Abstract
Strong interactions between decisions in the maintenance and production scheduling domains, and their impacts on the equipment yield rates necessitate maintenance and production decisions being optimized concurrently, with considerations of yield dependencies on the equipment conditions and production rates. This paper proposes an integrated decision-making policy for production and maintenance operations on a single machine under uncertain demand, with concurrent considerations and learning of yield dependencies on the equipment conditions and production rates. This policy is obtained through a two-stage stochastic programming model, which considers the variable demand, machine degradation, and maintenance times. This model incorporates outsourcing decisions and operational decisions regarding reworking, scraping of imperfect products to ensure the demand is adequately met. A closed-form reinforcement learning method is utilized to learn yield dependencies. Simulations confirm the necessity of yield learning and show the proposed method outperforms the traditional, fragmented approaches where the effects of production rates and machine conditions on the resulting yield rates are not considered. The two-stage stochastic setting is demonstrated by comparing with the traditional one-stage deterministic approach, where decisions are made based on the expected demand and production performance, with scrapping, reworking, and outsourcing decisions established before the demand and production performance are observed.
Similar content being viewed by others
Availability of data and material
The authors declare that the data supporting the findings of this study are available within the article.
Code availability
The code that supports the findings of this study are available from the corresponding author upon reasonable request.
Notes
Note, degradation is modeled as a unidirectional process and hence working condition of a machine can only remain the same or deteriorate.
I.e., in that case, no matter what the decision \(x_{RMO}\) is, machine will be restored to the perfect working condition after RM.
Notice that demand is modeled as a random variable with a known discrete distribution.
Excess products can also be put into inventory, but because of the holding cost this would incur, this can also be seen as equivalent to being sold at a lower price.
The model of yield rates and its learning process are explained in Sect. 3.
This cost parameter is taken to be big in order to penalize such events and prevent them from happening.
As will be seen in the simulation results, highest values of the first-stage objective function in Eq.(1) can be achieved under that assumption.
This, of course, is ideal and unrealistic.
I.e. when it does not change with the underlying machine condition, nor with the production rates.
References
Aktekin T, Ekin T (2016) Stochastic call center staffing with uncertain arrival, service and abandonment rates: a Bayesian perspective. Nav Res Logist (NRL) 63(6):460–478
Alaswad S, Xiang Y (2017) A review on condition-based maintenance optimization models for stochastically deteriorating system. Reliab Eng Syst Saf 157:54–63
Al-Turki UM, Ayar T, Yilbas BS, Sahin AZ (eds) (2014) Maintenance in manufacturing environment: an overview. In: Integrated maintenance planning in manufacturing systems. Springer, Cham, pp 5–23
Aramon Bajestani M, Banjevic D, Beck JC (2014) Integrated maintenance planning and production scheduling with Markovian deteriorating machine conditions. Int J Prod Res 52(24):7377–7400
Batun S, Maillart LM (2012) Reassessing tradeoffs inherent to simultaneous maintenance and production planning. Prod Oper Manag 21(2):396–403
Bearda T, Mertens PW, Beaudoin SP (2018) Overview of wafer contamination and defectivity. Handbook of silicon wafer cleaning technology, 3rd edn. Elsevier, Amsterdam, pp 87–149
Birge JR, Louveaux F (2011) Introduction to stochastic programming. Springer, New York
Celen M (2016). Joint maintenance and production operations decision making in flexible manufacturing systems (Doctoral dissertation). Retrieved from University of Texas Libraries (OCLC number: 979556469)
Christensen R, Johnson W, Branscum A, Hanson TE (2011) Bayesian ideas and data analysis: an introduction for scientists and statisticians. CRC Press, Boca Raton
Derman C (1970). Finite state Markovian decision processes (No. 04; T57. 83, D47.)
Djurdjanovic D, Mears L, Niaki FA, Haq AU, Li L (2018) State of the art review on process, system, and operations control in modern manufacturing. J Manuf Sci Eng 140(6):061010
Ekin T (2018) Integrated maintenance and production planning with endogenous uncertain yield. Reliab Eng Syst Saf 179:52–61
Ekin T, Polson NG, Soyer R (2014) Augmented Markov chain Monte Carlo simulation for two-stage stochastic programs with recourse. Decis Anal 11(4):250–264
Ekin T, Polson NG, Soyer R (2017) Augmented nested sampling for stochastic programs with recourse and endogenous uncertainty. Nav Res Logist (NRL) 64(8):613–627
Homem-de-Mello T, Bayraksan G (2014) Monte Carlo sampling-based methods for stochastic optimization. Surv Oper Res Manag Sci 19(1):56–85
Iravani SM, Duenyas I (2002) Integrated maintenance and production control of a deteriorating production system. IIE Trans 34(5):423–435
Khouja M, Mehrez A (1994) Economic production lot size model with variable production rate and imperfect quality. J Oper Res Soc 45(12):1405–1417
Lattimore T, Szepesvári C (2020) Bandit algorithms. Cambridge University Press, Cambridge
Russo D, Van Roy B, Kazerouni A, & Osband, I. (2017) A Tutorial on Thompson Sampling. arXiv preprint arXiv:1707.02038
Shapiro A, Dentcheva D, Ruszczyński A (2014) Lectures on stochastic programming: modeling and theory. Society for Industrial and Applied Mathematics
Sloan TW (2004) A periodic review production and maintenance model with random demand, deteriorating equipment, and binomial yield. J Oper Res Soc 55(6):647–656
Sloan T (2008) Simultaneous determination of production and maintenance schedules using in-line equipment condition and yield information. Nav Res Logist (NRL) 55(2):116–129
Sloan TW, Shanthikumar JG (2000) Combined production and maintenance scheduling for a multiple-product, single-machine production system. Prod Oper Manag 9(4):379–399
Sloan TW, Shanthikumar JG (2002) Using in-line equipment condition and yield information for maintenance scheduling and dispatching in semiconductor wafer fabs. IIE Trans 34(2):191–209
Sutton RS, Barto AG (2018) Reinforcement learning: an introduction. MIT press, Cambridge
Terwiesch C, Bohn RE (2001) Learning and process improvement during production ramp-up. Int J Prod Econ 70(1):1–19
Terwiesch C, Xu Y (2004) The copy-exactly ramp-up strategy: trading-off learning with process change. IEEE Trans Eng Manage 51(1):70–84
Thompson WR (1933) On the likelihood that one unknown probability exceeds another in view of the evidence of two samples. Biometrika 25(3/4):285–294
Thompson WR (1935) On the theory of apportionment. Am J Math 57(2):450–456
Wang H (2002) A survey of maintenance policies of deteriorating systems. Eur J Oper Res 139(3):469–489
Yano CA, Lee HL (1995) Lot sizing with random yields: a review. Oper Res 43(2):311–334
Funding
This study was funded by Department of Mechanical Engineering, University of Texas at Austin.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
In order to formulate the degradation process in the two-stage stochastic programming problem, based on the DTMC model in Sect. 2.2, constraint (4) represents a block of constraints as follows.
We use binary vectors \({{\varvec{\upxi}}}_{{{\mathbf{PRM}}}}^{{\mathbf{t}}} ,{{\varvec{\upxi}}}_{{{\mathbf{MRM}}}}^{{\mathbf{t}}} ,{{\varvec{\upxi}}}_{{{\mathbf{PM}}}}^{{\mathbf{t}}} \in {\mathbb{Z}}_{2}^{{N_{w} + N_{m} }}\) to respectively denote the uncertain perfect reactive maintenance, minimum reactive maintenance, and preventive maintenance time with elements satisfying \(\sum\limits_{j = 1}^{{N_{w} }} {\xi_{*M}^{t,j} } = 0\) and \(\sum\limits_{{j = N_{w} + 1}}^{{N_{w} + N_{m} }} {\xi_{*M}^{t,j} } = 1\). Element \(\xi_{*M}^{t,j} = 1\) indicates that the realization of maintenance time is \((N_{m} + 1 - j)\). For example, in Eq. (14), vector \({{\varvec{\upxi}}}_{{{\mathbf{PRM}}}}^{{\mathbf{t}}}\) indicates that if perfect reactive maintenance starts to be conducted at time \(t\), the realization of maintenance time is 2.
We use \(\Pr_{i,j}^{t}\) to denote the element of the realization of the Markovian transition matrix \({\mathbf{P}}_{{{\mathbf{trans}}}} ({\mathbf{x}},x_{PM} ,x_{RMO} )\) at time \(t\). Constraint (15) defines the Failure Transition Submatrix \({\mathbf{P}}_{{\mathbf{F}}} ({\mathbf{x}},x_{RMO} )\) in the Markovian transition matrix, where \(F_{qi}\) is the failure probability of the machine when it is working under production rate \(q\) and degradation level \(i\). This constraint aligns the vector of failure probability with the column determined by the realization of maintenance time. Other columns in the submatrix are zeros. The block of constraints (16)–(19) defines the Recovery Transition Submatrix \({\mathbf{P}}_{{\mathbf{R}}} (x_{PM} ,x_{RMO} )\). Constraint (16) defines the probability of restoring to the perfect working condition at time \(t\). If we decide to perform PRM, the probability is one. In the case of MRM, if the machine is working, the probability is \(\Pr_{{N_{w} + N_{m} ,1}}^{t - 1}\), if not, it equals to \(s_{1}^{t}\). Constraint (17) defines the probability of restoring to other working conditions at time \(t\). Constraint (18) and (19) define the initial probability of restoring to perfect and other working conditions (i.e. \(t = 1\)).
The second block of constraints (20)–(26) performs the transition of states to obtain the history of states of the machine over time \([T]\) represented by a binary matrix \({\mathbf{s}} \in {\mathbb{Z}}_{2}^{{(N_{w} + N_{m} ) \times T}}\). Constraint (20) defines the initial vector of state according to the decision for PM. Constraint (21)–(26) formulate the state transition over time \([T]\) using random variables \(\xi^{t}\) as a trigger of transition at each time \(t\). \({\mathbf{a}}^{{\mathbf{t}}} \in {\mathbb{Z}}_{2}^{{\left( {N_{w} + N_{m} } \right) \times \left( {N_{w} + N_{m} } \right)}}\) is a binary vector where only one element has value 1. The element \(a_{i,j}^{t} = 1\) indicates that at time \(t\), the machine is in state \(i\) and transition trigger is larger than \(\sum\limits_{j^{\prime} = 1}^{j - 1} {\Pr_{i,j^{\prime}}^{t} } ,\forall j \in [N_{w} + N_{m} ]/\{ 1\}\) or larger than 0 for \(j = 1\). \({\mathbf{b}}^{{\mathbf{t}}} \in {\mathbb{Z}}_{2}^{{\left( {N_{w} + N_{m} } \right) \times \left( {N_{w} + N_{m} } \right)}}\) is a binary vector where only one element has value 1. The element \(b_{i,j}^{t} = 1\) indicates that at time \(t\), the machine is in state \(i\) and transition trigger is smaller than or equal to \(\sum\limits_{j^{\prime} = 1}^{j} {p_{i,j^{\prime}}^{t} } ,\forall j \in [N_{w} + N_{m} ]\). \(M_{l\arg e}\) is a large positive number.
Rights and permissions
About this article
Cite this article
Zhang, H., Djurdjanovic, D. Integrated production and maintenance planning under uncertain demand with concurrent learning of yield rate. Flex Serv Manuf J 34, 429–450 (2022). https://doi.org/10.1007/s10696-021-09417-8
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10696-021-09417-8