Global warming poses a threat to the natural environment and human economic development [1
]. Many studies have shown that global warming is mainly due to the increasing emissions of greenhouse gases (GHGs) caused by human activities, prompting companies to focus on carbon emission from production [3
]. Human impact on the environment is estimated by the carbon emissions generated during a products life. The assessment of carbon emissions can help companies provide more green products and services.
With the development of society and the increasing population, the consumption of garments is increasing year by year, thus, the carbon emissions of garment production cannot be ignored. In particular, the apparel industry is an important energy consumer and leads to environmental and health-related risks [6
]. According to reports, the apparel industry accounts for 10% of global carbon emissions, it is becoming the second largest industrial pollution source following the oil industry [7
]. However, the pollution caused by the garment industry has not been fully discussed in the literature, which is probably due to the environmental impact of the oil and gas industry. Sewing is one of the key processes in garment production and involves many operations [8
]. As shown in Figure 1
, the carbon emission sources of sewing process include fabrics, accessories, sewing machines and operators. At present, there are some obstacles in the evaluation methods of garment sewing production, such as the difficulty of tracking and monitoring the detailed production processes. So, there are few studies on carbon emissions in the textile and garment field [10
]. Nowadays, most research on carbon emission reduction focuses on spinning, weaving and materials. For example, in [11
], it was found that 1 kg of fabric emitted 12.5 kg of CO2
from the spinning, weaving, dyeing, cutting, sewing, finishing to transportation. The carbon emissions of the T-shirt production process were 12 times more than their own weight. In [12
], it was found that a cotton T-shirt weighing about 250 g emitted about 7 kg of CO2
during its service life, which was 28 times more than its own weight. For a pair of trousers with a service life of about 2 years and 10% polyester in composition, the total carbon emissions of the production and consumption were about 47 kg, which was 117 times more than its own weight. In [13
], it was found that the carbon emissions of the production process of apparel products could account for 70% of the total carbon emissions of the product. In [14
], carbon emissions for cotton and wool fabric products were evaluated. The results pointed out that the industrial carbon emission of wool fabrics was almost three times that of cotton fabrics. The industrial carbon emissions of yarn-dyed fabric was higher than that of dyed fabric, and the industrial carbon emissions of plain weave fabric was higher than that of rib fabric. The consumption of energy, for instance electricity, steam and coal, was the main source of industrial carbon emissions. In [15
], the energy consumption of the plants in the entire knitted garment production chain was investigated. Specific energy consumption and carbon emission for producing one piece of knitted garment from dyed to finished fabric was found as 0.78–1.44 MJ/piece and 0.09–0.17 kgCO2
/piece, respectively. The steam production, compressors and lighting equipment had a considerable share in total energy consumption and cost. Application of energy efficient lighting equipment was found to have the highest energy saving potential. In [16
], the ecological footprint of a garment factory was studied. The evolution of environmental impacts caused by equipment performance was evaluated and the environmental performance of different manufacturing processes was compared. The results indicated that the value of the carbon footprint was influenced by the type of fabric used in the manufacturing process. In [17
], the carbon footprint of an apparel plant producing jackets was studied. The cutting process was the most energy intensive stage. Other apparel companies [18
] also reduced carbon emissions of organic cotton T-shirts by actively implementing measures.
In terms of textile and garment production, many companies have a large space for carbon emission reduction, but there is no convincing calculation model to guide them. Therefore, it is very necessary to establish an evaluation model for greenhouse gas emissions from garment sewing production and seek potential energy conservation measures.
The balanced assembly line can not only improve the production efficiency, but also can ensure the effective timing of the machines [19
] and reduce the carbon emission in the production process. The balance study of garment sewing assembly lines is mainly a single model [21
]. Most research mainly focuses on the mechanical and electronic field, while little research focuses on the balance optimization of the sewing line [23
]. In [24
], a genetic algorithm was applied to optimize operator assignment and minimize maximum completion time for apparel assembly lines. In [25
], a universal mathematical model of the job shop scheduling for the apparel assembly process was constructed. A genetic optimization process was presented to solve this model. In [26
], a genetic algorithm was applied to solve operator assignment in predefined workstations of an assembly line. In [27
], a grouping genetic algorithm was developed for sewing lines with different labor skill levels in the garment industry.
In the real world of assembly lines, it is highly desirable to achieve two or more objectives simultaneously. In most cases, these objectives may conflict with each other, and the performance of each objective may be not improved without sacrificing the performance of at least one objective. Therefore, most solutions of multi-objective problems are to find the trade-offs between objectives. In the Pareto set, many solutions give different values for multiple objectives in each solution. From these solutions, a solution that gives the most optimistic objective value for all objectives is considered a better solution. In other manufacturing industries, [28
] considered simultaneously minimizing the separation time, workload changes and worker costs. A multi-objective genetic algorithm was proposed to obtain the solution. In [29
], multi-objective optimization of random disassembly lines was proposed. The aim was to minimize the number of workstations and minimize the design costs associated with labor and equipment. In addition, a new multi-objective genetic algorithm was proposed to obtain the solution. In [30
], a multi-objective genetic algorithm was proposed to solve a U-shaped assembly line balance problem. A two-stage genetic algorithm was derived by [31
] to solve the mixed model U-shaped assembly line. In [32
], a hybrid genetic algorithm was proposed to solve a mixed model assembly line balance problem. There were three objectives to be achieved: minimize the number of workstations, maximize the workload smoothness between workstations and maximize the workload smoothness within workstations.
Several multi-objective optimization studies on the assembly line problem consider minimizing pitch time as one of their objectives. In the assembly line, the minimization of pitch time may not be enough, and all tasks assigned to different workstations may not be completed in the expected pitch time. Therefore, smooth workload allocation in the assembly line is very important to reduce workload changes in workstations. In this research, the minimum pitch time and smoothness index of the assembly line are studied.
Generally, the assembly line is arranged by three types of workstation layouts, as shown in Figure 2
. The order of processes is where the workstations are arranged according to the processing order. The type of machine workstation layout is where machines required for the same processing content are arranged in the same workstation. The components of the garment workstation layout is where each workstation is arranged so it can produce each garment component.
The purpose of this research was to establish carbon emission evaluation models for garment materials, sewing machines and operators. The assembly line balance results were obtained by the NSGA-II (fast and elitist multi-objective genetic algorithm) algorithm. The effect of different garment sewing assembly lines on carbon emissions was studied. This paper is organized as follows: the carbon emission evaluation models for fabrics, accessories, sewing machines and operators are established in Section 2
; the assumptions, constraints and objective function of the sewing assembly line balance model are described in Section 3
; the NSGA-II algorithm design is specifically described in Section 4
; men’s shirt sewing lines are used for application analysis in Section 5
; the study is discussed in Section 6
; in Section 7
, several conclusions are summarized.
4. Multi-Objective Balance Optimization Method
In this paper, the NSGA-II algorithm is used to calculate the multi-objective balance model of the assembly line. The optimization procedure using NSGA-II is summarized in Figure 3
, following that the steps of the genetic algorithm are shown as words.
- Step 1
- Step 2
The overall size is set, and the initial population is generated based on the constraints (Section 4.2
- Step 3
According to the objective function value of each individual, the fast-non-dominant sorting of the contemporary population (Section 4.3
) and the congestion distance (Section 4.3
) are calculated.
- Step 4
Based on the results of the ranking and congestion distance calculations, the tournament mechanism is adopted for selection (Section 4.4
), then two individuals are randomly selected for crossing (Section 4.5
) and mutation (Section 4.6
- Step 5
Elite retention strategy is adopted (Section 4.4
). The father and the subpopulation are combined. A new generation of populations is generated by combining the populations by rapid non-dominant sequencing and virtual congestion distances.
- Step 6
The number of iterations is increased by 1, and the operation returns to step 3. This loop will continue until the maximum number of iterations is reached.
The NSGA-II algorithm can be simply divided into six steps: (1) Coding, (2) Population Initialization and Decoding, (3) Fast-non-dominated sorting and Calculating congestion (4) Selecting operation, (5) Crossing operation (6) Mutation operation.
In this research, real numbers are used for coding. The processes are arranged in a row according to the priority relationship, and each process corresponds to a gene position in the chromosome. As shown in Figure 4
, the largest number of processes is set as the backbone, while the other tributary processes are set as branches. If there is no priority relationship between the branch (process 6 and 7) and backbone (process 1–5), the order of branch (process 6 and 7) can be arbitrarily arranged.
4.2. Population Initialization and Decoding
In this research, the initial population is generated randomly. Before the fitness value of the chromosome is calculated, chromosome coding needs to be segmented according to the requirements of the optimization objective, and the allocation of processes in each workstation is obtained. The decoding method can be illustrated as following steps:
- Step 1
The minimum theoretical production pitch time, the value range of the pitch time, and the minimum number of workstations are calculated.
- Step 2
The processes are assigned to the workstations according to the sequence of processes in the chromosome. If the operating time of each workstation is less than the theoretical pitch time, the pitch time is the smallest in the chromosome arrangement, the iteration is stopped. Otherwise, the next step is taken.
- Step 3
when the ith process is assigned to the jth workstation, the operating time is less than the minimum pitch time, the (i+1)th process is added to the jth workstation. Then it is judged whether operation time of the jth workstation meets the pitch time limit after the (i+1)th process is added.
- Step 4
If the operating time of the jth workstation is greater than the maximum pitch time, the ith process is assigned into the (j+1)th workstation.
- Step 5
The above allocation processes are repeated until all the processes in the chromosome have been assigned.
4.3. Fast Non-Dominated Sorting and Calculating Congestion
The fast-non-dominated sorting solution is found and classified into the first-level non-dominated solution. After that, a new non-dominated solution is found in the remaining solutions and classified into the second-level non-dominated solution. The operation is repeated until all solutions are assigned.
If the ranks of the non-dominated solutions are the same, it is necessary to distinguish the advantages and disadvantages according to the congestion distance of the non-dominated solutions. And the evenly distributed solution is preferentially selected because of the relatively large crowding distance. The solutions with a large congestion distance and uniform distribution are preferentially selected.
4.4. Operation Selection
In order to ensure the average dispersion of the non-dominated solution frontier, the population evolution is chosen to be close to the position of the non-dominated optimal solution. The principle of binary tournament selection is to arbitrarily select two non-dominated solutions (i and j) into the mating pool. The individual i is better than j when the non-dominated level ; if , the congestion distance is .
Elite retention strategies are adopted to accelerate the rate of convergence.
4.5. Crossing Operation
In this research, a PMX-like (Like Partial-Mapped Crossover) approach is used to deal with real-coded problems. The specific operations are as follows:
- Step 1
A crossover region is randomly selected in the parent chromosome. It is assumed that the intersection of two parental chromosomes is: A = 984 ∣ 256 ∣ 137; B = 871 ∣ 436 ∣ 529, where “|” represents the intersection area.
- Step 2
The intersection region of the A chromosome is inserted in front of the B chromosome. The intersection region of the B chromosome is inserted in front of the A chromosome. Two intermediate individuals A’ and B’ are obtained.
- Step 3
The same genes in A’ and B’ are deleted, and the two new chromosomes A’’ and B’’ are obtained. Figure 5
shows the crossing operation.
4.6. Mutation Operation.
In this research, the inversion mutation operator is used. The specific operations are as follows:
- Step 1
Two mutation points are randomly selected on the chromosome. It is assumed that chromosome A = 87239546, and 2 and 5 are selected as mutation points, namely: A = 87 ∣ 2359 ∣ 46, where “|” represents the intersection area.
- Step 2
The gene of the variant region is inserted into the original gene position in reverse order to obtain a new chromosome A’. The mutation operation is shown in Figure 6
Much literature only adopts one workstation layout to solve the balance optimization of the assembly line [38
]. In this research, the workstations are arranged in three different ways: the order of processes, the type of machines and the components of the garment, which can provide various solutions for the assembly line layout of the enterprise.
After the assembly lines are optimized, the number of workstations, operators and sewing machines in three workstation layouts are reduced. The time loss rate of the three optimized assembly lines (the order of processes, the type of machines and the components of the garment) meets the requirements of the minimum production efficiency of the assembly line. Assembly line balance can improve production efficiency and reduce production costs. This is basically consistent with the conclusions of other research scholars on the optimization problem of assembly line balance [40
When the workstations are arranged according to the order of processes and the type of machines, the numbers of workstations on the assembly line are same. However, the time loss rate and smoothness index of BLP is less than that of BLW. The transfer distance of the production materials in BLP is shorter than that of BLW, but the machines utilization rate of BLW is higher. There are seven workstations that need only one sewing machine and one workstation that needs five sewing machines in the BLP and BLW. However, there are only three workstations that need one sewing machine in BLG. The production of each garment component requires different sewing machines in the BLG [42
]. And the assembly line has great flexibility to adapt to the change of the garment variety [43
According to the carbon emission evaluation models, the types and quantities of fabrics and accessories are related to the garment production plan. The carbon emissions from the sewing machines consuming electricity are mainly related to the processing time and the time loss of the assembly line. The carbon emissions from the operators’ breathing are related to the number of operators and the operating time. The number of workstations, sewing machines and operators in the optimized assembly line are reduced. Therefore, the change in carbon emissions is mainly caused by the power consumption of the sewing machines and operator’s breathing. The number of workstations, sewing machines and operators are the same in BLP and BLW. However, the time loss rate and smoothness index of BLP are lower, and the sewing machines consume less energy and produce less carbon emissions. Therefore, the balance of the assembly line can reduce the carbon emission.
The apparel industry is the sixth largest energy consuming industry sector in China. Therefore, it is necessary to conduct a detailed analysis of garment sewing production to find relevant carbon emission characteristics for the purpose of the reduction of carbon emissions. In this research, carbon emission evaluation models for garment sewing production were established, while the carbon emissions of fabrics, accessories, sewing machines and operators in men’s shirt sewing assembly lines were calculated. A multi-objective assembly line balance optimization model was also constructed to reduce carbon emissions. The evaluation models and assembly line balance method could provide a basis for the quantification of carbon emissions of assembly lines and provide a reference for reducing carbon emissions of garment manufacturers. The main conclusions are listed as follows.
(1) According to the calculation results of the evaluation models, fabrics and accessories were the main sources of carbon emissions in garment sewing production, followed by sewing machines and operators.
(2) According to the evaluation models, the carbon emissions from fabrics and accessories were mainly related to the materials. The main factors affecting the carbon emissions generated by the sewing machines were the type of machines, the number of machines and the effective running time. The carbon emissions of operators were mainly related to the number of operators, labor intensity and working time. More importantly, the number of sewing machines and operators, and the working time were mainly affected by the balance of the assembly line.
(3) Compared to the normal assembly line, the number of workstations, sewing machines and operators of optimized assembly lines in the three workstation layouts (the order of processes, the type of machines and the components the of garment) were reduced, while the time loss rate of the optimized assembly lines was reduced. The BLP was the most effective assembly line with the shortest time loss.
(4) The quantities of carbon emissions in balancing and normal assembly line were calculated based on carbon evaluation models, they indicated that the total quantities of carbon emissions of the optimized assembly lines were lower than that of normal assembly line. Among the three optimized assembly lines, BLP has the lowest carbon emissions. Balancing the assembly line based on the order of process workstation layout was the best one since it was most effective assembly line with the lowest carbon emissions.
The carbon emission evaluation models of the garment sewing production provide a calculation basis for the quantitative study of carbon emissions in garment sewing production, while the multi-objective balance optimization model of garment sewing assembly lines provides a method for reducing carbon emissions in garment sewing production. In this study, the carbon emission evaluation models of sewing processes are established, but garment production also includes cutting and packaging processes. So, a more complete carbon emission model for garment production will be established in the future.