A method for balancing a human-robot collaborative disassembly line with detection-driven adaptation

Abstract

The application discloses a kind of detection driven adaptive man-machine cooperation disassembly line balancing method, this method includes: obtaining the disassembly task set of waste product, priority relationship, standard man-hour of each task under different operation mode, and robot resource upper limit;Build two-stage disassembly task model, determine the task before detection node as pre-disassembly mandatory task set, determine the task after detection node and the task of work time affected by part integrity state as time variable task set, the rest is detection normal task set;Establish multi-objective optimization mathematical model, and set constraint condition;For the task in time variable task set, different working condition adjustment coefficient is introduced according to different integrity state, to unify the working position work time under the constraint of each state;Improved multi-objective parrot optimization algorithm is used to solve the mathematical model, and man-machine cooperation disassembly scheme is obtained.The application realizes the adaptive balancing of man-machine cooperation disassembly line by detection driving.

Description

Technical Field

[0001] This invention belongs to the field of production system optimization and intelligent manufacturing scheduling technology, specifically involving a detection-driven adaptive human-machine collaborative disassembly line balancing method. Background Technology

[0002] The disassembly line balancing problem is an important research direction in the field of production system optimization. It aims to rationally allocate the product disassembly process to workstations on the assembly line in order to optimize cycle time, balance workstation load, and improve resource utilization efficiency.

[0003] In terms of disassembly logic models, the mainstream disassembly logic models can be summarized as complete disassembly models, goal-oriented disassembly models, and sequence generation models. All of them are built on the static assumption that the disassembly depth or goal is predetermined. Their process set and process path are fixed during planning. In theory, they maximize equipment utilization and personnel efficiency, and achieve static optimization of resource allocation.

[0004] In the face of unavoidable fluctuations in working hours during actual disassembly, existing probabilistic, fuzzy, and robust optimization methods are mainly dedicated to handling random fluctuations in task working hours. They also ensure robustness by establishing multiple separate constraints for different scenarios, thereby improving the adaptability of production line design to fluctuations in the actual operating environment.

[0005] In the field of remanufacturing and resource recycling, the uncertainty of product condition is the biggest challenge. Due to long-term use, parts may become unremovable or require special treatment due to wear and corrosion. Task execution time is no longer a fixed value but follows a probability distribution. Although existing uncertainty handling methods can handle the randomness of time, they have failed to effectively establish a causal quantitative relationship between physical condition and operation time.

[0006] With the development of collaborative robot technology, human-robot collaborative disassembly has attracted attention due to its ability to combine human flexibility with machine precision. Existing research has established a basic framework for human-robot task allocation, explored principles for dividing human and robot roles based on task characteristics, fixed-role collaboration modes, and rule-based dynamic adjustment modes, and analyzed the potential of human-robot collaboration in improving disassembly efficiency and reducing labor intensity, laying the foundation for achieving more efficient and intelligent disassembly operations.

[0007] In terms of economic benefit assessment, existing studies typically calculate costs based on task time, assuming that the dismantling operation can be completed 100% successfully and generate expected benefits. Benefit calculations use a fixed recovery value, and cost models primarily consider operation time costs, resulting in a time-centric economic benefit assessment method.

[0008] After constructing a mathematical model of a human-robot collaborative dismantling line that considers uncertainty and complex constraints, efficiently solving this model becomes another technical bottleneck. The balancing of a human-robot collaborative dismantling line is a typical NP-hard discrete combinatorial optimization problem, containing strict constraints on process priority relationships, complex upper limits on workstation resources (number of robots), and multi-dimensional optimization objectives. Traditional exact solution algorithms are prone to combinatorial explosion when facing medium-to-large-scale problems, resulting in low solution efficiency. Conventional metaheuristic algorithms (such as genetic algorithms and particle swarm optimization) often struggle to balance global exploration and local development when dealing with multi-layered discrete decision variables (task ordering, mode selection, resource allocation), and are prone to generating illegal solutions that violate process sequence logic or getting trapped in local optima during iterative mutation, leading to low optimization efficiency and poor solution quality. The Parrot Optimization Algorithm (POA) is an emerging swarm intelligence algorithm, but it still falls short in handling sequence optimization problems with complex hard constraints. Summary of the Invention

[0009] The main objective of this invention is to provide a detection-driven adaptive human-machine collaborative dismantling line balancing method to solve the problems of low production efficiency and unreasonable resource allocation caused by the uncertainty of part status during the dismantling of waste products, and to achieve adaptive balancing and efficient scheduling of the dismantling line under multiple working conditions.

[0010] The present invention achieves the above objectives through the following technical solutions: In a first aspect, the present invention provides a detection-driven adaptive human-machine collaborative disassembly line balancing method, the method comprising: Obtain the set of dismantling tasks for waste products, their priority, the standard working hours of each task under different operating modes, and the upper limit of robot resources; A two-stage disassembly task model is constructed. Tasks located before the inspection node are defined as the pre-disassembly forced task set, tasks located after the inspection node whose operation time is affected by the integrity status of the parts are defined as the time-variable task set, and the rest are the normal task set after inspection. A multi-objective optimization mathematical model is established with the optimization objectives of minimizing production cycle time, maximizing expected comprehensive profit, and minimizing load imbalance, and constraints are set. For tasks in the time-variable task set, different working condition adjustment coefficients are introduced according to different integrity states to uniformly constrain the workstation operation time under each state. The mathematical model is solved using an improved multi-objective parrot optimization algorithm, including: An initial population was generated using a three-layer collaborative coding strategy. Through the decoding strategy, the expected operation time of each task is calculated based on the probability of occurrence of different integrity states and the corresponding work condition adjustment coefficient. Based on the expected operation time, the code is converted into a workstation allocation scheme to form a fixed workstation physical layout, and the multi-objective fitness value is calculated. A progeny population is generated by performing update operations on individuals that simulate four parrot behaviors: foraging, loitering, communicating, and fear. The parent and offspring populations are merged, and the new generation population is retained after non-dominated sorting and crowding distance screening. After iterating to the termination condition, the Pareto front solution set is output as a human-machine collaborative disassembly scheme.

[0011] Following the above technical solution, the objective function of the multi-objective optimization mathematical model is: Minimize production cycle time : ; Maximize expected total profit : ; ; ; Minimize load imbalance : ; ; In the formula, For production cycle time; In order to recover profits, For operating costs; The dismantling task number is used. To pre-disassemble the forced task set; Number the workstation. Maximum number of workstations; Number the working mode. This indicates that the work is done entirely manually. This indicates purely robotic operation. Indicates human-machine collaboration; Assign variables to tasks, if the task At the workstation The above pattern If executed, the value is 1; otherwise, it is 0. For the task Recovery revenue, For the task In mode The success rate of the following The probability that discarded products have recycling value without requiring destructive disassembly. For the task In mode The operating costs below; For workstations k Expected workload For the task In mode Standard operating time; The integrity status number is used for identification. A set of complete states with recycling value. Completeness status The probability of occurrence in the set of complete states with recyclable value and satisfying , For the task In mode and integrity status Predicted job time; Among them, the predicted operation time for: ; In the formula, A set of time-variable tasks This is the set of normal tasks after testing; This is the operating condition adjustment coefficient, which is applied when the task belongs to the set of normal tasks after detection. When the task belongs to a time-variable task set, This indicates the percentage increase in operation time due to changes in the physical state of the parts.

[0012] The constraints of the above technical solution include: (1) Unique task assignment constraint: ensure that each disassembly task is uniquely assigned. They were all assigned to the only workstation And determine the unique processing mode. No omissions or duplicate allocations are allowed: ; In the formula, For dismantling task sets; (2) Task priority constraint: There are priority constraints between tasks, and the preceding task must be completed first: ; In the formula, For the set of priority relationships of dismantling tasks, This indicates that task i is the direct predecessor of task j; (3) Workstation robot configuration constraints: If in the workstation For any task requiring robot or human-robot collaboration to be assigned, this workstation must be equipped with a robot. ; In the formula, As a decision variable, if the workstation If a robot is equipped on it, the value is 1; otherwise, the value is 0. For workstation assembly, It is a maximum number; (4) Robot resource limit constraint: The total number of robot workstations in use across the entire line shall not exceed the robot resource limit. : ; (5) Workstation cycle time constraint: The end time of all tasks in the assigned mode cannot exceed the production line cycle time. : ; In the formula, Let be the decision variable, representing the task. i The start time of the work cycle at its assigned workstation; (6) Preceding task constraint at the same workstation: If the task It is a task If the task is a preceding process and is assigned to the same workstation, then the task... The process must be later than the task. Processing steps: ; (7) Parallel constraints at the same workstation: For any two tasks that do not have a direct priority relationship i and tasks If the two are assigned to the same workstation Only when the processing modes of two tasks are one person per machine or one machine per person can they be processed in parallel: ; ; ； In the formula, Represents the set of patterns that can be parallelized: ; Let be the decision variable. If task i and task j are assigned to the same workstation and have no absolute priority relationship, then when task i is executed before task j, then... It is 1 if it is true, otherwise it is 0.

[0013] Following the above technical solution, the three-layer collaborative coding strategy includes: The first layer is a sequence of disassembly tasks that satisfy priority constraints, with a length equal to the total number of tasks. The second layer consists of the operation modes corresponding to each task, and is the same length as the first layer. The third layer is a binary sequence indicating whether each workstation is configured with a robot. Its length is equal to the maximum number of workstations, and the total number of 1s does not exceed the robot resource limit.

[0014] Following the above technical solution, the decoding strategy includes: Construct a state-task time matrix: This matrix addresses the different integrity states of a product after passing through inspection nodes. Calculate the actual operation time for each task, including: for the completeness set of states with recovery value. The state, time-variable task set The task operation time is extended according to the corresponding working condition adjustment coefficient, while the operation time of other tasks remains unchanged at the standard working time. For low integrity status, when the product integrity is lower than the recycling threshold and destructive disassembly is triggered, some tasks do not need to be executed, the corresponding operation time is set to zero, and they are not included in the recycling revenue calculation. Determine the expected operation time of each task: Based on the probability of occurrence of each integrity state, the operation time of each task under different states is weighted and summed to obtain the expected operation time of each task, which serves as a unified basis for workstation allocation. Task allocation and workstation configuration: Starting from the initial reference cycle time, the workstation configuration is gradually optimized through iterative iterations until the cycle time can no longer be compressed. This includes: allocating tasks to workstations sequentially according to the first-level disassembly task sequence; calculating the workstation working time after adding a task based on the expected operation time; if it does not exceed the current reference cycle time, it is allocated to the current workstation; otherwise, a new workstation is opened, and the task is set as the first task of the new workstation; during the allocation process, if the robot configuration of the current workstation conflicts with the task operation mode, the operation mode of the task is forcibly downgraded to pure manual mode; after all tasks are allocated, if the maximum workstation working time is less than the current reference cycle time, the reference cycle time is updated and the tasks are reallocated until the cycle time can no longer be compressed, forming a fixed workstation physical layout, i.e., the mapping relationship between tasks and workstations.

[0015] Following the above technical solution, the calculation method for workstation working time includes: Initialize the manual progress time for the current workstation. Time with robot progress =0; Determining the task start time includes: if the task has no preceding task in the current workstation, or only has a preceding task with the same executor, then its start time is directly taken as the current progress time of that executor; if there are preceding process constraints for different executors in the workstation, then the task start time is taken as the larger value between the corresponding executor's progress time and the completion time of the preceding task; for human-machine collaboration mode, its start time is forcibly aligned to the maximum value between the current human progress time and the robot progress time. After determining the start time, accumulate processing time according to the task's operation mode. If it is in purely manual mode, only the manual time will be updated. Robot progress time Keep it unchanged; if it is a pure robot mode, only update the robot time. Manual progress time The time remains unchanged; if it is a human-machine collaboration mode, the manual progress time is... Time with robot progress Simultaneous increase ; Once all tasks at the workstation have been completed, the maximum of the manual processing time and the robot processing time is taken as the total working time for that workstation; where... This refers to the operation time of the corresponding task in the current mode and completeness state.

[0016] Following the above technical solution, foraging behavior includes: Legal interval insertion of process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement interval according to the priority relationship matrix, move the task out of its original position and randomly insert it into any other position within the interval; Operation mode reset disturbance: Randomly select a task and change its operation mode to another mode other than the current mode; Robot allocation layer exchange disturbance: Randomly select a pair of binary bits in the robot allocation layer and exchange their positions to adjust the distribution of robots between workstations; Staying behavior includes: Feasible domain exchange of process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement range according to the priority relationship matrix, and randomly select another task within the range to exchange the positions of the two. Operation mode coordination and exchange: Synchronously exchange the operation modes corresponding to the two tasks mentioned above. Robot allocation layer swap perturbation: Randomly select a pair of binary bits in the robot allocation layer and swap their positions; Communication behaviors include: Fragment preservation and recombination of the process sequence: Randomly select a parrot individual from the non-dominated solution set of the current population as the guide; randomly select two intersection points on the task sequence, directly copy the task in the fragment of the current individual to the corresponding position of the offspring, and fill the remaining empty positions in the offspring sequence in the relative order of the remaining tasks in the guide. Operation mode collaborative mapping: The operation mode of the offspring is migrated synchronously with the corresponding task. That is, the task inherited from the current individual retains its operation mode, and the task inherited from the facilitator adopts the facilitator's operation mode. Robot allocation layer replication and perturbation: The offspring robot allocation layer is inherited from the current individual and a random bit swap perturbation is performed; Fear behaviors include: Local reconstruction of the process sequence: Randomly select a continuous subsequence in the current individual's task sequence and move the tasks in it into a temporary task pool; Based on the priority relationship matrix, randomly select one task from the currently executable tasks and fill it back into the original empty position in the sequence until all tasks in the task pool are filled back. Operation mode randomization reset: Randomly reselect the operation mode for all tasks within the reconstruction area; Robot allocation layer exchange perturbation: The robot allocation layer performs a random bit exchange perturbation.

[0017] Following the above technical solution, the calculation method for multi-target fitness values ​​is as follows: Production cycle time Under a fixed workstation physical layout, substitute the actual operation time matrix of each integrity state, calculate the actual cycle time of each state, and take the worst performance of the system as the overall cycle time. Expected comprehensive profit Based on the probability of occurrence of each integrity state, calculate the expected difference between the task recovery revenue and the operating cost to obtain the expected comprehensive profit; Load imbalance Based on the workload of each workstation under the expected state, the overall load variance of the system is calculated to obtain the load imbalance degree. The operating condition adjustment coefficient The probability that the waste product has recycling value and does not require destructive disassembly is determined by calibrating the ratio of measured working time to benchmark working time under different physical states of parts; The probability of occurrence of the completeness state s is obtained through frequency statistics of historical production data, expert Delphi scoring, or classification prediction models based on machine learning. The success probability of task i in mode m is obtained by classifying and statistically calibrating historical samples that have entered the flexible phase. The ratio of the number of successful executions of task i in mode m to the total number of executions is obtained from long-term production data.

[0018] In a second aspect, the present invention provides a computer device / apparatus / system, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0019] Thirdly, the present invention provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0020] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) It achieves precise allocation of large-scale tasks under complex resource upper limit constraints.

[0021] For the dismantling of waste electromechanical products, the method of this invention can effectively achieve the rational allocation and scheduling of tasks across multiple candidate workstations. Leveraging a designed three-layer collaborative coding and adaptive decoding mechanism, this invention strictly adheres to the hard resource constraint that the total number of collaborative robots does not exceed a preset upper limit during iterative optimization. By introducing a detection decision node, the task is decoupled into a set of pre-dismantling procedures before detection and a set of conditionally triggered procedures, thus recreating the actual physical process. The final generated solution ensures a one-to-one correspondence between the task execution mode and the workstation resource configuration, highly consistent with the actual physical resource state, and effectively achieves feasible solution search under complex constraints.

[0022] (2) The algorithm has excellent optimization performance and provides high-quality multidimensional Pareto decision schemes.

[0023] The improved multi-objective parrot optimization algorithm designed in this invention exhibits superior performance in solving large-scale discrete combinatorial optimization problems with strict priority relationships. The algorithm relies on four behavior-driven mechanisms to demonstrate a fast convergence speed and can generate a Pareto front solution set with excellent distribution. This solution set, while satisfying the feasibility of the process logic, successfully achieves the optimal balance between the three core optimization objectives of minimizing production cycle time, maximizing expected total profit, and minimizing load imbalance, providing high-quality and diverse decision-making solutions for multi-objective dismantling line balancing problems.

[0024] (3) Significantly improved the adaptability and scheduling robustness of the production line under extreme working conditions.

[0025] The unified boundary constraint model constructed in this invention, which includes operating condition adjustment coefficients and success probabilities, possesses excellent operating condition coverage. By modeling stochastic working hours based on the integrity status of parts, a quantitative correlation is established between the wear and corrosion degree of parts and the uncertainty of disassembly time, providing a key input for robust optimization. Experiments demonstrate that this method can adaptively handle uncertain scenarios ranging from detection interruption to extreme corrosion. Even under complex conditions such as unclear status of scrapped products, random detection results, and drastic fluctuations in working hours, it can still ensure the rationality and stability of the disassembly line scheduling scheme. Attached Figure Description

[0026] Figure 1 This is a process flow diagram of an embodiment of the present invention; Figure 2 This is a flowchart of an improved multi-objective parrot algorithm according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an encoding method according to an embodiment of the present invention; Figure 4 This is a process priority relationship diagram according to an embodiment of the present invention; Figure 5 This is a scatter plot of an approximate Pareto front according to an embodiment of the present invention; Figure 6 This is a Gantt chart of a process allocation scheme according to an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0028] Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, modifications to design, manufacturing, or production based on the technical content disclosed in this invention are merely conventional technical means and should not be construed as insufficient disclosure of the present invention.

[0029] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention may be combined with other embodiments without conflict.

[0030] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "a," "an," "an," "the," and similar words used in this invention do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," "coupled," and similar words used in this invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "A plurality" used in this invention refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship. The terms "first," "second," and "third" used in this invention are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0031] This invention provides a detection-driven adaptive human-machine collaborative dismantling line balancing method, specifically a human-machine collaborative dismantling line balancing and multi-objective optimization scheduling method for waste product recycling. It aims to address the systematic deficiencies of existing technologies in handling the uncertainty of waste product quality flow, and specifically provides a complete integrated solution to address the limitations of existing technologies in three aspects: dismantling process planning, uncertainty handling methods, and human-machine collaboration arrangements.

[0032] Existing models typically assume that the disassembly path is unique and predetermined, making it difficult to flexibly stop unnecessary subsequent disassembly work, resulting in wasted manpower and time. The two-stage disassembly planning model based on detection proposed in this invention, by introducing detection decision nodes, clearly divides the disassembly task into mandatory tasks that must be completed and flexible tasks whose execution is determined based on detection results. This allows the design of the disassembly line to flexibly adapt to the actual condition of the product.

[0033] Existing technologies generally neglect the inherent correlation between the physical state of parts and their disassembly time, assuming that all task times are a fixed constant, or using a coarse, uniform probability distribution to describe the time fluctuations of all tasks. This leads to a serious disconnect between production plans and actual conditions, making it impossible to conduct refined capacity forecasting and management. This invention establishes a quantitative correlation between the physical state of parts and production cycle time. By calibrating the time extension coefficients for different states, it improves the accuracy and guiding value of production planning.

[0034] In terms of personnel and machine allocation, existing research often assigns tasks based on the fixed characteristics of the tasks themselves, without simultaneously considering the probability of the task occurring and the difficulty of its execution. This may result in expensive equipment being assigned to tasks that rarely occur or are easily completed even when they do occur. With a limited number of machines, this fails to optimize the efficiency and stability of the entire production line. The multi-mode adaptive matching strategy proposed in this invention calculates the frequency of task occurrence and the maximum possible difficulty, prioritizing the allocation of key resources to high-frequency or high-risk bottleneck processes, thus achieving the optimal balance between overall efficiency and stability.

[0035] Existing human-machine collaboration scheduling models only focus on time allocation, neglecting the significant differences in success rates under different human-machine collaboration modes. For example, manual disassembly of complex parts may result in damage due to misjudgment, while robot disassembly of rusted parts may require task rework due to gripping failure. This makes it impossible to assess and mitigate the economic risks arising from task failure. This invention incorporates the success probabilities of different human-machine collaboration modes into the model, enabling scheduling that includes risk assessment and aims to maximize overall expected economic benefits, thereby achieving economically optimal results.

[0036] This invention further aims to address the technical problems existing in current intelligent optimization algorithms when solving such complex multi-objective disassembly models: (1) The conventional encoding and crossover mutation operators of the existing algorithm are very likely to destroy the strict priority logic of the disassembly process, resulting in a large number of infeasible solutions, which seriously wastes the algorithm's computing power on invalid search. (2) Existing algorithms have difficulty effectively handling the upper limit constraint of the total number of devices (such as the global maximum number of collaborative robots), which often leads to the allocation scheme exceeding the actual physical resource limit; (3) Conventional multi-objective algorithms are prone to getting stuck in local optima in the high-dimensional solution space of multi-level decision-making (task sequence, operation mode, machine allocation), and cannot efficiently provide decision-makers with uniform and high-quality Pareto front solutions.

[0037] The detection-driven adaptive human-machine collaborative disassembly line balancing method of the present invention mainly includes the following parts: 1. Overview of the technical solution.

[0038] The adaptive human-machine collaborative disassembly line balancing method based on detection-driven and unified boundary constraints of this invention first reveals the physical law of decoupling, detection, and decision-making in the disassembly process of waste products. That is, the forced disassembly of external components (such as shells and fasteners) must be performed first to expose the internal core components. Then, the integrity of these components is judged by detection nodes, and a dynamic decision is made on whether to proceed to the subsequent flexible disassembly stage. The detection node design is as follows: Figure 1 To address the dual uncertainties of workflow interruption (failure to pass inspection) and drastic fluctuations in working hours (passing inspection but internal corrosion) caused by this physical process, this invention constructs a unified boundary constraint model that includes a working condition adjustment coefficient, and realizes the specific implementation steps and processes for adaptive balancing of the dismantling line for different quality flows.

[0039] This method, as a computer-implemented optimization scheduling method, must execute the following core steps in sequence: Step A (Model Building): Based on the physical law of "detect first, decide later" for waste products, construct a unified boundary constraint model that includes operating condition adjustment coefficients to quantitatively evaluate the cycle time, expected comprehensive profit, and load imbalance of different task allocation schemes.

[0040] Step B (Population Initialization): A three-layer collaborative sequence encoding strategy is used to initialize and generate an initial population that satisfies all hard constraints using the Improved Multi-Objective Parrot Optimization Algorithm (IMPOA).

[0041] Step C (Fitness Evaluation): Using a specific decoding strategy, the encoding is transformed into a specific dismantling station allocation scheme under preset constraints, and a multi-objective fitness vector is calculated.

[0042] Step D (Behavior-Driven Evolution): The algorithm guides the population to perform four behaviors that satisfy priority constraints—foraging, staying, communicating, and fear—according to a set probability, and explores and develops in the solution space.

[0043] Step E (Elite Retention and Output): Environment selection is performed through non-dominated sorting and crowding distance maintenance. Steps C and D are iterated repeatedly until the maximum number of iterations is met. Finally, an approximate Pareto front solution set is output as the recommended human-machine collaborative disassembly scheme.

[0044] 2. Construct a mathematical model.

[0045] This invention models the human-machine collaborative disassembly line balancing problem (HRC-DLBP) with state detection nodes. After the waste product passes through the necessary disassembly station, it enters the detection node to obtain its integrity and damage status. The system calculates the expected execution time of subsequent specific processes based on the status and autonomously decides whether the task should be assigned to a human operator or a collaborative robot, and whether to trigger destructive disassembly. To simplify the model, the following basic assumptions are proposed: (1) A single waste product flows continuously on the assembly line at a fixed pace. (2) There are strict priority constraints between disassembly tasks. (3) The disassembly line consists of a finite number of workstations, each of which can be equipped with a human operator or a collaborative robot, and the basic operating skills and time consumption of the two are different. (4) The detection node can accurately identify the physical state of the product, and the product state follows a specific probability distribution. Based on the product state, the actual processing time of subsequent processes can be predicted.

[0046] 2.1 Symbol Explanation.

[0047] 1. Sets and Indexes.

[0048] Disassembly task number, disassembly task set The collection of tasks that need to be pre-disassembled before the detection node. A set of time-variable processes affected by the integrity status of parts The set of other processes that are not affected by the integrity status of the part after the inspection node are also called normal processes after inspection. . , .

[0049] : Set of priority relationships for dismantling tasks This indicates that task i is the direct predecessor of task j. .

[0050] ： Workstation number, workstation set ,in This represents the maximum number of workstations.

[0051] : Working mode number, These represent manual operation, purely robotic operation, and human-machine collaboration, respectively.

[0052] ： Integrity status number, The set of integrity with low integrity, no recycling value, and requiring destructive dismantling. The set of complete states of nodes that have been tested and have recycling value. , 。

[0053] Completeness scenario The probability of occurrence in the set of completeness scenarios with recycling value. And satisfy: .

[0054] : Time-variable processes in scenarios The time dilation factor represents the proportion of time dilation caused by rusting and deformation. hour This indicates extremely high integrity with no extended processing time. hour This indicates moderate integrity, but also indicates extended disassembly time.

[0055] Process i In mode The standard operating time is as follows.

[0056] Process In mode and scenarios Predicted job time: .

[0057] The probability that discarded products have recycling value and do not require destructive disassembly.

[0058] Process In mode The operating costs include labor costs and energy consumption.

[0059] Process In mode The probability of success.

[0060] Process Recycling revenue after successful dismantling.

[0061] : Maximum number.

[0062] Robot resource limit.

[0063] 2. Decision variables.

[0064] If the process At the workstation The above pattern If executed, the value is 1; otherwise, it is 0.

[0065] If the workstation If a robot is equipped on it, the value is 1; otherwise, the value is 0.

[0066] CT Production cycle time variable CT .

[0067] :Task i At the start of the work cycle of its assigned workstation. .

[0068] If processes i and j are assigned to the same workstation and have no absolute priority, then when process i is executed before process j, then... It is 1 if it is true, otherwise it is 0.

[0069] 2.2 Objective function.

[0070] (1) Minimize production cycle time: ; (2) Maximize expected overall profit: ; ; ; (3) Minimize load imbalance: ; in It is a workstation k Expected workload: .

[0071] 2.2 Constraints.

[0072] (1) Constraint 1: Ensure that each disassembly process They were all assigned to a single workstation k and a unique processing mode was determined. No omissions or duplicate allocations are allowed: ; (2) Constraint 2: There is a priority relationship between tasks, and the preceding task must be completed first: ; (3) Constraint 3: If at the workstation If any task requiring robot or human-robot collaboration is assigned, this workstation must be equipped with a robot. ; (4) Constraint 4: The total number of robot workstations in use across the entire network shall not exceed the robot resource limit: ; (5) Constraint 5: The end time of all tasks in the assigned mode cannot exceed the production line cycle time. : ; (6) Constraint 6: If yes If a process is a preceding process and is assigned to the same workstation, then process j will be processed later than process i. ; (7) Constraint 7: For any two processes that do not have a direct priority relationship i and If they are assigned to the same work station Only two-stage processing modes can be performed in parallel, with one person per machine or one machine per person: ; ; ； Represents the set of patterns that can be parallelized: .

[0073] Step S3: Improvement and design of the multi-objective parrot optimization algorithm.

[0074] The Parrot Optimizer (PO) algorithm is a swarm intelligence algorithm that searches by simulating four typical behaviors of parrots in nature: foraging, dwelling, communicating, and fear. This invention addresses the complex characteristics of the human-robot collaborative disassembly line balancing problem, such as process priority relationships, work mode selection, and robot resource upper limit constraints. It designs a discrete multi-objective parrot optimization algorithm for combinatorial optimization problems, integrating biomimetic behavior with problem constraints and introducing an elite retention strategy based on non-dominated ranking and crowding distance. The algorithm flowchart is shown below. Figure 2 As shown, the specific implementation steps are as follows: 3.1 Parameter import and population initialization.

[0075] First, input basic parameters such as the disassembly task priority matrix, the standard working time matrix for each mode, and the robot resource limit. Preset a set of variable-time operations. Combined with the pre-disassembly process before testing Set the algorithm's running parameters, such as the maximum number of iterations and the population size. Randomly generate an initial parrot population, with each individual using a three-layer cooperative coding structure. First layer (process sequence layer): Length equal to the total number of tasks Based on the priority relation matrix, a feasible task permutation that satisfies the disassembly process logic is generated, ensuring that any task must appear after all its immediate predecessors.

[0076] The second layer (operation mode layer): is the same length as the first layer and represents the execution mode adopted for the corresponding task. 1 represents pure manual operation, 2 represents pure robot operation, and 3 represents human-machine collaborative operation.

[0077] Third layer (robot allocation layer): Length equal to the preset number of workstations 0-1 encoding indicates whether a collaborative robot is configured at this workstation, and the sequence contains... The total number of robots must not exceed the robot resource limit.

[0078] The three-layer coding sequence together constitutes a complete unit, preserving both the logical order of the disassembly task and encompassing the work mode and workstation resource configuration information. An example of this coding strategy is shown below. Figure 3 As shown.

[0079] 3.2 Adaptive Decoding and Fitness Evaluation.

[0080] Under preset constraints, the decoding process assigns each disassembly task in the disassembly sequence to each disassembly workstation and specifies the operation type for each task. This invention does not allocate tasks individually for a single scenario; instead, it first fixes the workstation layout using the expected working time and then verifies the actual performance under each scenario. This ensures the generated disassembly scheme has physical layout stability while also considering production feasibility under different scenarios. The specific logic is as follows: 3.2.1: Construct the scenario task time matrix.

[0081] Different integrity statuses of products after passing through testing points This includes situations with no recycling value that could trigger destructive dismantling; the actual operation time for each process is calculated separately: Set of states with recycling value Recovery can continue through detection nodes. For time-variable processes, the corresponding coefficient should be applied. ( Delayed work time, i.e. The time for the remaining processes remains unchanged; Low completeness scenarios When the product's integrity falls below the recycling threshold, destructive disassembly is triggered. At this time, some processes do not need to be performed, the corresponding operation time is reset to zero, and they are not included in the subsequent normal recycling revenue calculation.

[0082] 3.2.2 Determine the expected working hours for the task.

[0083] Combining the actual occurrence probability vectors of various scenarios The weighted sum of the actual working time of each process under different scenarios is used to obtain the expected working time of the process, which serves as a unified basis for workstation allocation.

[0084] 3.2.3: Calculation of workstation duration.

[0085] Before assigning tasks, it is necessary to clarify the time calculation criteria for parallel human-machine operations within the workstation. Initialize the manual progress time for the current workstation. Time with robot progress The value is 0, and the calculation logic is as follows: If the process If there are no preceding tasks at the current workstation, or only preceding tasks of the same executor, then its start time is directly taken as the current progress time of that executor (i.e., human or robot); if there are preceding processes of different executors within the workstation, the process... The start time needs to be updated to the larger value between the corresponding executor's progress time and the completion time of the preceding task.

[0086] For human-machine collaboration modes, due to process requirements that both parties must participate simultaneously, their starting times are forcibly aligned with the current time. After determining the start time, the processing time is accumulated according to the process mode. If it is in purely manual mode, only the manual time will be updated. Robot Time Keep it unchanged; if it is a pure robot mode, only update the robot time. If it is a human-machine collaboration mode, then and Simultaneous increase Once all tasks at the workstation are completed, the larger of the final progress times for manual and robotic work is taken as the total working time for that workstation. .

[0087] 3.2.4: Task allocation and workstation configuration.

[0088] Based on the initial reference production cycle Starting with this, the workstation configuration is gradually optimized through iterative iterations until the cycle time can no longer be compressed. The specific logic is as follows: (1) Initialize the workstation index Read the corresponding workstation from the third layer of encoding. Robot configuration variables. Extract processes according to the first-level sequence. The system retrieves the corresponding operating mode. If the current workstation robot configuration conflicts with the mode, it will be forcibly downgraded to manual mode.

[0089] (2) Calculate the input process using the method in S3.2.2. Back station Expected working hours ,like If the task is successful, it will be assigned to the current workstation; otherwise, a new workstation will be created, and the task will be the first task in the new workstation.

[0090] (3) Repeat (2) until all processes are assigned. If the maximum expected working time of all workstations is... Then Updated to The tasks are then redistributed until the cycle time can no longer be compressed. At this point, the mapping relationship between tasks and workstations is the fixed physical layout.

[0091] 3.2.5 Multi-objective fitness calculation.

[0092] With a fixed physical workstation layout, we input the actual operation time matrices for three different scenarios. Then, using the method in section 3.2.2, the actual working hours of each workstation are obtained, and the multi-objective fitness is calculated: Target 1 (Production Cycle Time) ): Calculate the actual beat for each scenario Take the worst performance of the system as the overall beat rate: .

[0093] Target 2 (Expected Overall Profit) ): Calculate the expected difference between the task recovery revenue and the operating cost based on the probability of occurrence of each scenario.

[0094] Objective 3 (Load Imbalance) ): Calculate the overall load variance of the system based on the workload of each workstation under the expected state.

[0095] 3.3 Initial fitness evaluation and non-dominated ranking.

[0096] Calculate the fitness vector of all individuals in the initial population. Employ a strategy based on non-dominated ranking and crowding distance to rapidly stratify the population, providing a ranking basis for subsequent elite retention.

[0097] 3.4 Evolutionary iteration based on parrot biological behavior.

[0098] The iterative process updates individual sequence features by simulating the survival behavior of parrots in nature. During each generation of evolution, the algorithm randomly selects one of four behaviors with equal probability for each parrot individual to perform the update. Each behavior is adaptively designed to address the priority constraints of the disassembly problem, ensuring that newly generated individuals always satisfy the feasibility of the disassembly sequence.

[0099] 3.4.1 Foraging behavior.

[0100] This behavior simulates the leaping movement of parrots when searching for food resources in a wide area. By introducing large-span random perturbations, it guides the population to move towards the globally potential optimal area. ① Insertion of legal interval in process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement interval (i.e., the start and end positions of the movement) according to the priority relationship matrix, move the task out of the original position and randomly insert it into any other position within the interval; ② Operation mode reset disturbance: Randomly select a task and change its operation mode to another mode other than the current mode; ③ Robot allocation layer exchange disturbance: Randomly select a pair of 0-1 variables to exchange positions and adjust the distribution of robots between workstations.

[0101] 3.4.2 Staying behavior.

[0102] This behavior mimics the characteristic of parrots to linger in localized areas and conduct detailed searches, exploring a small area near the current solution: ① Feasible domain exchange of process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement range, and randomly select another task within that range to exchange their positions; ②Operation mode coordination and exchange: The operation modes corresponding to the two tasks above are exchanged synchronously to maintain the "task-mode" correspondence; ③ Robot allocation layer exchange perturbation: Perform the same exchange perturbation as the foraging behavior to achieve fine-tuning of resource allocation.

[0103] 3.4.3 Communication behavior.

[0104] This behavior simulates the guiding effect of dominant individuals on a parrot population, accelerating the population's convergence towards the Pareto front through the inheritance and recombination of superior gene fragments. ① Fragment preservation and recombination of the process sequence: Randomly select a parrot individual from the non-dominated solution set of the current population as a guide and recombine its features with the current individual. Randomly select two intersection points on the task sequence, and directly copy the task within that fragment of the current individual to the corresponding position in the offspring. The remaining empty positions in the offspring sequence are filled in the relative order of the remaining tasks in the guide. ②Operation mode collaborative mapping: The operation mode of the offspring individual is migrated synchronously with the corresponding task. Tasks inherited from the current individual retain their operation mode, while tasks inherited from the facilitator adopt the facilitator's operation mode, thus fully preserving the combined experience of "process-mode". ③ Robot allocation layer replication and perturbation: The offspring robot allocation layer is inherited from the current individual and a swap perturbation is performed.

[0105] 3.4.4 Fear-related behaviors.

[0106] This behavior simulates the action of a parrot actively deviating from its current path and reorganizing when it senses extreme danger. By reconstructing the local sequence, the algorithm is forced to jump out of the current search area. ① Local Restructuring of Process Sequence: Randomly select a continuous subsequence from the current individual's task sequence and store the tasks within it into a temporary task pool. Based on the priority relation matrix, randomly select one task from the currently executable tasks and fill it back into the original empty position in the sequence until all tasks in the pool are filled back, thereby completely breaking the original local order; ② Randomized Reset of Operation Mode: Randomly reselect the operation mode for all tasks within the reconstruction area, breaking the original mode combination; ③ Robot allocation layer exchange disturbance: Perform an exchange disturbance to adjust the robot layout.

[0107] 3.5 Elite Preservation and Population Renewal.

[0108] The offspring population generated in this iteration is merged with the parent population to form a population of size [missing information]. The mixed population. The non-dominated sorting is re-applied to hierarchically divide the mixed population, and individuals are selected into the next generation of the population sequentially according to the non-dominated hierarchy from low to high. If the number of selected individuals exceeds the preset size... At that time, priority is given to retaining individuals with larger crowding distances, and the final selection is made up of a size of [number missing]. A new generation of parrots is entering the next generation.

[0109] 3.6 Algorithm Termination and Solution Set Output.

[0110] Determine the current iteration state; the algorithm stops when the set maximum number of iterations is reached. Output the sequence representing all parrot individuals in the first non-dominated sort (Rank=1), as the final recommended dismantling solution set.

[0111] To further verify the feasibility of the improved algorithm proposed in this invention on the problem, this section takes a specific test case of a dismantling line for waste electromechanical products as an example to illustrate the implementation process and technical effect of the detection-driven adaptive human-machine collaborative dismantling line balancing method of this invention.

[0112] This embodiment selects a certain type of single-cylinder gasoline engine as the disassembly object. This product contains a total of 39 disassembly tasks, and the physical priority constraints between the tasks are shown in the priority relationship diagram. Figure 4 Based on the process documents and on-site measured data, a standard time matrix for each task under the three operating modes was established, as shown in Table 1.

[0113] Table 1 Standard working hours for each of the 39 disassembly tasks under different disassembly modes

[0114] Time-variable process set This includes tasks {6, 9, 15, 18, 21, 22, 24, 33, 35, 36}. These tasks involve brittle plastic parts (such as spark plug covers), tiny precision springs (such as speed regulating springs), or high-torque fasteners (such as cylinder head bolts).

[0115] Pre-disassembly process set before inspection This includes tasks {1, 2, 3}. These tasks correspond to core components inside the engine crankcase (such as the camshaft, piston connecting rod assembly, and crankshaft). According to process requirements, these tasks must be completed before the inspection milestones, regardless of whether the product ultimately has any recycling value.

[0116] The system acquires the physical state of waste products through detection nodes. To simplify implementation, this embodiment uses the set of states with recycling value from the mathematical model. The scenarios are simply divided into two types: high completeness and medium completeness, and the set of states with no recycling value is also included. All three scenarios are considered as low-completeness scenarios. The probability settings for the three scenarios are as follows: Scenario 1: High completeness (scenario probability) (This represents the ideal state, where the product is in good condition, without any obvious physical deformation or damage, and all disassembly procedures are performed in accordance with standard operating hours.)

[0117] Scenario 2: Moderate completeness (scenario probability) During service or recycling, products undergo physical deformation, corrosion, or scale buildup, increasing the difficulty of disassembly. This applies to time-variable process sets affected by the condition of the parts. The expected working hours need to be multiplied by the working condition adjustment factor and corrected to... .

[0118] Scenario 3: Low completeness (scenario probability) This scenario corresponds to an extreme state where the integrity is extremely low or core components are missing, and the product has lost its recycling value. Under such conditions, since some parts are no longer available or are severely damaged, the corresponding disassembly process does not need to be performed, and its working hours are recorded as 0.

[0119] The parameters of the discrete multi-objective parrot optimization algorithm used in this embodiment were determined by referring to the commonly used settings of other intelligent optimization algorithms in combinatorial optimization problems, and were determined after multiple rounds of pre-experiments. The basic problem parameters are set according to the typical disassembly line case and are listed in Table 2 along with the algorithm parameters.

[0120] Table 2 Problem and Algorithm Parameter Table

[0121] The experimental environment for this embodiment was configured as follows: Intel Core i7-12700K processor (3.6GHz), 16GB of memory, and Windows 11 operating system. The algorithm was written in MATLAB R2020b and run independently 10 times. A representative result from one of the runs was then analyzed.

[0122] After the algorithm finished running, the parrot population obtained a total of 50 non-dominated solutions, forming a Pareto optimal front. To illustrate the distribution of the non-dominated solutions in the three-dimensional target space, an approximate Pareto front scatter plot was drawn as follows: Figure 5 As shown.

[0123] A dismantling and allocation scheme is randomly selected from the non-dominated solution set for in-depth analysis, and Gantt charts are drawn for three typical scenarios, such as... Figure 6 As shown. In Scenario 1 (high integrity), the product is in good condition, and each process is executed according to standard time. The overall cycle time for this scenario is 108 seconds, with the bottleneck located at station 1, indicating that this station bears the highest workload under standard conditions. In Scenario 2 (medium integrity), the time-variable process set is affected by the condition of the parts. The task completion time was extended by a coefficient, causing the cycle time of the solution to rise to 122 seconds. It is worth noting that the bottleneck station shifted from station 1 to station 4, indicating that fewer tasks at station 1 were affected by the time extension, while station 4 contained more variable-time processes sensitive to operational conditions, making the cumulative effect more significant under the delay treatment. In scenario 3 (low integrity), the product has no recycling value and only the pre-disassembly process set before inspection needs to be performed. All other tasks require no further operation, thus significantly reducing the overall line cycle time and involving only a few workstations. The above results demonstrate that, with a fixed physical layout of workstations, this disassembly scheme can effectively adapt to fluctuations in work time corresponding to different product integrity levels. Even under unfavorable conditions with moderate product integrity, it can maintain orderly line balance. In scenarios with no recycling value, it can quickly complete the necessary disassembly procedures before inspection, avoiding unnecessary work. Overall, this scheme possesses strong adaptability and operational stability, meeting the robustness requirements of actual disassembly lines for varying product quality.

[0124] The results of this embodiment demonstrate that the discrete multi-objective parrot optimization algorithm proposed in this invention can effectively handle medium-sized computational examples containing 39 dismantling tasks. Under the strict constraints of dismantling priority relationships, robot resource limits, and workstation operation time not exceeding the cycle time, the algorithm successfully obtains a series of Pareto optimal solutions that balance the three objectives of cycle time, expected comprehensive profit, and load imbalance for various typical production scenarios with a fixed physical layout. This provides scientific decision support for the high-value recycling of waste electromechanical products and can offer a multi-objective trade-off decision-making basis for actual dismantling line planning, demonstrating clear engineering applicability.

[0125] To ensure the effectiveness and accuracy of the model, this invention provides a systematic, data-driven method for parameter acquisition and calibration.

[0126] 1. Probability of entering the flexible phase ( ) calibration.

[0127] Method 1: Historical data frequency statistics method (most practical).

[0128] Description: In the absence of complex machine learning models, this invention employs a method of direct frequency statistics on historical production data. That is, within a certain period (such as a quarter), statistical processes are analyzed. i In mode s The procedure was executed a total of N times, of which the number of times the high-quality standard was met was [number missing]. Next, then can be This is an approximate result. It is a simple calibration method that yields objective results.

[0129] Method 2: Expert assessment and Delphi method.

[0130] Description: On new production lines or processes without historical data, this invention can organize 3-5 senior engineers or front-line technical experts to conduct a back-to-back scoring method (Delphi method) to make predictions. The values ​​are then taken as the average or weighted average as the initial input to the model.

[0131] Method 3: Employ machine learning methods based on historical data.

[0132] (1) Data acquisition: Collect a large number of physical state characteristics of parts after inspection and their final disposal decision (whether to enter the flexible stage).

[0133] (2) Model training: A prediction model is trained by applying classification algorithms such as Logistic Regression. The model takes the state of the part as input and the probability of it entering the flexible stage as output.

[0134] (3) Dynamic determination: In actual production, based on the inspection status of new parts, this model is called to dynamically calculate their... value.

[0135] 2. Conditional state probability ( ) calibration.

[0136] Method: Conditional frequency statistics based on historical data.

[0137] (1) Data screening: Screen all samples that have entered the flexible phase from historical data; (2) Status classification: These samples are classified according to their specific faults or physical states; (3) Frequency calculation: Statistical analysis of each state The frequency of occurrence, that frequency is The estimated value.

[0138] 3. State time extension factor ( ) calibration.

[0139] Method: Benchmark comparison method based on measured working hours.

[0140] (1) Data acquisition calibration benchmark: measuring an ideal state or a brand new part in mode m Next completed process i Baseline working hours .

[0141] (2) Measure and record the parts that can be detected at the nodes in different physical states. The actual working hours to complete the same operation are as follows. .

[0142] (3) Coefficient calculation: = / .

[0143] 4. Success rate of human-machine collaboration ( ) calibration.

[0144] Method: Statistical analysis of long-term production data.

[0145] (1) Data recording: long-term recording of processes i In mode m Total number of executions and the number of times it was successfully completed .

[0146] (2) Probability calculation: = / .

[0147] Alternative parameter acquisition: If dynamic updates are not used, a static preset method can be adopted, that is, always using the average historical data of the past three years as a fixed parameter, without updating with each batch. This approach is simple to implement, but it is poorly adaptable to short-term quality fluctuations.

[0148] Constraint substitution: This involves using the dual independent constraint method from the original solution, listing two inequalities for verification. Although mathematically equivalent, the formulas are highly redundant, which is not conducive to subsequent expansion (such as expanding to three or more operating conditions).

[0149] Algorithm Framework Alternatives: The IMPOA algorithm used in this invention can be replaced by improved NSGA-II or MOEA / D multi-objective optimization algorithms. These alternatives also require the introduction of the three-layer cooperative coding and priority-constrained operators designed in this invention to achieve feasible solution search. However, IMPOA's four-behavior random switching mechanism has unique advantages in balancing global and local searches.

[0150] Parameter configuration alternative (optional structure): The third robot allocation layer is a mandatory structure for scenarios with severely limited resources (such as the total number of robots being less than the number of workstations). If the enterprise workshop has extremely abundant automation resources (the number of robots is greater than or equal to the maximum number of workstations), then this layer is an optional structure and can be omitted to reduce the algorithm's dimensionality, allowing direct decoding through the first and second layers.

[0151] This invention has the following outstanding technical advantages and beneficial effects: 1. Achieved precise allocation of large-scale tasks under complex resource upper limit constraints: For complex electromechanical products involving 39 disassembly tasks, the method of this invention can effectively achieve reasonable allocation and scheduling of tasks across a maximum of 15 candidate workstations. Leveraging a designed three-layer collaborative coding and adaptive decoding mechanism, this invention strictly adheres to the hard resource limit constraint of no more than 10 collaborative robots during iterative optimization. By introducing a detection decision node, the task is decoupled into a set of pre-disassembly procedures before detection and a set of conditionally triggered procedures, thus recreating the actual physical process. The final generated solution ensures a one-to-one correspondence between the task execution mode and the workstation resource configuration, highly consistent with the actual physical resource state, effectively achieving feasible solution search under complex constraints.

[0152] 2. The algorithm boasts excellent optimization performance, providing high-quality multidimensional Pareto decision solutions: The improved multi-objective parrot optimization algorithm (IMPOA) designed in this invention exhibits superior performance in solving large-scale discrete combinatorial optimization problems with strict priority relationships. With a maximum of 200 iterations and a population size of 100, the algorithm demonstrates a fast convergence speed based on four behavior-driven mechanisms and can generate a Pareto front solution set with excellent distribution. This solution set, while satisfying the feasibility of the process logic, successfully achieves the optimal balance between the three core optimization objectives: minimizing production cycle time (CT), maximizing expected total profit (P), and minimizing load imbalance (SI), providing high-quality and diverse decision-making solutions for multi-objective dismantling line balancing problems.

[0153] 3. Significantly improved production line adaptability and scheduling robustness under extreme operating conditions: The unified boundary constraint model constructed in this invention, which includes operating condition adjustment coefficients, possesses excellent operating condition coverage. By modeling stochastic working hours based on the integrity state of parts, a quantitative correlation is established between the wear and corrosion degree of parts and the uncertainty of disassembly time, providing a key input for robust optimization. Experiments demonstrate that this method can adaptively handle uncertain scenarios ranging from detection interruption to extreme corrosion. Even under complex conditions such as unclear status of scrap products, random detection results, and drastic fluctuations in working hours, the rationality and stability of the disassembly line scheduling scheme can still be guaranteed.

[0154] Furthermore, the present invention also provides a computer device / apparatus / system, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0155] The present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0156] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or block diagrams.

[0157] These computer program instructions may also be stored in a computer-readable storage medium capable of directing a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more blocks in a block diagram.

[0158] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, thereby providing steps for implementing the functions specified in one or more flowcharts and / or one or more blocks in a block diagram.

[0159] It should be noted that, depending on the implementation needs, the various steps / components described in this invention can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0160] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A detection-driven adaptive human-machine collaborative disassembly line balancing method, characterized in that, The method includes: Obtain the set of dismantling tasks for waste products, their priority, the standard working hours of each task under different operating modes, and the upper limit of robot resources; A two-stage disassembly task model is constructed. Tasks located before the inspection node are defined as the pre-disassembly forced task set, tasks located after the inspection node whose operation time is affected by the integrity status of the parts are defined as the time-variable task set, and the rest are the normal task set after inspection. A multi-objective optimization mathematical model is established with the optimization objectives of minimizing production cycle time, maximizing expected comprehensive profit, and minimizing load imbalance, and constraints are set. For tasks in the time-variable task set, different working condition adjustment coefficients are introduced according to different integrity states to uniformly constrain the workstation operation time under each state. The mathematical model is solved using an improved multi-objective parrot optimization algorithm, including: An initial population was generated using a three-layer collaborative coding strategy. Through the decoding strategy, the expected operation time of each task is calculated based on the probability of occurrence of different integrity states and the corresponding work condition adjustment coefficient. Based on the expected operation time, the code is converted into a workstation allocation scheme to form a fixed workstation physical layout, and the multi-objective fitness value is calculated. A progeny population is generated by performing update operations on individuals that simulate four parrot behaviors: foraging, loitering, communicating, and fear. The parent and offspring populations are merged, and the new generation population is retained after non-dominated sorting and crowding distance screening. After iterating to the termination condition, the Pareto front solution set is output as a human-machine collaborative disassembly scheme.

2. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 1, characterized in that, The objective function of the multi-objective optimization mathematical model is: Minimize production cycle time : ； Maximize expected total profit : ； ； ； Minimize load imbalance : ； ； In the formula, For production cycle time; In order to recover profits, For operating costs; The dismantling task number is used. To pre-disassemble the forced task set; Number the workstation. Maximum number of workstations; Number the working mode. This indicates that the work is done entirely manually. This indicates purely robotic operation. Indicates human-machine collaboration; Assign variables to tasks, if the task At the workstation The above pattern If executed, the value is 1; otherwise, it is 0. For the task Recovery revenue, For the task In mode The success rate of the following The probability that discarded products have recycling value without requiring destructive disassembly. For the task In mode The operating costs below; For workstations k Expected workload For the task In mode Standard operating time; The integrity status number is used for identification. A set of complete states with recycling value. Incompleteness status The probability of occurrence in the set of complete states with recyclable value and satisfying , For the task In mode and integrity status Predicted job time; Among them, the predicted operation time for: ； In the formula, A set of time-variable tasks This is the set of normal tasks after testing; This is the operating condition adjustment coefficient, which is applied when the task belongs to the set of normal tasks after detection. When the task belongs to a time-variable task set, This indicates the percentage increase in operation time due to changes in the physical state of the parts.

3. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 2, characterized in that, The constraints include: (1) Task unique assignment constraint: ensure that each disassembly task is uniquely assigned. They were all assigned to the only workstation And determine the unique processing mode. No omissions or duplicate allocations are allowed: ； In the formula, For dismantling task sets; (2) Task priority constraint: There are priority constraints between tasks, and the preceding task must be completed first: ； In the formula, For the set of priority relationships of dismantling tasks, This indicates that task i is the direct predecessor of task j; (3) Workstation robot configuration constraints: If in the workstation For any task requiring robot or human-robot collaboration to be assigned, this workstation must be equipped with a robot. ； In the formula, As a decision variable, if the workstation If a robot is equipped on it, the value is 1; otherwise, the value is 0. For workstation assembly, It is a maximum number; (4) Robot resource limit constraint: The total number of robot workstations in use across the entire line shall not exceed the robot resource limit. : ； (5) Workstation cycle time constraint: The end time of all tasks in the assigned mode cannot exceed the production line cycle time. : ； In the formula, Let be the decision variable, representing the task. i The start time of the work cycle at its assigned workstation; (6) Preceding task constraint at the same workstation: If the task It is a task If the task is a preceding process and is assigned to the same workstation, then the task... The process must be later than the task. Processing steps: ； (7) Parallel constraints at the same workstation: For any two tasks that do not have a direct priority relationship i and tasks If the two are assigned to the same workstation Only when the processing modes of two tasks are one person per machine or one machine per person can they be processed in parallel: ； ； ； In the formula, Represents the set of patterns that can be parallelized: ; Let be the decision variable. If task i and task j are assigned to the same workstation and have no absolute priority relationship, then when task i is executed before task j, then... It is 1 if it is true, otherwise it is 0.

4. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 1, characterized in that, Three-layer collaborative coding strategies include: The first layer is a sequence of disassembly tasks that satisfy priority constraints, with a length equal to the total number of tasks. The second layer consists of the operation modes corresponding to each task, and is the same length as the first layer. The third layer is a binary sequence indicating whether each workstation is configured with a robot. Its length is equal to the maximum number of workstations, and the total number of 1s does not exceed the robot resource limit.

5. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 4, characterized in that, Decoding strategies include: Construct a state-task time matrix: This matrix addresses the different integrity states of a product after passing through inspection nodes. Calculate the actual operation time for each task, including: for the completeness set of states with recovery value. The state, time-variable task set The task operation time is extended according to the corresponding working condition adjustment coefficient, while the operation time of other tasks remains unchanged at the standard working time. For low integrity status, when the product integrity is lower than the recycling threshold and destructive disassembly is triggered, some tasks do not need to be executed, the corresponding operation time is set to zero, and they are not included in the recycling revenue calculation. Determine the expected operation time of each task: Based on the probability of occurrence of each integrity state, the operation time of each task under different states is weighted and summed to obtain the expected operation time of each task, which serves as a unified basis for workstation allocation. Task allocation and workstation configuration: Starting from the initial reference cycle time, the workstation configuration is gradually optimized through iterative iterations until the cycle time can no longer be compressed. This includes: allocating tasks to workstations sequentially according to the first-level disassembly task sequence; calculating the workstation working time after adding a task based on the expected operation time; if it does not exceed the current reference cycle time, it is allocated to the current workstation; otherwise, a new workstation is opened, and the task is set as the first task of the new workstation; during the allocation process, if the robot configuration of the current workstation conflicts with the task operation mode, the operation mode of the task is forcibly downgraded to pure manual mode; after all tasks are allocated, if the maximum workstation working time is less than the current reference cycle time, the reference cycle time is updated and the tasks are reallocated until the cycle time can no longer be compressed, forming a fixed workstation physical layout, i.e., the mapping relationship between tasks and workstations.

6. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 5, characterized in that, The methods for calculating workstation working hours include: Initialize the manual progress time for the current workstation. Time with robot progress =0; Determining the task start time includes: if the task has no preceding task in the current workstation, or only has a preceding task with the same executor, then its start time is directly taken as the current progress time of that executor; if there are preceding process constraints for different executors in the workstation, then the task start time is taken as the larger value between the corresponding executor's progress time and the completion time of the preceding task; for human-machine collaboration mode, its start time is forcibly aligned to the maximum value between the current human progress time and the robot progress time. After determining the start time, accumulate processing time according to the task's operation mode. If it is in purely manual mode, only the manual time will be updated. Robot progress time Keep it unchanged; if it is a pure robot mode, only update the robot time. Manual progress time The time remains unchanged; if it is a human-machine collaboration mode, the manual progress time is... Time with robot progress Simultaneous increase ; Once all tasks at the workstation have been completed, the maximum of the manual processing time and the robot processing time is taken as the total working time for that workstation; where... This refers to the operation time of the corresponding task in the current mode and completeness state.

7. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 4, characterized in that, Foraging behaviors include: Legal interval insertion of process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement interval according to the priority relationship matrix, move the task out of its original position and randomly insert it into any other position within the interval; Operation mode reset disturbance: Randomly select a task and change its operation mode to another mode other than the current mode; Robot allocation layer exchange disturbance: Randomly select a pair of binary bits in the robot allocation layer and exchange their positions to adjust the distribution of robots between workstations; Staying behavior includes: Feasible domain exchange of process sequence: Randomly select a task in the current individual's task sequence, determine its legal movement range according to the priority relationship matrix, and randomly select another task within the range to exchange the positions of the two. Operation mode coordination and exchange: Synchronously exchange the operation modes corresponding to the two tasks mentioned above. Robot allocation layer swapping perturbation: Randomly select a pair of binary bits in the robot allocation layer and swap their positions; Communication behaviors include: Fragment preservation and recombination of the process sequence: Randomly select a parrot individual from the non-dominated solution set of the current population as the guide; randomly select two intersection points on the task sequence, directly copy the task in the fragment of the current individual to the corresponding position of the offspring, and fill the remaining empty positions in the offspring sequence in the relative order of the remaining tasks in the guide. Operation mode collaborative mapping: The operation mode of the offspring is migrated synchronously with the corresponding task. That is, the task inherited from the current individual retains its operation mode, and the task inherited from the facilitator adopts the facilitator's operation mode. Robot allocation layer replication and perturbation: The offspring robot allocation layer is inherited from the current individual and a random bit swap perturbation is performed; Fear behaviors include: Local reconstruction of the process sequence: Randomly select a continuous subsequence in the current individual's task sequence and move the tasks in it into a temporary task pool; Based on the priority relationship matrix, randomly select one task from the currently executable tasks and fill it back into the original empty position in the sequence until all tasks in the task pool are filled back. Operation mode randomization reset: Randomly reselect the operation mode for all tasks within the reconstruction area; Robot allocation layer exchange perturbation: The robot allocation layer performs a random bit exchange perturbation.

8. The detection-driven adaptive human-machine collaborative disassembly line balancing method according to claim 2, characterized in that, The multi-objective fitness value is calculated as follows: Production cycle time Under a fixed workstation physical layout, substitute the actual operation time matrix of each integrity state, calculate the actual cycle time of each state, and take the worst performance of the system as the overall cycle time. Expected comprehensive profit Based on the probability of occurrence of each integrity state, calculate the expected difference between the task recovery revenue and the operating cost to obtain the expected comprehensive profit; Load imbalance Based on the workload of each workstation under the expected state, the overall load variance of the system is calculated to obtain the load imbalance degree. The operating condition adjustment coefficient The probability that the waste product has recycling value and does not require destructive disassembly is determined by calibrating the ratio of measured working time to benchmark working time under different physical states of parts; The probability of occurrence of the completeness state s is obtained through frequency statistics of historical production data, expert Delphi scoring, or classification prediction models based on machine learning. The success probability of task i in mode m is obtained by classifying and statistically calibrating historical samples that have entered the flexible phase. The ratio of the number of successful executions of task i in mode m to the total number of executions is obtained from long-term production data.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images