TL;DR
Researchers have developed a Modified Grey Wolf Optimizer (MGWO) to solve the complex, NP-hard problem of Disassembly Sequence Planning (DSP). By introducing specialized operators for feasibility and search, the algorithm significantly outperforms traditional meta-heuristics in finding the most efficient, cost-effective, and stable ways to disassemble end-of-life products.
What Problem Does This Paper Address?
As modern products reach their "end-of-life" (EOL), the manufacturing industry faces a massive challenge: how do we take them apart efficiently for recycling and remanufacturing? This isn't just about unscrewing bolts; it’s a high-stakes puzzle called Disassembly Sequence Planning (DSP).
The goal is to find a sequence that minimizes disassembly time, tool changes, and material mixing while maximizing the stability of the product during the process. Because the number of possible sequences grows exponentially with the number of components, DSP is a classic NP-hard combinatorial optimization problem. Traditional exact algorithms fail as products get complex, and existing "intelligence algorithms" often require tedious parameter tuning that doesn't adapt well to different types of hardware.
Methods
The authors propose the Modified Grey Wolf Optimizer (MGWO), building on the original Grey Wolf Optimizer (GWO) which mimics the social hierarchy and hunting behavior of wolves.
Standard GWO is designed for continuous math problems, so the authors introduced three novel operators to make it work for the discrete, constraint-heavy world of mechanical disassembly:
- Feasible Solution Generator (FSG): Mechanical parts have "precedence constraints" (you can't remove a piston before the engine head). FSG uses a decimal interference matrix to ensure every generated sequence is actually physically possible.
- Neighborhood Search Operator (NSO): This allows the "leader wolves" ($\alpha, \beta$, and a newly generated $\delta$) to self-evolve. It randomly re-inserts components within a valid range to find local improvements in the sequence.
- Guided Search Operator (GSO): This manages the global search. It uses self-adaptive factors based on fitness values to determine how much the rest of the "pack" should be influenced by the three leaders.
The algorithm evaluates sequences using a multi-indicator fitness function that weights four factors: changes in disassembly direction, tool changes, material changes (to group similar recyclables), and structural stability.
Datasets
The paper evaluates the MGWO using three distinct engineering benchmarks:
- Case 1: A Vise (11 components): A standard test case to verify the effectiveness of the modified strategies.
- Case 2: A Ball Valve (17 components): A more complex assembly used to measure the algorithm's significance and robustness.
- Case 3: Azimuth Thruster Propeller (17 components): A real-world industrial application from a Chinese boat company, representing a high-value product where efficient recovery is economically vital.
Key Insights and Ideas
- The "Delta" Wolf Innovation: In standard GWO, the $\delta$ wolf is simply the third-best solution. In this paper, the $\delta$ wolf is a newly generated feasible solution. This prevents the pack from getting stuck in "local optima" (good-but-not-great solutions) by maintaining population diversity.
- Self-Adaptive Parameters: Unlike algorithms like Particle Swarm Optimization (PSO) or Genetic Algorithms (GA), which require manual tuning of inertia weights or mutation rates, MGWO’s parameters are self-adaptive based on the fitness of the wolves. This makes it much more robust for different types of machinery.
- Balancing Local and Global Search: The combination of NSO (local refinement) and GSO (global pack movement) allows the algorithm to converge on the optimal solution much faster than its predecessors.
Key Results
The MGWO was compared against five major algorithms: Genetic Algorithm (GA), Ant Colony Optimization (ACO), Simplified Swarm Optimization (SSO), Teaching-Learning-Based Optimization (TLBO), and Improved Harmony Search (IHS).
- Higher Reliability: For the Vise case, MGWO found the optimal solution with a 63.33% probability, while GA and ACO only managed 13.33% and 16.67%, respectively.
- Superior Convergence: In all cases, MGWO reached the optimal fitness value (e.g., 3.7500 for the ball valve) in significantly fewer iterations than the other methods.
- Real-World Success: In the industrial propeller case, MGWO achieved the best mean fitness (5.5669) and the highest probability of hitting the absolute optimum (76.67%), proving its practicality for complex engineering.
Limitations and Future Work
The authors identify two main areas for improvement:
- Leader Dependency: Currently, only the three leading wolves perform the neighborhood search. The remaining wolves simply follow the leaders, which limits their individual "self-exploration" potential.
- Model Complexity: The current mathematical model primarily focuses on component interference. Real-world disassembly often involves more nuanced constraints, such as varying labor costs or specialized environmental hazards, which the authors plan to integrate into future versions of the DSP model.
Summary updated