A method for optimizing real-time production lead time of reconfigurable systems based on model predictive control
By adopting a model prediction control method in a reconfigurable production system, using max-plus algebra to generate time-varying state space equations of streamlines, optimizing the real-time job release plan, solving the risk of real-time production lead-update optimization in the existing technology , achieving cost minimization and operational efficiency improvement.
Patent Information
- Application Number
- CN202111658328.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-30
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-12-30
AI Technical Summary
The prior art is difficult to effectively optimize real-time production lead times in reconfigurable production systems, especially in the case of limited buffer capacity of workstations and reconfigurable production systems, and ignores coordination between products in work, advance and delays, resulting in Optimization risk is high.
Using a model prediction control method, the time-varying state space equation of streamlines is generated using max-plus algebra, an optimization model for reconstructed production system is established, and the real-time job release plan is optimized through the hidden enumeration algorithm, and the real-time response is combined with the event-driven mechanism. Interrupt, minimizing the cost of product in-product inventory, final product inventory and delayed penalty.
The total expected cost of work-in-one product inventory, final product inventory and delayed penalty costs is achieved, which reduces the tension of the production system, improves the operational efficiency of the enterprise, reduces operating costs, and enhances resource coordination and predictive control capabilities.
Smart Images

Figure CN114418197B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of enterprise production operation, and in particular to a method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control. Background Art
[0002] With the rapid development of production technology, enterprises pay more and more attention to the design and scheduling of production systems. The planning and scheduling of resource management systems play an important role in improving the efficiency of resource management systems and are a hot topic in previous research. The production planning problem aims to adjust the scalable production capacity and system functions to meet dynamic customer needs at the lowest cost.
[0003] Many studies have seriously addressed the scheduling problem in reconfigurable production systems, but most of them focus on revealing the relationship between machine configuration and job completion time, so as to minimize the total expected cost. At the same time, real-time production lead time optimization is also very important for coordinating existing operating costs (i.e., inventory costs of work-in-process and final products and penalty costs for lateness), but few studies on scheduling focus on the optimization of real-time production lead time. Moreover, most popular studies assume that the intermediate buffer capacity between workstations is sufficient, so there will be no transitional idleness or congestion of workstations, but this assumption is unrealistic because the intermediate buffer must be finite for real-time process control. At the same time, previous studies have rarely paid attention to the reconfigurable behavior of production systems; in these studies, the coordination between work-in-process, lead and delay is also ignored. In actual situations, most production systems require reconfiguration of production lines and real-time regulation of production lead time. The real-time production lead time optimization problem of reconfigurable systems obtained by the calculation methods of existing studies still has high risks. How to provide enterprises with a practical and feasible real-time production lead time optimization solution in reconfigurable production systems is of great practical significance.
[0004] Therefore, the present invention proposes a real-time production lead time optimization method for a reconfigurable system based on model predictive control, so that the total expected cost of work-in-process inventory, final product inventory and delay penalty is minimized. Taking into account the predictive multi-stage transition and reconfigurable behavior, the model uses max-plus algebra to generate the time-varying state space equations of the streamline. In this regard, a systematic method for permanent production loss identification can be formulated for real-time interruptions in the production process. The identification helps to provide feedback signals for the optimization model, thereby realizing closed-loop feedback control for real-time production lead time optimization. More importantly, the optimization model can effectively combine real-time production lead time optimization with machine reconstruction, thereby minimizing the total expected cost, and adopts an event-driven mechanism to respond to real-time interruptions, which can significantly reduce the tension of the production system and reduce the operating costs of the enterprise. Summary of the invention
[0005] In view of the existing problems, the present invention proposes a method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control. The time-varying state space equation of the streamline is generated by max-plus algebra. The real-time job release plan with the goal of minimizing the total expected cost of work-in-process inventory, final product inventory and delay penalty is established to perform predictive control. The model is optimized and solved using an implicit enumeration algorithm, and a real-time job release plan with a lower total expected cost is calculated.
[0006] In order to solve the technical problems existing in the prior art, the technical solution of the present invention is:
[0007] A method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control comprises the following steps:
[0008] Step S1: Collecting real-time data of materials and operations and real-time data of workstation status in the production system assembly line to be optimized, and determining parameters required for model calculation, wherein the real-time data of materials and operations at least include operation transfer data, operation location data and processing time; the real-time data of workstation status at least include machine failure event data, operation processing data, workstation reconfiguration time and workstation buffer capacity;
[0009] Step S2: Considering the predictive multi-stage transition and reconfigurable behavior, a reconfigurable production system model based on max-plus algebra is established. The completion time is predicted by establishing a time-varying transition model, a time-varying state space equation and a state transfer matrix for the time variation of the planned and real-time discrete events between multiple workstations, and various constraints in the real-time planning lead time optimization are handled;
[0010] Step S3: Considering the environment of the reconfigurable production system, a new optimization method and an optimization model of the real-time production lead time of the reconfigurable system are established to perform predictive control with the goal of minimizing the real-time job release plan of the total expected cost of the work-in-process inventory, the final product inventory and the delay penalty;
[0011] Step S4: solving the optimization model using an implicit enumeration algorithm, and finally obtaining a real-time job release plan with a smaller total expected cost;
[0012] In S2, the reconfigurable system described by max-plus algebra is further as follows:
[0013] Step S21: setting the assumption that there is enough buffer before the first workstation or after the last workstation;
[0014] Step S22: Draw a time event graph such as Figure 2 As shown in the figure, a time-varying transition model of a reconfigurable production system is established. The specific description is as follows: i The location represents the i-th workstation and the location where the reconstruction occurs, m represents the number of workstations, and p i The position represents the i-th buffer, p′ i (2≤i≤M) where the position controls the amount of real-time work being processed in the buffer, p″ i (1≤i≤M) controls the restart of the corresponding workstation, u(k) represents the start time of the kth job released by the production system, and the symbol p i It represents the real-time number of jobs being performed by workstation i, symbolized by p′ i represents the remaining capacity of the i-th buffer, σ i (k) represents the pure operation time of the kth job of workstation i, represents the machine reconfiguration time before the kth job of workstation i starts, σ′ i (k) represents the start time of the kth job of workstation i, and the calculation formula is:
[0015]
[0016] N i represents the capacity of the i-th buffer, the time-varying transition behavior of the reconfigurable production system can be expressed by the following equation:
[0017]
[0018]
[0019]
[0020] Step S23: According to the time-varying transformation, a time-varying state space equation based on max-plus algebra is established as follows:
[0021]
[0022]
[0023] Among them, x(k) represents the column vector of workstation status, which is given by
[0024] [x1(k),x2(k),...,x M (k)] T y(k) represents the completion time of the kth job, A s(k) represents the relationship between workstation states, B(k) represents the relationship between job start time and workstation state, and C(k) represents the relationship between job completion time and workstation state. The state transition matrix is as follows:
[0025]
[0026]
[0027] [A s (k)] ij Indicates the buffer capacity N j = s and there is a reconstruction path from j to workstation i, equal to the job start time, otherwise equal to ε; when the workstation has system input or output [B(k)] i1 and [C(k)] 1j is equal to 0, otherwise equal to ε;
[0028] In order to develop a predictive controller based on the above equations, the above state space equations are converted to standard state space equations as follows:
[0029]
[0030]
[0031] in, The state transfer matrix is as follows:
[0032]
[0033]
[0034]
[0035] The matrix and It can be generated by the following formula:
[0036]
[0037]
[0038] Step S24: Switch the state transition matrix to represent the real-time interruption in the reconfigurable production system. This interruption is defined as It means that when workstation i processes the kth n When there are multiple jobs, the interruption event can last for d n , workstation i for the kth n The processing time of a job can be expressed as σ″ i (k n )=σ′i (k n )+d n ,The updated state space equation is then used to represent the real-time production state of the reconfigurable production system;
[0039] The real-time production lead time model of the reconfigurable system described in step S3 is further as follows:
[0040] Step S31: Set the assumptions of the model: each order contains only one product, each workstation can only process one job at a time, any job can only be processed on the downstream workstation after it is processed on the current workstation, and interruption of processing is not allowed, if the products of two jobs are the same, reconstruction and reconfiguration are not required, only the first stage needs to set the planning lead time, while the other stages are determined by the upstream and downstream work-in-progress, the jobs are operated according to the first-in-first-out (FIFO) rule, and the buffer capacity before the first stage and after the last stage is sufficient;
[0041] Step S32: Set the symbols of the known parameters or sets in the model and the decision variables, as follows: Ω represents the current unfinished job set, q j Indicates the number of jobs, a j The earliest start time of job j is the control variable of the model, b j represents the latest start time of job j, M represents the number of workstations, i represents the workstation index, and p ij represents the processing time of job j on workstation i, d j Indicates the delivery date, e j represents the benefit of job j, w 1j represents the delay penalty coefficient, w 2j represents the inventory cost coefficient of the final product, w 3j represents the inventory cost coefficient of the work-in-progress, Φ represents the set of job release sequences, t i (k, k′) represents the reconfiguration time of workstation i processing job k to k′, k0 represents the time step;
[0042] Step S33: The objective function minimizes the total expected cost, wherein the total expected cost is composed of two parts: the work-in-process inventory cost and the final product inventory and delay penalty cost; the intermediate variable or set sequence at least includes the operation start time sequence, the operation completion time sequence, the work-in-process inventory cost, the final product inventory and delay cost, and the sum of the number of operations in the decision period; the step S33 further includes the following steps:
[0043] Step S331: The published job start time series is represented by μ(k0+k|k0), which is also the decision variable of the model, and the job completion time series is represented by y(k0+k|k0);
[0044] Step S332: Work in process inventory cost j in The work-in-process inventory cost coefficient w 3j , the deviation between the completion time of the operation and the deviation between the start time of the operation, and the calculation formula is:
[0045]
[0046] Step S333: The delay cost of the final product is related to the deviation between the completion time of the operation and the planned delivery time. The inventory cost of the final product is related to the deviation between the actual delivery time and the completion time of the operation between adjacent products. The calculation formula is:
[0047]
[0048] Step S334: The sum of the number of operations within the decision period is calculated by express;
[0049] Step S335: The optimization goal is to minimize the sum of the costs of S332 and S333: min j = j in +j out ;
[0050] Step S34: Establishing the constraint conditions of the model. Step S34 further includes:
[0051] Step S341: Demonstrate the relationship between the job release time and its corresponding completion time, with the following constraints:
[0052]
[0053]
[0054] Step S342: Ensure that the job can be released only when the first workstation of the reconfigurable production system is available, with the following constraints:
[0055] a(k0+k|k0)<μ(k0+k|k0) <b(k0+k|k0),1≤k≤N p
[0056] The parameters a(k0+k|k0) and b(k0+k|k0) are determined by the previous assignment;
[0057] Step S343: constrain the job start time series μ(k0+k|k0) to be monotonically increasing, and the constraints are as follows:
[0058] μ(k0 + k|k0) - μ(k0 + k - 1|k0) ≥ 0, 1 ≤ k ≤ N p
[0059] Step S344: When N c < k, increase the job release time equally, and the constraints are as follows:
[0060] μ(k0 + k|k0) - μ(k0 + k - 1|k0) = μ(k0 + k + 1|k0) - μ(k0 + k|k0)
[0061] N c < k < N p -1
[0062] Among them, the symbol N c represents the control range, and its purpose is to reduce the number of optimization variables;
[0063] Step S345: Decision variable constraints to ensure the non - negativity and integrity of decision variables, and the constraints are as follows:
[0064] μ(k0 + k|k0) ≥ 0, μ(k0 + k|k0) ∈ Z, k = 1,..., N p
[0065] Step S346: Additional constraints for real - time work - in - process, which are generated by the kanban mechanism, the continuous work - in - process mechanism, or the drum - buffer - rope mechanism, and the constraints are as follows:
[0066]
[0067] Among them, L(k0 + k|k0) represents the processing time of the (k0 + k) - th job, and F(k0 + k|k0) is determined by the allowed number of work - in - process;
[0068] The specific method for optimizing the real - time production lead time of the reconfigurable system based on model predictive control described in Step S3 is as follows:
[0069] Step S35: To reduce the tension of the reconfigurable production system, a new optimization method that does not select continuous re - optimization is established, that is, re - optimization is triggered only when the interruption affects the permanent throughput of the system;
[0070] Step S351: Construct a production loss identification method centered on the bottleneck. Production losses are divided into permanent production losses and non - permanent production losses, and the interruption event with the longest interference time will not cause permanent production losses at the end - of - line workstation;
[0071] Step S3511: If the bottleneck is fixed, then closest to the end of the production line, assuming there may be one or more slowest workstations, production loss identification can be divided into three categories: unexpected interruption The opportunity windows are as follows:
[0072] Table 1 Unexpected interruption Window of opportunity
[0073]
[0074] in, It is the last and slowest workstation in the reconfigurable production system;
[0075] Step S3512: If the bottleneck is shifting, use steady-state analysis to identify production losses. Based on the state space equation of step S23, use the deviation c to determine whether there is a permanent production loss. The calculation formula is as follows:
[0076]
[0077]
[0078] Among them, if c is greater than 0, there is a permanent production loss, otherwise, there is no permanent production loss;
[0079] Sometimes unexpected interruptions It will not cause any permanent production loss, but it may cause the reconfigurable production system to maintain excessive flow for a long time during the production process. In this case, the increase in inventory costs caused by work-in-progress should be considered, and the system recovery time t should be calculated. The calculation formula is as follows:
[0080]
[0081]
[0082] Step S352: If there is a permanent production loss or the system recovery time is long, the production lead time of the job must be regenerated to minimize the total expected cost. To this end, a new processing parameter switching model must be used, and the state transfer matrix in step S23 is used for event switching. By analyzing the new state space equation, a new state space equation is given for re-optimization, and its parameter k″ is calculated as follows:
[0083]
[0084] The implicit enumeration algorithm described in step S4 is further as follows:
[0085] Step S41: The size of the enumeration space significantly affects the search efficiency of the algorithm. First, generate the enumeration space. This step S41 further includes the following steps:
[0086] Step S411: The lower bound of each variable μ(k + k0|k0), 1 ≤ k ≤ N can be determined by the processing time of the first workstation in the reconfigurable production system. The calculation formula is as follows: p The lower bound, and the calculation formula is as follows:
[0087]
[0088] where μ lower (k0|k0) = μ(k0);
[0089] Step S412: The upper bound of each variable μ(k + k0|k0), 1 ≤ k ≤ N is obtained based on the order due date and the late penalty cost. For each order, the relationship between the latest delivery time and the due date is as follows: p The upper bound, and for each order, the relationship between the latest delivery time and the due date is as follows:
[0090]
[0091] Furthermore, the expression formula for the latest delivery time is obtained as follows:
[0092]
[0093] Considering the processing parameters of order j, the upper bound of each variable μ(k + k0|k0), 1 ≤ k ≤ N can be calculated. The calculation formula is as follows: p The upper bound, and the calculation formula is as follows:
[0094]
[0095] Step S42: Determine the enumeration strategy: To improve the search efficiency, based on the analysis of the variable decision space, a variable search step is designed. This step S42 further includes the following steps:
[0096] Step S421: Set μ(k + k0|k0) equal to its lower bound, specify the search step size δ, determine the initial operating cost f0, and start the search from 1 → k;
[0097] Step S422: Calculate the operating cost f for each search through the system throughput and then according to the space state equation;
[0098] Step S423: If f > f0, continue the search with the step size; if f = f0, return to step S422; if f < f0, execute step S424;
[0099] Step S424: adjust the step length to δ / 2. If the obtained variable is less than the unit production time, execute S425; otherwise, return to step S422;
[0100] Step S425: If k>N p , stop searching and export function f0 otherwise, return to step S4222.
[0101] Compared with the prior art, the present invention solves the real-time production lead time optimization problem of reconfigurable systems in enterprises from the perspective of predictive control. The technical solution of the present invention can solve the problems of limited workstation buffer capacity, reconfigurable production system, and real-time predictive control of production lead time, which helps to improve the operating efficiency of the enterprise production system, reduce the enterprise's operating costs, and enable the production system to have stronger resource coordination and predictive control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0102] Figure 1 A flowchart of a method for optimizing real-time production lead time of a reconfigurable system based on model predictive control provided by the present invention;
[0103] Figure 2 To reconstruct the time event graph of the production system;
[0104] Figure 3 The physical layout and series model diagram of the door welding line in the specific experimental case;
[0105] Figure 4 It is a time event diagram in a specific experimental case; DETAILED DESCRIPTION
[0106] The specific implementation modes of the present invention are described in detail below with reference to the accompanying drawings.
[0107] See also Figure 1 The flowchart is a method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control, which includes the following steps:
[0108] Step S1: collect the real-time data of materials and operations and the real-time data of workstation status in the production system assembly line to be optimized through time study, and determine the parameters required for model calculation, wherein the real-time data of materials and operations at least include operation transfer data, operation location data and processing time; the real-time data of workstation status at least include machine failure event data, operation processing data, workstation reconfiguration time and workstation buffer capacity;
[0109] Step S2: Considering the predictive multi-stage transition and reconfigurable behavior, a reconfigurable production system model based on max-plus algebra is established. The completion time is predicted by establishing a time-varying transition model, a time-varying state space equation and a state transfer matrix for the time variation of the planned and real-time discrete events between multiple workstations, and various constraints in the real-time planning lead time optimization are handled;
[0110] Step S3: Considering the environment of the reconfigurable production system, a new optimization method and an optimization model of the real-time production lead time of the reconfigurable system are established to perform predictive control with the goal of minimizing the real-time job release plan of the total expected cost of the work-in-process inventory, the final product inventory and the delay penalty;
[0111] Step S4: solving the optimization model using an implicit enumeration algorithm, and finally obtaining a real-time job release plan with a smaller total expected cost;
[0112] In S2, the reconfigurable system described by max-plus algebra is further as follows:
[0113] Step S21: setting the assumption that there is enough buffer before the first workstation or after the last workstation;
[0114] Step S22: Draw a time event graph such as Figure 2 As shown in the figure, a time-varying transition model of a reconfigurable production system is established. The specific description is as follows: i The location represents the i-th workstation and the location where the reconstruction occurs, m represents the number of workstations, and p i The position represents the i-th buffer, p′ i (2≤i≤M) where the position controls the amount of real-time work being processed in the buffer, p″ i (1≤i≤M) controls the restart of the corresponding workstation, u(k) represents the start time of the kth job released by the production system, and the symbol p i It represents the real-time number of jobs being performed by workstation i, symbolized by p′ i represents the remaining capacity of the i-th buffer, σ i (k) represents the pure operation time of the kth job of workstation i, represents the machine reconfiguration time before the kth job of workstation i starts, σ′ i (k) represents the start time of the kth job of workstation i, and the calculation formula is:
[0115]
[0116] N irepresents the capacity of the i-th buffer, the time-varying transition behavior of the reconfigurable production system can be expressed by the following equation:
[0117]
[0118]
[0119]
[0120] Step S23: According to the time-varying transformation, a time-varying state space equation based on max-plus algebra is established as follows:
[0121]
[0122]
[0123] Among them, x(k) represents the column vector of the workstation status, which is composed of [x1(k),x2(k),...,x M (k)] T y(k) represents the completion time of the kth job, A s (k) represents the relationship between workstation states, B(k) represents the relationship between job start time and workstation state, and C(k) represents the relationship between job completion time and workstation state. The state transition matrix is as follows:
[0124]
[0125]
[0126] [A s (k)] ij Indicates the buffer capacity N j = s and there is a reconstruction path from j to workstation i, equal to the job start time, otherwise equal to ε; when the workstation has system input or output [B(k)] i1 and [C(k)] 1j is equal to 0, otherwise equal to ε;
[0127] In order to develop a predictive controller based on the above equations, the above state space equations are converted to standard state space equations as follows:
[0128]
[0129]
[0130] in, The state transfer matrix is as follows:
[0131]
[0132]
[0133]
[0134] The matrix and It can be generated by the following formula:
[0135]
[0136]
[0137] Step S24: Switch the state transition matrix to represent the real-time interruption in the reconfigurable production system. This interruption is defined as It means that when workstation i processes the kth n When there are multiple jobs, the interruption event can last for d n , workstation i for the kth n The processing time of a job can be expressed as σ″ i (k n )=σ′ i (k n )+d n ,The updated state space equation is then used to represent the real-time production state of the reconfigurable production system;
[0138] The real-time production lead time model of the reconfigurable system described in step S3 is further as follows:
[0139] Step S31: Set the assumptions of the model: each order contains only one product, each workstation can only process one job at a time, any job can only be processed on the downstream workstation after it is processed on the current workstation, and interruption of processing is not allowed, if the products of two jobs are the same, reconstruction and reconfiguration are not required, only the first stage needs to set the planning lead time, while the other stages are determined by the upstream and downstream work-in-progress, the jobs are operated according to the first-in-first-out (FIFO) rule, and the buffer capacity before the first stage and after the last stage is sufficient;
[0140] Step S32: Set the symbols of the known parameters or sets in the model and the decision variables, as follows: Ω represents the current unfinished job set, q j Indicates the number of jobs, a j The earliest start time of job j is the control variable of the model, b j represents the latest start time of job j, M represents the number of workstations, i represents the workstation index, and p ij represents the processing time of job j on workstation i, d jIndicates the delivery date, e j represents the benefit of job j, w 1j represents the delay penalty coefficient, w 2j represents the inventory cost coefficient of the final product, w 3j represents the inventory cost coefficient of the work-in-progress, Φ represents the set of job release sequences, t i (k, k′) represents the reconfiguration time of workstation i processing job k to k′, k0 represents the time step;
[0141] Step S33: The objective function minimizes the total expected cost, wherein the total expected cost is composed of two parts: the work-in-process inventory cost and the final product inventory and delay penalty cost; the intermediate variable or set sequence at least includes the operation start time sequence, the operation completion time sequence, the work-in-process inventory cost, the final product inventory and delay cost, and the sum of the number of operations in the decision period; the step S33 further includes the following steps:
[0142] Step S331: The published job start time series is represented by μ(k0+k|k0), which is also the decision variable of the model, and the job completion time series is represented by y(k0+k|k0);
[0143] Step S332: Work in process inventory cost j in The work-in-process inventory cost coefficient w 3j , the deviation between the completion time of the operation and the deviation between the start time of the operation, and the calculation formula is:
[0144]
[0145] Step S333: The delay cost of the final product is related to the deviation between the completion time of the operation and the planned delivery time. The inventory cost of the final product is related to the deviation between the actual delivery time and the completion time of the operation between adjacent products. The calculation formula is:
[0146]
[0147] Step S334: The sum of the number of operations within the decision period is calculated by express;
[0148] Step S335: The optimization goal is to minimize the sum of the costs of S332 and S333: min j = j in +j out ;
[0149] Step S34: Establishing the constraint conditions of the model. Step S34 further includes:
[0150] Step S341: Demonstrate the relationship between the job release time and its corresponding completion time, with the following constraints:
[0151]
[0152]
[0153] Step S342: Ensure that the job can only be released when the first workstation of the reconfigurable production system is available, with the following constraints:
[0154] a(k0 + k|k0) < μ(k0 + k|k0) < b(k0 + k|k0), 1 ≤ k ≤ N p
[0155] where the parameters a(k0 + k|k0) and b(k0 + k|k0) are determined by the previous job;
[0156] Step S343: Constrain the job start time series μ(k0 + k|k0) to be monotonically increasing, with the following constraints:
[0157] μ(k0 + k|k0) - μ(k0 + k - 1|k0) ≥ 0, 1 ≤ k ≤ N p
[0158] Step S344: When N c < k, increase the job release time equally, with the following constraints:
[0159] μ(k0 + k|k0) - μ(k0 + k - 1|k0) = μ(k0 + k + 1|k0) - μ(k0 + k|k0)
[0160] N c < k < N p -1
[0161] where the symbol N c represents the control range, the purpose of which is to reduce the number of optimization variables;
[0162] Step S345: Decision variable constraints to ensure the non-negativity and integrity of the decision variables, with the following constraints:
[0163] μ(k0 + k|k0) ≥ 0, μ(k0 + k|k0) ∈ Z, k = 1,..., N p
[0164] Step S346: Additional constraints on the real-time work-in-process, which are generated by the kanban mechanism, the continuous work-in-process mechanism, or the drum buffer rope mechanism, with the following constraints:
[0165]
[0166]
[0167] Where L(k0+k|k0) represents the processing time of the k0+kth job, and F(k0+k|k0) is determined by the allowed number of work-in-progress;
[0168] The method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control described in step S3 is specifically as follows:
[0169] Step S35: In order to reduce the tension of the reconfigurable production system, a new optimization method is established that does not select continuous re-optimization, that is, re-optimization is triggered only when the interruption affects the permanent throughput of the system;
[0170] Step S351: construct a bottleneck-centered production loss identification method, where production losses are divided into permanent production losses and non-permanent production losses, where the interruption event with the longest interference time will not cause permanent production losses at the end workstation of the production line;
[0171] Step S3511: If the bottleneck is fixed, then closest to the end of the production line, assuming there may be one or more slowest workstations, production loss identification can be divided into three categories: unexpected interruption The opportunity windows are as follows:
[0172] Table 1 Unexpected interruption Window of opportunity
[0173]
[0174] in, It is the last and slowest workstation in the reconfigurable production system;
[0175] Step S3512: If the bottleneck is shifting, use steady-state analysis to identify production losses. Based on the state space equation of step S23, use the deviation c to determine whether there is a permanent production loss. The calculation formula is as follows:
[0176]
[0177]
[0178] Among them, if c is greater than 0, there is a permanent production loss, otherwise, there is no permanent production loss;
[0179] Sometimes unexpected interruptions It will not cause any permanent production loss, but it may cause the reconfigurable production system to maintain excessive flow for a long time during the production process. In this case, the increase in inventory costs caused by work-in-progress should be considered, and the system recovery time t should be calculated. The calculation formula is as follows:
[0180]
[0181]
[0182] Step S352: If there is a permanent production loss or the system recovery time is long, the production lead time of the job must be regenerated to minimize the total expected cost. To this end, a new processing parameter switching model must be used, and the state transfer matrix in step S23 is used for event switching. By analyzing the new state space equation, a new state space equation is given for re-optimization, and its parameter k″ is calculated as follows:
[0183]
[0184] The implicit enumeration algorithm described in step S4 is further as follows:
[0185] Step S41: The size of the enumeration space obviously affects the search efficiency of the algorithm. First, the enumeration space is generated. Step S41 further includes the following steps:
[0186] Step S411: Each variable μ(k+k0|k0), 1≤k≤N can be determined from the processing time of the first workstation in the reconfigurable production system. p The lower bound of is calculated as follows:
[0187]
[0188] where μ lower (k0|k0)=μ(k0);
[0189] Step S412: derive each variable μ(k+k0|k0), 1≤k≤N according to the order due date and the late penalty cost p For each order, the relationship between the latest delivery time and the due date is as follows:
[0190]
[0191] The latest delivery time can be further expressed as follows:
[0192]
[0193] Considering the processing parameters of order j, each variable μ(k+k0|k0), 1≤ k≤N can be calculated p The upper bound of is calculated as follows:
[0194]
[0195] Step S42: Determine the enumeration strategy: Based on the analysis of the variable decision space, a variable search step is designed, and this step S42 further includes the following steps:
[0196] Step S421: Set μ(k + k0|k0) equal to its lower bound, specify the search step size δ, determine the initial operating cost f0, and start the search from 1→k;
[0197] Step S422: Calculate the operating cost f for each search through the system throughput and then according to the space state equation;
[0198] Step S423: If f > f0, continue the search by the step size; if f = f0, return to step S422; if f < f0, execute step S424;
[0199] Step S424: Adjust the step size to δ / 2. If the obtained variable is less than the unit production time, execute S425; otherwise, return to step S422;
[0200] Step S425: If k > N p , stop the search and export the function f0; otherwise, return to step S4222.
[0201] To verify the technical effects of the present invention, the superiority of the method of the present invention will be verified through specific experimental cases:
[0202] Select a local welding line. The welding line includes seven workstations and six buffers. The operations between adjacent workstations are transported by a conveyor belt. The description is as follows: Workstation 1 - steel plate welding, Workstation 2 - window frame welding, Workstation 3 - bumper welding, Workstation 4 - bonding and curling, Workstation 5 - induction curing, Workstation 6 - hinge welding, and Workstation 7 - testing.
[0203] Due to the variety of products, robots with flexible tools and fixtures are introduced on this welding line. These robots can quickly reconfigure tools and fixtures to respond to different types of car doors, including the doors of multiple vehicles and the front and rear doors of a single vehicle. For example, stations 1 to 6 in the case have four reconfigurable modes so that four types of car doors can be produced, as Figure 3 shown. Station 7 is a worker - based testing station with full flexibility. The selection of the configurable mode for each station is determined by the order scheduling. To determine the parameters of this welding line, a five - day time study was conducted to collect data during the active production time (k = 500);
[0204] The data listed in Tables 2 to 5 and the order set Ω are selected from the five - day time study.
[0205] Order set Ω
[0206]
[0207]
[0208] Table 2 Buffering capacity of welding line
[0209]
[0210] Table 3 Machine failure event table
[0211]
[0212] Table 4 Weld processing schedule
[0213]
[0214] Table 5 Welding line reconstruction schedule
[0215]
[0216]
[0217] According to the serial model of the weld line (such as Figure 3 As shown in Figure 2 and the buffer capacity listed in Table 2, the timing event diagram of the weld can be expressed as Figure 4 By numbering the transitions and positions from 1 to 7 from left to right (excluding the input), the state-space model for this weld line can be written as:
[0218]
[0219]
[0220] in X(k)=[x1(k),x1(k),x1(k),x1(k),x1(k),x1(k),x1(k)] T
[0221] Then, according to step S23, A can be calculated. s (k),s∈[1,6],B0(k),
[0222] In order to evaluate the effectiveness and efficiency of the proposed optimization method in solving real-time problems, seven sets of problem instances are generated based on the range of real production data. The order quantity is changed from 20 to 80 during the decision period, and the detailed information about these processing orders is given in the Appendix.
[0223] The proposed optimization method is a nonlinear integer programming and Lingo 12.0 is used to solve the global optimal solution. Table 6 shows the results obtained from Lingo and IEA. The deviation of the results between IEA and Lingo is also given in this table.
[0224] Table 6 Performance of implicit algorithms in different problem instances
[0225]
[0226]
[0227] As can be seen from Table 6, Lingo has a long running time compared to the processing time of each product. Such a long running time may delay job release, especially for real-time production lead time optimization. The advantage of IEA is that it can quickly obtain a satisfactory feasible solution for small-scale instances. For example, IEA can find a satisfactory feasible solution |Ω|≤30 in 185 seconds. This result means that IEA can be used to make real-time PLT decisions by: N p ≤6 hours. N of this decision range p Enough to make real-time decisions for production lead time optimization.
[0228] Next, two comparative experiments are set up to verify the performance of the optimization method. One is to optimize the production lead time without reconfiguring the machine, and the other is to optimize with reconfiguring the machine. After finding the best job release plan, these two plans are used to operate the welding line and calculate the corresponding costs C1, C2, C3, and J as shown in Table 7:
[0229] Table 7 Operational costs with and without considering reconfiguration
[0230]
[0231] Experimental results show that real-time production lead time optimization based on reconfigurable systems can significantly reduce the delay penalty cost of orders while reducing the inventory cost of final products and work-in-process.
[0232] Then two classic job start mechanisms are introduced to compare with the two mechanisms in the optimization method of the present invention: one is the job start mechanism based on the bottleneck; the other is the job start mechanism based on the first workstation. Mechanism one releases the job based on the bottleneck workstation, while mechanism two executes this job based on the status of the first workstation. If the first workstation is blocked, the job is stopped; otherwise, the job is started after the previous job is completed.
[0233] The two mechanisms considered in the optimization method of the present invention are: one is an optimization method without in-process work constraints, and the other is an optimization method with in-process work constraints.
[0234] The experimental results are shown in Table 8:
[0235] Table 8 Operating costs of the four mechanisms without interruption
[0236]
[0237] It can be seen that the mechanism three and mechanism four used in the present invention reduce the delay penalty cost, but will also significantly increase the inventory cost. Compared with other mechanisms, the method of the present invention can generate a more appropriate job release plan for the job to respond to these predetermined machine reconfigurations. If there is no unexpected interruption, the additional work-in-process constraints are unnecessary.
[0238] If there is an unexpected interruption, corresponding experiments are also conducted to verify the performance of the method of the present invention by identifying permanent production losses for real-time interruptions. The experimental results are shown in Tables 9 and 10:
[0239] Table 9 Identification of permanent production losses
[0240]
[0241]
[0242] With this identification information, it can determine when to update its job release schedule to minimize expected costs.
[0243] Table 10 Total expected costs of four mechanisms with interruptions
[0244]
[0245] The experimental results show that real-time interruptions increase the expected cost of the weld line. If the interruption does not last too long, mechanism three shows better performance. However, if the interruption lasts for a long time, the higher mechanism will stand out. For example, when and When an interruption occurs, mechanisms 2 and 4 perform better than mechanism 3. In comparison, the method of the present invention still has very good performance.
[0246] Combined with the conclusions of the above case experiments, it can be concluded that the present invention is superior in solving the real-time production lead time problem of reconfigurable systems, providing enterprises with better job release solutions, helping to improve the operating efficiency of enterprise production systems, reduce enterprise operating costs, and enable production systems to have stronger resource coordination and predictive control capabilities.
[0247] The above embodiments are only used to help understand the method and core idea of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.
[0248] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control, characterized in that: The steps include: Step S1: Collecting real-time data of materials and operations and real-time data of workstation status in the production system assembly line to be optimized, and determining parameters required for model calculation, wherein the real-time data of materials and operations at least include operation transfer data, operation location data and processing time; the real-time data of workstation status at least include machine failure event data, operation processing data, workstation reconfiguration time and workstation buffer capacity; Step S2: Considering the predictive multi-stage transition and reconfigurable behavior, a reconfigurable production system model based on max-plus algebra is established. The completion time is predicted by establishing a time-varying transition model, a time-varying state space equation and a state transfer matrix for the time variation of the planned and real-time discrete events between multiple workstations, and various constraints in the real-time planning lead time optimization are handled; Step S3: Considering the environment of the reconfigurable production system, a new optimization method and an optimization model of the real-time production lead time of the reconfigurable system are established to perform predictive control with the goal of minimizing the real-time job release plan of the total expected cost of the work-in-process inventory, the final product inventory and the delay penalty; Step S4: The optimization model is solved using an implicit enumeration algorithm to ultimately obtain a real-time job release plan with a smaller total expected cost.
2. The method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control according to claim 1, characterized in that: In S2, the reconfigurable system described by max-plus algebra further follows the following steps: Step S21: setting the assumption that there is enough buffer before the first workstation or after the last workstation; Step S22: Establish a time-varying transition model of the reconfigurable production system; the specific description is as follows: i The location represents the i-th workstation and the location where the reconstruction occurs, m represents the number of workstations, and p i The position represents the i-th buffer, p′ i (2≤i≤M) where the position controls the amount of real-time work being processed in the buffer, p″ i (1≤i≤M) controls the restart of the corresponding workstation, u(k) represents the start time of the kth job released by the production system, and the symbol p i It represents the real-time number of jobs being performed by workstation i, symbolized by p′ i represents the remaining capacity of the i-th buffer, σ i (k) represents the pure operation time of the kth job of workstation i, represents the machine reconfiguration time before the kth job of workstation i starts, σ′ i (k) represents the start time of the kth job of workstation i, and the calculation formula is: N i represents the capacity of the i-th buffer, the time-varying transition behavior of the reconfigurable production system can be expressed by the following equation: Step S23: According to the time-varying transformation, a time-varying state space equation based on max-plus algebra is established as follows: Among them, x(k) represents the column vector of the workstation status, which is composed of [x1(k),x2(k),...,x M (k)] T y(k) represents the completion time of the kth job, A s (k) represents the relationship between workstation states, B(k) represents the relationship between job start time and workstation state, and C(k) represents the relationship between job completion time and workstation state. The state transfer matrix is as follows: [A s (k)] ij Indicates the buffer capacity N j = s and there is a reconstruction path from j to workstation i, equal to the job start time, otherwise equal to ε; when the workstation has system input or output [B(k)] i1 and [C(k)] 1j is equal to 0, otherwise equal to ε; The above state space equation is converted into the standard state space equation as follows: in, The state transfer matrix is as follows: The matrix and It can be generated by the following formula: Step S24: Switch the state transition matrix to represent the real-time interruption in the reconfigurable production system; define this interruption as It means that when workstation i processes the kth n When there are multiple jobs, the interruption event can last for d n , workstation i for the kth n The processing time of a job can be expressed as σ″ i (k n )=σ′ i (k n )+d n ,The updated state-space equations are then used to represent the real-time production status of the reconfigurable production system.
3. The method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31: Set the assumptions of the model: each order contains only one product, each workstation can only process one job at a time, any job can only be processed on the downstream workstation after it is processed on the current workstation, and interruption of processing is not allowed, if the products of two jobs are the same, reconstruction and reconfiguration are not required, only the first stage needs to set the planning lead time, and the other stages are determined by the upstream and downstream work-in-progress, the jobs are operated according to the first-in-first-out rule, and the buffer capacity before the first stage and after the last stage is sufficient; Step S32: Set the symbols of the known parameters or sets in the model and the decision variables, as follows: Ω represents the current unfinished job set, q j Indicates the number of jobs, a j The earliest start time of job j is the control variable of the model, b j represents the latest start time of job j, M represents the number of workstations, i represents the workstation index, and p ij represents the processing time of job j on workstation i, d j Indicates the delivery date, e j represents the benefit of job j, w 1j represents the delay penalty coefficient, w 2j represents the inventory cost coefficient of the final product, w 3j represents the inventory cost coefficient of the work-in-progress, Φ represents the set of job release sequences, t i (k, k′) represents the reconfiguration time of workstation i processing job k to k′, k0 represents the time step; Step S33: The objective function minimizes the total expected cost, wherein the total expected cost is composed of two parts: the work-in-process inventory cost and the final product inventory and delay penalty cost; the intermediate variable or set sequence at least includes the operation start time sequence, the operation completion time sequence, the work-in-process inventory cost, the final product inventory and delay cost, and the sum of the number of operations in the decision period; step S33 further includes the following steps: Step S331: The published job start time series is represented by μ(k0+k|k0), which is also the decision variable of the model, and the job completion time series is represented by y(k0+k|k0); Step S332: Work in process inventory cost j in The work-in-process inventory cost coefficient w 3j , the deviation between the completion time of the operation and the deviation between the start time of the operation, and the calculation formula is: Step S333: The delay cost of the final product is related to the deviation between the completion time of the operation and the planned delivery time. The inventory cost of the final product is related to the deviation between the actual delivery time and the completion time of the operation between adjacent products. The calculation formula is: Step S334: The sum of the number of operations within the decision period is calculated by express; Step S335: The optimization goal is to minimize the sum of the costs of S332 and S333: minj=j in +j out ; Step S34: Establishing the constraint conditions of the model. Step S34 further includes: Step S341: Demonstrate the relationship between the job release time and its corresponding completion time, with the following constraints: Step S342: Ensure that the job can be released only when the first workstation of the reconfigurable production system is available, with the following constraints: a(k0+k|k0)<μ(k0+k|k0) <b(k0+k|k0),1≤k≤N p The parameters a(k0+k|k0) and b(k0+k|k0) are determined by the previous assignment; Step S343: constrain the job start time series μ(k0+k|k0) to be monotonically increasing, and the constraints are as follows: μ(k0+k|k0)-μ(k0+k-1|k0)≥0.1≤k≤N p Step S344: When N c < is less than k, increase the job release time equally, with the following constraints: μ(k0+k|k0)-μ(k0+k-1|k0)=μ(k0+k+1|k0)-μ(k0+k|k0)N c <k<N p -1 Among them, the symbol N c Represents the control range, the purpose of which is to reduce the number of optimization variables; Step S345: Decision variable constraints, ensuring the non-negativity and integrity of decision variables, the constraints are as follows: μ(k0+k|k0)≥0,μ(k0+k|k0)∈Z,k=1,...,N p Step S346: Additional constraints of real-time WIP, which are generated by the kanban mechanism, the continuous WIP mechanism or the drum buffer rope mechanism, are as follows: Among them, L(k0+k|k0) represents the processing time of the k0+kth job, and F(k0+k|k0) is determined by the allowed number of work-in-progress; Step S35: re-optimization is triggered only when the interruption affects the permanent throughput of the system; Step S351: construct a bottleneck-centered production loss identification method, where production losses are divided into permanent production losses and non-permanent production losses, where the interruption event with the longest interference time will not cause permanent production losses at the end workstation of the production line; Step S3511: If the bottleneck is fixed, then closest to the end of the production line, assuming there may be one or more slowest workstations, production loss identification can be divided into three categories: unexpected interruption The windows of opportunity are as follows: hour, When O(k n )=0; hour, in, It is the last and slowest workstation in the reconfigurable production system; Step S3512: If the bottleneck is shifting, use steady-state analysis to identify production losses. Based on the state space equation of step S23, use the deviation c to determine whether there is a permanent production loss. The calculation formula is as follows: Among them, if c is greater than 0, there is a permanent production loss, otherwise, there is no permanent production loss; Considering the increase in inventory costs caused by work-in-progress, and calculating the system recovery time t, the calculation formula is as follows: Step S352: If there is a permanent production loss or the system recovery time is long, the production lead time of the job must be regenerated to minimize the total expected cost; for this purpose, a new processing parameter switching model must be used, and the state transfer matrix in step S23 is used for event switching. By analyzing the new state space equation, a new state space equation is given for re-optimization, and its parameter k″ is calculated as follows:
4. The method for optimizing the real-time production lead time of a reconfigurable system based on model predictive control according to claim 1, characterized in that: Step S4 specifically includes the following steps: Step S41: The size of the enumeration space obviously affects the search efficiency of the algorithm. First, the enumeration space is generated. Step S41 further includes the following steps: Step S411: Determine each variable μ(k+k0|k0), 1≤k≤N, from the processing time of the first workstation in the reconfigurable production system p The lower bound of is calculated as follows: where μ lower (k0|k0)=μ(k0); Step S412: derive each variable μ(k+k0|k0), 1≤k≤N according to the order due date and the late penalty cost p For each order, the relationship between the latest delivery time and the due date is as follows: The latest delivery time can be further expressed as follows: Considering the processing parameters of order j, each variable μ(k+k0|k0), 1≤k≤N can be calculated p The upper bound of is calculated as follows: Step S42: Determine the enumeration strategy: Based on the analysis of the variable decision space, a variable search step is designed. Step S42 further includes the following steps: Step S421: Set μ(k+k0|k0) equal to its lower bound, specify the search step length δ, determine the initial operating cost f0, and start searching from 1→k; Step S422: Calculate the operating cost f for each search based on the system throughput and then according to the space state equation; Step S423: If f > f0, continue the search in steps; if f = f0, return to step S422; if f < f0, execute step S424; Step S424: Adjust the step size to δ / 2. If the obtained variable is less than the unit production time, execute S425; otherwise, return to step S422; Step S425: If k>N p , stop searching and export function f0 otherwise, return to step S4222.
