Professional Documents
Culture Documents
Automation in Construction
journal homepage: www.elsevier.com/locate/autcon
Heuristic algorithm for ready-mixed concrete plant scheduling with multiple MARK
mixers
Zhenyuan Liua,b,⁎, Yakun Zhanga, Minghui Yua,b, Xiaolu Zhoua
a
School of Automation, Huazhong University of Science and Technology, Wuhan 430074, China
b
Key Laboratory of Education, Ministry for Image Processing and Intelligent Control, Wuhan 430074, China
A R T I C L E I N F O A B S T R A C T
Keywords: In this study, the problem of integration of ready-mixed concrete (RMC) production scheduling and truck and
Ready-mixed concrete pump dispatching in an RMC plant with multiple mixers is focused upon. A time-space network model, which
Multiple mixers combines RMC production and vehicle dispatching, is proposed. A heuristic algorithm with eight sets of conjoint
Production scheduling priority rules for production scheduling, truck and pump dispatching, and mixer scheduling is developed to solve
Vehicle dispatching
the problem. Computational experiments are conducted on actual cases collected from an RMC company; these
Conjoint-rule based heuristic algorithm
cases are classified into four types based on the quantity of orders from construction sites and the discrete degree
of construction sites. The advantages and disadvantages of adopting various conjoint-rules in the various types of
cases are determined. The experimental results demonstrate that this heuristic solution is capable of enabling
managers of RMC plants to develop more suitable schedules in various types of practical cases.
⁎
Corresponding author at: School of Automation, Huazhong University of Science and Technology, Wuhan 430074, China.
E-mail address: zyliu@mail.hust.edu.cn (Z. Liu).
http://dx.doi.org/10.1016/j.autcon.2017.08.013
Received 8 July 2016; Received in revised form 3 June 2017; Accepted 9 August 2017
Available online 23 August 2017
0926-5805/ © 2017 Elsevier B.V. All rights reserved.
Z. Liu et al. Automation in Construction 84 (2017) 1–13
proposed a mixed-integer programming model, which included an RMC obtain the EOQ-JIT cost indifference point. Deligiannis and Manesis
mixer, several construction sites, and a few trucks and pumps with [24] developed a hybrid model for a concrete batching and mixing
various specifications. Kinable et al. [4] presented a fundamental ver- plant, which can be used to adopt a development environment and
sion of concrete delivery problem incorporating a mixed-integer pro- actualize the influence of the operation parameters on RMC plant.
gramming model and solved using constraint programming. Sawhney A variety of algorithms such as genetic algorithm, heuristic algo-
et al. [5] concentrated their attention on a Petri network with advanced rithm, and a few hybrid algorithms are developed. Zegordi and Beheshti
features, which was used as a model and analysis tool to study the [25] presented a genetic algorithm with three populations to solve the
operation processes in a ready-mixed concrete plant. Wang et al. [6] problem of integration of production and transportation. Naso et al.
developed a simulation model based on data collected in Singapore. In [26] proposed a genetic algorithm combined with the heuristic method
order to analyze the influence on productivity of concrete plant and to solve delivery problems in the RMC operation. Liu et al. [2] also
proper arrival times of trucks, they calculated charts with computers. applied a genetic algorithm to determine a suitable sequence for dis-
Feng and Wu [7] presented a model based on genetic algorithm to patching trucks and production for construction sites. Silva et al. [27]
obtain a near-optimal solution of a delivery problem. This solution is presented a new meta-heuristic algorithm, a combination of the GA and
capable of minimizing the total waiting time of RMC trucks at con- ACO algorithm, to solve the RMC dispatching problem. By analyzing a
struction sites and satisfying the demand from each site. In another practical case, the advantage of this method was evident. Ko and Wang
study, they [8] introduced a system model that adopted fmGA and [28] introduced a multi-objective genetic algorithm to balance pro-
CYCLONE to ascertain the optimal scheduling solution. Considering the duction resources and optimize the buffer size between stations.
process of RMC production and delivery, Matsatsinis [9] analyzed a Maghrebi et al. [29] proposed a sequential meta-heuristic method,
routing problem with time windows, multiple plants, and multiple ve- which can solve large-scale RMC problems efficiently. They developed a
hicles including trucks and pumps. Graham et al. [10] addressed a robust sequential genetic algorithm and a new formulation for mini-
delivery problem with a specialized neural network including a feed- mizing the number of trucks. Maghrebi et al. [30] also developed a
forward network and Elman neural network. Park et al. [11] estab- column generation algorithm for vehicle routing problem with time
lished a dynamic model based on the analysis of the RMC delivery windows (VRPTW) problem. They employed the Dantzig-Wolfe method
process; this model offered a scientific procedure to instruct the RMC to reformulate the problem. In another study [31], they tested a robust
operation. Feng et al. [12] presented a systematic model aimed at op- genetic algorithm and column generation with various sizes of actual
timizing the scheduling of dispatching trucks, which could minimize RMC problems. In order to compare the optimization models and ex-
the total waiting duration of trucks. Yan and Lai [13] constructed a pert-based decisions, Maghrebi et al. [32] selected four examples of
network flow model that employed a time-space network technique to various sizes and tested them. It could be concluded that optimization
solve the problem of production and dispatching of ready-mixed con- models only attempt to achieve the lowest cost, whereas the expert
crete that took into consideration the overtime in a plant in Taiwan. Lin prefers a more stable dispatching system with a marginally higher cost.
et al. [14] developed a multi-objective programming model for the Cheng and Tran [33] developed a DES-COMODE algorithm, which in-
dispatching operation of RMC trucks as a job shop problem with re- tegrated discrete event simulation and chaotic initialized opposition
circulation. In this study, intrinsic and imposed constraints that affected multi-objective differential evolution to solve an RMC dispatching truck
truck dispatching of RMC plants were presented. Garcia and Lozano problem. Low et al. [34] presented an integrated scheduling problem in
[15] addressed the problem of selecting and scheduling RMC orders to which RMC was delivered to retailers within time windows, which was
satisfy the demands of sites expeditiously. The objective was to max- also a zero-inventory concept similar to RMC in production and de-
imize the total value of orders served under two scenarios: arbitrary and livery problem. Two types of genetic-algorithm-based heuristics were
uniform profit margins for orders. Sarker et al. [16] reviewed various designed to solve the problems. For the case of multiple plants, Schmid
planning and design models developed for various aspects of the con- et al. [35] employed an effective hybrid approach for RMC delivery,
struction industry, which incorporated mathematical models and si- which consisted of an integer multi-commodity flow optimization
mulation based on certain actual data including those on scheduling component and a variable neighborhood search component.
and dispatching of trucks at an RMC plant. For the exception in op- With respect to the solution method, Lu et al. [36,37] developed a
erations management of an RMC plant, Yan et al. [17] considered the computer software for simulation and used it to analyze the manage-
adaptations for supplying ready-mixed concrete following incidents ment of ready-mixed concrete in Hong Kong. With respect to the study
with a time-space network flow model. In another study of theirs [18], a of scheduling processes, Zayed and Halpin [38] proposed a replaceable
systematic scheduling model was developed that would enable RMC solution to simulate RMC delivery operations. Schmid et al. [39] de-
carriers/suppliers recalibrate schedules after temporary breakdown of veloped a method to solve the concrete delivery problem involving
mixers. Azambuja and Chen [19] provided a systematic method to as- multiple plants, multiple construction sites, and a variety of trucks and
sess supply chain risks, identify vulnerabilities, and measure the impact pumps, which aimed to minimize the overall traveling fee. In this re-
of disruptions in an RMC supply chain. search, he proposed certain combinatorial algorithms such as the
Furthermore, there are a few related studies on production and combination of variable neighborhood search (VNS) and exact method.
delivery scheduling of RMC. Ghasri et al. [20] investigated concrete Matsatsinis [9] developed a Decision Support System using a heuristic
pouring duration affected by construction site features and supply chain algorithm to analyze the model. A vehicle routing problem with soft
parameters including supply process, location of the project, and traffic time-windows (VRPSTW) in a fuzzy random environment was studied
conditions. Based on the linear regression model and hazard-based by Xu et al. [40]. They proposed a GLNPSO-ep to solve this problem.
model, their study provided decision-makers with a practical tool to Asbach et al. [41] studied the problem involving multiple plants,
enhance concrete pouring productivity. Thawongklang and Tanwa- multiple construction sites, and various vehicles; however, he did not
nichkul [21] mainly concentrated on the total tardiness of truck dis- consider specialized pumps. He developed a local search algorithm to
patching plans and optimum number of trucks. Wang et al. [22] de- solve the model and obtained a perfect solution. Feng et al. [12] de-
veloped an integrated model, which considered the mold veloped a computer program to enable managers to dispatch trucks.
manufacturing, precast components storage, and transportation pro- Pierre et al. [42] used CPLEX to solve mixed-integer programs with a
cesses. Two case studies were conducted to test the validity of the variable neighborhood search heuristic. Maghrebi et al. [43,44] pre-
model, and the results demonstrate that optimal scheduling is related to sented a machine learning based method to select techniques using
the traveling time of precast components. Wu [23] studied a ready- various approaches. This method had demonstrated the potential to
mixed concrete plant in Singapore and attempted to expand the clas- solve RMC dispatching problems to match experts' decisions with high
sical economic order quantity (EOQ) with a price discount model to accuracy. They also used bender decomposition to solve RMC
2
Z. Liu et al. Automation in Construction 84 (2017) 1–13
dispatching problems; this procedure was referred to as “large scale 3.1.1. Production and loading arc
mixed-integer problemming” [45]. The Lagrangian approach was ap- In order to satisfy the demands of the construction sites, the RMC
plied by Narayanan et al. [46] to achieve solutions with a marginal plant organizes RMC production at appropriate times as planned. After
optimality gap albeit within a practical time period. the specified mixer completes production, RMC is loaded on the truck
So far, a number of studies have been conducted on RMC production that has been selected from the set of idle trucks according to a certain
and delivery problems. A number of algorithms, such as genetic algo- scheduling rule. This arc connects nodes of the set of idle trucks and
rithms and heuristics, are employed for solving these problems. mixers, which indicates the beginning of the production and delivery of
However, compared to the preceding research studies, this investiga- a batch of RMC.
tion is concerned more with a conjoint-rule based heuristic algorithm
for solving the integrated production and delivery problem in the RMC 3.1.2. Delivery arc
plant with multiple mixers, wherein both truck and pump dispatching A delivery arc connects nodes of a mixer in the RMC plant and a
are considered. In addition, unlike a majority of the research studies, construction site, and the truck selected in the production and loading
each of which investigated a single case, this study takes into con- arc follows a planned route to deliver RMC. The arcs between the nodes
sideration various cases, classifies them into various types, and con- of the construction sites are RMC casting durations, and they are likely
ducts separate experiments for each case type. This could enable to include waiting duration of trucks or pumps.
managers to select a more suitable conjoint-rule to develop schedules
for various cases. 3.1.3. Return arc
After completing RMC casting, trucks return to the RMC plant ex-
peditiously. They are added to the set of idle trucks and wait for the
3. Problem description next delivery task. This arc connects nodes of a construction site and the
set of idle trucks.
3.1. Operation process
3.1.4. Pump-transfer arc
The production and delivery process of RMC is impelled by demands Pumps are dispatched to construction sites that require pumps to
from construction sites. The managers are required to collect orders assist in casting according to the scheduled plan. After casting, pumps
from construction sites, which are to be validated. After validation, the are not required to return to the RMC plant; therefore, this arc is likely
valid orders are set as a guide for production and delivery. In this study, to connect nodes of two construction sites. However, all the pumps are
it is assumed that all orders are to be collected at the RMC plant a day required to return to the RMC plant after completing all the tasks in a
before the work starts, and there is no temporary order. Considering the workday.
absence of finished inventory of RMC, JIT (Just in Time) is employed
for RMC production and delivery. Certain assumptions related to the 3.2. Vehicles
RMC plant are made as follows:
3.2.1. Trucks
(1) RMC plant is capable of producing all types of RMC that con- Each truck, when it receives a delivery task, starts from the RMC
struction sites require. plant and delivers RMC to the corresponding construction site.
(2) RMC plant is capable of producing and loading more than one batch Preferably, the arrival time of each truck is to fall within the operation
of RMC simultaneously, aided by multiple mixers in the plant. time-window of this construction site; otherwise, a penalty fee is gen-
(3) Various types of RMC exhibit similar hardening duration. erated. After completing the unloading of RMC at the construction site,
the truck is required to return to the RMC plant, and prepares for the
The procedure for RMC production and vehicle dispatching can be next delivery batch of RMC. If two consecutive batches of RMC trans-
described as follows. First, according to production scheduling priority ported by this truck are not of similar type, the truck is required to be
rules, the production sequence for the various construction sites is de- cleaned between the two deliveries.
termined, and suitable mixers are selected to produce RMC. After When a truck arrives at the corresponding construction site, it is
production, a few trucks are selected to be loaded with RMC. Following likely to be in the following situations:
an appropriate route, trucks deliver RMC to corresponding construction
sites. Pumps are dispatched at appropriate times to certain construction (1) If the preceding truck just completes unloading and starts to return
sites that require such pumps to assist casting. After the trucks reach the to the plant, the truck under consideration can immediately begin
construction sites, workers on those sites begin to unload RMC from the unloading without any wait.
trucks; moreover, the process is likely to be completed with the aid of a (2) If the preceding truck is in the process of unloading RMC at the
pump. After RMC unloading is completed, trucks are to return to the construction site, the truck under consideration is to wait until the
plant instead of staying at the construction sites. On the other hand, the preceding truck completes its work.
pumps remain at the construction sites, and they may be dispatched to (3) If the truck under consideration arrives a significant time after the
the next site that requires a pump. All trucks and pumps are to return to preceding one, which has completed unloading a long time in time
the RMC plant after completing the tasks scheduled for the workday. in the past, the former is required to unload immediately. This si-
Time-space network flow models constructed by Yan et al. [1,13,41] tuation generates a penalty fee as a result of the waiting by the
provide references for the description of operation processes in an RMC personnel and equipment at the construction site.
plant. As we can observed in Fig. 1, there are several nodes and arcs
following the timeline. Each node represents an activity at a specific 3.2.2. Pumps
time, and an arc between nodes implies the duration of an activity or When a pump receives its first task, it departs from the RMC plant to
the waiting duration between two activities. Although Fig. 1 illustrates a construction site and assists in the RMC unloading process. In prac-
a marginal fraction of a workday in an RMC plant, it comprises all the tice, the arrival time of each pump is to coincide with or precede the
activities of the mixers, trucks, and pumps occurring in RMC production arrival time of the corresponding truck dispatched to this site. A pump
and delivery. In the following paragraphs, the detailed operation pro- can depart from the construction site only after all the trucks dispatched
cess in a practical RMC plant is described with the aid of the time-space to the site have been unloaded. In the scheduling process, the pump
network model. with specification that exactly satisfies the demand of the construction
site is most likely to be selected. If such a pump is not available, a pump
3
Z. Liu et al. Automation in Construction 84 (2017) 1–13
Pumps
Time Flow Set
Trucks Mixer Mixer
Set 1# 2# Site1 Site2 Site 3
···
Other
Sites
of higher specification can be selected owing to downward compat- Compared with the model constructed by Yan et al. [1], the model
ibility. If the available pump resource is inadequate, the construction developed in this study adds an additional “truck start delivery node”
site is required to wait for the release of other operational pumps. (as there is more than one mixer in the plant) and a new type of arc
In contrast to the operation mode of trucks, a pump can be sched- named as “Pump-Transfer Arc.” Yan set Truck start node and Garage
uled for dispatch from a construction site directly to another site after node as the beginning and ending of trucks in a working day; in this
completion of work in the former. After all the tasks of the pumps have study, Trucks Set and Pumps Set, which are located in the plant, are set
been completed, they are to return to the RMC plant. as the beginning and ending. The time interval between various nodes
When a pump arrives at a site, three situations are likely to occur: is set as 2 min in Yan's model, while in the present model, the time
interval between various nodes is presented as durations of production,
(1) The truck has not arrived; therefore, the pump is required to wait. loading, delivery, casting, etc.; therefore, it is not constant. Moreover,
This generates a penalty fee as a result of the wait. the production and loading arc represents the operating duration of the
(2) The truck arrives simultaneously; then, the pump can begin op- mixers.
eration immediately. To summarize, the highlights of the time-space network model are
(3) The truck has been at the site for a while, and the pump can begin more numerous mixer nodes and a new type of arcs, which imply
operation immediately. However, this generates a penalty fee as a connections of higher complexity between the various nodes. This time-
result of the waiting by the personnel and equipment at the site and space network model is more complicated and can be characterized as
the truck. NP-hard in terms of optimization.
4
Z. Liu et al. Automation in Construction 84 (2017) 1–13
Table 1 ▪ When a truck's arrival time is later than the beginning time of work
Notations of the problem. at the site
Notation Description Unit Penalty1 = ∑ max {ArriveTij − a (Ci ), 0} × PenaltyRate1
i ∈ C,j =1 (3)
P The plant –
[a(p),b(p)] The time-window of the plant –
M The set of mixers – ▪ When a truck's arrival time is later than the ending time of work at
MBusym The state of Mixer m (1 = busy, 0 = idle) – the site
MCm The unit cost of Mixer m –
MCost Total cost of all the mixers – Penalty2 = ∑ max {ArriveTij − b (Ci ), 0} × PenaltyRate2
T The set of trucks – i ∈ C , j = CBatchi (4)
TBusyt The state of Truck t (1 = busy, 0 = idle) –
TCt The unit cost of Truck t –
TCapacityt The capacity of Truck t m3
▪ Penalty owing to discontinuity of casting at the site
TCost Total cost of all the trucks –
Pump The set of pumps –
Penalty3 = ∑ ∑ max {ArriveTij + 1 − FinishTij − timelag 0}
i ∈ C j ∈ {1 … , CBatchi}
PBusyp The state of Pump p (1 = busy, 0 = idle) –
PCp The unit cost of Pump p – × PenaltyRate3 (5)
PCost Total cost of all the pumps –
xii′ The state of pump-transfer arc from site i to site i′ –
(It is a binary variable. xii′ = 1 implies a pump transfer from ▪ When the returning time of all the vehicles is later than the plant
site i to site i′; otherwise, xii′ = 0) ending time
C The set of construction sites –
CBatchi The total batches of site i Penalty4 = ∑ max {FinishTime − b (p), 0} × PenaltyRate4
CPi The state of need for pump at site i (1 = yes, 0 = no) – i∈C (6)
CPcurrenti The state of availability of pump at site i (1 = yes, 0 = no) –
[a(Ci),b(Ci)] The time-window at site i –
▪ Penalty as a result of the waiting by the personnel and equipment at
CDi The demand quantity of RMC at site i m3
Cdemandij The RMC quantity of the jth batch at site i m3 the sites
CRestDij The demand quantity of RMC at site i after the jth batch m3
timelag The longest work break-off time at the site Min
Penalty5 = ∑ ∑ max {ArriveTij + 1 − FinishTij 0} × PenaltyRate5
Timeii′ Traveling time from site i to i′ (i = 0 represents the plant) Min i ∈ C j ∈ {1, … , CBatchi}
Penalty Total penalty costs – (7)
Penaltyk The cost of penalty k (k = 1, 2, 3, 4, 5, 6)
PenaltyRatek The unit fee of penalty k (k = 1, 2, 3, 4, 5, 6) –
▪ Penalty owing of the waiting of trucks
Truckij The available time of the truck dispatching the jth batch to
site i.
ProduceTij Production time of the jth batch at site i –
Penalty6 = ∑ ∑ max {FinishTij − ArriveTij + 1 , 0} × PenaltyRate6
i ∈ C j ∈ {1, … , CBatchi}
Produceij Duration of the jth batch producing at site i Min
Loadij Duration of loading of the jth batch for site i Min (8)
SendTij Departure time of truck with the jth batch toward site i –
ArriveTij
PArrivei
Arrival time of truck with the jth batch at site i
Arrival time of the pump at site i
–
–
• The cost of mixers
BeginTij Dealing time of the jth batch at site i – Mcost = ∑ ∑ ∑ Produceij × MCm
CastTij Duration of casting of the jth batch at site i Min m ∈ M i ∈ C j ∈ {1, … , CBatchi } (9)
FinishTij Finishing dealing time of the jth batch in site i –
The finishing time of work at site i
• The cost of trucks
EndTimei
FinishTime The time when all the vehicles return to the plant –
TravelTime The total traveling time of all the trucks Min
Tcost = ∑∑ ∑ (ArriveTij − Sendij ) × 2 × TCt
t ∈ T i ∈ C j ∈ {1, … , CBatchi } (10)
RMC production and vehicle dispatching, a heuristic algorithm is de-
veloped to solve the problem. • The cost of pumps (The pump site i needed is dispatching from i)
′
5
Z. Liu et al. Automation in Construction 84 (2017) 1–13
No
Yes
Site need pump? Site exist pump?
No Yes
No Yes Yes
Mixers Pumps
available? available?
Yes No
The first batch of
the site
of construction sites, mixers, trucks, pumps, and penalty coef- to next step.
ficient. finish is set as 0. Step 3 Examine sites for which CRestDi ≠ 0; then, add site i into the
Step 2 Examine ending flag finish. If finish equals 1, end scheduling; service sequence with the corresponding priority rule.
then, calculate SumCost and construct Gantt chart; otherwise, go Step 4 Examine CPi and CPcurrenti.
6
Z. Liu et al. Automation in Construction 84 (2017) 1–13
One of key point of the heuristic algorithm is to set priority rules for Starting point Terminal point Distance (km) Average traveling time (min)
planning production sequence for construction sites, choosing trucks, Plant C1 6.2 15
pumps and mixers for production and delivery. The details are pre- C2 4.0 9
sented as follows. C3 5.5 12
C4 3.4 8
C5 12.0 27
4.4.1. Rules for production scheduling
C6 4.1 10
C7 6.6 16
(1) Rule S1: According to starting time of time-window [a(C), b(C)],
deliver RMC to the site with the smallest a(C) among unsatisfied
sites. Table 7
(2) Rule S2: According to ending time of time-window [a(C), b(C)], Traveling time between two construction sites with need for pumps.
deliver RMC to the site with the smallest b(C) among unsatisfied
Starting point Terminal point Distance (km) Average traveling time (min)
sites.
C2 C3 3.0 9
4.4.2. Rules for truck dispatching C3 C6 3.1 9
C6 C7 5.4 15
C7 C2 5.4 16
(1) Rule T1: Select available truck of which the TCapacity is the nearest
C6 C2 0.09 0.3
to the CRestD (in the following paragraphs, this truck is defined as C3 C7 3.7 9
Opt. truck); if all the trucks are busy, select the earliest available
truck.
(2) Rule T2: Unlike Rule T1, when all the trucks are unavailable, record higher specification can also be dispatched to the site.
the earliest available truck return time t; then, select the truck of
which the TCapacity is the nearest to the CRestD from the idle trucks 4.4.4. Rules for mixer scheduling
within [t, t + Tt].
(1) Rule M1: Select the mixer whose mix rate is the highest.
4.4.3. Rules for pump dispatching (2) Rule M2: Select the mixer whose production cost per cubic meter is
In this study, a pump whose specification just satisfies the demand the lowest.
of the construction site is selected. If there is no such pump, a pump of
7
Z. Liu et al. Automation in Construction 84 (2017) 1–13
Load
Pump2
13 • 3
Delivery
Pump1
2 • 2 40 • 7
Unload
Truck8
9 • 2 26 • 6 34 • 5
Return
Truck7
8 • 2 17 • 3 25 • 6 33 • 5
Truck6
1 • 1 10 • 2 24 • 6 32 • 5
Truck5
7 • 2 16 • 3 23 • 6 31 • 5 39 • 5
Truck4
6 • 2 15 • 3 22 • 6 30 • 5 38 • 5
Truck3
5 • 2 14 • 3 21 • 6 29 • 4 37 • 5 43 • 7
Truck2
4 • 2 12 • 2 20 • 6 28 • 6 36 • 5 42 • 7
Truck1
3 • 2 11 • 2 19 • 6 27 • 6 35 • 5 41 • 7
Mixer2
Mixer1
8:00 8:30 9:00 9:30 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00 Time
5. Computational experiments located at the same place. However, they are different construction
parts with different time-windows. Therefore, C2 and C6 are treated as
The problem of the RMC production scheduling and vehicle dis- different sites here.
patching with multiple mixers in RMC plants, several trucks, and pumps In this model, all the penalty rates are 2 per min. The average
with various specifications, several construction sites with varying de- loading rate is 2 m3/min, average unloading and placement rates are
mands and various geographic positions is the focus in this study. In 2 m3/min, time-window of the plant [a(p), b(p)] is 05:00–23:00, and
order to compare various conjoint priority rules for production and discontinuity duration of RMC is 90 min.
vehicle dispatching, computational experiments are conducted on a few This experiment is conducted with MATLAB R2012a on a PC with
actual cases collected from records of certain workdays of an RMC Microsoft Window 7 (CPU 2.93 GHz, 2 GB RAM).
company in Wuhan, China, whose annual production capacity of RMC Fig. 3 illustrates the operation process of RMC production and ve-
is 600,000 m3. hicle dispatching on this workday. A rectangle in the chart represents
activity duration. For example, a rectangle in the timeline of a pump
implies a transfer of the pump from the plant or a construction site to
5.1. An illustrative example another site, a rectangle of one of four colors in the timeline of a truck
represents one of four activities, and a rectangle in the timeline of a
The following case is an example of data collected on a workday. mixer represents the production duration of the mixer (In order to
There are an RMC plant, two mixers, seven construction sites, eight distinguish between two continuous productions, two colors were used
heterogeneous trucks, and two heterogeneous pumps. The seven con- to highlight the interval). From the chart, the schedules of mixers,
struction sites are labeled C1–C7, and their information is illustrated in trucks, and pumps are evident. In this example, the conjoint-rule of S1,
Table 2, which includes pump specifications that construction sites T1, and M1 is selected. Each batch includes four parts: load, delivery,
require (0 = absence of need for pump), their demands for RMC, and unload, and return. The numbers below the rectangles of vehicles re-
time-windows. The information of the eight trucks, two pumps, and two present the sequence and destination of this batch. For example, the
mixers is also presented in Table 3, Table 4, and Table 5, respectively. text ‘3 → 2’ in Fig. 3 implies that this production and delivery batch is
In addition, certain information about the traveling distances and tra- the third batch in the total sequence and that the destination of this
veling times between the RMC plant and the seven construction sites batch is site 2. In addition, the penalty fees, SumCost, and traveling
and among those construction sites that require pump to assist casting is times of vehicles can be calculated in the computational test.
presented in Tables 6 and 7. From Table 7, it can be determined that the
distance between C2 and C6 is 0.09 km. This is because C2 and C6 are
8
Z. Liu et al. Automation in Construction 84 (2017) 1–13
Table 8 Table 10
Congestion degrees in various areas. Various types of cases.
9
Z. Liu et al. Automation in Construction 84 (2017) 1–13
mixers: S1-T1-M1, S1-T1-M2, S1-T2-M1, S1-T2-M2, S2-T1-M1, S2-T1- experiments conducted for analyzing the performances of various
M2, S2-T2-M1, and S2-T2-M2. conjoint-rules in the various types of cases. There are an RMC plant,
The following paragraphs describe a few computational two types of mixers, several construction sites, eight types of trucks, and
10
Z. Liu et al. Automation in Construction 84 (2017) 1–13
two types of pumps in these cases. The main standards of evaluation are rules are approximately equally effective on March 12. On March
the penalty, gross cost, and traveling time of vehicles. 10, conjoint-rule S2-T2-M2 is the optimum option, and conjoint-
As illustrated in Figs. 5, 6, and 7: rule S1-T1-M1 is the least suitable.
(4) In Case type 4 (concentration-sites and few-orders case):- On both
(1) In Case type 1 (decentralization-sites and more-orders cases):- the days, in terms of penalty, conjoint–rule S1-T1-M1 is the op-
Conjoint–rules S1-T1-M1 and S1-T2-M1 are the optimum options, timum selection, and S2-T2-M2 is the least suitable. From the per-
and S2-T1-M2 and S2-T2-M2 are the least suitable in terms of spective of sum cost, conjoint–rules S1-T1-M2 and S1-T2-M2 are the
penalty. In terms of sum cost, conjoint–rules that incorporates rule optimum options, and S2-T2-M1 is the least suitable on both the
M2 are more effective than those with rule M1. In terms of days. On the basis of TravelTime, all the conjoint-rules are ap-
TravelTime, conjoint-rules S1-T1-M2, S1-T2-M1, and S1-T2-M2 are proximately equally effective on February 25. On February 26,
similar and are the optimum options on March 7. Conjoint-rules S1- conjoint-rule S1-T2-M2 is the optimum option, and conjoint-rules
T1-M2 and S1-T2-M2 are the optimum options on March 9, and S2-T1-M2 and S2-T2-M2 are the least suitable.
conjoint-rule S2-T1-M1 is the least suitable on the two days.
(2) In Case type 2 (decentralization-sites and few-orders cases):- On As can be observed in all the above, these sets of conjoint-rules
March 3, conjoint–rules S1-T1-M1 and S2-T1-M1 are the optimum exhibit varied performance in various types of cases. Therefore, man-
options, and S2-T2-M2 is the least suitable one. On March 11, agers can select suitable conjoint-rules to schedule production and de-
conjoint–rules S2-T1-M1 and S2-T1-M2 are the optimum options, livery of RMC according to penalty fee, sum cost, or travel time.
and S1-T2-M2 is the least suitable in terms of penalty. On March 3, Decision-making depends on the practical situation and the manager's
in terms of sum cost, conjoint–rules that include rule M2 are the individual preference. Certain general conclusions can be drawn as
optimum options, and conjoint–rules that include rule M1 are the follows:
least suitable. On March 11, conjoint–rule S2-T1-M2 is the optimum
option, and S1-T1-M1 is the least suitable. On the basis of (1) The traveling times of vehicles under various sets of conjoint-rules
TravelTime, all the conjoint-rules are approximately equally effec- are affected to a significantly higher extent by discrete degree of
tive on March 11. On March 3, conjoint-rule S1-T2-M2 is the op- construction sites than quantity of orders.
timum option, and conjoint-rules S1-T2-M1 and S2-T1-M1 are the (2) Generally, selecting rule M2 can reduce sum cost; however, it is
least suitable. likely to increase penalty fee.
(3) In Case type 3 (concentration-sites and more-orders case):- On
March 10, conjoint–rules S1-T1-M1 and S2-T1-M1 are the optimum 6. Conclusion
options, and S1-T2-M2 is the least suitable. On March 12, in terms
of penalty, conjoint–rules S1-T1-M1 and S1-T2-M1 are the optimum It is necessary that managers have the capability to develop an ef-
options, and conjoint-rule S2-T2-M2 is the least suitable. On March fective schedule for RMC production and delivery with the aid of
10, from the perspective of sum cost, conjoint–rule S2-T2-M2 is the computers, particularly in various types of practical cases. Few previous
optimum option, and conjoint-rule S1-T1-M1 is the least suitable. studies have considered RMC production scheduling and vehicle (trucks
On March 12, conjoint–rule S1-T1-M2 is the selection, and S2-T2- and pumps) dispatching in a plant with multiple mixers and classified
M1 is the least suitable. On the basis of TravelTime, all the conjoint- practical cases into various types of cases.
11
Z. Liu et al. Automation in Construction 84 (2017) 1–13
12
Z. Liu et al. Automation in Construction 84 (2017) 1–13
04015034-1-11http://dx.doi.org/10.1061/(ASCE)ME.1943-5479.0000398. 1016/j.tre.2011.04.002.
[34] C. Low, R.K. Li, C.M. Chang, Integrated scheduling of production and delivery with [41] L. Asbach, U. Dorndorf, E. Pesch, Analysis, modeling and solution of the concrete
time windows, Int. J. Prod. Res. 51 (2013) 897–909 http://dx.doi.org/10.1080/ delivery problem, Eur. J. Oper. Res. 193 (2009) 820–835 http://dx.doi.org/10.
00207543.2012.677071. 1016/j.ejor.2007.11.011.
[35] V. Schmid, K.F. Doerner, R.F. Hartl, M.W.P. Savelsbergh, W. Stoecher, A hybrid [42] H. Pierre, N. Mladenovi'c, D. Uroševi'c, Variable neighborhood search and local
solution approach for ready-mixed concrete delivery, Transp. Sci. 43 (2009) 70–85 branching, Comput. Oper. Res. 33 (2006) 3034–3045 http://dx.doi.org/10.1016/j.
http://dx.doi.org/10.1287/trsc.1080.0249. cor.2005.02.033.
[36] M. Lu, M. Anson, S.L. Tang, Y.C. Ying, HKCONSIM: a practical simulation solution [43] M. Maghrebi, T. Waler, C. Sammut, Matching experts' decisions in concrete delivery
to planning concrete plant operations in Hong Kong, J. Constr. Eng. Manag. 129 dispatching centers by ensemble learning algorithms: tactical level, Autom. Constr.
(2003) 547–554 http://dx.doi.org/10.1061/(ASCE)0733-9364(2003)129:5(547. 68 (2016) 146–155 http://dx.doi.org/10.1016/j.autcon.2016.03.007.
[37] K.C. Ying, S.L. Tang, M. Anson, M. Lu, An experiment to explore the potential of [44] M. Maghrebi, S.T. Waller, C. Sammut, Feasibility study of automatically performing
simulation for improving ready mixed concrete delivery to construction sites, HKIE the concrete delivery dispatching by implementing machine learning techniques,
Trans. 12 (2005) 6–13 http://dx.doi.org/http://dx.doi.org/10.1080/1023697X. Eng. Constr. Archit. Manag. 22 (2015) 573–590 http://dx.doi.org/10.1108/ECAM-
2005.10668005. 06-2014-0081.
[38] T.M. Zayed, D.W. Halpin, Simulation of concrete batch plant production, J. Constr. [45] M. Maghrebi, V. Periaraj, S.T. Waller, C. Sammut, Using benders decomposition for
Eng. Manag. 127 (2001) 132–141 http://dx.doi.org/10.1061/(ASCE)0733- solving ready mixed concrete dispatching problems, in: A. Akbarnejad (Ed.), 31th
9364(2001)127:2(132. International Symposium on Automation and Robotics in Construction and Mining,
[39] V. Schmid, K.F. Doerner, R.F. Hartl, J.J.S. González, Hybridization of very large IAARC, Sydney, Australia, 2014, pp. 145–154. Available from: http://www.iaarc.
neighborhood search for ready-mixed concrete delivery problems, Comput. Oper. org/publications/fulltext/isarc2014_submission_171.pdf.
Res. 37 (2010) 559–574 http://dx.doi.org/10.1016/j.cor.2008.07.010. [46] P.K. Narayanan, D. Rey, M. Maghrebi, S.T. Waller, Using Lagrangian relaxation to
[40] J.P. Xu, F. Yan, S. Li, Vehicle routing optimization with soft time windows in a fuzzy solve ready mixed concrete dispatching problems, Transp. Res. Rec: J. Res. Boards
random environment, Transp. Res. E 47 (2011) 1075–1091 http://dx.doi.org/10. 2498 (2015) 84–90 http://dx.doi.org/10.3141/2498-10.
13