Bending process planning method based on suboptimal experience inheritance

Abstract

The invention discloses a bending process planning method based on suboptimal experience inheritance, and belongs to the technical field of intelligent manufacturing, and the method comprises the steps: constructing an N-layer graph model, calculating a fitness value, screening suboptimal experience, inheriting suboptimal experience and random experience, and outputting an approximate optimal bending process. The technical problem of improving the efficiency and reliability of process planning by constructing a multilayer graph model and introducing a dynamic fitness evaluation mechanism is solved, systematic expression and management of complex workpieces are realized, the bending sequence is gradually screened and optimized by adopting a dynamic fitness function, and the efficiency and reliability of process planning are improved. Compared with the prior art, unnecessary search paths and collision detection times are reduced, the risk of falling into local optimum is effectively avoided, understanding globality is improved, the method adapts to sheet metal part planning scenes with different complexity and requirements, the search process is more simplified, and computing resources and time cost are effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent manufacturing, and particularly relates to a bending process planning method based on sub-optimal experience inheritance. Background Art

[0002] Traditional bending process planning usually adopts heuristic search algorithms. Such methods guide the search path through specific rules to quickly generate bending sequences.

[0003] Heuristic search algorithms:

[0004] Use experience rules and heuristic functions to prune the search space and try to find a sub-optimal solution within a limited time; typical methods include genetic algorithms, simulated annealing algorithms, and ant colony optimization algorithms, etc.

[0005] Disadvantages of the prior art:

[0006] Huge search space: With the increase in the complexity of the workpiece geometric structure, the possible sequences of the bending process increase exponentially, resulting in a sharp expansion of the search space and huge consumption of computing resources.

[0007] Low efficiency: Heuristic search algorithms need to perform collision detection and fitness evaluation on each possible bending sequence, which greatly increases the computing time.

[0008] Prone to local optimum: Traditional methods are prone to staying at sub-optimal solutions in complex search spaces and are difficult to further jump out and find the global optimum solution.

[0009] Insufficient generalization ability: For sheet metal parts with variable shapes, fixed heuristic rules are difficult to effectively adapt to, resulting in unsatisfactory planning effects. Summary of the Invention

[0010] The purpose of the present invention is to provide a bending process planning method based on sub-optimal experience inheritance, which solves the technical problem of improving the efficiency and reliability of process planning by constructing a multi-layer graph model and introducing a dynamic fitness evaluation mechanism.

[0011] To achieve the above purpose, the present invention adopts the following technical solutions:

[0012] A bending process planning method based on sub-optimal experience inheritance, comprising the following steps:

[0013] Step 1: The model construction module constructs a graph model with N layers according to the number of bending times N required for the workpiece, and each layer represents a bending process;

[0014] The first layer represents the Nth bending, the second layer represents the (N - 1)th bending, and so on, until the Nth layer represents the first bending;

[0015] Each layer contains several nodes, and each node stores the following information: the current bending sequence number Ind, the position Pos corresponding to the current bending, all possible local bending sequences Process corresponding to the current node, and the fitness value Fit of the local bending sequence;

[0016] The data storage module saves the information of each node, including Ind, Pos, Process, and Fit;

[0017] Step 2: The local bending sequence is input at the data entry of the fitness calculation module. The fitness calculation module calculates the fitness values of all local bending sequences in the nodes layer by layer. The comprehensive factors for calculating the fitness value include the number of flips F, the number of U-turns R, and the number of die changes M;

[0018] The screening module screens out excellent local bending sequences according to the fitness value, and the screening is carried out according to the weighted score of the comprehensive factors for calculating the fitness value;

[0019] Step 3: The inheritance module inherits the sequences that do not include the current bending operation from the excellent local bending sequences screened out from the previous layer to generate new sequences;

[0020] The random introduction module introduces a random inheritance mechanism, selects some sequences whose fitness values do not reach the expected value but are feasible to expand the search space;

[0021] Step 4: The optimal solution evaluation module selects the path with the highest sorted fitness value as the approximate optimal bending process among all the complete bending sequences screened out in the Nth layer. The optimal path meets the conditions including minimizing the number of flips F, the number of U-turns R, and the number of die changes M, and meets the process constraints and processing feasibility requirements.

[0022] Preferably, when performing Step 1, the geometric shape and process information of the workpiece are input at the data entry of the model construction module. The model construction module generates a graph model containing N layers according to the geometric shape and process information: specifically, the nodes of each layer represent the feasible bending operations of the current process, the local bending sequence of each node is initially empty, and the fitness value has not been calculated.

[0023] Preferably, when performing Step 2, the fitness calculation module traverses all the nodes of each layer, calculates the fitness values of their local bending sequences, sorts them according to the size of the fitness values, and screens out the sequences with fitness values greater than the threshold as excellent experiences to be passed to the next layer;

[0024] The fitness calculation formula is as follows:

[0025] Fit = w 1 ×(1 - (F ÷ F max )) + w 2 ×(1 - (R ÷ R max )) + w3 ×(1 -

[0026] (M÷M max ));

[0027] Wherein, w 1 , w 2 , w 3 are weight parameters, Fit is the fitness value, indicating the quality of a certain bending sequence, ranging from 0 to 1, and the larger the value, the better the bending sequence; F max is the theoretically maximum number of flips; R max is the theoretically maximum number of U-turns; M max is the theoretically maximum number of die replacements;

[0028] The fewer the number of flips F, the better; the fewer the number of U-turns R, the better; the fewer the number of die replacements M, the better.

[0029] Preferably, when performing step 3, after the inheritance module generates a new sequence, it conducts a feasibility check on the new sequence, and the check content includes process collision, the number of flips, and the number of U-turns.

[0030] Preferably, when performing step 3, the random introduction module starts from the third layer and randomly introduces some low-fitness sequences when generating a new bending sequence, expanding the search scope and avoiding the local optimum trap.

[0031] A bending process planning method based on sub-optimal experience inheritance according to the present invention solves the technical problem of improving the efficiency and reliability of process planning by constructing a multi-layer graph model and introducing a dynamic fitness evaluation mechanism. The present invention constructs a multi-layer graph model based on the bending sequence, and each layer corresponds to a specific process serial number, realizing the systematic expression and management of complex workpieces. A dynamic fitness function is used to gradually screen and optimize the bending sequence, reducing unnecessary search paths and the number of collision detections. Utilizing the global characteristics of the graph model and the weight dynamic adjustment mechanism, the risk of falling into the local optimum is effectively avoided, improving the globality of the solution. Through the comprehensive evaluation of multiple factors such as the number of flips, the number of U-turns, and the number of die replacements, it adapts to the planning scenarios of sheet metal parts with different complexities and requirements. Through normalization processing and weight allocation, the search process is made more concise, effectively reducing the computational resources and time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is the main flow chart of the present invention;

[0033] Figure 2 is a schematic diagram of the process planning graph model of the present invention;

[0034] Figure 3 is a schematic diagram of the 4-time bending operation of the workpiece of the present invention;

[0035] Figure 4 is the Node of the present invention 43 , Node 44 Feasible folding local bending sequence judgment of Node, schematic diagram showing the feasibility of the 3rd bend at the 4th fold

[0036] Figure 5 is the Node of the present invention 43 , Node 44 Feasible folding local bending sequence judgment of Node, schematic diagram showing the infeasibility of the 4th bend at the 4th fold

[0037] Figure 6 is the Node of the present invention 31 Schematic diagram showing the feasibility of the 1st bend at the 3rd fold and the 2nd bend at the 4th fold in the feasible bending local sequence judgment of Node

[0038] Figure 7 is the Node of the present invention 31 Schematic diagram showing the infeasibility of the 1st bend at the 3rd fold and the 3rd bend at the 4th fold in the feasible bending local sequence judgment of Node

[0039] In the figure: workpiece 1, bending machine 2 Detailed implementation method

[0040] Consisting of Figures 1-7 A bending process planning method based on sub - optimal experience inheritance as shown, includes the following steps

[0041] Step 1: The model construction module constructs a graph model with N layers according to the number of bending times N required for the workpiece. Each layer represents a bending process

[0042] The first layer represents the Nth bending, the second layer represents the (N - 1)th bending, and so on, until the Nth layer represents the 1st bending

[0043] Each layer contains several nodes, and each node stores the following information: the current bending serial number Ind, the position Pos corresponding to the current bending, all possible local bending sequences Process corresponding to the current node, and the fitness value Fit of the local bending sequence

[0044] Input the geometric shape and process information of the workpiece at the data entry of the model construction module. The model construction module generates a graph model with N layers according to the geometric shape and process information: specifically, the nodes of each layer represent the feasible bending operations of the current process. The local bending sequence of each node is initially empty, and the fitness value is not calculated

[0045] Such as Figure 3A specific example of this embodiment is shown. In this example, a total of 4 bending operations are required for workpiece 1, all of which are performed by bending machine 2. The creases generated successively by the 4 bending operations are B1, B2, B3, and B4. In this embodiment, a graph model composed of four layers is constructed. This model is generated based on the fact that the workpiece has four bending operations, and each layer corresponds to a bending serial number. Specifically, the first layer represents the 4th bending, and the fourth layer represents the 1st bending.

[0046] Each layer of nodes Nodeij in the graph model contains four key pieces of information:

[0047] Ind: Represents the current bending serial number (i.e., the i-th bending);

[0048] Pos: Represents the position of the current bending (i.e., the j-th bending);

[0049] Process: Stores k feasible local bending sequences in the current node;

[0050] Fit: Stores the fitness values corresponding to the k feasible bending sequences.

[0051] Therefore, the node Nodeij can clearly describe that it is the i-th bending, the j-th bend, and all feasible local bending sequences and their fitness values in this state.

[0052] The feasible local bending sequences in these nodes are called "experiences". At the same time, each layer of nodes in the graph inherits the experiences of the previous layer, thus forming a systematic structure.

[0053] The data storage module saves the information of each node, including Ind, Pos, Process, and Fit;

[0054] Step 2: The local bending sequence is input at the data entry of the fitness calculation module. The fitness calculation module calculates the fitness values of all local bending sequences in the nodes layer by layer. The comprehensive factors for calculating the fitness values include the number of flips F, the number of turns R, and the number of die replacements M;

[0055] The fitness calculation module traverses all nodes in each layer, calculates the fitness values of their local bending sequences, and sorts them according to the magnitudes of the fitness values. The sequences with fitness values greater than the threshold are screened out and passed to the next layer as excellent experiences;

[0056] The fitness calculation formula is as follows:

[0057] Fit = w 1 ×(1 - (F ÷ F max )) + w 2 ×(1 - (R ÷ R max )) + w 3 ×(1 -

[0058] (M÷M max ));

[0059] Among them, w 1 , w 2 , w 3 are weight parameters, Fit is the fitness value, representing the quality of a certain bending sequence, ranging from 0 to 1, and the larger the value, the better the bending sequence; F max is the theoretically maximum number of flips; R max is the theoretically maximum number of U-turns; M max is the theoretically maximum number of die replacements;

[0060] The screening module screens out excellent local bending sequences according to the fitness value, and the screening is carried out according to the weighted score of comprehensive factors of the fitness value calculation;

[0061] The fewer the number of flips F, the better; the fewer the number of U-turns R, the better; the fewer the number of die replacements M, the better.

[0062] In this embodiment, the search direction of the graph model starts from the last bending (the 4th bending) and is passed forward layer by layer. In the first-layer nodes, all possible local bending sequences need to be determined. For example, at the 4th bending, some positions may be feasible bending points (such as bending No. 3), while other positions are not feasible (such as bending No. 4). In this case, the feasible local bending sequences of the first-layer nodes only contain one element, and all fitness values are initialized to 1, and the fitness threshold is also set to 1. This is to enable the second-layer nodes to inherit all the experiences of the first layer when generated.

[0063] For example, in the nodes Node 41 , Node 42 ,... in the first layer, the fitness values of each feasible bending sequence stored in the field Fit are all 1.

[0064] Starting from the second layer, the fitness of the feasible local bending sequences stored in each node is calculated layer by layer.

[0065] The fitness value is used to measure the quality of the current bending sequence, and it is based on process characteristics such as the number of flips, the number of U-turns, and the number of die replacements. The fewer the occurrences of the above situations, the higher the fitness value, and the better the bending sequence.

[0066] For example, in the nodes Node 31 , Node 32 ,... in the second layer, the field Fit stores the fitness values of each feasible bending sequence.

[0067] After calculating the fitness values of the nodes in the current layer, store the feasible local bending sequences of all nodes in the list Lis and sort them in ascending order of fitness values. Subsequently, according to the preset fitness threshold (e.g., FitThreshold = 0.5), filter out the bending sequences with fitness values greater than or equal to the threshold to form the "sub-optimal experience sequences" P s . These sub-optimal experience sequences will be used as references for generating the nodes in the next layer, thereby improving the search efficiency. For example, the sub-optimal experience sequence may be Second_best_experience = {{1, 2}, {4, 3}}.

[0068] Step 3: The inheritance module inherits the sequences that do not contain the current bending operation from the excellent local bending sequences screened from the previous layer to generate new sequences;

[0069] After generating the new sequences, the inheritance module conducts feasibility detection and verification on the new sequences. The detection contents include process collision, number of flips, and number of U-turns.

[0070] The random introduction module introduces a random inheritance mechanism to select some sequences with fitness values not reaching the expected value but being feasible to expand the search space;

[0071] The random introduction module starts from the third layer. When generating new bending sequences, randomly introduce some low-fitness sequences to expand the search range and avoid local optimum traps.

[0072] In this embodiment, starting from the second layer, each layer of nodes Node ij will inherit the sub-optimal experience of the previous layer, but does not include the bending position j corresponding to the current layer.

[0073] On the basis of inheritance, detect whether the combined bending sequences collide. If there is no collision, classify them as the feasible local bending sequences of the current node.

[0074] For example, at the 3rd bending, the node Node 31 can inherit the sub-optimal experience {2}, {3} of the previous layer. After collision detection, it is determined that only the combination {1, 2} is feasible, and thus store it as the feasible bending sequence of the node Node 31 .

[0075] The introduction of sub-optimal experience ensures that the quality of the bending sequences in each layer is relatively high and reduces the number of collision detections, thereby effectively improving the search efficiency

[0076] To avoid the algorithm falling into a local optimal solution, a random inheritance mechanism is introduced starting from the third layer. The specific operation is as follows: Set a random number rand and a threshold R. The random number is generated within the range of [0, 1]. If the random number is greater than the threshold R, select the bending sequence with a lower fitness value from the previous layer as the reference for the current layer.

[0077] For example, in a certain planning, the sequence {2, 3} with a lower fitness value does not belong to the sub-optimal experience originally, but is adopted by nodes Node 21 and Node 24 due to the random inheritance mechanism. This randomness expands the search space to a certain extent and improves the global search ability of the algorithm.

[0078] Step 4: Among all the complete bending sequences screened by the optimal solution evaluation module in the Nth layer, select the path with the highest sorted fitness value as the approximate optimal bending process. The optimal path satisfies the conditions including minimizing the number of flips F, the number of U-turns R, and the number of die changes M, and meets the process constraints and processing feasibility requirements.

[0079] In this embodiment, after four layers of node search and inheritance, when the algorithm reaches the last layer (i.e., the first bending), traverse the sub-optimal bending processes stored in all nodes, calculate their fitness values and sort them, and finally output the complete bending process with the highest fitness as the approximate optimal solution.

[0080] For example, in the last layer nodes of the graph model, nodes Node 11 and Node 14 store the complete bending sequences, and the calculated optimal bending sequence is {1, 2, 4, 3}.

[0081] This bending process not only meets the requirements of the bending process, but also has high processing efficiency and product quality, and can significantly reduce the number of flips, U-turns, and die changes in the process. The finally output bending path is the approximate optimal solution under the current constraint conditions and can provide effective guidance for actual production.

[0082] A bending process planning method based on sub-optimal experience inheritance according to the present invention solves the technical problem of improving the efficiency and reliability of process planning by constructing a multi-layer graph model and introducing a dynamic fitness evaluation mechanism. The present invention constructs a multi-layer graph model based on the bending sequence, with each layer corresponding to a specific process number, realizing the systematic expression and management of complex workpieces. A dynamic fitness function is used to gradually screen and optimize the bending sequence, reducing unnecessary search paths and the number of collision detections. By utilizing the global characteristics of the graph model and the weight dynamic adjustment mechanism, the risk of falling into local optimality is effectively avoided, improving the globality of the solution. Through comprehensive evaluation of multiple factors such as the number of flips, the number of turnarounds, and the number of die replacements, it adapts to the planning scenarios of sheet metal parts with different complexities and requirements. Through normalization processing and weight allocation, the search process is made more concise, effectively reducing the computational resources and time costs.

Claims

1. A bending process planning method based on suboptimal experience inheritance, characterized by: The steps include: Step 1: The model building module builds a graph model containing N layers according to the number of bending times N that the workpiece needs to complete, and each layer represents a bending process; The first layer represents the Nth bend, the second layer represents the N-1th bend, and so on, until the Nth layer represents the 1st bend; Each layer contains several nodes, each of which stores the following information: the current bending sequence number Ind, the position Pos corresponding to the current bending, all possible local bending sequences Process corresponding to the current node, and the fitness value Fit of the local bending sequence; The data storage module saves the information of each node, including Ind, Pos, Process and Fit; Step 2: The data entry of the fitness calculation module inputs the local bending sequence, and the fitness calculation module calculates the fitness values ​​of all local bending sequences in the node hierarchically. The comprehensive factors for calculating the fitness value include the number of flips F, the number of U-turns R, and the number of mold changes M; The screening module selects excellent local bending sequences according to the fitness value, and the screening is performed based on the weighted score of the comprehensive factors calculated by the fitness value; Step 3: The inheritance module inherits the sequence that does not contain the current bending operation from the excellent local bending sequences screened out in the previous layer to generate a new sequence; The random introduction module introduces a random inheritance mechanism to select some sequences whose fitness values ​​do not reach the expected values ​​but are feasible, so as to expand the search space; Step 4: The optimal solution evaluation module selects the path with the highest fitness value as the approximate optimal bending process from all the complete bending sequences screened out in the Nth layer. The optimal path satisfies the conditions including minimizing the number of flips F, the number of U-turns R and the number of mold changes M, and meets the process constraints and processing feasibility requirements.

2. A bending process planning method based on suboptimal experience inheritance as claimed in claim 1, characterized in that: When executing step 1, the geometric shape and process information of the workpiece are input into the data entry of the model building module. The model building module generates a graph model containing N layers based on the geometric shape and process information: specifically, the nodes of each layer represent the bending operations that are feasible for the current process, the local bending sequence of each node is initially empty, and the fitness value is not calculated.

3. A bending process planning method based on suboptimal experience inheritance as claimed in claim 1, characterized in that: When executing step 2, the fitness calculation module traverses all nodes in each layer, calculates the fitness value of its local bending sequence, sorts them by the size of the fitness value, selects the sequence with a fitness value greater than the threshold, and passes it to the next layer as good experience; The fitness calculation formula is as follows: Fit=w1×(1-(F÷F max ))+w2×(1-(R÷R max ))+w3×(1- (M÷M max )); Among them, w1, w2, w3 are weight parameters, and Fit is the fitness value, which indicates the quality of a certain bending sequence, ranging from 0 to 1. The larger the value, the better the bending sequence. max is the theoretical maximum number of flips; R max is the theoretical maximum number of U-turns; M max is the theoretical maximum number of mold changes; The fewer the number of flipping times F, the better; the fewer the number of U-turns R, the better; and the fewer the number of mold changes M, the better.

4. A bending process planning method based on suboptimal experience inheritance as claimed in claim 1, characterized in that: When executing step 3, after generating a new sequence, the inheritance module performs a feasibility test on the new sequence, and the test content includes process collision, number of flips and number of U-turns.

5. The bending process planning method based on suboptimal experience inheritance as claimed in claim 1, characterized in that: When executing step 3, the random introduction module starts from the third layer. When generating new bending sequences, it randomly introduces some low-fitness sequences to expand the search range and avoid the local optimal trap.

Images