Abstract
This paper studies the scheduling problem for the manufacturing systems with uncertain job duration, and the possibility of planning due-date quotations for critical manufacturing tasks given a fixed contingency budget. We propose a due-date quotation model to measure the risk of delay in the manufacturing process in terms of the allocated contingency budget. The risk of delay is measured in the same unit as its corresponding milestone factor such that the decision makers could directly visualize and quantify the level of risks in units of hours or days. In addition, the proposed model possesses various great properties required by a convex risk measure and it represents a minimized certainty equivalent of the overall expected risk in achieving the manufacturing due-dates. Extensive computational experiments are conducted to evaluate the model performance. The results show that our proposed model, compared to various existing methods, provides a much more balanced performance in terms of success rate of due-date achievement, due-date quotation shortfall, as well as, robustness against uncertainties. The practical applicability of the proposed models are also tested with the job scheduling problem in a real stamping industry application.
Similar content being viewed by others
References
Andreoni J, Sprenger C (2011) Uncertainty equivalents: testing the limits of the independence axiom, Technical Report. National Bureau of Economic Research
Artzner P, Delbaen F, Eber JM, Heath D (1999) Coherent measures of risk. Math Finance 9:203–228
Chen G, Zhang L, Arinez J, Biller S (2013) Energy-efficient production systems through schedule-based operations. IEEE Trans Autom Sci Eng 10:27–37
Cioffi DF, Khamooshi H (2009) A practical method of determining project risk contingency budgets. J Oper Res Soc 60(4):565–571
Cohen I, Golany B, Shtub A (2007) The stochastic time–cost tradeoff problem: a robust optimization approach. Networks 49:175–188
Demeulemeester E, Vanhoucke M, Herroelen W (2003) Rangen: a random network generator for activity-on-the-node networks. J Sched 6:17–38
Follmer H, Schied A (2002) Convex measures of risk and trading constraints. Finance Stoch 6:429–447
Hall NG, Posner ME (2001) Generating experimental data for computational testing with machine scheduling applications. Oper Res 49:854–865
Herroelen W, Leus R (2005) Project scheduling under uncertainty: survey and research potentials. Eur J Oper Res 165:289–306
Hu H, Zhou M, Li Z (2011) Supervisor optimization for deadlock resolution in automated manufacturing systems with petri nets. IEEE Trans Autom Sci Eng 8:794–804
Kerzner H (2013) Project management: a systems approach to planning, scheduling, and controlling. New York, Wiley
Klastorin T (2004) Project management: tools and trade-offs. Wiley, Hoboken
Li J, Morrison JR, Zhang MT, Nakano M, Biller S, Lennartson B (2013) Automation in green manufacturing. IEEE Trans Autom Sci Eng 10:1–4
Malcolm DG, Roseboom JH, Clark CE, Fazar W (1959) Application of a technique for research and development program evaluation. Oper Res 7:646–669
Meredith JR, Mantel SJ Jr (2011) Project management: a managerial approach. Wiley, New York
Might RJ, Fischer WA (1985) The role of structural factors in determining project management success. IEEE Trans Eng Manage 2:71–77
Miller R, Lessard D (2001) Understanding and managing risks in large engineering projects. Int J Proj Manag 19:437–443
Pach C, Berger T, Sallez Y, Trentesaux D (2015) Reactive control of overall power consumption in flexible manufacturing systems scheduling: a potential fields model. Control Eng Pract 44:193–208
Schonberger RJ (1981) Why projects are “always” late: a rationale based on manual simulation of a pert/cpm network. Interfaces 11:66–70
Shabtay D, Steiner G, Zhang R (2016) Optimal coordination of resource allocation, due date assignment and scheduling decisions. Omega 65:41–54
Sindicic I, Bogdan S, Petrovic T (2011) Resource allocation in free-choice multiple reentrant manufacturing systems based on machine-job incidence matrix. IEEE Trans Ind Inform 7:105–114
Wang Z, Pang CK, Ng TS (2018) Data-driven scheduling optimization under uncertainty using renyi entropy and skewness criterion. Comput Ind Eng 126:410–420
Yang F, Gao K, Simon IW, Zhu Y, Su R (2018) Decomposition methods for manufacturing system scheduling: a survey. IEEE/CAA J Autom Sin 5:389–400
Yue H, Xing K, Hu H, Wu W, Su H (2016) Petri-net-based robust supervisory control of automated manufacturing systems. Control Eng Pract 54:176–189
Zammori FA, Braglia M, Frosolini M (2009) A fuzzy multi-criteria approach for critical path definition. Int J Proj Manag 27:278–291
Acknowledgements
This research was supported by a grant from the National Research Foundation, Prime Minister Office, Singapore under its Energy and Environmental Sustainability for Megacities (E2S2), Campus of Research Excellence and Technological Enterprise (CREATE) programme.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This article belongs to the Topical Collection: Smart Manufacturing - A New DES Frontier
Guest Editors: Rong Su and Bengt Lennartson
This research is supported by the National Research Foundation, Prime Minister’s Office, Singapore under its Energy and Environmental Sustainability for Megacities (E2S2), Campus for Research Excellence and Technological Enterprise (CREATE) programme.
Appendix
Appendix
1.1 A.1 Proof of Theorem 1
Proof
Each of the properties will be proved in detail here.
The proposed model in Eq. 1 can be equivalently written as:
-
1.
Normalization: ψτ(0) = 0.
By definition,
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\boldsymbol{0}):=\left\{\inf\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{i}+\tau\right\}: \sum\limits_{i=1}^{m}\omega_{i}\tau_{i}=\tau, \right.\ \forall i &\in& \mathcal{M}:\\ \mathbb{E}[0-\alpha_{i}]^{+} &\leq& \tau_{i},\\ \tau_{i} &\geq& 0 \left.\vphantom{\sum\limits_{i=1}^{m}}\right\}. \end{array} $$It is clear that αi = −τi is the smallest feasible value for the constraint.
As such \(\psi _{\tau }(0)=\min \limits _{\boldsymbol {\alpha }, \boldsymbol {\tau }}\{{\sum }_{i=1}^{m}\omega _{i}(-\tau _{i})+\tau \}=(-\tau +\tau )=0\).
-
2.
Monotonicity: If \(\boldsymbol {\tilde {z}_{1}} \leq \boldsymbol {\tilde {z}_{2}}\), then \(\psi _{\tau }(\boldsymbol {\tilde {z}_{1}}) \leq \psi _{\tau }(\boldsymbol {\tilde {z}_{2}}).\)
By the definition of Eq. 1,
Model in the case of \(\boldsymbol {\tilde {z}_{1}}\):
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\boldsymbol{\tilde{z}_{1}})&=&\inf\limits_{\boldsymbol{\alpha}_{1}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{1,i}+\tau\right\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{1,i}-\alpha_{1,i}]^{+} \leq \tau_{i}, \\ &&\sum\limits_{i=1}^{m}\omega_{i}\tau_{i}=\tau,\\ &&\tau_{i} \geq 0, \ \forall i. \end{array} $$Model in the case of \(\boldsymbol {\tilde {z}_{2}}\):
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\boldsymbol{\tilde{z}_{2}})&=&\inf\limits_{\boldsymbol{\alpha}_{2}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{2,i}+\tau\right\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{2,i}-\alpha_{2,i}]^{+} \leq \tau_{i}, \\ &&\sum\limits_{i=1}^{m}\omega_{i}\tau_{i}=\tau,\\ &&\tau_{i} \geq 0, \ \forall i. \end{array} $$Suppose the optimal α values are \(\boldsymbol {\alpha }_{1}^{*}\) and \(\boldsymbol {\alpha }_{2}^{*}\) respectively. As such, \(\psi _{\tau }(\boldsymbol {\tilde {z}_{1}})={\sum }_{i=1}^{m}\omega _{i}\alpha _{1,i}^{*}+\tau \) and \(\psi _{\tau }(\boldsymbol {\tilde {z}_{2}})={\sum }_{i=1}^{m}\omega _{i}\alpha _{2,i}^{*}+\tau \). For the same τ, the feasible values satisfy \(\alpha _{1,i}^{*} \leq \alpha _{2,i}^{*}\) since function \(\mathbb {E}[\tilde {z}_{i}-\alpha _{i}]^{+}\) is non-decreasing function w.r.t \(\tilde {z}_{i}\). As such, \(\psi _{\tau }(\boldsymbol {\tilde {z}_{1}}) \leq \psi _{\tau }(\boldsymbol {\tilde {z}_{2}})\).
-
3.
Convexity: \(\forall \boldsymbol {\tilde {z}_{1}}, \boldsymbol {\tilde {z}_{2}} \in \boldsymbol {\mathcal {Z}}\) and λ ∈ [0, 1], \(\psi _{\tau }(\lambda \boldsymbol {\tilde {z}_{1}}+(1-\lambda )\boldsymbol {\tilde {z}_{2}}) \leq \lambda \psi _{\tau }(\boldsymbol {\tilde {z}_{1}})+(1-\lambda )\psi _{\tau }(\boldsymbol {\tilde {z}_{2}})\).
The model is defined for \(\boldsymbol {\tilde {z}_{1}}\) as:
$$ \begin{array}{@{}rcl@{}} &&\psi_{\tau}(\boldsymbol{\tilde{z}_{1}})= \sum\limits_{i=1}^{m}\omega_{i} \alpha_{1,i}^{*} +\tau, \\ &&\sum\limits_{i=1}^{m}\omega_{i} \tau_{1,i} =\tau, \end{array} $$and for \(\boldsymbol {\tilde {z}_{1}}\) as:
$$ \begin{array}{@{}rcl@{}} &&\psi_{\tau}(\boldsymbol{\tilde{z}_{2}})= \sum\limits_{i=1}^{m}\omega_{i} \alpha_{2,i}^{*} +\tau, \\ &&\sum\limits_{i=1}^{m}\omega_{i} \tau_{2,i} =\tau. \end{array} $$It is also known that
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\lambda \boldsymbol{\tilde{z}_{1}}+(1-\lambda)\boldsymbol{\tilde{z}_{2}}) = \sum\limits_{i=1}^{m}\omega_{i} \alpha_{0,i}^{*} +\tau. \end{array} $$Proving the convexity is equivalent to proving that
$$ \begin{array}{@{}rcl@{}} &&\sum\limits_{i=1}^{m}\omega_{i} \alpha_{0,i}^{*} +\tau \leq \lambda \left( \sum\limits_{i=1}^{m}\omega_{i} \alpha_{1,i}^{*} +\tau\right) + (1-\lambda) \left( \sum\limits_{i=1}^{m}\omega_{i} \alpha_{2,i}^{*} +\tau\right), \end{array} $$which can be further simplified as \({\sum }_{i=1}^{m}\omega _{i} \alpha _{0,i}^{*} \leq \lambda {\sum }_{i=1}^{m}\omega _{i} \alpha _{1,i}^{*} + (1-\lambda ){\sum }_{i=1}^{m}\omega _{i} \alpha _{2,i}^{*}\).
Let \(\hat {\alpha _{i}}=\lambda \alpha _{1,i}^{*} + (1-\lambda )\alpha _{2,i}^{*}\). The function \(f(\tilde {z})=[\tilde {z}]^{+}=\max \limits \{0, \tilde {z}\}\) can be proved to be convex.
$$ \begin{array}{@{}rcl@{}} (\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}})^{+} &=&\max\{\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}}, 0\} \\ &\leq& \max\{\lambda \tilde{z_{1}}, 0\} + \max\{(1-\lambda) \tilde{z_{2}}, 0\}\\ &=& \lambda \max\{\tilde{z_{1}}, 0\} + (1-\lambda) \max\{\tilde{z_{2}}, 0\}\\ &=& \lambda \{\tilde{z_{1}}\}^{+} + (1-\lambda) \{\tilde{z_{2}}\}^{+}.\\ \end{array} $$As such, \(\mathbb {E}(\tilde {z})^{+}\) is also convex. Then the following holds:
$$ \begin{array}{@{}rcl@{}} &&\mathbb{E}[\lambda \tilde{z}_{1,i}+(1-\lambda) \tilde{z}_{2,i} - \hat{\alpha_{i}}]^{+}\\ &=&\mathbb{E}[\lambda \tilde{z}_{1,i}+(1-\lambda) \tilde{z}_{2,i} - \lambda \alpha_{1,i}^{*} - (1-\lambda)\alpha_{2,i}^{*}]^{+}\\ &\leq& \lambda \mathbb{E}[\tilde{z}_{1,i}-\alpha_{1,i}^{*}]^{+} + (1-\lambda) \mathbb{E}[\tilde{z}_{2,i}-\alpha_{2,i}^{*}]^{+} \\ &\leq& \lambda \tau_{1,i} + (1-\lambda) \tau_{2,i} \end{array} $$Since both τ1,i,τ2,i ∈Γ and set Γ is convex, λτ1,i + (1 − λ)τ2,i ∈Γ for any λ ∈ [0, 1].
$$ \begin{array}{@{}rcl@{}} &&\sum\limits_{i=1}^{m}\omega_{i} \{\lambda \tau_{1,i} + (1-\lambda) \tau_{2,i}\}\\ &=&\lambda \sum\limits_{i=1}^{m}\omega_{i} \tau_{1,i}+ (1-\lambda)\sum\limits_{i=1}^{m}\omega_{i} \tau_{2,i}\\ &=&\lambda \tau + (1-\lambda) \tau\\ &=&\tau. \end{array} $$As a result, \(\hat {\alpha _{i}}\) is a feasible threshold value for \(\psi _{\tau }(\lambda \tilde {z}_{1,i}+(1-\lambda )\tilde {z}_{2,i})\). Nevertheless, \(\alpha _{0,i}^{*}\) is the optimal value that minimizes the objective function among all the feasible solutions. As a result, it is clear to conclude that \(\alpha _{0,i}^{*} \leq \hat {\alpha _{i}} \ \forall i=1,2,\dots ,m\). Further, it is not difficult to see that \({\sum }_{i=1}^{m}\omega _{i} \alpha _{0,i}^{*} \leq \lambda {\sum }_{i=1}^{m}\omega _{i} \alpha _{1,i}^{*} + (1-\lambda ){\sum }_{i=1}^{m}\omega _{i} \alpha _{2,i}^{*}\).
In conclusion, \(\psi _{\tau }(\lambda \boldsymbol {\tilde {z}_{1}}+(1-\lambda )\boldsymbol {\tilde {z}_{2}}) \leq \lambda \psi _{\tau }(\boldsymbol {\tilde {z}_{1}})+(1-\lambda )\psi _{\tau }(\boldsymbol {\tilde {z}_{2}})\) and the convexity of ψ is proved.
-
4.
Translation Invariance: If \(\boldsymbol {\delta }=(\delta _{1}, \delta _{2}, \dots , \delta _{m}) \in \mathbb {R}^{m}\), then \(\psi _{\tau }(\boldsymbol {\tilde {z}}-\boldsymbol {\delta })=\psi _{\tau }(\boldsymbol {\tilde {z}})-{\sum }_{i=1}^{m}\omega _{i} \delta _{i}\).
By definition,
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\boldsymbol{\tilde{z}}-\boldsymbol{\delta})&=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{i} + \tau\right\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{i}-\delta_{i}-\alpha_{i}]^{+} \leq \tau_{i},\\ &&\sum\limits_{i=1}^{m}\omega_{i}\tau_{i}=\tau. \end{array} $$For the objective function,
$$ \begin{array}{@{}rcl@{}} \psi_{\tau}(\boldsymbol{\tilde{z}}-\boldsymbol{\delta})&=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{i} + \tau\right\}\\ &=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{i} + \sum\limits_{i=1}^{m}\omega_{i} \delta_{i}+ \tau\right\} - \sum\limits_{i=1}^{m}\omega_{i} \delta_{i} \\ &=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}(\alpha_{i} + \delta_{i})+ \tau\right\} - \sum\limits_{i=1}^{m}\omega_{i} \delta_{i} \\ &=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i} \hat{\alpha_{i}}+ \tau\right\} - \sum\limits_{i=1}^{m}\omega_{i} \delta_{i}. \end{array} $$For the constraint,
$$ \begin{array}{@{}rcl@{}} \mathbb{E}[\tilde{z}_{i}-\delta_{i}-\alpha_{i}]^{+} &\leq& \tau_{i}\\ \mathbb{E}[\tilde{z}_{i}-(\delta_{i}+\alpha_{i})]^{+} &\leq& \tau_{i}\\ \mathbb{E}[\tilde{z}_{i}-\hat{\alpha_{i}}]^{+} &\leq& \tau_{i} \end{array} $$As a result, the model can be written as:
$$ \begin{array}{@{}rcl@{}} &&\min\limits_{\boldsymbol{\tilde{\alpha}}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\hat{\alpha_{i}} + \tau\right\} - \sum\limits_{i=1}^{m}\omega_{i} \delta_{i} = \psi(\boldsymbol{\tilde{z}})-{\sum}_{i=1}^{m}\omega_{i} \delta_{i}\\ &&s.t. \ \ \mathbb{E}[\tilde{z}_{i}-\hat{\alpha_{i}}]^{+} \leq \tau_{i}. \end{array} $$ -
5.
Contingency Sensitivity: if τ1 ≤ τ2, then \(\psi _{\tau _{2}}(\boldsymbol {\tilde {z}}) \leq \psi _{\tau _{1}}(\boldsymbol {\tilde {z}})\).
Since τ1 ≤ τ2, suppose τ2 = τ1 + Δτ,Δτ ≥ 0 without loss of generality.
-
(P1)
$$ \begin{array}{@{}rcl@{}} \psi_{\tau_{1}}(\boldsymbol{\tilde{z}})&=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{1,i} + \tau_{1}\right\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{i}-\alpha_{1,i}]^{+} \leq \tau_{1,i},\\ &&\sum\limits_{i=1}^{m}\omega_{i}\tau_{1,i}=\tau_{1}, \\ &&\tau_{1,i} \geq 0, \forall i. \end{array} $$
-
(P2)
$$ \begin{array}{@{}rcl@{}} \psi_{\tau_{2}}(\boldsymbol{\tilde{z}})&=&\min\limits_{\boldsymbol{\alpha}, \boldsymbol{\tau}}\left\{\sum\limits_{i=1}^{m}\omega_{i}\alpha_{2,i} + \tau_{2}\right\}\\ &s.t. \quad&\mathbb{E}[\tilde{z}_{i}-\alpha_{2,i}]^{+} \leq \tau_{2,i},\\ &&\sum\limits_{i=1}^{m}\omega_{i}\tau_{2,i}=\tau_{2}, \\ &&\tau_{2,i} \geq 0, \forall i. \end{array} $$
-
(P1)
Suppose that \(\alpha _{1,i}^{*}, \tau _{1,i}^{*}\) are optimal solutions to (P1). The idea is to construct one feasible solution to (P2) using expressions of the optimal solutions to (P1). Constructing the following:
Since \(\alpha _{1,i}^{*}, \tau _{1,i}^{*}\) are optimal solutions to (P1), it is true that
For the case that i≠j,
As such, \(\{\alpha _{1,j}^{*}-\frac {\Delta \tau }{\omega _{j}}, \alpha _{1,i}^{*}\}\) is one feasible solution to (P2). In this case, the value of the objective function of (P2) can be calculated as:
In other words, the minimal solution of (P1) is one feasible solution of (P2). As such, \(\psi _{\tau _{2}}(\boldsymbol {\tilde {z}}) \leq \psi _{\tau _{1}}(\boldsymbol {\tilde {z}})\). This completes the proof. □
1.2 A.2 Single due-date Quotation Model (Theorem 1 under m = 1)
In the special case that m = 1 in the model (1), i.e. there exists only a single milestone variable, the single milestone quotation model can be formulated as
where \([\tilde {z}-\alpha ]^{+}\) is defined as a positive function:
Theorem 2
Single milestone quotation model (12) possesses the following properties:
-
1.
Normalization: ρ(0) = 0.
-
2.
Monotonicity: If z1 ≤ z2, then ρ(z1) ≤ ρ(z2).
-
3.
Convexity: \(\forall z_{1}, z_{2} \in \mathcal {Z}\) and λ ∈ [0, 1], ρ(λz1 + (1 − λ)z2) ≤ λρ(z1) + (1 − λ)ρ(z2).
-
4.
Translation Invariance: If \(m \in \mathbb {R}\), then ρ(z − m) = ρ(z) − m.
-
5.
Contingency Sensitivity: If τ1 ≤ τ2, then ρ(τ1) ≥ ρ(τ2).
The first four properties in Theorem 2 together proves that the proposed milestone quotation model in Eq. 12 is a convex risk measure defined in Follmer and Schied (2002). The fifth property reveals that this proposed milestone quotation model possesses is contingency budget sensitive. In other words, the proposed risk measure will return a smaller risk value when there is a larger amount of contingency budget and vice versa, which coincides with the decision maker’s intuition.
Proof
Each of the properties will be proved in detail here.
-
1.
Normalization: ρ(0) = 0.
By definition,
$$ \begin{array}{@{}rcl@{}} &&\rho(0)=\min\limits_{\alpha}\{\alpha+\tau\}\\ &s.t. \ \ &\mathbb{E}[0-\alpha]^{+} \leq \tau. \end{array} $$It is clear that α = −τ is the smallest feasible value for the constraint.
As such \(\rho (0)=\min \limits _{\alpha }\{\alpha +\tau \}=(-\tau +\tau )=0\).
-
2.
Monotonicity: If \(\tilde {z}_{1} \leq \tilde {z}_{2}\), then \(\rho (\tilde {z}_{1}) \leq \rho (\tilde {z}_{2}).\)
By the definition of Eq. 12,
Model in the case of \(\tilde {z}_{1}\):
$$ \begin{array}{@{}rcl@{}} &&\rho(\tilde{z}_{1})=\min\limits_{\alpha_{1}}\{\alpha_{1}+\tau\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{1}-\alpha_{1}]^{+} \leq \tau. \end{array} $$Model in the case of \(\tilde {z}_{2}\):
$$ \begin{array}{@{}rcl@{}} &&\rho(\tilde{z}_{2})=\min\limits_{\alpha_{2}}\{\alpha_{2}+\tau\}\\ &s.t. \quad &\mathbb{E}[\tilde{z}_{2}-\alpha_{2}]^{+} \leq \tau. \end{array} $$Define sets of feasible α1,α2 values that the 2 models can take as \(C_{\alpha _{1}}\) and \(C_{\alpha _{2}}\).
$$ \begin{array}{@{}rcl@{}} && C_{\alpha_{1}}=\{\alpha_{1} | \mathbb{E}[\tilde{z}_{1}-\alpha_{1}]^{+} \leq \tau\},\\ && C_{\alpha_{2}}=\{\alpha_{2} | \mathbb{E}[\tilde{z}_{2}-\alpha_{2}]^{+} \leq \tau\}. \end{array} $$Suppose the optimal α values are \(\alpha _{1}^{*}\) and \(\alpha _{2}^{*}\) respectively. As such, \(\rho (\tilde {z}_{1})=\alpha _{1}^{*}+\tau \) and \(\rho (\tilde {z}_{2})=\alpha _{2}^{*}+\tau \). For the same τ, the feasible values satisfy \(\alpha _{1}^{*} \leq \alpha _{2}^{*}\) since function \(E[\tilde {z}-\alpha ]^{+}\) is non-decreasing function w.r.t z. As such, \(\rho (\tilde {z}_{1}) \leq \rho (\tilde {z}_{2})\).
-
3.
Convexity: \(\forall \tilde {z}_{1}, \tilde {z}_{2} \in \mathcal {Z}\) and λ ∈ [0, 1], \(\rho (\lambda \tilde {z}_{1}+(1-\lambda )\tilde {z}_{2}) \leq \lambda \rho (\tilde {z}_{1})+(1-\lambda )\rho (\tilde {z}_{2})\).
Firstly, the function \(f(\tilde {z})=[\tilde {z}]^{+}=\max \limits \{0, \tilde {z}\}\) can be proved to be convex.
$$ \begin{array}{@{}rcl@{}} (\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}})^{+} &=&\max\{\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}}, 0\} \\ &\leq& \max\{\lambda \tilde{z_{1}}, 0\} + \max\{(1-\lambda) \tilde{z_{2}}, 0\}\\ &=& \lambda \max\{\tilde{z_{1}}, 0\} + (1-\lambda) \max\{\tilde{z_{2}}, 0\}\\ &= &\lambda \{\tilde{z_{1}}\}^{+} + (1-\lambda) \{\tilde{z_{2}}\}^{+}.\\ \end{array} $$Then any function \(\mathbb {E}(\tilde {z})^{+}\) is also convex.
Now suppose that:
$$ \begin{array}{@{}rcl@{}} \rho(\lambda \tilde{z}_{1}+(1-\lambda)\tilde{z}_{2})&=&\alpha^{*}+\tau, \\ \rho(\tilde{z}_{1})&=& \alpha_{1}^{*}+\tau, \\ \rho(\tilde{z}_{2})&=& \alpha_{2}^{*}+\tau . \end{array} $$Proving the convexity is equivalent to proving that \(\alpha ^{*}+\tau \leq \lambda (\alpha _{1}^{*}+\tau ) + (1-\lambda ) (\alpha _{2}^{*}+\tau ) \), which can be further simplified as \(\alpha ^{*} \leq \lambda \alpha _{1}^{*} + (1-\lambda )\alpha _{2}^{*}\).
Let \(\hat {\alpha }=\lambda \alpha _{1}^{*} + (1-\lambda )\alpha _{2}^{*}\). Then the following holds:
$$ \begin{array}{@{}rcl@{}} &&\mathbb{E}[\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}} - \hat{\alpha}]^{+}\\ &=&\mathbb{E}[\lambda \tilde{z_{1}}+(1-\lambda) \tilde{z_{2}} - \lambda \alpha_{1}^{*} - (1-\lambda)\alpha_{2}^{*}]^{+}\\ &\leq& \lambda \mathbb{E}[\tilde{z_{1}}-\alpha_{1}^{*}]^{+} + (1-\lambda) \mathbb{E}[\tilde{z_{2}}-\alpha_{2}^{*}]^{+} \\ &\leq& \lambda \tau + (1-\lambda) \tau \\ &\leq& \tau \end{array} $$As a result, \(\hat {\alpha }\) is a feasible threshold value for \(\rho (\lambda \tilde {z}_{1}+(1-\lambda )\tilde {z}_{2})\). Nevertheless, α∗ is the optimal value that minimizes the objective function among all the feasible solutions. As a result, it is clear to conclude that \(\alpha ^{*} \leq \hat {\alpha }\) which in return will prove that the convexity holds for ρ.
-
4.
Translation Invariance: If \(\delta \in \mathbb {R}\), then \(\rho (\tilde {z}-\delta )=\rho (\tilde {z})-\delta \).
By definition,
$$ \begin{array}{@{}rcl@{}} &&\rho(\tilde{z}-\delta)=\min\limits_{\alpha}\{\alpha + \tau\}\\ s.t. \quad &&\mathbb{E}[\tilde{z}-\delta-\alpha]^{+} \leq \tau. \end{array} $$For the objective function,
$$ \begin{array}{@{}rcl@{}} \rho(\tilde{z}-\delta)&=&\min\limits_{\alpha}\{\alpha + \tau\}\\ &=&\min\limits_{\alpha}\{\alpha + \delta + \tau\}-\delta \\ &=&\min\limits_{\hat{\alpha}}\{\tilde{\alpha} + \tau\}-\delta. \end{array} $$For the constraint,
$$ \begin{array}{@{}rcl@{}} \mathbb{E}[\tilde{z}-\delta-\alpha]^{+} &\leq& \tau\\ \mathbb{E}[\tilde{z}-(\delta+\alpha)]^{+} &\leq& \tau\\ \mathbb{E}[\tilde{z}-\hat{\alpha}]^{+} &\leq& \tau. \end{array} $$As a result, the model can be written as:
$$ \begin{array}{@{}rcl@{}} \min\limits_{\hat{\alpha}}\{\tilde{\alpha} + \tau\} - \delta &=& \rho(\tilde{z})-\delta\\ s.t. \quad \mathbb{E}[\tilde{z}-\hat{\alpha}]^{+} &\leq& \tau. \end{array} $$ -
5.
Contingency Sensitivity: If τ1 ≤ τ2, then ρ(τ1) ≥ ρ(τ2).
Suppose
$$ \begin{array}{@{}rcl@{}} &&\rho(\tau_{1})= \alpha_{1}^{*}+\tau_{1}, \\ &&\rho(\tau_{2})= \alpha_{2}^{*}+\tau_{2}. \end{array} $$In order to prove that ρ(τ1) ≥ ρ(τ2), the following condition should hold:
$$ \begin{array}{@{}rcl@{}} &&\alpha_{1}^{*}+\tau_{1} \geq \alpha_{2}^{*}+\tau_{2}. \end{array} $$Let \(\tilde {\alpha }=\alpha _{1}^{*}+\tau _{1}-\tau _{2}\),
$$ \begin{array}{@{}rcl@{}} \mathbb{E}[\tilde{z}-\tilde{\alpha}]^{+}&=&\mathbb{E}[\tilde{z}-(\alpha_{1}^{*}+\tau_{1}-\tau_{2})]^{+}\\ &\leq& \mathbb{E}[\tilde{z}-\alpha_{1}^{*}]^{+} + \mathbb{E}[\tau_{2}-\tau_{1}]^{+}\\ &\leq& \tau_{1} + \tau_{2}-\tau_{1}\\ &=&\tau_{2}. \end{array} $$Hence, \(\tilde {\alpha }\) is a feasible solution of ρ(τ2). Besides, \(\alpha _{2}^{*}\) is the optimal (minimum) value for ρ(τ2). As a result, \(\alpha _{2}^{*} \leq \tilde {\alpha }\) and further it is proved that ρ(τ1) ≥ ρ(τ2). This completes the proof.
□
Rights and permissions
About this article
Cite this article
Wang, Z., Ng, T.S. & Pang, C.K. Due-date quotation model for manufacturing system scheduling under uncertainty. Discrete Event Dyn Syst 31, 271–293 (2021). https://doi.org/10.1007/s10626-020-00332-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10626-020-00332-y