1,721,000 research outputs found
An entity embeddings deep learning approach for demand forecast of highly differentiated products
Full text linkThe paper deals with Deep Learning architectures applied to demand forecasting in a complex environment. The focus is on a famous Italian Fashion Company, which periodically performs a sales campaign, to presents its new products' line and to collect customers' orders. Although production follows an MTO strategy, fabrics must be purchased in advance and a forecasting system is required to predict the total quantity sold for each product, at the early stages of the campaign. Due to high product variability, the forecasting system must consider products' similarities and the evolution of customers taste. Additionally, customer and product data are mostly described by categorical variables (hard to reconcile with a predictive task) and, unfortunately, time-series techniques cannot be used because of a sparse dataset. Given these criticalities, we propose an end-to-end approach based on Deep Neural Networks and on Entity Embeddings. A first neural network is trained to predict the total quantity of a given product ordered by a specific customer. Different Embeddings are learned for each customer and product categorical attribute. This gives the network the ability to effectively learn the complex and evolving relationships between products characteristics and customers taste. Next, freezing the learned product's embeddings, a second Recurrent Neural Network is trained to predict the total amount ordered for a given product, incorporating real-time data of customers' orders of the ongoing sales campaign. Ten years of sales have been analyzed and the approach, tested on unseen sales campaigns, has outperformed the forecasting algorithm currently adopted by the fashion firm
Exploiting Machine Learning and Industry 4.0 traceability technologies to re-engineering the seasoning process of traditional Parma’s Ham
No full textThe work presents a Machine Learning approach for predicting the quality of the curing process of Parma ham, combined with a study of business process re-engineering, based on RFID and Deep Learning technologies for automatic recognition and tracking of the hams along the curing process. Quality management has proven to be crucial for efficient and effective processes, even more so for the food industry, both for commercial and regulatory purposes. This is even more evident in artisanal-based processes, such as the one concerning traditional Prosciutto di Parma seasoning. The work proposes and compares a Feed-Forward Neural Network and a Random Forest for predicting the distribution of the number of hams by commercial quality class of a given aging lot. Such a prediction, based on origin, process, and curing data, can provide early indications of process output, enabling strategic commercial competitive advantages. The importance of the genetic component in the determination of the final quality is also evaluated, as it is considered one of the most influential external variables. Moreover, following the AS-IS description of the current process, a redesign is proposed, to enable data collection and tracking of individual ham in order to propose a future precision prediction system that would allow even finer control of the process
Defining accurate delivery dates in make to order job-shops managed by workload control
Full text linkWorkload control (WLC) is a lean oriented system that reduces queues and waiting times, by imposing a cap to the workload released to the shop floor. Unfortunately, WLC performance does not systematically outperform that of push operating systems, with undersaturated utilizations levels and optimized dispatching rules. To address this issue, many scientific works made use of complex job-release mechanisms and sophisticated dispatching rules, but this makes WLC too complicated for industrial applications. So, in this study, we propose a complementary approach. At first, to reduce queuing time variability, we introduce a simple WLC system; next we integrate it with a predictive tool that, based on the system state, can accurately forecast the total time needed to manufacture and deliver a job. Due to the non-linearity among dependent and independent variables, forecasts are made using a multi-layer-perceptron; yet, to have a comparison, the effectiveness of both linear and non-linear multi regression model has been tested too. Anyhow, if due dates are endogenous (i.e. set by the manufacturer), they can be directly bound to this internal estimate. Conversely, if they are exogenous (i.e. set by the customer), this approach may not be enough to minimize the percentage of tardy jobs. So, we also propose a negotiation scheme, which can be used to extend exogenous due dates considered too tight, with respect to the internal estimate. This is the main contribution of the paper, as it makes the forecasting approach truly useful in many industrial applications. To test our approach, we simulated a 6-machines job-shop controlled with WLC and equipped with the proposed forecasting system. Obtained performances, namely WIP levels, percentage of tardy jobs and negotiated due dates, were compared with those of a set classical benchmark, and demonstrated the robustness and the quality of our approach, which ensures minimal delays
Deep learning and WLC: How to set realistic delivery dates in high variety manufacturing systems
Full text linkThe focus is on workload control, a production planning and control technique that reduces and stabilizes the total throughput time. In these conditions, defining realistic delivery dates should become easier, yet the use of basic techniques often proves to be ineffective. Hence, we propose using statistical and/or neural network techniques to estimate, starting from the current workload of the job shop, the expected lead time of entry jobs, and to use this estimation to define reliable delivering dates. To test the approach, we simulated a 6-machines job-shop and we make predictions using a multi-regressive linear model and a multi-layer neural network. In terms of tardy jobs, both approaches performed very well, with the neural network providing the best results
Cycle time calculation of shuttle-lift-crane automated storage and retrieval system
No full textThis article deals with cycle time calculation of Automated Storage and Retrieval Systems (AS/RS). Cycle time has a high impact on the operating performance of an AS/RS, and its knowledge is essential, both at the operational and design level. The novelty of this work concerns the peculiar kind of system that is considered, as the focus is on the Shuttle-Lift-Crane AS/RS. This solution, common in the steel sector, is used to store bundles of long metal bars, which are automatically handled by cranes, lifts, and shuttles. The functioning of these machines, which can operate in parallel and independently, is stochastically modeled, and the probability distribution function of the cycle time is computed, both for single and dual command cycles. The model, assessed via discrete event simulation, ensures a high average accuracy of 96% and 98%, under single and dual command cycles, respectively
Allocation of items considering unit loads balancing and joint retrieving
No full textIn the last years, the diffusion of lean thinking had a big impact, not only in manufacturing, but in logistics too. Because of one-piece-flow production and the point of view on inventory that considers it as inefficiency, purchasing and shipping batches have become smaller and more varied, requiring to the suppliers more shipments per day, a shorter throughput time, and, in general, higher performances. To improve retrieving performance in automated warehouses, many routing and scheduling procedures are presented in literature, although retrieving can be speeded up starting from the input phase using a correct allocation policy. In this paper, we present a procedure inspired by Genetic Algorithm (GA) for allocation of items inside unit loads. The procedure considers two aspects that are hardly studied in literature, such as unit load weight balancing and market basket analysis aimed at closed allocation of items that are usually jointly retrieved. The first one is a physical necessity, especially required in the steel sector, where objects stocked are heavy. The second one improves the retrieving performance and it increases the possibility to satisfy more order lines with fewer travels. The algorithm proposed was tested using the digital twin of an existing warehouse and comparing the results with the current performances of the real system
A new perspective on Workload Control by measuring operating performances through an economic valorization
Full text linkWorkload Control (WLC) is a production planning and control system conceived to reduce queuing times of job-shop systems, and to offer a solution to the lead time syndrome; a critical issue that often bewilders make-to-order manufacturers. Nowadays, advantages of WLC are unanimously acknowledged, but real successful stories are still limited. This paper starts from the lack of a consistent way to assess performance of WLC, an important burden for its acceptance in the industry. As researchers often put more focus on the performance measures that better confirm their hypotheses, many measures, related to different WLC features, have emerged over years. However, this excess of measures may even mislead practitioners, in the evaluation of alternative production planning and control systems. To close this gap, we propose quantifying the main benefit of WLC in economic terms, as this is the easiest, and probably only way, to compare different and even conflicting performance measures. Costs and incomes are identified and used to develop an overall economic measure that can be used to evaluate, or even to fine tune, the operating features of WLC. The quality of our approach is finally demonstrated via simulation, considering the 6-machines job-shop scenario typically adopted as benchmark in technical literature
Hybrid heuristic for the one-dimensional cutting stock problem with usable leftovers and additional operating constraints
Full text linkThe One-Dimensional Cutting Stock Problem consists in cutting long bars into smaller ones, to satisfy customers’ demand, minimizing waste and cost. In this paper the standard problem is extended with the inclusion of additional constraints that are generally neglected in scientific literature, although relevant in many industrial applications. We also modified the standard objective function, by assuming that bars may have a different economical value and a different processing or shipping priority. Moreover, in line with business requirements, among solutions that generate the same cutting waste, we prefer the ones that generate a low number of leftovers, especially if leftovers are long, so that the likelihood of their reuse is high. To solve the problem, we propose a Simulated Annealing based heuristic, which exploits a specific neighbor search. The heuristic is implemented in a parametric way that allows the user to set the priorities of the bars and to choose the specific sub-set of constraints he or she wants to consider. The heuristic is finally tested on many problem instances, and it is compared to three benchmarks and to one commercial software. The outcomes of this comparative analysis demonstrate both its quality and effectiveness
Comparison of new metaheuristics, for the solution of an integrated jobs-maintenance scheduling problem
Full text linkThis paper presents and compares new metaheuristics to solve an integrated jobs-maintenance scheduling problem, on a single machine subjected to aging and failures. The problem, introduced by Zammori et al. (2014), was originally solved using the Modified Harmony Search (MHS) metaheuristic. However, an extensive numerical analysis brought to light some structural limits of the MHS, as the analysis revealed that the MHS is outperformed by the simpler Simulated Annealing by Ishibuchi et al. (1995). Aiming to solve the problem in a more effective way, we integrated the MHS with local minima escaping procedures and we also developed a new Cuckoo Search metaheuristic, based on an innovative Levy Flight. A thorough comparison confirmed the superiority of the newly developed Cuckoo Search, which is capable to find better solutions in a smaller amount of time. This an important result, both for academics and practitioners, since the integrated job-maintenance scheduling problem has a high operational relevance, but it is known to be extremely hard to be solved, especially in a reasonable amount of time. Also, the developed Cuckoo Search has been designed in an extremely flexible way and it can be easily readapted and applied to a wide range of combinatorial problems. (C) 2018 Elsevier Ltd. All rights reserved
Development and Stress Test of a New Serious Game for Food Operations and Supply Chain Management: Exploring Students’ Responses to Difficult Game Settings
No full textSerious games (SGs) in engineering education are a proven topic, whose implementation has been significantly growing in the last decades. They are recognized as effective tools to teach and learn subjects like Operations and Supply Chain Management. The research on SGs, however, is primarily focused on displaying applications and teaching results of particular games to achieve given purposes. In this paper, we provide an exploratory research and a stress test of a new SG on a specific target group in the field of food operations and supply chain management. We provide an overview of the SG and detail its mechanics. Also, we explain how the mechanics has been implemented, by means of a set of parameters and indicators that better explain the roles available to players in the game. We conclude by reporting and discussing the results of a game session played by a class of Vocational Education and Training students under stress conditions generated by an accelerated game time
- …
