Large-scale Manufacturing Systems Research Articles

A sensor-driven operations and maintenance planning approach for large-scale leased manufacturing systems

The rise of mass customisation drives manufacturers to adopt leasing agreements for their machinery rather than outright their ownership. This industrial trend results in large-scale leased manufacturing systems, where the asset conditions and maintenance requirements for a fleet of machines and/or production lines from several manufacturers are continuously monitored and managed by a lessor. In this paper, we propose an operations and maintenance planning model that explicitly models (i) dynamic real-time failure rate predictions for machines based on sensor-driven degradation data, (ii) optimal routes for maintenance teams across a number of geographically-distributed manufacturing sites, and (iii) operational outcomes. The model can handle complex maintenance and operational outcome interdependencies between multiple machines, and incorporate demand for different products, heterogeneous machine capacities and capabilities, factory topology, and the resulting production flow capacities of products. To demonstrate the model’s practicality, it is applied to three experimental case studies with various numbers of machines, operating schedules, product types, and flow capacities. The results show that the proposed model significantly outperforms the traditional reliability-based maintenance model in key metrics such as the number of unexpected failures, the cost of preventive maintenance actions, machine downtime, the percentage of unsatisfied demand, and total cost.

Hyperparameter Tuning with Gaussian Processes for Optimal Abstraction Control in Simulation-based Optimization of Smart Semiconductor Manufacturing Systems

Smart manufacturing utilizes digital twins that are virtual forms of their production plants for analyzing and optimizing decisions. Digital twins have been mainly developed as discrete-event models (DEMs) to represent the detailed and stochastic dynamics of productions in the plants. The optimum decision is achieved after simulating the DEM-based digital twins under various what-if decision candidates; thus, simulation acceleration is crucial for rapid optimum determination for given problems. For the acceleration of discrete-event simulations, adaptive abstraction-level conversion approaches have been previously proposed to switch active models of each machine group between a set of DEM components and a corresponding lookup table-based mean-delay model during runtime. The switching is decided by detecting the machine group’s convergence into (or divergence from) a steady state. However, there is a tradeoff between speedup and accuracy loss in the adaptive abstraction convertible simulation (AACS), and inaccurate simulation can degrade the quality of the optimum (i.e., the distance between the calculated optimum and the actual optimum). In this paper, we propose a simulation-based optimization (SBO) that optimizes the problem based on a genetic algorithm (GA) while tuning specific hyperparameters (related to the tradeoff control) to maximize the speedup of AACS under a specified accuracy constraint. For each individual, the proposed method distributes the overall computing budget for multiple simulation runs (considering the digital twin’s probabilistic property) into the hyperparameter optimization (HPO) and fitness evaluation. We propose an efficient HPO method that manages multiple Gaussian process models (as speedup-estimation models) to acquire promising optimal hyperparameter candidates (that maximize the simulation speedups) with few attempts. The method also reduces each individual’s exploration overhead (as the population evolves) by estimating each hyperparameter’s expected speedup using previous exploration results of neighboring individuals without actual simulation executions. The proposed method was applied to optimize raw-material releases of a large-scale manufacturing system to prove the concept and evaluate the performance under various situations.

Coordinated scheduling of production and logistics for large-scale closed-loop manufacturing using Benders decomposition optimization

This paper studies coordinated scheduling of production and logistics for a large-scale closed-loop manufacturing system by integrating its manufacturing and recycling process. In addition to the forward manufacturing process, different recycling units in reverse recycling process are also studied. A decentralized network is designed to formulate the coordinated scheduling problem as a mixed integer programming model with both binary and integer variables. As the problem for closed-loop manufacturing is large-scale and computational-consuming in nature, the model is divided into integer variable sub-models and complex binary variable sub-models for preprocessing and reprocessing respectively. An iterative solution approach by Benders decomposition is developed to accelerate the solving efficiency in large-scale case by updating custom constraints. A case study is conducted to investigate the managerial implications of the decentralized network for the closed-loop manufacturing system. Computational experiments demonstrate the validity and efficiency of the proposed iterative solution approach for the large-scale scenarios.

An Effective 4–Phased Framework for Scheduling Job-Shop Manufacturing Systems Using Weighted NSGA-II

Improving the performance of manufacturing systems is a vital issue in today’s rival market. For this purpose, during the last decade, scientists have considered more than one objective function while scheduling a production line. This paper develops a 4-phased fuzzy framework to identify effective factors, determine their weights on multi-objective functions, and, accordingly, schedule manufacturing systems in a fuzzy environment. The aim is to optimize product completion time and operational and product defect costs in a job-shop-based multi-objective fuzzy scheduling problem. In the first and second phases of the proposed framework, it was shown that the existing uncertainty of the internal factors for the studied cases causes the weights of factors to change up to 44.5%. Then, a fuzzy-weighted NSGA-II is proposed (FW-NSGA-II) to address the developed Non-linear Fuzzy Multi-objective Dual resource-constrained scheduling problem. Comparing the outcomes of the proposed method with other solving algorithms, such as the Sine Cosine Algorithm, Simulated Annealing, Tabu Search, and TLBO heuristic, using seven series of comprehensive computational experiments, indicates the superiority of the proposed framework in scheduling manufacturing systems. The outcomes indicated that using the proposed method for the studied cases saved up to 5% in the objective function for small-scale, 11.2% for medium-scale, and 3.8% for large-scale manufacturing systems. The outcomes of this study can help production planning managers to provide more realistic schedules by considering fuzzy factors in their manufacturing systems. Further investigating the proposed method for dynamic product conditions is another direction for future research.

Realtime Calibration of an Industrial Robot

In large scale, complex and low volume manufacturing systems, robotics are now considered unavoidable for automating the factory operations. The aerospace industry focuses on a high variety and quality but extremely low volume. The precision it requires for numerous tasks is unique and distinct from any other manufacturing industry. This can comprise accurate position, module assembly, inspection, fastening, etc. The scale of the robot invites different types of errors during operation, which can be either because of the kinematics of the robot or because of the environment (noise, temperature, load, etc.). There are packages available from robot manufacturers for the correction and compensation of errors on the robot to achieve accuracy. There are two associated problems: 1. cost and 2. static nature. They are very costly and they do not provide correction in realtime fashion (dynamic); the robot stops, waits for the correction, and then moves to the next position. The external tool to monitor the accuracy also requires attaching with the robot permanently. These are dedicated resources. These tools for accurate measurement are expensive and attached permanently to a robot, which means wastage of resources. These measuring tools are called metrology devices and attaching these devices and the robot to the network means that other robots/machines can also use these expensive tools for measurement. Our aim was to address two problems in this project: 1. calibration (error correction and compensation of robot) and 2. dynamic and realtime processing. It helped to perform the dynamic error correction and the compensation of an industrial robot. The results showed the error correction was achieved in the region of 0.02 mm.

Evaluation of a High-Temperature Pre-Heating System Design for a Large-Scale Additive Manufacturing System

Additive Manufacturing (AM) of titanium (Ti6Al4V) material using Selective Laser Melting (SLM) may generate significant residual stresses of a tensile nature, which can cause premature component failure. The Aeroswift platform is a large volume AM machine where a high-temperature substrate preheating system is used to mitigate high thermal gradients. The current machine platform is unable to achieve a target build-plate temperature of 600 °C. This study focuses on the analysis of the preheating system design to determine the cause of its inefficiency, and the experimental testing of key components such as the heater and insulation materials. A Finite Element Analysis (FEA) model shows the ceramic heater achieves a maximum temperature of 395 °C, while the substrates (build-plates) only attain 374 °C. Analysis showed that having several metal components in contact and inadequate insulation around the heater caused heat loss, resulting in the preheating system’s inefficiency. Additionally, experimental testing shows that the insulation material used was 44% efficient, and a simple insulated test setup was only able to obtain a maximum temperature of 548.8 °C on a 20 mm thick stainless steel 304 plate, which illustrated some of the challenges faced by the current pre-heating design. New design options have been developed and FEA analysis indicates that a reduction in heat loss through improved sub-component configurations can obtain 650 °C degrees above the substrate without changing the heating element power. The development and challenges associated with the large-scale preheating system for AM are discussed, giving an insight into improving its performance.

Weighted Voting in Physical Layer Authentication for Industrial Wireless Edge Networks

Edge computing (EC) is an essential component of large-scale intelligent manufacturing systems for Industry 4.0, which promises to provide a pre-processing platform for the massive data generated by the terminals and guarantee lower delay and more security compared to directly processing data in cloud computing. Nevertheless, access authentication is a crucial security issue of current EC systems, and thus this paper presents a solution to enhance the access authentication accuracy by exploiting the physical layer information. Our method employs a weighted voting scheme for channel state information (CSI) based authentication using a single sample which include sample segmentation, grouping, and weighted voting, and finally achieves the fast and low complexity secure-access requirement of the EC system without increasing the individual devices' sample size and computational complexity. Experimental results utilizing public datasets and field-measured datasets demonstrate that the proposed weighted voting method has higher accuracy and robustness than existing methods.

Ultra‐Fast Click Modification of Self‐Assembled Zwitterionic Copolymer Membranes for Enhanced Ion Selectivity

AbstractMembranes that can separate small molecules and ions are of immense interest in various separations. Self‐assembling zwitterionic amphiphilic copolymers (ZACs) are extremely promising as exceptionally fouling‐resistant membrane selective layers with narrow size cut‐offs. However, it is difficult to tune their effective pore size within time frames compatible with roll‐to‐roll manufacturing. We report a novel, ultra‐fast approach for tuning the pore size of ZAC‐based membranes through thiol‐ene click chemistry that brings the needed post‐modification time from tens of minutes to only 10–40 s, making its integration feasible in large scale manufacturing systems. Resultant membranes have enhanced organic molecule and salt rejections, resulting in tunable mono‐/divalent ion selectivity. For instance, cross‐linking increases Na2SO4 rejections from <20% to up to 83%, while NaCl rejection remains below 30%. Furthermore, these membranes completely resist fouling by oil and proteins, outperforming commercial membranes. These results suggest that our cross‐linked membranes have the potential to be used in separations in the food industry, pharmaceutical manufacturing, and wastewater treatment.

Deep multi-agent reinforcement learning for multi-level preventive maintenance in manufacturing systems

Designing preventive maintenance (PM) policies that ensure smooth and efficient production for large-scale manufacturing systems is non-trivial. Recent model-free reinforcement learning (RL) methods shed lights on how to cope with the non-linearity and stochasticity in such complex systems. However, the action space explosion impedes RL-based PM policies to be generalized to real applications. In order to obtain cost efficient PM policies for a serial production line that has multiple levels of PM actions, a novel multi-agent modeling is adopted to support adaptive learning by modeling each machine as cooperative agent. The evaluation of system-level production loss is leveraged to construct the reward function. An adaptive learning framework based on value-decomposition multi-agent actor–critic algorithm is utilized to obtain PM policies. In simulation study, the proposed framework demonstrates its effectiveness by leading other baselines on a comprehensive set of metrics whereas the centralized RL-based methods struggles to converge to stable policies. Our analysis further demonstrates that our multi-agent reinforcement learning based method learns effective PM policies without any knowledge about the environment and maintenance strategies.

Specification governor for fault tolerant control of large-scale manufacturing systems

In this paper, a novel approach for fault tolerant control of large-scale manufacturing systems is presented. A specification governor is introduced, which is inspired by the concept of ”reference governor” in feedback control systems. The goal is to modify an infeasible sequence of specifications, which can be used to generate a distributed logic controller, in such a way that the desired product requirements are satisfied in case of faults. The procedure considers redundancies in the manufacturing system and has the capability of keeping the modification of the infeasible specification sequence as less as possible. A modeling process is proposed that automatically adapts fault-free models in form of signal interpreted Petri nets (SIPNs) according to the occurring faults. Based on these faulty models, it is possible to find out which of the natural language specifications in the specification sequence cannot be executed.

A loosely-coupled deep reinforcement learning approach for order acceptance decision of mass-individualized printed circuit board manufacturing in industry 4.0

Printed Circuit Board (PCB) manufacturing is a kind of energy-intensive and pollution-intensive industries. With the increment of individualized requirements, PCB manufacturers face massive customized orders with a variety of specifications. The individualized customization on orders results in large differentials of the profit, energy consumption, and environmental pollution. Making energy-efficient order acceptance decisions can reduce carbon consumption and improve material utilization during the whole manufacturing process. An order acceptance decision model is established based on a loosely-coupled integration of deep learning and reinforcement learning techniques. Firstly, different from a simple assumption of the linear cost function in a small-scale manufacturing system, the deep learning algorithm is presented for accurately predicting the production cost, makespan, and carbon consumption of incoming PCB orders in the large-scale manufacturing system. Secondly, these predicted cleaner production indicators are combined with original order features to perform a reinforcement learning-based order acceptance decision. The proposed loosely-coupled deep reinforcement learning approach is verified with a dataset built based on data collected from a PCB manufacturer in China. This research is expected to provide an environment-friendly order acceptance decision-making approach for sustainable manufacturing in the Industry 4.0 context.

Using CP/SMT Solvers for Scheduling and Routing of AGVs

An improved method for solving conflict-free scheduling and routing of automated guided vehicles is proposed in this article, with promising results. This is achieved by reformulating the mathematical model of the problem, including several improvements and speedup strategies of an existing Benders decomposition method. A new heuristic is also presented that quickly yields high-quality solutions. Moreover, a real-large-scale industrial instance is solved using an open-source satisfiability module theories solver and a commercial constraint programming solver. According to the results, both of these general-purpose solvers can effectively solve the proposed models. Note to Practitioners-The problem of conflict-free routing and scheduling of automated guided vehicles (AGVs) in large-scale manufacturing systems has been an ever-present challenge for many AGV companies. Although these companies have developed rather efficient control policies and algorithms, retrofitting the existing heuristic to future's denser, more complicated, and more demanding AGV layouts is not guaranteed to be easy. Furthermore, the installed system will not necessarily be as efficient as expected. Currently, it is common to use heuristics to allocate vehicles to orders and route them. There are also rules of thumbs to avoid collisions and deadlocks. However, with increasing demand for high-performance AGV solutions, it is of interest to employ optimization algorithms that handle the order allocation, scheduling, and routing in a more efficient way. In this article, we present an improved method to tackle this issue, with promising results. We have developed our work in collaboration with a Swedish AGV company, and we have investigated a real-large-scale industrial instance as our case study.

Machine learning for predictive scheduling and resource allocation in large scale manufacturing systems

The digitalization processes in manufacturing enterprises and the integration of increasingly smart shop floor devices and software control systems caused an explosion in the data points available in Manufacturing Execution Systems. The degree in which enterprises can capture value from big data processing and extract useful insights represents a differentiating factor in developing controls that optimize production and protect resources. Machine learning and Big Data technologies have gained increased traction being adopted in some critical areas of planning and control. Cloud manufacturing allows using these technologies in real time, lowering the cost of implementing and deployment. In this context, the paper offers a machine learning approach for reality awareness and optimization in cloud.Specifically, the paper focuses on predictive production planning (operation scheduling, resource allocation) and predictive maintenance. The main contribution of this research consists in developing a hybrid control solution that uses Big Data techniques and machine learning algorithms to process in real time information streams in large scale manufacturing systems, focusing on energy consumptions that are aggregated at various layers. The control architecture is distributed at the edge of the shop floor for data collecting and format transformation, and then centralized at the cloud computing platform for data aggregation, machine learning and intelligent decisions. The information is aggregated in logical streams and consolidated based on relevant metadata; a neural network is trained and used to determine possible anomalies or variations relative to the normal patterns of energy consumption at each layer. This novel approach allows for accurate forecasting of energy consumption patterns during production by using Long Short-term Memory neural networks and deep learning in real time to re-assign resources (for batch cost optimization) and detect anomalies (for robustness) based on predicted energy data.

Unsupervised Learning for Opportunistic Maintenance Optimization in Flexible Manufacturing Systems

Large scale manufacturing systems with a high degree in automation and the ability to produce several product variants in parallel meet current requirements of a highly flexible and at the same time productive manufacturing process. In practice, however, the non-transparency as well as the complexity of these systems overwhelm the maintenance department in the effective planning and implementation of maintenance tasks. As a result, major maintenance tasks are postponed to non-production times which causes increased maintenance cost as well as a decrease in system availability. This research explores a method that uses unsupervised learning algorithms to analyze type mixes and related process performances inside the system. The information is used to determine the optimal master production schedule prior to maintenance activities which leads to more frequent and extended time windows for maintenance activities during production time and thus to an increase in system availability.

A new data analytics framework emphasising preprocessing of data to generate insights into complex manufacturing systems

Recent emphasis has been placed on improving the processes in manufacturing by employing early detection or fault prediction within production lines. Whilst companies are increasingly including sensors to record observations and measurements, this brings challenges in interpretation as standard approaches do not highlight the presence of unknown relationships. To address this, we have proposed a new data analytics framework for predicting faults in a large-scale manufacturing system and validated it using both a publicly available Bosch manufacturing dataset with a focus on preprocessing of the data and the open-source SECOM industrial dataset. This paper is an extension to the work presented at International Conference on Intelligent Manufacturing and Internet of Things. The additional material includes a detailed focus on feature selection and the various approaches for identifying important features in the data, an updated framework methodology and description, an extension of XGBoost to allow this model to be used for prediction/classification and a comparison for classification with a Random Forest, tree-based model. The framework was used to explore two public manufacturing datasets and successfully identified the most influential features related to product failure in each production line data.

A Global Manufacturing Big Data Ecosystem for Fault Detection in Predictive Maintenance

Artificial intelligence, big data, machine learning, cloud computing, and Internet of Things (IoT) are terms which have driven the fourth industrial revolution. The digital revolution has transformed the manufacturing industry into smart manufacturing through the development of intelligent systems. In this paper, a big data ecosystem is presented for the implementation of fault detection and diagnosis in predictive maintenance with real industrial big data gathered directly from large-scale global manufacturing plants, aiming to provide a complete architecture which could be used in industrial IoT-based smart manufacturing in an industrial 4.0 system. The proposed architecture overcomes multiple challenges including big data ingestion, integration, transformation, storage, analytics, and visualization in a real-time environment using various technologies such as the data lake, NoSQL database, Apache Spark, Apache Drill, Apache Hive, OPC Collector, and other techniques. Transformation protocols, authentication, and data encryption methods are also utilized to address data and network security issues. A MapReduce-based distributed PCA model is designed for fault detection and diagnosis. In a large-scale manufacturing system, not all kinds of failure data are accessible, and the absence of labels precludes all the supervised methods in the predictive phase. Furthermore, the proposed framework takes advantage of some of the characteristics of PCA such as its ease of implementation on Spark, its simple algorithmic structure, and its real-time processing ability. All these elements are essential for smart manufacturing in the evolution to Industry 4.0. The proposed detection system has been implemented into the real-time industrial production system in a cooperated company, running for several years, and the results successfully provide an alarm warning several days before the fault happens. A test case involving several outages in 2014 is reported and analyzed in detail during the experiment section.

An intelligent scheduling algorithm for complex manufacturing system simulation with frequent synchronizations in a cloud environment

For cloud-based, large-scale complex manufacturing system simulation (CMSS), allocating appropriate service instances (virtual machines or nodes) is a promising way to improve execution efficiency. However, the complex interactions among and frequent aperiodic synchronizations of the entities of a CMSS make it challenging to estimate the influence of service instances’ computing power and network latency on the execution efficiency. This hinders the appropriate allocation of service instances for CMSS. To solve this problem, we construct a performance estimation model (PEM) using the executed events and synchronization algorithms to evaluate the running time of CMSS on different service instance combinations. Further, an intelligent scheduling algorithm that introduces PEM as fitness function is proposed to search for a near-optimal allocation scheme of CMSS service instances. To be specific, the PEM-based optimization algorithm (PEMOA) incorporates simulated annealing into the mutation phase of a genetic algorithm to strengthen its local searching ability. A series of experiments were performed on a computer cluster to compare the proposed PEMOA with two representative algorithms: an adapted first-come-first-service-based and the max-min-based allocation algorithms. The experimental results demonstrate that the PEMOA can reduce the running time by more than 7%. In particular, the improvement of PEMOA increases when the manufacturing system simulation is communication-intensive or spans a small number of service instance combinations.

Historical survey and emerging challenges of manufacturing automation modeling and control: A systems architecting perspective

An architecting perspective is derived as part of a more global perspective, related to manufacturing automation modeling and control through four industrial revolutions. This historical survey (Pereira et al., 2017) highlights the impact of digitalization both on isolated machines and devices, and on the architecting of large-scale manufacturing and logistics systems. To distribute the holistic control over related components, two main paradigms have been followed for the engineering of those System of Systems (SoS): (1) technology-enhanced, and (2) bio-inspired. In both, the cognitive orchestration of the knowledge and skills involved in a system project remains challenging for the interdisciplinary togetherness and harmony required beyond the disciplinary boundaries. Finally, lessons learned, as well as acquired knowledge through this innovative history are addressed with several open questions and emerging challenges related to the cyber-physical systems evolution.

The evolving digital factory – new chances for a consistent information flow

Within product development process data and information management is an essential task. Data and information are generated at every stage of this process and stored in software tools used to generate that data. For large scale manufacturing systems this leads to manual work, which is error-prone. This issue intensifies due to increasing influence of computer science on those systems (e.g. Industrie 4.0), leading to a rise of complexity. New approaches are needed for a better information flow.This contribution introduces an approach of an evolving Digital Factory containing and visualizing all generated information based on an open standard meta model.

FMS Scheduling under Availability Constraint with Supervisor Based on Timed Petri Nets

This paper proposes an optimal solution to large-scale Flexible Manufacturing System (FMS) scheduling problems under availability constraints based on Timed Petri Nets (TPNs). First a decomposition method of TPNs is proposed, then a mathematical model is derived based on their properties. The mathematical model is built to determine the optimal firing sequence of TPN transitions to minimize the total manufacturing time. The resulting firing sequence of TPN transitions is used to generate the manufacturing system supervisor operated by TPN and digital controllers. Several numerical examples and comparative studies are provided in this paper in order to prove the new approach’s efficiency.