulations, optimization techniques and predictive engineering. With the help of all these concepts, the smart factory integrates the manufacturing assets and represents industrial networks. The aim of smart factory industrial networks is mass customization, on-demand supply chain management, optimal and flaexible processing solutions, and parallel processing. Smart factory faces many limitations in the current age and is need of research solutions for issues such as environmental hazards, energy consumption, productivity, efficient planning, task management, job scheduling, machine utilization, reliable infrastructure and integrated solutions. In this thesis, we put our efforts to find integrated solutions for smart factory concerns by proposing an efficient task management mechanism based on learning to scheduling in smart factory. The scope of the proposal is to efficiently plan tasks execution, maximize machines' resource utilization, maximize productivity, minimize production delays, efficiently handle exceptions and efficiently control smart factory actuators. The proposed learning to scheduling mechanism focuses on both machine structure and tasks modeling for efficient scheduling. We design and develop an integrated solution of learning to scheduling based on sub-modules of prediction and learning for prediction mechanism, optimization and learning for optimization mechanism and inference engine based control mechanism in this thesis work. The scheduling algorithm used for the efficient task management is hybrid of the two scheduling approaches as agent cooperation mechanism (ACM) and fair emergency first (FEF) scheduling scheme. ACM is a decentralized scheduling approach which focuses on the production maximization goals per machine and also centers the production goals of all the machine networks involved in the smart factory. FEF scheduling scheme focuses on minimizing the tasks starvation rate and maximizing the machine utilization by efficiently using the machine slots. In FEF scheduling scheme, two predictive learning based factors are used to improve the scheduling performance; UM (Urgency Measure) and FM (Failure Measure). Both UM and FM use ANN prediction algorithm to learn from scheduler's history decisions and put the learnings in context to wisely use the free machine slots; aiming to increase machine utilization without risking timely execution of any high priority task. The learning to prediction mechanism takes scheduler history data as input and predicts the future tasks completion status and machine utilization rate under varying tasks' loads. The prediction algorithm used is artificial neural network (ANN) and learning algorithm used is particle swarm optimization (PSO). The learning algorithm of PSO tunes the ANN's weights during training iterations to optimize the ANN weights and maximize the prediction accuracy. The learning to optimization mechanism aims to maximize the machine utilization for machines involved in the smart factory, in order to efficiently use the machine resources. The optimization algorithm used is PSO and the learning algorithm used is ANN. First, PSO history is built to train the ANN algorithm and then based on ANN training the PSO particle's velocities are tuned in order to enhance the optimization results. The control mechanism for smart factory actuators is based on the inference engine. The inference engine is fed with rule base which contains list of rules for existing actuators based on incoming sensing and system values. The inference engine matches the rules and generates the control tasks. The control tasks are sent to the scheduler to be executed; and on execution of control tasks, the control commands are sent to the actuators via control unit. The proposed task management mechanism is evaluated based on multiple scenario simulations and performance analysis. The comparisons analysis shows that proposed task management system, referred as learned predictive and optimized hybrid scheduling scheme, significantly improves the machine utilization rate and drastically drops the tasks instances missing rate and tasks starvation rate. Overall, we observe that the learned predictive FEF scheduling in comparison to basic FEF scheduling scheme shows an average of 72.23% reduction in tasks starvation rate and an average of 54.17% reduction in tasks instances missing rate reduction. Also, the learned predictive and optimized hybrid scheduling scheme demonstrates an average of 27.28% increase in machine utilization, and an average of 36.38% improvement in response times.