Research Project: H2020 Industrial Leadership - Leadership in Enabling and Industrial Technologies-Public - private leadership - The Joint Technology Initiative on Electronic Components and Systems for European Leadership (ECSEL)
Loading...
Contributors
Funders
ID
EC.00101
Authors
Tan, Barış
Faculty Member
Publications
Supervised learning-based approximation method for single-server open queueing networks with correlated interarrival and service times
(Taylor _ Francis, 2021) Khayyati, Siamak; Tan, Barış; Department of Business Administration; Department of Industrial Engineering; Graduate School of Sciences and Engineering; Yes; College of Administrative Sciences and Economics; College of Engineering; GRADUATE SCHOOL OF SCIENCES AND ENGINEERING
Efficient performance evaluation methods are needed to design and control production systems. We propose a method to analyse single-server open queueing network models of manufacturing systems composed of delay, batching, merge and split blocks with correlated interarrival and service times. Our method (SLQNA) is based on using a supervised learning approach to determine the mean, the coefficient of variation, and the first-lag autocorrelation of the inter-departure time process as functions of the mean, coefficient of variation and first-lag autocorrelations of the interarrival and service times for each block, and then using the predicted inter-departure time process as the input to the next block in the network. The training data for the supervised learning algorithm is obtained by simulating the systems for a wide range of parameters. Gaussian Process Regression is used as a supervised learning algorithm. The algorithm is trained once for each block. SLQNA does not require generating additional training data for each unique network. The results are compared with simulation and also with the approximations that are based on Markov Arrival Process modelling, robust queueing, and G/G/1 approximations. Our results show that SLQNA is flexible, computationally efficient, and significantly more accurate and faster compared to the other methods.
Modelling and analysis of the impact of correlated inter-event data on production control using Markovian arrival processes
(Springer, 2019) Tan, Barış; Dizbin, Nima Manafzadeh; Department of Business Administration; Department of Industrial Engineering; Graduate School of Business; Yes; College of Administrative Sciences and Economics; College of Engineering; GRADUATE SCHOOL OF BUSINESS
Empirical studies show that the inter-event times of a production system are correlated. However, most of the analytical studies for the analysis and control of production systems ignore correlation. In this study, we show that real-time data collected from a manufacturing system can be used to build a Markovian arrival processes (MAP) model that captures correlation in inter-event times. The obtained MAP model can then be used to control production in an effective way. We first present a comprehensive review on MAP modeling and MAP fitting methods applicable to manufacturing systems. Then we present results on the effectiveness of these fitting methods and discuss how the collected inter-event data can be used to represent the flow dynamics of a production system accurately. In order to study the impact of capturing the flow dynamics accurately on the performance of a production control system, we analyze a manufacturing system that is controlled by using a base-stock policy. We study the impact of correlation in inter-event times on the optimal base-stock level of the system numerically by employing the structural properties of the MAP. We show that ignoring correlated arrival or service process can lead to overestimation of the optimal base-stock level for negatively correlated processes, and underestimation for the positively correlated processes. We conclude that MAPs can be used to develop data-driven models and control manufacturing systems more effectively by using shop-floor inter-event data.
Data-driven control of a production system by using marking-dependent threshold policy
(Elsevier, 2020) Khayyati, Siamak; Tan, Barış; Department of Business Administration; Graduate School of Sciences and Engineering; Yes; College of Administrative Sciences and Economics; GRADUATE SCHOOL OF SCIENCES AND ENGINEERING
As increasingly more shop-floor data becomes available, the performance of a production system can be improved by developing effective data-driven control methods that utilize this information. We focus on the following research questions: how can the decision to produce or not to produce at any time be given depending on the real-time information about a production system?; how can the collected data be used directly in optimizing the policy parameters?; and what is the effect of using different information sources on the performance of the system? In order to answer these questions, a production/inventory system that consists of a production stage that produces to stock to meet random demand is considered. The system is not fully observable but partial production and demand information, referred to as markings is available. We propose using the marking-dependent threshold policy to decide whether to produce or not based on the observed markings in addition to the inventory and production status at any given time. An analytical method that uses a matrix geometric approach is developed to analyze a production system controlled with the marking-dependent threshold policy when the production, demand, and information arrivals are modeled as Marked Markovian Arrival Processes. A mixed integer programming formulation is presented to determine the optimal thresholds. Then a mathematical programming formulation that uses the real-time shop floor data for joint simulation and optimization (JSO) of the system is presented. Using numerical experiments, we compare the performance of the JSO approach to the analytical solutions. We show that using the marking-dependent control policy where the policy parameters are determined from the data works effectively as a data-driven control method for manufacturing.
Optimal control of production-inventory systems with correlated demand inter-arrival and processing times
(Elsevier, 2020) Tan, Barış; Dizbin, Nima Manafzadeh; Department of Business Administration; Graduate School of Business; Department of Industrial Engineering; Yes; College of Administrative Sciences and Economics; College of Engineering; GRADUATE SCHOOL OF BUSINESS
We consider the production control problem of a production-inventory system with correlated demand inter-arrival and processing times that are modeled as Markovian Arrival Processes. The control problem is minimizing the expected average cost of the system in the steady-state by controlling when to produce an available part. We prove that the optimal control policy is the state-dependent threshold policy. We evaluate the performance of the system controlled by the state-dependent threshold policy by using the Matrix Geometric method. We determine the optimal threshold levels of the system by using policy iteration. We then investigate how the autocorrelation of the arrival and service processes impact the performance of the system. Finally, we compare the performance of the optimal policy with 3 benchmark policies: a state-dependent policy that uses the distribution of the inter-event times but assumes i.i.d.inter-event times, a single-threshold policy that uses both the distribution and also the autocorrelation, and a single-threshold policy that uses the distribution of the inter-event times but assumes they are not correlated. Our analysis demonstrates that ignoring autocorrelation in setting the parameters of the production policy causes significant errors in the expected inventory and backlog costs. A single-threshold policy that sets the threshold based on the distribution and also the autocorrelation performs satisfactorily for systems with negative autocorrelation. However, ignoring positive correlation yields high errors for the total cost. Our study shows that an effective production control policy must take correlations in service and demand processes into account.
An integrated data-driven method using deep learning for a newsvendor problem with unobservable features
(Elsevier, 2022) Karaesmen, Fikri; Pirayesh Neghab, D.; Khayyati, S.; Department of Industrial Engineering; Yes; College of Engineering
We consider a single-period inventory problem with random demand with both directly observable and unobservable features that impact the demand distribution. With the recent advances in data collection and analysis technologies, data-driven approaches to classical inventory management problems have gained traction. Specially, machine learning methods are increasingly being integrated into optimization problems. Although data-driven approaches have been developed for the newsvendor problem, they often consider learning from the available data and optimizing the system separate tasks to be performed in sequence. One of the setbacks of this approach is that in the learning phase, costly and cheap mistakes receive equal attention and, in the optimization phase, the optimizer is blind to the confidence of the learner in its estimates for different regions of the problem. To remedy this, we consider an integrated learning and optimization problem for optimizing a newsvendor's strategy facing a complex correlated demand with additional information about the unobservable state of the system. We give an algorithm based on integrating optimization, neural networks and hidden Markov models and use numerical experiments to show the efficiency of our method. In an empirical experiment, the method outperforms the best competitor benchmark by more than 27%, on average, in terms of the system cost. We give further analyses of the performance of the method using a set of numerical experiments.
