A robust optimization scheduling method and system for complex manufacturing systems

Abstract

This invention relates to a robust optimization scheduling method and system for complex manufacturing systems. The method includes: acquiring real-time production status and production status data of the production line; generating a sample set based on the production status data; generating scheduling rules that integrate equipment maintenance and workpiece dispatch based on the sample set; generating a scheduling scheme by combining the real-time production status and scheduling rules; applying the scheduling scheme to the production line and obtaining the production performance feedback from the production line; updating the scheduling rule parameters based on the feedback production performance and regenerating the scheduling rules; repeating the above steps until the production performance feedback from the production line reaches the expected standard. Compared with the prior art, this invention adopts the idea of ​​closed-loop optimization, considering the robustness of the production line scheduling scheme while comprehensively considering the flexible maintenance of production line equipment, process constraints, and dynamic dispatch, thereby improving the process constraint satisfaction rate of the production line and thus improving the on-time delivery rate of workpieces.

Description

Technical Field

[0001] This invention relates to the field of production line scheduling technology, and in particular to a robust optimization scheduling method and system for complex manufacturing systems. Background Technology

[0002] In the production process, preventative maintenance of production machines can more rationally and effectively mitigate the impact of machine failures on the entire manufacturing system. Existing production scheduling methods that consider equipment maintenance are mostly based on single-machine systems and assembly line workshops, which does not match the actual manufacturing systems and lacks research on integrated scheduling optimization problems of complex manufacturing systems under uncertain random factors.

[0003] Current research on scheduling robustness typically focuses on the ability of scheduling to maintain its original state and / or performance under uncertain environments. Little attention is paid to the criteria and basis for the measurement methods and their role in the research and application of robust production scheduling. Furthermore, scheduling is an NP-hard problem. To reduce computational difficulty and better align with practical engineering applications, it is necessary to centrally optimize equipment maintenance, dynamic work assignment, and the robustness of scheduling schemes. This would enable scheduling schemes that consider equipment maintenance to adapt to dynamic processing environments, thus possessing a certain degree of robustness.

[0004] Extensive research has been conducted on data-driven closed-loop optimization scheduling problems, but existing methods are usually targeted at specific production processes and rarely consider issues such as scheduling optimization, control integration, and model consistency. Although the idea of ​​"closed-loop optimization" is adopted, the universality and robustness of the scheduling scheme are not considered, and the equipment maintenance, scheduling scheme robustness, and manufacturing system performance are not integrated and optimized, resulting in poor scheduling performance and failing to maximize the production efficiency of the production line. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a robust optimization scheduling method and system for complex manufacturing systems.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A robust optimization scheduling method for complex manufacturing systems includes the following steps:

[0008] S1. Obtain the real-time production status and production status data of the production line;

[0009] S2. Generate a sample set based on the production status data;

[0010] S3. Based on the sample set, generate scheduling rules for fusion equipment maintenance and workpiece dispatch;

[0011] S4. Combine the real-time production status and the scheduling rules to generate a scheduling scheme;

[0012] S5. Apply the scheduling scheme to the production line and obtain the production performance feedback from the production line;

[0013] S6. Based on the feedback of production performance, update the scheduling rule parameters and regenerate the scheduling rules;

[0014] S7. Repeat steps S4 to S6 until the production performance feedback from the production line reaches the expected standard.

[0015] Furthermore, the production performance includes workpiece processing cycle, workpiece on-time delivery rate, equipment availability index, and scheduling scheme robustness.

[0016] Furthermore, a feature selection algorithm based on an improved immune algorithm is used to select features from the sample set to reduce redundant features and computation time. Specifically, this includes the following steps:

[0017] S301, antibody and antigen initialization; where antibodies are different combinations of the production line's production status attributes.

[0018] S302. Calculate the affinity between the antibody and the antigen;

[0019] S303. Based on antibody affinity, antibodies are cloned and mutated. The higher the antibody affinity, the more clones there are and the lower the probability of mutation.

[0020] S304. Determine whether the maximum number of iterations has been reached. If yes, proceed to step S307; otherwise, proceed to step S305.

[0021] S305. The average similarity of antibodies is calculated using the reciprocal of the Euclidean distance;

[0022] S306. If the average similarity of antibodies is less than the threshold, sort the antibodies in the antibody group by affinity, select d antibodies with the highest affinity to add to the antibody library, and replace an equal number of antibodies with the lowest affinity, then proceed to step S302; otherwise, proceed to step S307.

[0023] S307. Screen the antibodies in the antibody group and select the antibody with the highest affinity as the optimal solution.

[0024] Furthermore, the process for generating scheduling rules for the integrated equipment maintenance and workpiece dispatch includes the following steps:

[0025] S401. Determine whether the workpiece meets the process constraints. If yes, proceed to the next step; otherwise, the workpiece enters the scrap area.

[0026] S402. Calculate the urgency level P1 of the process constraints of the workpiece;

[0027] S403, Calculate the urgency level of workpiece delivery P3;

[0028] S404, Calculate the workpiece dispatch priority P LOT ;

[0029] S405. Determine if the current time is within the equipment maintenance period. If it is, proceed to the next step; otherwise, follow P. LOT Assign tasks based on the size of the workpiece;

[0030] S406, Urgency level of computer equipment maintenance P4;

[0031] S407, Calculate the load level P5 of downstream equipment;

[0032] S408. The unavailability of computing devices;

[0033] S409, Computer Equipment Maintenance Priority P PM ;

[0034] S410, Assign work priority P to the workpiece LOT and equipment maintenance priority P PM Perform normalization processing, and denot the normalization results as P. LOT ’ and P PM ’ ;

[0035] S411, Compare the normalized job dispatch priority P LOT ’ and equipment maintenance priority P PM ’ If P LOT ’ >P PM ’ If necessary, dispatch the workpiece; otherwise, perform equipment maintenance.

[0036] Furthermore, based on the feedback production performance, the IWOA-MLP algorithm outputs the expected performance that matches the current production state; the IWOA-MLP algorithm is used to optimize the scheduling rule parameters and update the scheduling rules.

[0037] A robust optimization scheduling system for complex manufacturing systems includes a production line simulation module, a sample generation module, an offline training module, and an online scheduling module;

[0038] The production line simulation module is connected to the sample generation module, the sample generation module is connected to the offline training module, the offline training module is connected to the online scheduling module, and the online scheduling module is connected to the production line simulation module. The system as a whole forms a closed-loop feedback.

[0039] The production line simulation module is used to generate real-time production status and production status data, and send the real-time production status to the online scheduling module and the production status data to the sample generation module.

[0040] The sample generation module generates a sample set based on the production status data and sends the sample set to the offline training module;

[0041] The offline training module generates scheduling rules that integrate equipment maintenance and workpiece dispatch based on the sample set.

[0042] The online scheduling module generates a scheduling scheme based on the scheduling rules and sends the scheduling scheme to the production line simulation module;

[0043] The production line simulation module applies the scheduling scheme and feeds back the production performance to the online scheduling module. The online scheduling module updates the scheduling rule parameters based on the feedback production performance and sends the updated scheduling rule parameters to the offline training module. The offline training module regenerates the scheduling rules based on the updated scheduling rule parameters.

[0044] Furthermore, the production performance considered in scheduling includes workpiece processing cycle, workpiece on-time delivery rate, equipment availability indicators, and the robustness of the scheduling scheme.

[0045] Furthermore, the offline training module employs a feature selection algorithm based on an improved immune algorithm to select features from the sample set, thereby reducing redundant features and computation time. Specifically, this includes the following steps:

[0046] S301, antibody and antigen initialization; where antibodies are different combinations of the production line's production status attributes.

[0047] S302. Calculate the affinity between the antibody and the antigen;

[0048] S303. Based on antibody affinity, antibodies are cloned and mutated. The higher the antibody affinity, the more clones there are and the lower the probability of mutation.

[0049] S304. Determine whether the maximum number of iterations has been reached. If yes, proceed to step S307; otherwise, proceed to step S305.

[0050] S305. The average similarity of antibodies is calculated using the reciprocal of the Euclidean distance;

[0051] S306. If the average similarity of antibodies is less than the threshold, sort the antibodies in the antibody group by affinity, select d antibodies with the highest affinity to add to the antibody library, and replace an equal number of antibodies with the lowest affinity, then proceed to step S302; otherwise, proceed to step S307.

[0052] S307. Screen the antibodies in the antibody group and select the antibody with the highest affinity as the optimal solution.

[0053] Furthermore, the offline module generates scheduling rules for fusion equipment maintenance and workpiece dispatching, including the following steps:

[0054] S401. Determine whether the workpiece meets the process constraints. If yes, proceed to the next step; otherwise, the workpiece enters the scrap area.

[0055] S402. Calculate the urgency level P1 of the process constraints of the workpiece;

[0056] S403, Calculate the urgency level of workpiece delivery P3;

[0057] S404, Calculate the workpiece dispatch priority P LOT ;

[0058] S405. Determine if the current time is within the equipment maintenance period. If it is, proceed to the next step; otherwise, follow P. LOT Assign tasks based on the size of the workpiece;

[0059] S406, Urgency level of computer equipment maintenance P4;

[0060] S407, Calculate the load level P5 of downstream equipment;

[0061] S408. The unavailability of computing devices;

[0062] S409, Computer Equipment Maintenance Priority P PM ;

[0063] S410, Assign work priority P to the workpiece LOT and equipment maintenance priority P PM Perform normalization processing, and denot the normalization results as P. LOT 'and P PM ';

[0064] S411, Compare the normalized job dispatch priority P LOT and equipment maintenance priority P PM ', if P LOT >P PM If the work order is not specified, then work order assignment will be carried out; otherwise, equipment maintenance will be performed.

[0065] Furthermore, the online scheduling module, based on the feedback production performance, outputs the expected performance that matches the current production state through the IWOA-MLP algorithm; it also optimizes the scheduling rule parameters and updates the scheduling rules through the IWOA-MLP algorithm.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] 1. The production line scheduling rule of this invention adopts the idea of ​​closed-loop optimization. While considering the robustness of the production line scheduling scheme, it comprehensively considers the flexible maintenance of production line equipment, process constraints, and dynamic work assignment. It takes into account factors related to process constraints, as well as factors related to equipment flexible maintenance and work assignment, so as to improve the process constraint satisfaction rate of the production line and thus improve the on-time delivery rate of workpieces.

[0068] 2. This invention employs a feature selection algorithm based on an improved immune algorithm to reduce the attributes of the manufacturing system, thereby reducing redundant features and computation time. It uses the IWOA-MLP algorithm to achieve performance prediction and scheduling parameter optimization, and updates the newly generated scheduling parameters to the combined dispatching rules to form optimized scheduling rules, thereby achieving adaptive scheduling.

[0069] 3. This invention uses an improved Whale Optimization Algorithm (IWOA) to achieve performance prediction and scheduling parameter optimization, resulting in a more uniform population distribution, faster convergence speed, improved ability to escape local optima, and increased optimization accuracy. Attached Figure Description

[0070] Figure 1 This invention provides a closed-loop adaptive optimization scheduling framework.

[0071] Figure 2 It is a closed-loop adaptive optimization scheduling system driven by multiple objectives;

[0072] Figure 3 Antibody encoding based on a feature selection algorithm derived from an improved immune algorithm;

[0073] Figure 4 A flowchart for the execution of work assignment rules that comprehensively considers the combination of equipment and workpieces;

[0074] Figure 5 Flowchart of the improved whale optimization algorithm;

[0075] Figure 6 The flowchart is for the IWOA-MLP algorithm. Detailed Implementation

[0076] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0077] The goal of robust optimization is to find a solution that satisfies the constraints for all possible scenarios and optimizes the objective function in the worst-case scenario. Its core idea is to transform the original problem into a convex optimization problem with polynomial computational complexity, approximating it to a certain degree. The key to robust optimization is establishing a robust equivalence model, then using relevant optimization theory to transform it into a solvable "approximate" robust equivalence problem, and finally providing a robust optimal solution.

[0078] To address the shortcomings of existing production line scheduling systems in considering scheduling optimization, control integration, and model consistency, this invention proposes a robust optimization scheduling system for complex manufacturing systems, comprising a production line simulation module, a sample generation module, an offline training module, and an online scheduling module.

[0079] like Figure 1 As shown, the production line simulation module is connected to the sample generation module, the sample generation module is connected to the offline training module, the offline training module is connected to the online scheduling module, and the online scheduling module is connected to the production line simulation module. The system as a whole forms a closed-loop feedback.

[0080] The production line simulation module is used to generate real-time production status and production status data, and sends the real-time production status to the online scheduling module and the production status data to the sample generation module.

[0081] The sample generation module generates a sample set based on production status data and sends the sample set to the offline training module;

[0082] The offline training module generates scheduling rules that integrate equipment maintenance and workpiece dispatching based on the sample set;

[0083] The online scheduling module generates a scheduling plan based on scheduling rules and sends the scheduling plan to the production line simulation module;

[0084] The production line simulation module applies a scheduling scheme and feeds back production performance to the online scheduling module. The online scheduling module updates the scheduling rule parameters based on the feedback production performance and sends the updated scheduling rule parameters to the offline training module. The offline training module regenerates the scheduling rules based on the updated scheduling rule parameters.

[0085] Based on this system, a robust optimization scheduling method for complex manufacturing systems can be implemented, such as... Figure 2 As shown, it includes the following steps:

[0086] S1. Obtain the real-time production status and production status data of the production line;

[0087] S2. Generate a sample set based on production status data;

[0088] S3. Based on the sample set, generate scheduling rules for fusion equipment maintenance and workpiece dispatch;

[0089] S4. Generate a scheduling plan by combining real-time production status and scheduling rules;

[0090] S5. Apply the scheduling scheme to the production line and obtain the production performance feedback from the production line;

[0091] S6. Based on the feedback of production performance, update the scheduling rule parameters and regenerate the scheduling rules;

[0092] S7. Repeat steps S4 to S6 until the production performance feedback from the production line reaches the expected standard.

[0093] In this embodiment, a feature selection algorithm based on an improved immune algorithm is used to select features from the sample set to reduce redundant features and computation time. Specifically, the algorithm includes the following steps:

[0094] 1): Antibody and antigen initialization; where antibodies are different combinations of production status attributes of the production line; i.e., candidate set, antibody encoding such as Figure 3 As shown.

[0095] Where "1" indicates that the production attribute is selected as a feature attribute, and "0" indicates that the production attribute is not selected as a feature attribute. Initial antibody population A b M antibodies were generated, and a portion of these antibodies were randomly selected and added to the antibody memory library; antigen group A was obtained from the sample data. g Each antigen includes the antigen itself. (Performance metrics of interest) and antigenic determinants (production line status attributes).

[0096] 2) Calculate the affinity between the antibody and the antigen (the strength of binding between the antibody and the antigenic determinant). The higher the value, the closer the candidate solution is to the true solution. Train a new antigen ontology using a BP neural network:

[0097]

[0098] The BP neural network takes as input a set of production line performance attributes selected by the immune algorithm antibodies and outputs the performance indicators of interest. It is trained using a test set to generate the representation of these antibody groups. One test item can get one There are k in total. The BP neural network here uses a three-layer structure, with one hidden layer and 13 neurons in the hidden layer.

[0099] and The similarity between them is calculated as follows:

[0100]

[0101] and The greater the similarity between the antibodies, the higher their affinity. With each The average affinity is taken as the final affinity of the antibody:

[0102]

[0103] Where Dim is A b (m) The number of dimensions selected, where ave_Dim is the average number of dimensions of the previous generation of antibodies. This operation can further reduce the dimensionality while improving the antigen-antibody fitness.

[0104] 3) Antibody cloning. Based on antibody affinity, antibodies are cloned and mutated (the higher the antibody affinity, the more clones are produced and the lower the probability of mutation). Progeny antibodies with increased affinity are selectively chosen to replace parent antibodies, thereby improving the overall affinity of the antibody population.

[0105] 4): Determine if the maximum number of iterations has been reached. If the maximum number of iterations has been reached, go to 7); otherwise, go to 5).

[0106] 5): According to the formula, the average similarity of antibodies is calculated using the reciprocal of the Euclidean distance. The closer the distance between antibodies, the greater their similarity.

[0107]

[0108] 6): If the average similarity of antibodies is less than the threshold, sort the antibodies in the antibody group by affinity, select d of them with the highest affinity, add them to the antibody library and replace an equal number of antibodies with the lowest affinity, then go to 2); otherwise, go to 7).

[0109] 7): Screen the antibodies in the antibody group and select the one with the highest affinity as the optimal solution.

[0110] like Figure 4 As shown, the process for generating scheduling rules that integrate equipment maintenance and workpiece dispatch includes the following steps:

[0111] 1) Determine whether the workpiece meets the process constraints. If the workpiece meets the process constraints, proceed to the next step; otherwise, put the workpiece into the scrap area.

[0112] 2) Calculate the urgency level p1 of the workpiece's process constraints;

[0113] 3) Calculate the urgency level of the workpiece delivery, P3;

[0114] 4) Calculate the workpiece dispatch priority P LOT ;

[0115] 5) Determine if the current time is within the equipment's maintenance period. If it is, proceed to the next step; otherwise, follow step P. LOT Assign tasks based on the size of the workpiece;

[0116] 6) Calculate the urgency level of equipment maintenance, P4;

[0117] 7) Calculate the load level P5 of the downstream equipment;

[0118] 8) Calculate the unavailability of the equipment;

[0119] 9) Calculate the equipment maintenance priority P PM ;

[0120] 10) Work order priority P LOT and equipment maintenance priority P PM Perform normalization and denote the results as P′. LOT and P′ PM ;

[0121] 11) Compare the normalized job scheduling priorities P′ LOT and equipment maintenance priority P′ PM If P′ LOT >P′ PM If necessary, perform workpiece scheduling; otherwise, perform equipment maintenance.

[0122] In step 1), it is determined whether the workpiece meets the process constraints:

[0123]

[0124] In step 2), the urgency level P1 of the workpiece constraint is calculated:

[0125]

[0126] Formula (2) means that at time t, for a workpiece with process constraints, the larger the ratio of the theoretical value to the actual value of the remaining processing time of the process constraints, the greater the probability that the next step will not meet the process constraints, and the easier it is to be selected by the equipment for priority processing; if the theoretical remaining processing time of the process constraints multiplied by the time constant O is greater than the actual remaining processing time of the process constraints, the probability that the workpiece does not meet the process constraints is extremely high, and the system may mark it as an emergency workpiece: it has the highest priority on the equipment.

[0127] In step 3), the formula for calculating the urgency level P3 of the workpiece delivery is as follows:

[0128]

[0129] Where, ∑L n D is the remaining machining time for the workpiece. n It is the theoretical delivery time of the workpiece (one of the properties of the workpiece, known, expressed in terms of a certain moment), T NOW It represents the current moment of decision-making.

[0130] In step 4), the workpiece dispatch priority P is calculated. LOT :

[0131] P LoT =α1·P1+α2·P2+α3·P3

[0132] Where (α1, α2, α3) are three random parameters of the scheduling algorithm; P2 is the client priority inherent to the workpiece (one of the workpiece's attributes, known, its value is a number between 1 and 4). Considering that P1, P2, and P3 are of different orders of magnitude, the value ranges of the three random numbers (α1, α2, α3) are different: α1 and α3 are in the range of (0-1), and α2 is in the range of (0-0.25).

[0133] In step 5), it is determined whether the current time is within the equipment maintenance period:

[0134]

[0135] Among them, T NoW This is the current decision-making moment; T PM-S The earliest start time for equipment maintenance; T PM-E It is the latest time to end equipment maintenance; T PM-MAX This is the longest maintenance time for the equipment.

[0136] In step 6), the urgency level P4 of equipment maintenance is calculated:

[0137]

[0138] In step 7), the load level P5 of the downstream equipment is calculated:

[0139]

[0140] in, T represents the time that workpiece n occupies on downstream device id; id This represents the daily processing capacity of the downstream equipment ID. Formula (7) indicates that the heavier the equipment load, the higher its information variable. If This indicates that the equipment's load has exceeded its available time, and the equipment is marked as a bottleneck. It should be noted that if multiple machines can complete a specific process for a certain workpiece, then T... id This represents the sum of the processing capacities of this type of equipment.

[0141] In step 8), the unavailability of the device is calculated:

[0142]

[0143] In the formula, BM i For machine M i Unavailability, BM min For the minimum unavailability in the machine, BM max This represents the maximum unavailability of the machine.

[0144] In step 9), the equipment maintenance priority P is calculated. PM :

[0145] P PM =α4·P4 + α5·P5 + α6·P6

[0146] Where (α4, α5, α6) are three random parameters of the scheduling algorithm.

[0147] In step 10), the workpiece assignment priority P is... LoT and equipment maintenance priority P PM Perform normalization and denote the results as P′. LOT and P′ PM .

[0148] In step 11), the normalized job scheduling priority P′ is compared. LOT and equipment maintenance priority P′ PM If P′ LOT >P′ PM If the task is not scheduled, workpiece scheduling is performed; otherwise, equipment maintenance is performed, where the duration of equipment maintenance is set to an interval (T). PM-MIN ,T PM-MAX The duration of the random number within )

[0149] The IWOA-MLP algorithm outputs the expected performance that matches the current production state; the IWOA-MLP algorithm is also used to optimize the scheduling rule parameters and update the scheduling rules.

[0150] Among them, the IWOA-MLP algorithm is an improved whale optimization algorithm, such as... Figure 5 As shown, it includes the following steps:

[0151] 1) The initial population position of the algorithm is randomly generated by chaotic Tent mapping, which makes the population distribution more uniform and speeds up the convergence speed of the algorithm.

[0152] 2) New nonlinear parameters are proposed to enable the whale optimization algorithm to adapt to complex nonlinear problems;

[0153] 3) Introduce a fitness control mechanism to prevent update stagnation by controlling population position updates and improve the algorithm's ability to escape local optima;

[0154] 4) Introduce the Harris Eagle swarm mechanism to accelerate the whale's hunting speed;

[0155] 5) At the end of each whale hunting iteration, a Gaussian detection position control mechanism is added to increase the optimization accuracy of the algorithm.

[0156] The Whale Optimization Algorithm is a swarm intelligence optimization algorithm developed by simulating the bubble-net foraging method of humpback whales. The algorithm mainly consists of two phases: shrinking the encirclement and updating the position.

[0157] (1) Shrink the enclosure

[0158] Whales can sense the area where prey is located and surround it. Since the optimal design's position in the hunting or search space is inconsistent with previous positions, the WOA optimization algorithm assumes that the current best candidate solution is the target prey or close to the optimal solution. In this case, the whale defines the best search agent; then, other search agents will attempt to change positions to move closer to the best search agent. The shrinking encirclement hunting behavior is described by the following formula:

[0159] X(t+1)=X*(t)-A·D1#(10)

[0160] D1=|C·X*(t)-X(t)|#(11)

[0161] t represents the current iteration number, A and C are vector coefficients, X(t) is the current position, X(t+1) is the next position, D1 is the absolute value of the difference between C times the prey position and the current whale position, and X*(t) is the position vector of the current optimal solution. If a better solution is found in each iteration, i.e., the fitness value of the current position is less than the fitness value of X*(t), then the whale position vector at this time is set to the new X*.

[0162] The formulas for calculating A and C are as follows:

[0163] A = 2·a·r1 - a#(12)

[0164] C = 2·r²#(13)

[0165]

[0166] r1 and r2 are random numbers within the range [0,1], T max The maximum number of iterations is T, which is chosen in this paper. max =500, the value of a is in the range of [0,2], and it changes linearly with the increase of t.

[0167] (2) Location update

[0168] There are two methods for updating the whale's location: spiral update and random search. To simulate the whale's location update at a given moment and ensure that the whale has an equal probability of choosing either spiral update or random search at any given time, a random number p is set to a value in the range [0,1]. The update method for the whale's location is randomly selected based on the value of p.

[0169] When p ≥ 0.5, the spiral position update method is selected. By simulating the way a whale spirals around its prey, a spiral position update equation is established to update the whale's position for the next time. The calculation formula is as follows:

[0170] X(t+1)=D2·e bl ·cos(2πl)+X*(t)#(15)

[0171] D2=|X*(t)-X(t)|#(16)

[0172]

[0173] l=(a1-41)*r3+1#(18)

[0174] Where D2 represents the distance between the prey and the whale; b is a parameter that controls the shape of the spiral, which is set to 1 in this paper; a1 is a parameter that changes linearly in the range of [-2, -1]; r3 is a random number between [0, 1]; and l takes values ​​in the range of [-2, 1].

[0175] When p < 0.5, the random search position update formula is selected. The random search is divided into two methods. When |A| < 1, it means that the whale is moving towards the prey position. At this time, the shrinking encirclement formula is used to simulate the whale's action behavior, that is, to encircle the prey through formula (10).

[0176] When |A|≥1, it means that the whale has moved beyond the location of the prey. At this time, the whale will abandon the previous direction of movement and randomly search for a new updated position in other directions to avoid getting trapped in local extrema, as shown in equations (19) and (20).

[0177] D rand =|C·X rand (t)-X(t)|#(19)

[0178] X(t+1)=X rand (t)-A·D rand #(20)

[0179] X rand D represents the vector of randomly selected whale positions. rand Indicates C times X rand The absolute value of the difference between X(t) and X(t).

[0180] The basic whale optimization algorithm suffers from several drawbacks. It cannot select the most efficient update method during whale position updates and may rapidly converge to local optima, degrading the quality of the solution. Therefore, an improved whale optimization algorithm based on a siege mechanism is adopted.

[0181] Y=X*(t)-A·D1#(21)

[0182] Z=Y+S*LF(D)#(22)

[0183] This embodiment combines the actual situation of the semiconductor production line and the needs of the enterprise, and selects the simulation platform of a semiconductor manufacturing enterprise in Shanghai to verify the effectiveness of the optimized scheduling method. Four performance indicators are selected as the objectives of scheduling optimization: average processing cycle of workpieces, on-time delivery rate of workpieces, equipment unavailability, and robustness of scheduling scheme.

[0184] 1) Comparison of combined dispatching rules and heuristic rules (FIFO: first-come, first-served rule, SPT: longer or shorter processing time, earlier delivery date of workpiece, earlier delivery date, earlier delivery date, earlier delivery date, earlier delivery date, etc.).

[0185] To further verify the effectiveness of the combined dispatching rule proposed in this chapter, this combined scheduling rule is compared with FIFO, SPT, and EDD. Statistical analysis is performed on the average processing cycle of workpieces, on-time delivery rate of workpieces, equipment availability, and robustness of the scheduling rule.

[0186] Average machining cycle time (CT) of workpiece:

[0187]

[0188] Among them, CT i This represents the processing cycle of workpiece i, that is, the time difference between workpiece i entering the line and exiting the line. N represents the number of workpieces completed per unit time.

[0189] On-time delivery rate (ODR):

[0190] ODR represents the ratio of the number of workpieces delivered on time to the total number of completed workpieces within a given time period.

[0191] ODR = n1 / n1+n2

[0192] Where n1 represents the number of workpieces delivered on time within a unit of time, and n2 represents the number of workpieces produced by the production line within a unit of time.

[0193] Let Tpoint(k) = {0, t1, t2, ..., t r C max},t1,t2,…,t r Machine M k The start time of preventive maintenance, r is the number of preventive maintenance operations, and t is the number of preventive maintenance operations. r+1 =C max This refers to the completion time of the last operation of machine k. Because machine unavailability is an increasing function, and the machine returns to its initial state after preventative maintenance, it only occurs at times t1, t2, ..., t... r+1 Calculate the unavailability of all devices.

[0194] A relaxation-time-based measure is used to describe the scheduling robustness index. For a scheduling scheme σ, the robustness of the scheduling scheme is defined by the following formula:

[0195]

[0196] in, For workpiece J i The number of immediately following workpieces is also the number of workpieces J. i The weight represents the impact of the workpiece's position on scheduling. For workpiece J i The relaxation time of workpiece J i The difference between the latest possible start time and the earliest possible start time. M(σ) The larger the value, the better the scheduling robustness.

[0197] The results are shown in Table 1.

[0198] Table 1 Performance Indicators under Different Dispatch Rules

[0199] performance Combined Dispatch Rules FIFO SPT EDD Processing cycle 42.037 42.853 39.374 42.761 On-time delivery rate 82.672 81.974 82.592 84.048 Equipment unavailability indicators 0.424 0 0 0 Robustness index 0.329 0.248 0.175 0.206

[0200] As shown in Table 1, the processing cycle of the combined dispatching rule is better than FIFO and EDD, but worse than SPT. However, SPT may cause the workpiece queue length to exceed expectations and may increase the workpiece rework rate. The on-time delivery rate of the combined dispatching rule is slightly worse than EDD, but EDD does not consider the overall state of the production line and cannot reflect real-time and dynamic characteristics. FIFO, SPT and EDD do not consider the predictive maintenance of equipment, and the equipment failure rate is 0. Therefore, the equipment unavailability index is 0 for all of them. The robustness index of the combined dispatching rule is the highest.

[0201] 2) Closed-loop optimization verification based on IWOA-MLP algorithm

[0202] The simulation experiments compared the scheduling system and composite dispatching rules based on IWOA-MLP, as shown in Table 2.

[0203] Table 2 Simulation results of the scheduling method based on PSO-SVR

[0204]

[0205] As shown in Table 2, this part of the experiment can simultaneously demonstrate the effectiveness of the "performance-driven" and "closed-loop optimization" links in this method, and can improve the corresponding performance indicators.

[0206] 3) Comparison of this scheduling system with self-organizing scheduling systems

[0207] To test the effectiveness of the "robust optimization scheduling system based on IWOA-MLP algorithm" proposed in this invention, the experimental results are shown in Table 3, compared with the self-organizing scheduling system and the predictive scheduling system based on MLP algorithm.

[0208] Table 3 Comparison of simulation results for different scheduling systems

[0209]

[0210] The robust optimization scheduling system based on the IWOA-MLP algorithm integrates equipment failure and preventative maintenance, factors that the other two scheduling systems do not consider. The processing cycle time and on-time delivery rate generated by the three scheduling systems are comparable, demonstrating the superiority of this system. Furthermore, it improves the robustness of the system compared to self-organizing and predictive scheduling systems.

[0211] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A robust optimization scheduling method for complex manufacturing systems, characterized in that, Includes the following steps: S1. Obtain the real-time production status and production status data of the production line; S2. Generate a sample set based on the production status data; S3. Based on the sample set, generate scheduling rules for fusion equipment maintenance and workpiece dispatch; S4. Combine the real-time production status and the scheduling rules to generate a scheduling scheme; The process for generating scheduling rules for fusion equipment maintenance and workpiece dispatch includes the following steps: S401. Determine whether the workpiece meets the process constraints. If yes, proceed to the next step; otherwise, the workpiece enters the scrap area. S402. Calculate the urgency of the workpiece's process constraints. : ； At time t, for a workpiece with process constraints, the larger the ratio of the theoretical value to the actual value of the remaining processing time for the process constraints, the greater the probability that the next step will not meet the process constraints, and the more likely it is to be selected by the equipment for priority processing; if the theoretical remaining processing time for the process constraints multiplied by the time constant O is greater than the actual remaining processing time for the process constraints, the probability that the workpiece will not meet the process constraints is extremely high, and the system may mark it as an emergency workpiece: it has the highest priority on the equipment. S403, Calculate the urgency of workpiece delivery : ， in, This is the remaining processing time for the workpiece. It is the theoretical delivery date of the workpiece (one of the properties of the workpiece, known, and expressed in terms of a certain moment). Represents the current decision-making moment; S404, Calculate workpiece dispatch priority : ， in, These are three random parameters for the scheduling algorithm; Prioritize customers who own the workpiece; S405. Determine if the current time is within the equipment maintenance period. If it is, proceed to the next step; otherwise, proceed as follows: Assign tasks based on the size of the workpiece; S406. Urgency of Computer Equipment Maintenance : ， in, This is the current moment for decision-making; This is the earliest time to begin equipment maintenance; This is the latest time to end equipment maintenance; This is the longest duration of equipment maintenance; S407, Calculate the load level of downstream equipment : ， in, Representative workpiece Downstream equipment The time occupied on; Representing downstream equipment The processing capacity on that day; S408. Unavailability of computing devices: ， in, For machines Unavailability, The minimum unavailability in the machine. Maximum machine unavailability; S409, Computer Equipment Maintenance Priority : ， in, These are three random parameters for the scheduling algorithm; S410, Prioritize the workpiece assignment. and equipment maintenance priority Perform normalization processing, and record the normalization results as follows: and ; S411. Compare the normalized job dispatch priorities. and equipment maintenance priority ,like If necessary, dispatch the workpiece; otherwise, perform equipment maintenance. S5. Apply the scheduling scheme to the production line and obtain the production performance feedback from the production line; S6. Based on the feedback of production performance, update the scheduling rule parameters and regenerate the scheduling rules; S7. Repeat steps S4 to S6 until the production performance feedback from the production line reaches the expected standard.

2. The robust optimization scheduling method for complex manufacturing systems according to claim 1, characterized in that, The production performance includes workpiece processing cycle, workpiece on-time delivery rate, equipment availability index, and scheduling scheme robustness.

3. The robust optimization scheduling method for complex manufacturing systems according to claim 1, characterized in that, A feature selection algorithm based on an improved immune algorithm is used to select features from the sample set to reduce redundant features and computation time. Specifically, the algorithm includes the following steps: S301, antibody and antigen initialization; where antibodies are different combinations of the production line's production status attributes. S302. Calculate the affinity between the antibody and the antigen; S303. Based on antibody affinity, antibodies are cloned and mutated. The higher the antibody affinity, the more clones there are and the lower the probability of mutation. S304. Determine whether the maximum number of iterations has been reached. If yes, proceed to step S307; otherwise, proceed to step S305. S305. The average similarity of antibodies is calculated using the reciprocal of the Euclidean distance; S306. If the average similarity of antibodies is less than the threshold, sort the antibody affinity in the antibody population and select the appropriate antibody. d Add the antibody with the highest affinity to the antibody library and replace an equal number of antibodies with the lowest affinity, then proceed to step S302; otherwise, proceed to step S307. S307. Screen the antibodies in the antibody group and select the antibody with the highest affinity as the optimal solution.

4. The robust optimization scheduling method for complex manufacturing systems according to claim 1, characterized in that, Based on the production performance feedback, the expected performance matching the current production state is output through the IWOA-MLP algorithm; the scheduling rule parameters are optimized and output through the IWOA-MLP algorithm, and the scheduling rules are updated.

5. A robust optimization scheduling system for complex manufacturing systems, characterized in that, It includes a production line simulation module, a sample generation module, an offline training module, and an online scheduling module; The production line simulation module is connected to the sample generation module, the sample generation module is connected to the offline training module, the offline training module is connected to the online scheduling module, and the online scheduling module is connected to the production line simulation module. The system as a whole forms a closed-loop feedback. The production line simulation module is used to generate real-time production status and production status data, and send the real-time production status to the online scheduling module and the production status data to the sample generation module. The sample generation module generates a sample set based on the production status data and sends the sample set to the offline training module; The offline training module generates scheduling rules that integrate equipment maintenance and workpiece dispatch based on the sample set. The online scheduling module generates a scheduling scheme based on the scheduling rules and sends the scheduling scheme to the production line simulation module; The production line simulation module applies the scheduling scheme and feeds back the production performance to the online scheduling module. The online scheduling module updates the scheduling rule parameters based on the feedback production performance and sends the updated scheduling rule parameters to the offline training module. The offline training module regenerates the scheduling rules based on the updated scheduling rule parameters.

6. A robust optimization scheduling system for complex manufacturing systems according to claim 5, characterized in that, The production performance considered in scheduling includes workpiece processing cycle, workpiece on-time delivery rate, equipment availability indicators, and the robustness of the scheduling scheme.

7. A robust optimization scheduling system for complex manufacturing systems according to claim 5, characterized in that, The offline training module employs a feature selection algorithm based on an improved immune algorithm to select features from the sample set, thereby reducing redundant features and computation time. Specifically, it includes the following steps: S301, antibody and antigen initialization; where antibodies are different combinations of the production line's production status attributes. S302. Calculate the affinity between the antibody and the antigen; S303. Based on antibody affinity, antibodies are cloned and mutated. The higher the antibody affinity, the more clones there are and the lower the probability of mutation. S304. Determine whether the maximum number of iterations has been reached. If yes, proceed to step S307; otherwise, proceed to step S305. S305. The average similarity of antibodies is calculated using the reciprocal of the Euclidean distance; S306. If the average similarity of antibodies is less than the threshold, sort the antibody affinity in the antibody population and select the appropriate antibody. d Add the antibody with the highest affinity to the antibody library and replace an equal number of antibodies with the lowest affinity, then proceed to step S302; otherwise, proceed to step S307. S307. Screen the antibodies in the antibody group and select the antibody with the highest affinity as the optimal solution.

8. A robust optimization scheduling system for complex manufacturing systems according to claim 5, characterized in that, The offline training module generates scheduling rules for fusion equipment maintenance and workpiece dispatching, including the following steps: S401. Determine whether the workpiece meets the process constraints. If yes, proceed to the next step; otherwise, the workpiece enters the scrap area. S402. Calculate the urgency of the workpiece's process constraints. ; S403, Calculate the urgency of workpiece delivery ; S404, Calculate workpiece dispatch priority ; S405. Determine if the current time is within the equipment maintenance period. If it is, proceed to the next step; otherwise, proceed as follows: Assign tasks based on the size of the workpiece; S406. Urgency of Computer Equipment Maintenance ; S407, Calculate the load level of downstream equipment ; S408. The unavailability of computing devices; S409, Computer Equipment Maintenance Priority ; S410, Prioritize the workpiece assignment. and equipment maintenance priority Perform normalization processing, and record the normalization results as follows: and ; S411. Compare the normalized job dispatch priorities. and equipment maintenance priority ,like If necessary, dispatch the workpiece; otherwise, perform equipment maintenance.

9. A robust optimization scheduling system for complex manufacturing systems according to claim 5, characterized in that, The online scheduling module outputs the expected performance that matches the current production status based on the feedback production performance using the IWOA-MLP algorithm; it also optimizes the scheduling rule parameters using the IWOA-MLP algorithm and updates the scheduling rules.

Images