Keywords: Machine learning, Reinforcement, AI-Based Methods
Abstract: The enormous interest in artificial intelligence and especially machine learning (ML) among scientists in different research fields has recently also influenced the focus of our CASE conference. This is manifested by the main themes at IEEE CASE 2018-2021: Knowledge-based Automation, Smart Automation, Automation Analytics, and Data-Driven Automation. Since learning is such an important tool in many automation solutions, including data-based model generation, online optimization, and adaptive control, it is crucial to increase our activities in this field even further, to become an important player in the tough scientific race around ML that is going on right now.

The goal of this workshop is therefore to create a deeper interest and understanding of ML, but also to identify niche areas of ML in automation, where our research community should take the lead. More specifically, we want to present some interesting ongoing research activities, but also to discuss and propose what we believe are important research directions where automation can play an important role in this dynamic research area.

The presentations in this workshop will be given by members of a recently created AdHoc on Machine Learning for Automation. This AdHoc is focused on how to strengthen research activities, but also organization and infrastructure around ML research within automation. The workshop will therefore conclude with an open discussion to get interesting inputs for future activities within this challenging research field.

Keywords: Deep Learning in Robotics and Automation
Abstract: Efficiency and sustainability will be the key for the future factory, whose main focus will be on efficient and sustainable industrial processes. A sustainable production, an efficient use of the resources, and an increase in the recovered and reused products will be crucial to reduce the impact of the production on the environment, in compliance with the upcoming Industry 5.0 paradigm. Artificial intelligence (AI) and robotics are leading to deep workplace innovation, optimizing human-machine interactions, and giving more importance to workers. But the environmental goals can only be achieved by rethinking the production processes in order to limit the environmental impact. Disassembly is an industrial process that will have to be continuously optimized to increase efficiency and sustainability in years to come. Disassembly extracts valuable components/materials from end-of-life goods for reuse and recycling. It is also used in product refurbishment when products are restored to full manufacturer conditions by running quality tests and replacing broken or defective parts. Refurbishing products is a great opportunity for sustainability as it gives new life to used products instead of producing new ones, thereby providing consumers with quality products at an affordable price. Statistics say that the refurbished market for consumer electronics is estimated to be 10 billion. Disassembly consists of a series of tasks performed in lines made up of workstations where workers may be assisted by robots. Making these lines as efficient and sustainable as possible includes the design, the optimization, and the improvement of the collaborations between workers and machines. Artificial Intelligence (AI) can help deal with the complexity of these problems to find and implement solutions that increase efficiency and reduce the impact of production on the environment. This Workshop aims to collect the latest research and achievements and discuss the progress regarding advanced AI techniques for optimal industrial disassembly processes.

Keywords: Control Architectures and Programming
Abstract: In the UAV field, the efforts to develop drones with more payload and flight time capacity are constant. Coaxial multirotor drones offer high payload capacity in a relatively small vehicle footprint. However, compared to regular 'flat' multirotors, they exhibit a much lower efficiency. The content covered in this proposed tutorial is based on a very recent work of the authors where they developed a control allocation method in which experimental results showed an increase in efficiency of up to 11% compared to the current state-of-the-art. Additionally, the tutorial also covers the operation of an open-source benchmarking platform developed by the authors with the purpose of testing and optimizing the performance of coaxial rotor systems. Therefore, the tutorial should provide the participants with all the tools needed to perform experiments, develop, and implement a control allocation methods to improve the efficiency of coaxial rotor systems in autonomous applications.

Keywords: Machine learning, Additive Manufacturing, Cyber-physical Production Systems and Industry 4.0
Abstract: Organizers: Qiang Huang, Zengxi Pan, YuMing Zhang

Date: Aug. 20, 2022, 2-5PM

Abstract: Quality and productivity are critical for additive manufacturing (AM). With increased availability of AM product data, Machine Learning for AM (ML4AM) has become a viable strategy for knowledge discovery and performance enhancement. Although general-purpose machine learning and geometric analysis in computer vision have been extensively studied, ML4AM differs in many ways. Shape accuracy control in AM needs a new engineering-informed data analytics framework to facilitate efficient machine learning of AM product data. Furthermore, new AM processes such as wire arc additive manufacturing (WAAM) introduce additional challenges. This workshop introduces recently development in ML4AM including control for WAAM.

Descriptions: The workshop covers the following topics:

• 3D shape accuracy representation; prescriptive modeling of shape accuracy through learning heterogeneous training data; optimal compensation of shape deformation, engineering-informed transfer learning in AM systems; Statistical transfer learning methods and applications in AM.

• Challenges in AM/WAAM and machine learning-based real-time monitoring and control solutions; deep Learning based monitoring of weld penetration; and development of process monitoring for WAAM using machine learning.

Mode: In-person/Mexico City

Keywords: Reinforcement, Autonomous Agents, Motion and Path Planning
Abstract: Transfer learning is potential for reducing expensive interactions with the physical environment in reinforcement learning (RL). Based on the correlation between scenarios, both the prior policy and historical experiences may be helpful to accelerate policy optimization in the target scenario. However, without manually designing proper relaying scenarios, the discrepancy between scenarios may give rise to sub-optimal policies or even negative transfer. In this paper, we firstly propose a continuously generated relaying predictive model (CRPM), which autonomously bridges the source and target scenarios with a series of gradually modifying relaying scenarios. Then, we experimentally show that the CRPM effectively reduces interactions required for policy optimization in the target scenario. Besides, we combine the CRPM with model-based RL, which further improves the performance from the aspects of both learning rate and gathered rewards. The CRPM also helps to improve the classical model-free RL by treating it as a special case of transfer learning in the same scenario. Experimental results also show that the CRPM helps to avoid sub-optimal policies and performs better on both the source and target scenarios.

Keywords: Diagnosis and Prognostics, Data fusion, AI-Based Methods
Abstract: With the advances in sensor technology, it has become applicable to use multiple sensor data to analyze health statuses and degradation processes of operating units in engineering systems. However, two challenging issues remain: one is how to combine sensor signals to quantify the health status of units under multiple operational conditions. The other is how to capture common and individual characteristics of the degradation processes of units according to their health status, which can contribute to accurate remaining useful lifetime (RUL) prediction. To address the above issues, this paper proposes a data fusion-based long short-term memory (LSTM) network for degradation modeling under multiple operational conditions. An iterative algorithm is proposed to estimate the fusion coefficients and degradation parameters in the proposed network. In the case study that utilizes the degradation data set of aircraft turbine engines operating under six conditions, the proposed method has better model performance for RUL prediction compared with the benchmark method that does not consider multiple operational conditions.

Keywords: Sensor Fusion, Industrial and Service Robotics
Abstract: In this paper, a multi-sensor fusion framework is proposed to solve the localization problem of mobile robot in indoor environments. To improve the localization accuracy, two kinds of fusion algorithms, namely EKF and MCL, are used and the motion model as well as the measurement model are selected according to the complexity of the environment, which is quantified by the minimum distance between the robot and the obstacles. In EKF fusion, the motion model is obtained by wheel odometer, and the measurement model is the combination of IMU and UWB sensor. While in MCL fusion, the motion model switches between odometry and the output of EKF, and the measurement model switches between lidar and the combination of IMU and UWB sensor. Experimental results show that the localization effect of multi-sensor fusion is better than that of a single sensor. The improved MCL algorithm proposed in this paper is superior to the traditional Monte Carlo localization algorithm in both localization accuracy and convergence speed.

Keywords: Diagnosis and Prognostics, Manufacturing, Maintenance and Supply Chains, AI-Based Methods
Abstract: Door system is the core part of the elevator. With an accurate fault diagnosis of door system, the decision support for engineers in troubleshooting and reducing maintenance costs can be provided. However, the research of fault diagnosis based on elevator operation and maintenance data is still in its infancy. With the development of the industrial Internet-of-things, real-time monitoring data of elevator equipment can be collected and used for modeling. This paper aims to investigate a machine learning-based approach to achieve accurate door fault diagnosis. The experimental results revealed that the Xgboost algorithm can achieve good performance.

Keywords: Big-Data and Data Mining
Abstract: 晶圆加工周期时间是晶圆制造系统生产过程中的重要指标，有助于制定更合理的加工调度，提高生产效率。晶圆加工车间监控数据维度高、参数冗余大等问题使得构建周期时间预测模型的过程具有挑战性。针对上述问题，提出一种基于相关特征的密度峰值聚类算法（DPC-Relief-F）混合特征提取方法，可以过滤出样本特征中的关键特征子集，从而减小模型输入数据维度，提高模型训练速度和预测精度。此外，提出一种将模糊C均值与反向传播网络（FCM-BPN）相结合的并行训练预测模型，以加快大规模数据下的训练过程。实'

Keywords: Computer Vision in Automation, Machine learning
Abstract: The detection of tiny targets on the surface with high efficiency and accuracy is significant for the current intelligent manufacturing. Visual inspection methods based on deep learning are widely utilized to detect tiny objects. However, the tiny objects appear less distinct, less wide, and less area occupied in the image. At the same time, there is a lot of object-like noise, which further increases the difficulty of detecting tiny objects. In response to the challenges brought by the complexity of the detection environment, this paper proposes a detection network architecture that combines the enhancement of pixel-level features at equal resolution and the introduction of full-scale features based on attention. The model utilizes the subtle differences between the tiny target and the background and the semantic information of the tiny target outline to enhance the features of the tiny target while significantly reducing its loss in the equal-resolution feature layer. Additionally, a gradual attention mechanism is proposed to guide the network to pay attention to tiny objects features on the full-scale feature layer. The performance of this network architecture is validated on a real dataset. Experiments show that the model exhibits superior performance and outperforms existing resNet50, DenseNet, Racki-Net, and SegDecNet in detecting tiny objects.

Keywords: Planning, Scheduling and Coordination, Cellular and Modular Robots, Industrial Robots
Abstract: In an environment with limited space and dense goal configuration, the path of robot team is forced to coincide without much adjustment space, which is a challenge for multi-robot collaborative path planning. In this work, a novel Optimal Path and Timetable Planning (OPTP) method is proposed. The OPTP firstly generates the near-shortest paths for each robot by an RRT*-based planner. Then the timetables for each robot in the path-time space are created by the improved Particle Swarm Optimization (PSO) method. A heuristic bias is added to the PSO optimizer to efficiently mediate the conflict near the goal configuration. The OPTP achieves the near-shortest moving distance of the multi-robot team, as well as the near-optimal navigation makespan in face of complex obstacles, narrow channels, and dense goal configurations. The compared simulations and real-world experiments verify the effectiveness of the OPTP method.

Keywords: Collaborative Robots in Manufacturing, Cognitive Automation, Reinforcement
Abstract: Human-robot collaboration (HRC) is an important method for manufacturing industry to realize intelligent and flexible production. While robots partially replace human labor, they improve production efficiency and accelerate the process of intellectualization. However, in the human-robot collaboration system, when humans and robots need to perform frequent collaborative operations, the execution of cobot actions is plagued by the robot's inability to know the trend of human behavior in advance, which in turn leads to decision delays or decision errors. In this regard, a cognition-driven robot decision-making method in a human-robot collaboration environment is proposed to divided acceptance, rejection and delay regions for decision values, and use a process of dynamic adjustment of decision values and region boundaries to simulate the human decision-making process to achieve robot decision cognition. At the same time, a reinforcement learning algorithm is used to optimize the decision boundary values based on the reward function to improve the cognitive efficiency and arrive at the final decision results as soon as possible. Finally, we take the assembly process of engine end cover as the goal of the cooperative task, and find that the efficiency is improved compared with the cooperative method based on attitude recognition.

Keywords: Task Planning, Planning, Scheduling and Coordination, Motion and Path Planning
Abstract: In many industrial applications, the robot is required to perform a set of repetitive tasks without collision as quickly as possible to maximize productivity. It is essential to find an optimal sequence of collision-free motions to visit a set of repetitive tasks and determine the optimal robot configuration used to complete each task, which is formulated as the Robotic Task Sequencing Problem (RTSP). In this paper, we propose an approach based on a typical decoupling strategy to solve RTSP efficiently. Firstly, the task execution sequence is obtained by solving a TSP in task space and candidates of the optimal configuration for each task are selected from the collision-free configuration space according to the self-designed optimality value derived from the similarity to the initial configuration in configuration space. Then the optimal configuration for each task is determined by finding the shortest path in a graph that is constructed according to the task execution sequence and optimal configuration candidates. Finally, collision-free motion trajectories required for the robot to complete each task with the optimal configuration are generated by running a motion planning algorithm. Through a series of experiments, we show that our approach outperforms the state-of-the-art approaches when applied to the RTSP instances in a cluttered 3D environment, with up to 29.6% reduction in computation time while providing comparable performance.

Keywords: Collaborative Robots in Manufacturing, Intelligent and Flexible Manufacturing, Intelligent Transportation Systems
Abstract: Cooperative transportation of large objects with a multi-AGV system is of great economic and practical significance. How to ensure the synchronized motion of the multiple AGVs, which are connected to each other by carriers, is the most critical and also the most difficult problem. In this paper, we propose a motion planning and control method for a cooperative transportation system, which consists of a carrier and two AGVs communicating with each other in 5G environment. In order to maintain a required constant distance between two AGVs during the movement, a leader-follower formation strategy based approach with the follower tracking the leader in real time is used. Specifically, the reference trajectory of the follower is calculated by the real-time trajectory of the leader. Then a tracking controller is designed for the follower so that it can follow the target trajectory. Experiments in 5G environment with actual industrial AGVs show that the proposed approach can ensure a smooth transportation motion and the maximum error of distance controlling of the two AGVs is within ±3cm.

Keywords: Logistics
Abstract: The integration of warehousing and distribution operations allows a company to become "constraint-aware" in both inventory management and transportation management, which helps reduce the total operational cost and improve the customer service. In this paper, we study how to plan the daily delivery routes in such an integrated warehousing and delivery logistics system. We consider a manufacturer that sells multiple products to its customers distributed nationwide. The products are stored in a cluster of warehouses with various assortments and inventory levels. On a daily base, the manufacturer receives customer orders and decides a delivery plan to fulfill these orders, considering the product inventories and outbound shipment capacities of the warehouses. We formulate this problem as a binary integer linear program and develop a branch-price-and-cut algorithm, which consists of an efficient label-setting pricing algorithm and a customized branching scheme. We perform a comprehensive numerical study using test instances based on the (masked) real data from a manufacturer in China. The results show the efficiency and the effectiveness of the branch-price-and-cut algorithm. We also study and obtain practical implications on the impact of product inventories and outbound shipment capacities at the warehouses.

Keywords: Intelligent and Flexible Manufacturing, Manufacturing, Maintenance and Supply Chains, Planning, Scheduling and Coordination
Abstract: In the automobile engine flow shop (AEFS), the equipment has more chance to work for a long time, so the manufacturing performance may deteriorate rapidly, thus causing operating unbalance and inefficiency. To tackle this problem at minimum cost, this study proposes a multi-thread constrained two-archive evolutionary algorithm (CTAEA) to enable workstation reconfiguration for multi-stage AEFS considering performance deterioration. First, a workstation reconfiguration model that selects the extra cost and alteration of cycle time as the objectives is developed to evaluate the reconfiguration scheme for the multi-stage AEFS. Then, a multi-thread CTAEA is designed to efficiently optimize the established mathematical model that involves a large number of variables making it hard to obtain globally optimal solutions. Finally, a case study is implemented to verify the feasibility and superiority of the proposed method via experimental validation and comparison, respectively.

Keywords: Optimization and Optimal Control, Planning, Scheduling and Coordination, Intelligent Transportation Systems
Abstract: Stowage planning is one of the most important stages in management of container terminals and depends the sequence of containers to be loaded on the ship. For non-clear containers, which are absence from the terminal, their slots will be selected manually by stowage planners, making it a time-consuming job. In order to optimize the slot reservation problem of non-clear containers, a mathematical model based on the knapsack problem is constructed. A Stack Selection Algorithm based on dynamic programming is proposed to solve the model. Further case study of Yangshan automatic container terminal demonstrates that the method can solve the non-clear containers reservation problem in a very short time, and the results are better than traditional heuristic approaches.

Keywords: Planning, Scheduling and Coordination
Abstract: Multimodal transport harbors, especially rail-water multimodal transport, have drawn much attention in recent years. There are two operation modes describing the flow of containers, namely "train-yard-vessel" mode and "vessel-train" mode. The two modes often coexist in reality, however, there is few research focusing on this hybrid mode. To solve the problem of collaborative scheduling of railway gantry cranes and trucks in a rail-water multimodal transport harbor under the hybrid mode of containers flow, a mixed integer programming model is established and a genetic algorithm (GA) is designed to generate the schedule. Results show that the model and algorithm proposed can solve the problems in a medium and large scale. When the number of containers is 40 and the proportion of quayside containers is 0.4 or 0.6, the optimal number of gantry cranes and trucks is 3 and 10 respectively. When the number of containers is 80 and the proportion of quayside containers is 0.4 or 0.6, the optimal number of gantry cranes and trucks is 3 and 13 respectively. It is found that the number of trucks has a marginal benefit for makespan and more trucks is not always better.

Keywords: Hybrid Strategy of Intelligent Manufacturing, Domain-specific Software and Software Engineering
Abstract: The digital twin (DT) connects the physical world with the virtual world and is an integral part of the smart workshop. Existing digital twin modeling methods focus on the detailed description of modules and the interaction between modules, without considering the subject status of data, which limits the practical application of digital twins.Data-centric digital twin modeling methods can be well combined with computer programming languages to guide the implementation of digital twin applications. With the continuous advancement of big data and artificial intelligence, there is a need for an approach to modeling multi-dimensional digital twins at the conceptual level. To this end, this paper proposes a digital twin concept modeling method based on Artifact, which connects the five-dimensional digital twin framework and Web service semantics, which not only describes complex digital twin components, but also discovers Web service semantics to guide the implementation of programming languages. Based on this model, this research designs a digital twin concept modeling prototype system and models the digital twin of the warehouse workshop.

Keywords: Planning, Scheduling and Coordination, AI-Based Methods, Optimization and Optimal Control
Abstract: The traditional material transportation vehicle scheduling adopts manual scheduling application and vehicle scheduling, which has low efficiency, high cost and waste of human resources. The existing scheduling system can not realize the information interaction and collaborative integration between the physical world and the virtual world, and the digital twin technology can effectively solve the problem of information interaction. This paper establishes a model of the transportation vehicle scheduling problem, proposes a transportation vehicle scheduling framework based on digital twin model, and uses Q-learning method to solve the dynamic scheduling problem. Finally, an example is given to verify the superiority and effectiveness of this method.

Keywords: Building Automation, Automation Technologies for Smart Cities, Energy and Environment-Aware Automation
Abstract: The energy saving problem of a variable air volume (VAV) system in heating, ventilation, and air-conditioning (HVAC) systems has an increasing attention. In this paper, we propose a distributed optimal dynamic duct static pressure method (DODSP) for the original non-convex energy saving problem in multi-zone variable air volume systems. Our method is based on the variable separation method and the alternating direction method of multipliers to solve an equivalent distributed convex problem. The whole system is fully distributed and each zone is controlled by an intelligent node. Every node makes the optimal control decision to meet the temperature setpoint. Our method is the first effort to solve the energy saving problem in a fully distributed manner and has the convergence guarantee. Even if the damper breaks down, the system can still be working. Numerical simulations on a distributed building simulation rig are given to illustrate the effectiveness of our method and compare with current centralized methods, which show that our method provides great flexibility and realizes the coordination among different zones. The energy saving can be up to 20%.

Keywords: Smart Home and City, Optimization and Optimal Control, Reinforcement
Abstract: The idea of using wind power to charge electric vehicles (EVs) has attracted more and more attention nowadays due to the potential in significantly reducing air pollution. However, this problem is challenging on account of the uncertainty in the wind power generation and the charging demand from the EVs. Simulation-based policy improvement (SBPI) has been an important method for decision-making in stochastic dynamic programming and, in particular, for charging decisions of EVs in microgrids. However, the problem of allocating the limited computing budget for the best decision-making in online applications is less discussed. We consider this important problem in this work and make the following three major contributions. First, we show that the significant uncertainty in wind power generation forecasting could make the policy that is the outcome of an SBPI worse than the base policy. Second, we apply two existing methods to address this issue, namely, the optimal computing budget allocation (OCBA) for maximizing the probability of correct selection (OCBA_PCS) and the OCBA for minimizing the expected opportunity cost (OCBA_EOC). The asymptotic optimality is briefly reviewed. Third, we numerically compare the performance of OCBA_PCS and OCBA_EOC with the equal allocation (EA), a principle-based method, and a stochastic scenario-based method on small-scale and large-scale experiments. This work sheds light on the EV charging decision in general. Note to Practitioners-Together with the growing adoption of EVs in modern societies, there goes the challenge of how to satisfy the charging demand. Given the high uncertainty both in the wind power generation and in the charging demand, it is important to make decisions online using up-to-date estimation on the renewable power gene

Keywords: Human-Centered Automation, Intelligent and Flexible Manufacturing, Learning and Adaptive Systems
Abstract: Nowadays, with the advancement of production technologies, the manufacturing paradigm has gradually shifted from mass production to a small-batch and high-variety personalized production manner, urged by high flexible automation capabilities. In this paradigm, the existing inspection and assembly processes after manufacturing still rely to a large extent on either human operators with low efficiency or machines with low flexibility. To solve this issue, human-robot collaboration (HRC) has been a prevailing topic of recent concerns. Current robot control strategies in human-machine collaboration are mainly through pre-defined programming and do not yet meet the need for flexible and adaptable tasks in individualised production. To address this challenge, this paper proposes a deep reinforcement learning (DRL) approach based on meta-learning to drive robots in HRC. It enables collaborative robots (cobots) to acquire basic skills and perform tasks based on personalised production requirements, improving learning efficiency and thus quickly adapting to new tasks for human operators. The robot control task was carried out in a simulated environment taken from a real production scenario to assess its efficacy. Experimental results show that our proposed method enables the robot to learn and perform HRC tasks quickly and outperforms the baseline DRL method in terms of success rate.

Keywords: Human Factors and Human-in-the-Loop, Task Planning
Abstract: Collaborative robots have been introduced into manufacturing workshops to collaborate with human and improve production flexibility and ergonomics. In human-robot collaboration (HRC) workshops, detecting the fatigue state of workers and reassigning tasks online quickly is the key to effectively avoid product quality defects, safety incidents, and diseases caused by fatigue of workers. Due to the differences among workers and the diversity of assembly tasks, previous methods for task allocation is not efficient enough. This paper proposes a task reassignment method based on online fatigue detection. This method uses electroencephalography (EEG) and image processing data to detect workers' fatigue firstly, and then updates workers' fatigue status flag online according to the detection data, and then uses the improved NSGA-III algorithm proposed in this paper to obtain a new task assignment. The experiment results show that the multi-modal fatigue detection method proposed in this paper outperforms the one using single-modal data, and the improved NSGA-III algorithm is superior to the other two optimization algorithms in convergence speed and quality of the Pareto solution set.

Keywords: Collaborative Robots in Manufacturing, Assembly
Abstract: In the context of industry 5.0, human-robot collaboration (HRC) in assembly, a flexible production mode, has been paid increasing attention. In the scenario of HRC in assembly, to perfect the efficiency and fluency of the whole assembly process, the leading point is to develop a more natural human-robot interaction (HRI). In that way, the robot has the access to predict the human’s intention earlier. The single process’s intention has been mainly focused on human intention prediction, however, is verified against the natural HRI, causing the robot insensitive to turn-taking among the successive process. Therefore, this paper enters a proposal that we can realize early prediction of turn-taking in HRC assembly tasks based on Izhikevich neuron model-based spiking neuron network (SNN). The proposal is also verified in a developed HRC gear assembly scenario. The results express that our method can greatly advance the recognition time of human-robot turn-taking, which improves the efficiency of human-robot collaborative assembly.

Keywords: Human Factors and Human-in-the-Loop, Planning, Scheduling and Coordination, Intelligent and Flexible Manufacturing
Abstract: Dynamic scheduling is one of the most important problems in the field of production scheduling. Existing ways to solve the problem are mainly based on experienced workers or automatic scheduling models (SMs). Because of the complementary advantages of workers and SMs, their combination has the potential to be a better solution. In this letter, a human-machine collaborative decision-making method based on confidence (HMCDM/C) is proposed. SMs are in charge of the automatic generation of decisions, and workers have the power to revise unreliable decisions. Furthermore, a threshold-based handover mechanism is proposed to determine when workers are involved in decision-making. Firstly, a measurement method is developed to quantify the confidence levels of SM decisions. Secondly, an evaluation method is presented to determine the threshold of confidence levels, which can be used to discriminate whether the SM decisions are acceptable or not at a certain confidence level. Finally, several experiments are conducted in a smart workshop. The results show that the HMCDM/C can effectively coordinate workers with different experience levels and SMs, and has a very competitive performance.

Keywords: Computer Vision in Automation, Collaborative Robots in Manufacturing, Learning and Adaptive Systems
Abstract: The assembly accuracy of aircraft skin directly affects the aerodynamics, air tightness and invisibility of aircraft. Extracting the butt joint feature area of aircraft skin can provide an effective basis for the assembly of aircraft skin. With the wide application of surface structured light measurement in many fields, based on surface structured light, we propose an aircraft skin seam point cloud extraction technology with adaptive matching calibration algorithm. Firstly, multi-stage high-precision calibration is carried out, and then the negative feedback of height difference is introduced to correct and calculate the calibration range of aircraft skin height by using the negative feedback of height difference to realize adaptive calibration. Finally, high-precision point cloud data of aircraft skin butt joint are obtained. The experimental results show that the adaptive matching calibration algorithm proposed in this paper can effectively extract the butt seam feature area of aircraft skin, and the scanning time and data processing time are less than that of single line structured light measurement. This study provides an efficient method for 3D point cloud data extraction of aircraft skin butt joint feature area.

Keywords: Intelligent and Flexible Manufacturing
Abstract: Constructing intelligent manufacturing systems and optimizing scheduling management is of economic significance. To apply the flexible job-shop scheduling problem to specific manufacturing fields, such as the food fermentation industry and aerospace industry, in which each product is required to be delivered within its due window, this paper establishes the mixed-integer nonlinear programming model for the flexible job-shop scheduling problem with due windows and develops a bi-level closed-loop heuristic algorithm to minimize the weighted earliness or tardiness. In the upper level, the genetic algorithm is used to optimize each job's processing path and sequence iteratively. In the lower level, the sequential quadratic programming determines the start time for each operation of each job by solving a nonlinear programming problem. The convergence of the algorithm is analyzed. Experimental results show that the algorithm can efficiently solve the optimal solution or approximate optimal solution of the problem. As a general idea of dividing integer and continuous solution space, the algorithm is robust to the field and details of the problem.

Keywords: Reinforcement, Intelligent and Flexible Manufacturing, Inventory Management
Abstract: In this research, the production inventory control problem in automobile manufacturers is investigated to keep the inventory at the ideal level and minimize the production cost. Firstly, we establish a linear-quadratic tracking (LQT) model for three serial production workshops. The reinforcement learning (RL) algorithm is employed to give a control policy of the problem with unknown parameters. Furthermore, an improved multi-objective differential evolution (MODE) algorithm is proposed to adjust the weight matrix and hyperparameters of RL so that the diversity of policies on conflicting operational indicators can be enhanced. Simulation results show that the proposed algorithm achieves better performance on both production inventory control and parameter optimization.

Keywords: Manufacturing, Maintenance and Supply Chains, Logistics
Abstract: For steel production, due to the characteristics of long manufacturing process, batch production model, limited capacity of bottleneck units and strict delivery requirements, how to balance the production capacity of different units in each stage is an important problem to be solved in production management. In this paper, we concentrate on steelmaking stage and study the capacity planning problem considering the demands of downstream processes. To solve the problem, we propose a hierarchical solution framework in which the demands of downstream stages are transmitted to steelmaking stage through planned time windows and a combination of mixed integer programing model and local search heuristics is employed to balance the capacity of units in steelmaking stage. To verify the performance of the proposed solution framework, computational tests are conducted on synthetic instances. The results show the efficiency of the proposed solution framework.

Keywords: Planning, Scheduling and Coordination
Abstract: We propose two formulations for the capacitated lot sizing problem with family-based setup and downstream-process-based demand (CLSP-FSDPD), one is the aggregate formulation (AGG) and the other one is the facility-location-based disaggregate formulation (FDG). We compare two formulations and extensive computational experiments show that FDG formulation outperforms AGG formulation in terms of solution quality when the planning horizon is relatively short, and AGG formulation consistently takes shorter CPU computation time that FDG formulation to solve in most cases. We show some of the differences in the performance of these various formulations arise from their different use of variables to represent production, setup or inventory states.

Keywords: Cyber-physical Production Systems and Industry 4.0
Abstract: Production systems include both discrete part and batch operations, where an individual part is manufactured in a discrete operation, and a group of parts are processed simultaneously, i.e., in a batch, on one machine for a batch operation. Many manufacturing industries, such as battery, aircraft, and automotive, consist of mixed batch and discrete part operations, referred to as batch-discrete lines. Although such operations are widely encountered, analytical studies of these systems are limited in current literature. In this paper, a systems approach is presented to model and analyze batch-discrete lines. First, a Bernoulli machine reliability model for a two-machine batch-discrete system is introduced. Using a virtual buffer to represent the batch processing feature, performance evaluation formulae are derived and system properties are investigated. Using them, improvement analyses and bottleneck identification are presented. Then, the model is extended to systems with a quality inspection device under different control policies. To illustrate the applicability of the model, a case study in a composite part production process is described. Such a work delivers a quantitative tool for production engineers and managers to design, analyze, and improve batch-discrete manufacturing systems.

Keywords: Planning, Scheduling and Coordination, Optimization and Optimal Control, Petri Nets for Automation Control
Abstract: This work investigates a scheduling problem of deadlock-prone flexible manufacturing systems. It proposes an Anytime Branch and Bound (AB&B) algorithm to minimize system makespan based on the branch tree of a net model and a highly permissive deadlock controller. This work develops two pruning rules, a lower bound of makespan, and a novel branching strategy to ensure AB&B's high search efficiency. Experimental results demonstrate that the proposed algorithm surpasses the state-of-the-art ones.

Keywords: Robust Manufacturing
Abstract: Empirical models that relate quality characteristic to a set of design variables play a vital role in many industrial process optimization methods. Many of the current modeling methods usually employ a single-response normal model to analyze industrial processes without taking into consideration model form uncertainty and the non-normality in the modeling process. Failure to account for these issues may result in a misleading prediction model and therefore poor process design. In this article, we will introduce a double-robust Bayesian process optimization approach, which accounts for the effect of both the two issues on the implementation of response surface methodology. Moreover, the proposed approach has variable selection consistency and more importantly it can hold for the entire class of spherically symmetric distributions, e.g., student t-distribution and scale mixtures of normals. The effectiveness of the above approach is illustrated with both simulation studies and a case study in a melt electrospinning process. The comparison results demonstrate that the proposed approach can achieve a better process performance than its classical counterparts in the literature.

Keywords: Failure Detection and Recovery, Machine learning
Abstract: Considering the reliability of the data storage system, it is essential to accurately and timely predict impending failures of hard disk drives (HDDs) so as to prevent data loss and reduce recovery cost. Over the past decades, taking as input the SMART (Self-Monitoring, Analysis and Reporting Technology) attributes, many supervised machine learning based methods have been proposed for HDD failure prediction. However, these methods are conducted on different datasets or different preprocessing treatments and thus lack comparative analysis. To fill this gap, we provide a systematic study in this paper on three key steps of the failure prediction, i.e., feature selection strategies, data preprocessing treatments and classification models. A feature selection strategy is proposed by testing the significance of difference between healthy and failed samples. Data relabeling, together with some other data preprocessing treatments are applied and proven to be effective in the case study. The performance of seven classification models are compared, among which the Random Forest model achieves the best performance with 53.95% failure detection rate (FDR) and 6.0% false alarm rate (FAR). Moreover, the Gini importance of SMART attributes is calculated, where two attributes, SMART 197 and SMART 187 are found closely related to the HDD failures.

Keywords: Process Control, Big-Data and Data Mining
Abstract: Industrial big data has provided an unprecedented data-driven opportunity for monitoring large-scale manufacturing and service systems. When a system process involves dozens of even hundreds of categorical variables each evaluated by attribute levels, which is commonly seen in industrial applications, the existing methods usually fail as they either only consider one single categorical variable or lack computational scalability for the high-dimensional (HD) cases. This work proposes a multiscale pattern mining and testing-based HD categorical process monitoring framework. To be specific, a frequency-based pattern is particularly defined and quickly mined to characterize both the significant individual behavior and the major joint behavior of all categorical variables. Then all these patterns are taken as informative surrogates of the original HD categorical process, and are monitored simultaneously via a principled and powerful multiple hypotheses testing procedure. The advantage of the new framework is verified by both numerical experiments and real case studies.

Keywords: Manufacturing, Maintenance and Supply Chains, Machine learning
Abstract: With the development of computer technology and artificial intelligence, various reinforcement learning (RL) algorithms are proposed to solve large state and action spaces of the Markov decision process (MDP). In this research, different RL algorithms are used to optimize the maintenance of multicomponent production systems with intermediate buffers. Results show that the application of RL methods to maintenance optimization is not straightforward. Some commonly used RL methods cannot obtain maintenance strategies as effective as control-limit type maintenance strategies. Results show that the multiagent RL method is a promising approach to optimize the maintenance of multicomponent systems. Furthermore, this research also finds that embedding domain knowledge of maintenance optimization into RL is a potential future research direction.

Keywords: Big-Data and Data Mining, Factory Automation, Machine learning
Abstract: The prediction of diesel engine power is a vital prerequisite for diesel engine quality promotion. A key issue of diesel engine power prediction is the selection of representative features for forecasting. However, current feature selection methods mainly rely on correlation analysis which cannot distinguish between direct correlation and indirect correlation. This paper presents a causal feature selection method for diesel engine power forecasting. Causalities distinguish direct influences from indirect ones. Therefore, this paper proposes a diesel engine power prediction framework based on using Markov Blanket-based feature selection approach and Gradient Boosting Decision Tree (GBDT) forecasting model. The proposed framework first applies Markov Blanket to identify causalities between manufacturing variables and diesel engine power and generates a causal feature set. Then, the quantitative relationship between causal features and the diesel engine power is established through GBDT. Finally, the proposed framework is tested by the experiment on a real diesel engine dataset. And the results show that the proposed framework delivers a satisfactory performance advantage for the validation condition in actual applications, the root mean squared error and the coefficient of variation of the root mean squared error of the GBDT model under the validation condition are 2.94kW and 1.17%, respectively.

Keywords: Diagnosis and Prognostics
Abstract: This study proposes a constraint linear model (CLM) to represent periodic signals. Different from the conventional linear model, linear constraints are added to the proposed model to fulfill the continuity constraints of signals. To estimate the period of the signals, a hypothesis testing based on the generalized likelihood ratio is proposed to statistically compare misleading peaks exhibited in the likelihood function of CLM. The misleading peaks usually arise because the likelihood function has a similar value in the true period and its multiples, and this misleading effect obstructs the effectiveness of the conventional maximum likelihood estimation method. An iterative likelihood ratio test (ILRT) is proposed, in which hypothesis testing is iteratively conducted to test the significance of an identified period until no candidate period with significant evidence is found. After the period is estimated by ILRT, an L1 penalty is added to CLM to construct a sparse representation of the signals, and an augmented Lagrangian shrinkage algorithm is proposed to extract sparse features from the signals. The effectiveness of the proposed method is verified through a simulation study on synthetic signals and a case study on real vibration signals.