Disassembly sequence planning using a Flatworm algorithm

Highlights

• A new Flatworm algorithm for disassembly sequence planning model is proposed.

• Novel growth, fracture and regeneration mechanisms are explored.

• Solution quality is efficient in improving as complexity of problems increase.

Abstract

Disassembly Sequence Planning (DSP) refers to a disassembly sequence based on the disassembly properties and restrictions of the product parts that meets the benefit goal. This study aims to reduce the number of changes in disassembly direction and disassembly tools so as to reduce the disassembly time. This study proposes a novel Flatworm algorithm that evolves through the regenerative properties of the flatworm. It is similar to the evolutionary concept of genetic algorithms, with evolution as the main idea, but without crossover, mutation or replication mechanisms in the evolutionary processes. Instead, it is based upon the characteristics of the growth, fracture and regeneration mechanisms of the flatworm. The Flatworm algorithm features a variety of disassembly combinations and excellent mechanisms to avoid the local optimal solution. In particular, it has the advantage of keeping a good disassembly combination from being destroyed. In this study, it is compared with two genetic algorithms and two ant colony algorithms and tested in three examples of different complexity: a ceiling fan, a printer, and 150 simulated parts. The solution searching ability and execution time are compared upon the same evaluation standard. The test results demonstrate that the novel Flatworm algorithm proposed in this study is superior to the two genetic algorithms and ant colony algorithms in solution quality.

Introduction

Disassembly Sequence Planning (DSP) represents the disassembly process of a product. When it comes to the end of a product’s life cycle, DSP can effectively evaluate the benefits of disassembly and obtain a disassembly process yielding the maximum profit or minimum cost. The issue of DSP is often combined with green design. The purpose of green design is to make the product meet the requirements of environmental protection so as to reduce the impact on the environment. This is also the reason why DSP research is increasingly valued [[1], [2], [3], [4], [5], [6]].

In the past, DSP research was often mixed with assembly sequence planning (ASP). ASP has to consider the relationship between parts as well as related constraints such as assembly time, geometric features, tools, and machines to generate the priority order of assembly. Traditionally, ASP solves this problem by first modeling the product into a so-called liaison graph and then solving the problem through a combination of graph theory and exhaustive search methods [3]. Homem De Mello and Sanderson (1991) once considered ASP as the inverse of DSP; this conception should be seen as the beginning of DSP research [7]. As far as the nature of DSP is concerned, disassembly can be divided into Total Disassembly and Selective Disassembly. Total disassembly occurs when the disassembly target is clear; thus, the focus lies on searching for the optimal disassembly sequence of all parts that matches the objective (minimum cost or maximum effectiveness) under the constraints. In contrast, selective disassembly focuses on searching for parts or modules that are worth disassembling at the end of a product’s life cycle [8]. At that time, the recycling company can process related materials according to a predetermined recycling plan. This study focuses on total disassembly.

The DSP problem is an NP-hard problem, and its solution cannot be found in polynomial time [3]. In early research, graph theory was combined with other methods, such as linear programming and branch-and-bound or shortest-path algorithms to find the optimal disassembly sequence. However, as the complexity of the problem increases, the result cannot be obtained in a reasonable time [[9], [10], [11]]. In light of this phenomenon, a heuristic algorithm cannot guarantee the optimal solution, but it is usually possible to get good results in a reasonable time, which is why the heuristic algorithm was popular in later research [[12], [13], [14], [15], [16], [17], [18], [19]].

Among the many heuristic algorithms used in DSP problems, genetic algorithms (GAs) are common. GAs were developed from Darwinian evolution, mimicking the principles of biological evolution through crossover, mutation, and replication mechanisms. Through generations, species continue to evolve, resulting in more adaptable offspring [20]. The span from the crossover mechanism to the replication mechanism is regarded as a generation. The chromosomes produce offspring chromosomes through the crossover mechanism, which makes the offspring chromosomes diverse while retaining some of the original gene fragments. The mutation mechanism helps to increase chromosome diversity. This algorithm considers the chromosome of the best adaptation as the near-optimal or optimal disassembly sequence. Among GA studies, Seo et al. pioneered the application of GAs in disassembly planning [21]. Later, Kongar and Gupta developed the so-called Precedence Preservative Crossover (PPX) mechanism [22], which makes the chromosomes have diverse characteristics in the crossover process. PPX was strongly supported by other scholars because it increases the search scope and avoids premature convergence [23]. Rickli and Camelio (2013) discussed DSP issues from the three viewpoints of cost, revenue, and environmental impact with multi-objective GAs [24]. Based on the directional search, a novel Block-based gene algorithm is offered to increase the efficiency of solution searching, in which the scope of the search is greatly reduced and the efficiency of the algorithm is improved [25,26]. All of these researchers used the concept of GAs to explore the DSP problem.

Ant Colony Optimization (ACO) is another method that has been employed in recent years. ACO constructs the next feasible path through state transformation in a feasible path, so its search mechanism is a non-blind search. The state transformation method includes both local search and local solution escape mechanisms, and it uses the pheromone matrix as feedback, featuring the dual benefits of fast convergence and good solution quality [27]. In the field of DSP, the Disassembly Completed Graph was proposed to represent all possible disassembly sequences where ACO is adopted to find the optimum solution [28]. Alternatively, Liu et al. dealt with DSP problems with an improved max–min ant system (MMAS) [29]. There are two main differences between the MMAS and traditional ACO. First, in the pheromone update phase, an iterative optimal solution update is used to avoid losses in exploration, and then a global best solution will be updated every few generations. Second, the pheromone update function is limited by the upper and lower limits. Moreover, a framework similar to the Bill of Material (BoM) hierarchical structure was used to solve the selective disassembly problem [30]. Kheder et al. compared ACO with GAs in the performance of DSP problems [31], while Tseng et al. explored the application of five different ant algorithms to DSP problems [32].

By observing the characteristics of the regeneration ability of the flatworm from broken parts, this study attempts to propose a more efficient novel flatworm algorithm. The flatworm is a multicellular organism that lives under the rocks of ponds and streams or in seawater. Some flatworms can live in soil. The regenerative ability of the flatworm comes from the pluripotent stem cells in the body. The pluripotent stem cells are cells with the differentiation ability. When the flatworm faces stimuli, such as autologous division, cleavage, or any injury, the original pluripotent stem cells are altered by stimulation and differentiated into specific cells to replenish the insufficient cells. In addition to the differentiation ability, the pluripotent stem cells also have a self-renewal ability, providing more pluripotent stem cells for differentiation and development. The pluripotent stem cells are different from normal cells. Normal cells will no longer differentiate after the differentiation process and stop the cell proliferation. But the pluripotent stem cells of the flatworm have growth and tissue repair functions. As shown in Fig. 1, the two cut flatworm sections will grow into two complete flatworms, each of which will grow the missing parts through their regeneration ability [33,34].

The Flatworm algorithm (FA) proposed in this study is very different from GAs and ACO. A FA regards the disassembly sequence as a flatworm. Only the splitting probability and the number of generations are considered as parameters in a FA. The generation evolves through the growth, splitting, and regeneration mechanisms. With the splitting mechanism, the randomly-generated flatworms begin to divide, and the weaker parts of the flatworm are more likely to break. In the disassembly sequence, there is discrimination of good and bad sequential combinations. In this study, a poor combination of parts in the disassembly sequence is considered to be easy to break. Via the destruction of the poor disassembly combination and then regeneration of the lost parts through the regenerative ability, a new disassembly sequence, as shown in Fig. 1, will be generated. As shown in the figure, the flatworms are constantly evolving, and the best disassembly sequence will be sought by continuous destruction of the poor disassembly sequential combinations and regeneration of better disassembly sequential combinations. In this study, DSP solution tests are executed in FAs and compared with GAs and ACO. First, the GAs proposed by Kongar and Gupta [22] (denoted as K&G) and Block-based genetic algorithms [25] are selected for comparison with the novel Flatworm algorithm in a DSP problem. Second, the Ant Colony Optimization [26] and Max–Min Ant System [28] are selected for the benchmark comparison.

Before executing the algorithm, we must understand the restrictions and evaluation criteria of disassembly in parts of the product. To keep the solution in the algorithm a feasible one, we need to consider the disassembly limits of the parts because the steps in the disassembly sequence will interfere with one another during the disassembly process. Otherwise, a large number of infeasible solutions will be generated in the algorithm process, decreasing the efficiency of calculation. The disassembly issue in this study adopts the disassembly direction and disassembly tools of the parts as two criteria in an attempt to find the fewest changes in disassembly direction and tools. Moreover, the concept of penalty value is integrated in the direction and tool changes so as to evaluate the advantages and disadvantages of certain disassembly sequences. In Section 2, we introduce the framework of the disassembly planning, and in Section 3, we describe the process and offer simple examples of the FAs. The results of the example tests are compared in Section 4, and the results and future recommendations are discussed in Section 5.