Decomposition/aggregation-based dynamic programming optimization of partially homogeneous unreliable transfer lines

The general problem of buffers sizing for mean work in process/inventory minimization in a particular class of single part unreliable manufacturing flow lines, subjected to a constant rate of demand for finished parts, is analyzed. Two variants of the problem are considered: buffers sizing for average work in process minimization when there is a fixed requirement on parts availability in the buffer next to last; minimization of an aggregate measure of average work in process and demand backlog when the complete flow line is considered. A fluid model of part production is employed. The production control policies of interest are suboptimal, strictly decentralized, and are unambiguously parameterized by the size of buffer levels. Optimization of policy parameters is based on the analysis of the structural properties of an associated dynamic program. The latter is built around an approximate, flow line decomposition based, buffer levels dependent theoretical expression of the policy performance measure. The nature of the related flow line approximations is discussed and numerical results of the dynamic programming procedure are reported. Scalability of the computations is demonstrated. The numerical results suggest that when parameters are optimal, both a form of flow line balancing and a just in time internal production principle, are in place.     

Self-balancing dynamic scheduling of electrical energy for energy-intensive enterprises

Balancing production and consumption with self-generation capacity in energy-intensive enterprises has huge economic and environmental benefits. However, balancing production and consumption with self-generation capacity is a challenging task since the energy production and consumption must be balanced in real time with the criteria specified by power grid. In this article, a mathematical model for minimising the production cost with exactly realisable energy delivery schedule is formulated. And a dynamic programming (DP)-based self-balancing dynamic scheduling algorithm is developed to obtain the complete solution set for such a multiple optimal solutions problem. For each stage, a set of conditions are established to determine whether a feasible control trajectory exists. The state space under these conditions is partitioned into subsets and each subset is viewed as an aggregate state, the cost-to-go function is then expressed as a function of initial and terminal generation levels of each stage and is proved to be a staircase function with finite steps. This avoids the calculation of the cost-to-go of every state to resolve the issue of dimensionality in DP algorithm. In the backward sweep process of the algorithm, an optimal policy is determined to maximise the realisability of energy delivery schedule across the entire time horizon. And then in the forward sweep process, the feasible region of the optimal policy with the initial and terminal state at each stage is identified. Different feasible control trajectories can be identified based on the region; therefore, optimising for the feasible control trajectory is performed based on the region with economic and reliability objectives taken into account.     

A Model Predictive Control framework for robust management of multi-product,multi-echelon demand networks

Model Predictive Control (MPC) is presented as a robust, flexible decision framework for dynamically managing inventories and meeting customer requirements in demand networks (a.k.a. supply chains). As a control-oriented framework, an MPC-based planning scheme has the advantage that it can be tuned to provide acceptable performance in the presence of significant uncertainty, forecast error, and constraints on inventory levels, production and shipping capacity. The translation of the supply chain problem into a formulation amenable to MPC implementation is initially developed for a single-product, two-node example. Insights gained from this problem are used to develop a partially decentralized MPC implementation for a six-node, two-product, three-echelon demand network problem developed by Intel Corporation that consists of interconnected assembly/test, warehouse, and retailer entities. Results demonstrating the effectiveness of this Model Predictive Control solution under conditions of demand forecast error, constraints on capacity, shipping and release, and discrepancies between actual and reported production throughput times (i.e. plant-model mismatch) are presented. The Intel demand network problem is furthermore used to evaluate the relative merits of various information sharing strategies between controllers in the network. Both the two-node and Intel problems show the potential of Model Predictive Control as an integral component of a hierarchical, enterprise-wide planning tool that functions on a real-time basis, supports varying levels of information sharing and centralization/decentralization, and relies on combined feedback–feedforward control action to enhance the performance and robustness of demand networks. These capabilities ultimately mitigate the “bullwhip effect” in the supply chain while reducing safety stocks to profitable levels and improving customer satisfaction.     

Projecting on-time performance for deterministic schedules in dynamic environments

Manufacturing in a world class environment demands a high level of customer service. The production control department is responsible for achieving this high level of service through accurate planning and scheduling. The ability to achieve such a high level of customer service is limited by the scheduling tools currently available. Production planning is typically performed using MRP infinite capacity, fixed lead-time, backward scheduling. The work in each MRP time-bucket is then sequenced to develop a schedule. The production floor however, is not a static environment. Dynamic events, that cannot be scheduled, degrade production performance relative to the projected schedule. In this paper the relationship between dynamic events and schedule degradation is examined. Common approaches to production scheduling underestimate the effect of capacity loading relative to unplanned events and schedule achievability. These dynamic events exhaust capacity previously allocated to production orders. To hedge against such known, but unscheduled events, production control can schedule resources to a level less than full capacity. The size of the capacity hedge and the duration of the unplanned event dictate the time to recover from the backlog created by these dynamic events. A performance metric is developed to measure the ability to achieve customer promise dates. A machine loading policy is also presented to achieve the optimal capacity hedge point that will maximize this performance measure. The results are compared to simulated failures to examine the accuracy of the predicted performance degradation. The results suggest a trade-off of throughput for improved performance to customer promise date.     

On serial and parallel information flows,incorporated in a multi-stage production-inventory-system

We consider a chain of production operations separated by inventories, where two different systems-for replenishing the inventories at each stage may be adopted. In the first, the serial information-flow system, each stage works against orders placed only by the following stage. In the second system, called the parallel information-flow system, every stage in the network works against the actual customer demand, materializing at the finished-goods stage.

The basic objective is to show that in the former system the levels of base stock to be maintained at the different stages are more sensitive to changes in the customer demand than in the second system. In fact, in the former case, base-stock fluctuations become larger and larger, progressively back in the chain. Consequently, the production rates also behave in the same way and their variations got multiplied successively back in the chain.     

Quantitative analysis of semiconductor supply chain contracts with order flexibility under demand uncertainty: A case study

The evidence base for the configuration of rolling horizon flexibility (RHF) contracts (a type of quantity flexibility contract) used in the semiconductor industry to coordinate production and demand remains meagre, more art than science. Informed by the characteristics of actual clauses and demand behaviors drawn from a company’s experience, a discrete-event simulation model is developed to represent the company’s supply chain. It comprises of three parties: a customer, a supplier (semiconductor manufacturer), and a capacity provider. Through analysis of customer forecasted demand the paper characterizes forecast demand as being under, over or unbiased. Models of these forecasted demands drives both long and short term planning. In long term planning, which is given twelve months before an order is delivered, capacity at the capacity provider is booked. Short term planning is also driven by this forecast which, within a binding period, is governed by an RHF contract. Results from the model report inventory levels, and delivery compliance, namely Delivery Performance (DP) and Delivery Reliability (DR), measures widely used in this sector. It is concluded from this work that on the balance of performance measures RHF contracts with asymmetrical flexibility bounds are substantially better than those with symmetrical boundaries, and that this conclusion is robust with regard to both over-planning and under-planning behaviors. This robustness is a critical attribute with respect to the endemic medium-term vacillation between both states experienced in practice in this sector.     

A multi-level incentive model for service organizations

Incentive programmes for production employees have been used for many years. However, more recently, there has been an increasing desire to use incentive programmes in service applications, particularly with the advent of electronic performance monitoring. This paper summarizes the methodology for developing a model that incorporates the parameters of quality (customer and supervisor evaluations), quantity of production, sales and individual goals. The model was used to compute incentive bonus payments for customer service employees who performed above standard. Employees were motivated by the incentive bonus, and thus productivity was increased. The model could be applied to any service organization. The model was developed to assist a customer service department where performance standards, quality measures or customer feedback were not firmly established. It measured objectively employee performance and rewarded the employee with individually calculated incentive bonuses through an electronically monitored system. The internal working of the model is a combination of the above parameters. Customized software was used to collect and tabulate customer feedback regarding a recent inquiry to the customer service department. This made up the first part of the quality measure for the department. The second part of the quality parameter was a regular evaluation by the supervisor. The quantity of production was measured by the number of inquiries handled per unit time. Sales were also calculated on a per-unit-time basis. The individual goals were the result of regular employee-supervisor meetings where past and future employee performance was discussed and mutually agreed upon. The five model parameters of quality (customer and supervisor), quantity, sales and individual goals are combined through the use of a linear equation. It was found that this form of employee evaluation worked very well in an electronic monitoring environment and was accepted by the participating work group. Overall, monthly productivity gains exceeded 15%.     

Measure the quality level for a supplier-demand system by a multicommodity stochastic-flow network

From the point of view of quality management, it is important to meet the customer's demand. The probability that the system can satisfy the customer's demand is an important performance index, and can be used to measure the quality level of the system. In this paper, we use a multicommodity stochastic-flow network to describe the relationship between the supplier and the customer. Each node as well as each arc has several possible capacities and may fail. The network allows multiple types of commodities to be transmitted from the source to the sink. Given the demand for each commodity at the sink, evaluation of the probability that the system meets the demands is performed. Such a probability, named the system reliability, is a performance index of quality level. At first, a simple algorithm is proposed to generate all lower boundary points for the demand, and the system reliability can be calculated in terms of such points. The computational complexity of the proposed algorithm is polynomial time in number of arcs, nodes and minimal paths.     

Coordination between strategic forest management and tactical logistic and production planning in the forestry supply chain

In this paper, we study the coordination mechanism in the forestry supply chain between strategic forest management and tactical production planning. We first formulate an integrated model to establish a theoretical benchmark for performance of the entire supply chain. It is a mixed integer programming model that involves harvesting, bucking, transportation, production, and sales decisions for both tactical and strategic planning levels. We then present two sequential approaches S‐A and S‐B where the coordination is done through internal pricing. S‐A is the approach currently used in practice where harvesting in the forest is the main driver of the supply chain activities and internal pricing is introduced to control bucking decision in a separate stage. In contrast, S‐B takes downstream demand information into consideration and internal pricing directly influences harvesting decision in the first stage. In order to find the appropriate setting of internal pricing that leads to the system optimum, we suggest two heuristics H‐I and H‐II. The internal pricing in H‐I is based on dual values and in H‐II, it is derived from a Lagrangian decomposition. A real‐life case study in the Chilean forestry industry is used to compare the results of different approaches. It is shown that the new sequential approach S‐B generates as good feasible solution as that obtained from the integrated approach but in much less time. Both heuristics H‐I and H‐II bring about near‐optimal feasible solutions. H‐II also provides optimistic bound of the optimal objective function value, which can be used as a measure of the solution quality.     

Measuring Information Technology's Indirect Impact on Firm Performance

It has been recognized that the link between information technology (IT) investment and firm performance is indirect due to the effect of mediating and moderating variables. For example, in the banking industry, the IT-value added activity helps to effectively generate funds from the customer in the forms of deposits. Profits then are generated by using deposits as a source of investment funds. Traditional efficiency models, such as data envelopment analysis (DEA), can only measure the efficiency of one specific stage when a two-stage production process is present. We develop an efficiency model that identifies the efficient frontier of a two-stage production process linked by intermediate measures. A set of firms in the banking industry is used to illustrate how the new model can be utilized to (i) characterize the indirect impact of IT on firm performance, (ii) identify the efficient frontier of two principal value-added stages related to IT investment and profit generation, and (iii) highlight those firms that can be further analyzed for best practice benchmarking.     

Development and evaluation of a Basestock-CONWIP pull production control strategy in balanced assembly systems

In this study, a new pull production control strategy called Basestock-Constant Work-in-Process (B-CONWIP) is proposed. It is used to control the flow of materials and information in balanced assembly production systems. This proposed control strategy uses one type of authorization cards called CONWIP card that limits the work-in-process (WIP) in the whole system. It has been applied in a single-product and a mixed-product assembly system balanced by two efficient Genetic algorithms introduced in literature. The performance of this control strategy is compared with another pull production control strategy called Basestock Kanban CONWIP (BK-CONWIP), which is a very promising production control strategy found in literature. The proposed strategy has two control parameters, CONWIP authorization cards and basestock levels while BK-CONWIP has three control parameters Kanban authorization cards, CONWIP authorization cards and basestock levels. The comparison is based on three performance measures average system WIP, percentage of satisfied customer demand (service level) and WIP variation between workstations. The performance of the proposed strategy B-CONWIP and BK-CONWIP is mainly similar in both types of assembly systems when mean demand rates are low with respect to mean service rates with the proposed strategy being easier to control and optimize. On the other hand, when mean demand rates are high with respect to mean service rates; B-CONWIP is preferable if service level is more important, while BK-CONWIP is preferable if WIP level is more important. Regarding WIP variation, it mainly depends on the efficiency of the balancing approach. The more efficient the balancing approach, the less WIP variation. Treating demand as lost instead of backordered results in decreased average system WIP and does not affect service levels in both PCSs. It is also shown that S-KDP is more flexible in dealing with situations of variable product mixes than d-KDP because control parameters can be used by any product which minimizes the effect of the unbalanced systems.     

Using genetic algorithm for lot sizing and scheduling problem with arbitrary job volumes and distinct job due date considerations

This paper considers an integrated lot sizing and scheduling problem for a production–distribution environment with arbitrary job volumes and distinct due dates considerations. In the problem, jobs are firstly batch processed on a batching machine at production stage and then delivered to a pre-specified customer at the subsequent delivery stage by a capacitated vehicle. Each job is associated with a distinct due date and a distinct volume, and has to be delivered to the customer before its due date, i.e. delay is not allowed. The processing time of a batch is a constant independent of the jobs it contains. In production, a constant set-up time as well as a constant set-up cost is required before the first job of this batch is processed. In delivery, a constant delivery time as well as a constant delivery cost is needed for each round-trip delivery between the factory and the customer. Moreover, it is supposed that a job that arrives at the customer before its due date will incur a customer inventory cost. The objective is to find a coordinated lot sizing and scheduling scheme such that the total cost is minimised while guaranteeing a certain customer service level. A mixed integer formulation is proposed for this problem, and then a genetic algorithm is developed to solve it. To evaluate the performance of the proposed genetic algorithm, a lower bound on the objective value is established. Computational experiments show that the proposed genetic algorithm performs well on randomly generated problem instances.     

Utilization of a reinforcement learning algorithm for the accurate alignment of a robotic arm in a complete soft fabric shoe tongues automation process

As usher in Industry 4.0, there has been much interest in the development and research that combine artificial intelligence with automation. The control and operation of equipment in a traditional automated shoemaking production line require a preliminary subjective judgment of relevant manufacturing processes, to determine the exact procedure and corresponding control settings. However, with the manual control setting, it is difficult to achieve an accurate quality assessment of an automated process characterized by high uncertainty and intricacy. It is challenging to replace handicrafts and the versatility of manual product customization with automation techniques. Hence, the current study has developed an automatic production line with a cyber-physical system artificial intelligence (CPS-AI) architecture for the complete manufacturing of soft fabric shoe tongues. The Deep-Q reinforcement learning (RL) method is proposed as a means of achieving better control over the manufacturing process, while the convolutional and long short-term memory artificial neural network (CNN + LSTM) is developed to enhance action speed. This technology allows a robotic arm to learn the specific image feature points of a shoe tongue through repeated training to improve its manufacturing accuracy. For validation, different parameters of the network architecture are tested, and the test convergence accuracy was found to be as high as 95.9 %. During its actual implementation, the production line completed 509 finished products, of which 349 products were acceptable due to the anticipated measurement error. This showed that the production line system was capable of achieving optimum product accuracy and quality with respect to the performance of repeated computations, parameter updates, and action evaluations.     

The integrated production–inventory–distribution–routing problem

The integration of production and distribution decisions presents a challenging problem for manufacturers trying to optimize their supply chain. At the planning level, the immediate goal is to coordinate production, inventory, and delivery to meet customer demand so that the corresponding costs are minimized. Achieving this goal provides the foundations for streamlining the logistics network and for integrating other operational and financial components of the system. In this paper, a model is presented that includes a single production facility, a set of customers with time varying demand, a finite planning horizon, and a fleet of vehicles for making the deliveries. Demand can be satisfied from either inventory held at the customer sites or from daily product distribution. In the most restrictive case, a vehicle routing problem must be solved for each time period. The decision to visit a customer on a particular day could be to restock inventory, meet that day’s demand or both. In a less restrictive case, the routing component of the model is replaced with an allocation component only. A procedure centering on reactive tabu search is developed for solving the full problem. After a solution is found, path relinking is applied to improve the results. A novel feature of the methodology is the use of an allocation model in the form of a mixed integer program to find good feasible solutions that serve as starting points for the tabu search. Lower bounds on the optimum are obtained by solving a modified version of the allocation model. Computational testing on a set of 90 benchmark instances with up to 200 customers and 20 time periods demonstrates the effectiveness of the approach. In all cases, improvements ranging from 10–20% were realized when compared to those obtained from an existing greedy randomized adaptive search procedure (GRASP). This often came at a three- to five-fold increase in runtime, however.     

高性能互联网络交换机研究与设计

高性能互联网络交换机是高性能计算机系统的核心部件.科学计算作为高性能计算机的上层应用,不仅要求交换机具有低延迟、高带宽的特性,还要求其在集合通信如广播、多播和同步操作等进行硬件级支持.HyperLink交换机,作为曙光5000计算机系统互联网络的重要组成部件,具有38.4ns单级延迟和160Gbps聚合带宽,并能够同时支持16组多播和16组同步操作.理想情况下,1024个节点多播和同步操作可以在2μs内完成,大大加速了科学计算的性能.为了对HyperLink交换机性能进行评价,建立了周期精确的仿真模型.通过模拟证明,对于16端口输入缓冲交换机,3个虚通道是性价比最好的选择;当MTU为1KB时,4KB大小的输入缓冲就可达到最高单播吞吐率.采用理论分析的方法比较了具有相同网络带宽的多轨网络和单轨网络,分析表明,前者可以有效降低网络延迟,因此能够比后者提供更高的网络吞吐率.采用LogP模型分析了HyperLink多播和Barrier的性能,分析表明,HyperLink交换机具有良好扩展性,能够很好支持到数千节点.     

Pricing and inventory management in a system with multiple competing retailers under (r, Q) policies

We consider a multi-retailer system operated on an infinite horizon, in which each retailer faces stochastic demand following a Poisson process and adopts a continuous-review (r, Q) policy for replenishing inventory to satisfy customer demand. The system involves decisions of pricing and inventory management with the goal of maximizing profit, which equals the sales revenue minus the purchase and inventory costs. Taking Cournot competition into account, models are formulated to optimize simultaneously the expected sales volumes and (r, Q) policies of all retailers. An efficient approach is proposed to calculate the approximate inventory cost. Based on this approach, solution methods for centralized and decentralized scenarios are developed. A great number of numerical computations are provided to evaluate the efficiency of the solution methods, and their performance in the two scenarios. Moreover, system performance under sequential decisions (first pricing and then inventory management) is also investigated.     

Optimal Measurement-based Pricing for an M/M/1 Queue

In this paper, we consider a system modelled as an M/M/1 queue. Jobs corresponding to different classes are sent to the queue and are characterized by a delay cost per unit of time and a demand function. Our goal is to design an optimal pricing scheme for the queue, where the total charge depends on both the mean delay at the queue and arrival rate of each customer. We also assume that those two values have to be (statistically) measured, introducing errors on the total charge that might avert jobs from using the system, and then decrease demand. This model can be applied in telecommunication networks, where pricing can be used to control congestion, and the network can be characterized by a single bottleneck queue; the throughput of each class would be determined through passive measurements while the delay would be determined through active measurements.     

Exact and metaheuristic algorithms for the vehicle routing problem with a factory-in-a-box in multi-objective settings

Emergencies, such as pandemics (e.g., COVID-19), warrant urgent production and distribution of goods under disrupted supply chain conditions. An innovative logistics solution to meet the urgent demand during emergencies could be the factory-in-a-box manufacturing concept. The factory-in-a-box manufacturing concept deploys vehicles to transport containers that are used to install production modules (i.e., factories). The vehicles travel to customer locations and perform on-site production. Factory-in-a-box supply chain optimization is associated with a wide array of decisions. This study focuses on selection of vehicles for factory-in-a-box manufacturing and decisions regarding the optimal routes within the supply chain consisting of a depot, suppliers, manufacturers, and customers. Moreover, in order to contrast the options of factory-in-a-box manufacturing with those of conventional manufacturing, the location of the final production is determined for each customer (i.e., factory-in-a-box manufacturing with production at the customer location or conventional manufacturing with production at the manufacturer locations). A novel multi-objective optimization model is presented for the vehicle routing problem with a factory-in-a-box that aims to minimize the total cost associated with traversing the edges of the network and the total cost associated with visiting the nodes of the network. A customized multi-objective hybrid metaheuristic solution algorithm that directly considers problem-specific properties is designed as a solution approach. A case study is performed for a vaccination project involving factory-in-a-box manufacturing along with conventional manufacturing. The case study reveals that the developed solution method outperforms the ε-constraint method, which is a classical exact optimization method for multi-objective optimization problems, and several well-known metaheuristics.     

Planning tools for managing the supply chain

In this paper, we introduce a new way to manage the supply chain. The proposed solution reduces the problem's complexity using a two-stage hierarchical production planning method and is applicable to realistic transportation optimization problems. The approach is based on planning and operations scheduling models, and is designed to minimize travel and production costs within a flexible organizational network. In the aggregate planning phase, a mathematical model involving an aggregation of products, demand and time periods is solved. It is at this initial stage that the size of the problem is reduced and its output is used as input to the detailed phase in order to improve resolution time. The second stage produces a detailed schedule. It is shown that the proposed approach generates good and feasible solutions to practical problems within a reasonable computational time.     

The impact of replenishment parameters and information sharing on the bullwhip effect: A computational study

Demand variability amplification across the supply chain, known as the bullwhip effect, results in serious inefficiencies across the chain. Managers are expected to minimize this phenomenon in their chain in order to reduce costs and increase customer satisfaction by making critical decisions on replenishment policy. We study how specific replenishment parameters affect order variability amplification, product fill rates and inventory levels across the chain. Furthermore, we study how demand information sharing can help towards reducing order oscillations and inventory levels in upper nodes of a supply chain. A two-stage supply chain consisting of a warehouse and stores that face customer demand is modeled. Real demand data are used as the underlying customer demand during the experiments.