A method for automatically calculating bending of a bending die library

Abstract

This invention discloses an automatic quick calculation method for bending dies using a bending die library, relating to the field of workpiece bending processing technology. For each candidate die, the springback correction amount is calculated based on pre-stored bending compensation lookup table data using a three-dimensional index of die-material-thickness. When the lookup table data is not found, a regression prediction model trained using historical processing data is invoked to calculate the springback correction amount. This invention, through parameterized die records and a three-layer index structure, enables sub-second retrieval of candidate sets that meet geometric and fit conditions from a large-scale die library for input bending parameters. This eliminates the manual search and inefficient traversal required for die library retrieval, avoids the computational bottleneck caused by candidate explosion, significantly shortens die matching time, improves the relevance of candidate dies, reduces operator time costs, and enhances system response speed.

Description

Technical Field

[0001] This invention relates to the field of workpiece bending technology, specifically to a method for automatic quick calculation of bending using a bending die library. Background Technology

[0002] In traditional industries, sheet metal bending process design relies heavily on experience-based judgment. Technicians must manually match suitable upper and lower dies from a vast inventory of dies, estimate springback based on experience, and manually adjust angles, while simultaneously considering machine tool bending force and stroke limitations, as well as die changeover time, to arrange the sequence. This manual process is time-consuming and highly subjective. In production environments with a wide variety of dies and complex combinations of materials and thicknesses, it easily leads to dies mismatch, incorrect bending force estimation, frequent die changes, and rework, thereby reducing production efficiency and increasing costs. Existing automation solutions often focus on single steps, rarely treating die library retrieval, pre-calculation compensation, mechanical calculation, and sequence optimization with die changeover costs as a unified, closed-loop engineering system. There is a lack of comprehensive solutions that can both guarantee online calculation performance and progressively improve compensation accuracy.

[0003] Patent CN114261081B discloses a method for calculating bending trajectory and a bending device. The above patent enables a high yield rate for flexible displays.

[0004] The aforementioned patent can calculate the bending trajectory that prevents damage to the internal circuitry of the flexible circuit board during bending, but the selection and replacement of the mold relies on manual experience, which is time-consuming and labor-intensive.

[0005] To address this, this application proposes a method for automatically calculating bending in a large-scale mold library by retrieving candidate sets that meet geometric and fitting conditions in sub-second time, thus eliminating the need for manual searching and inefficient traversal required for mold library retrieval. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic quick calculation method for bending dies, in order to solve the technical problems mentioned in the background art, such as the time-consuming and subjective nature of manual processes; the tendency for dies to mismatch, incorrect bending force estimation, frequent die changes and rework in production environments with a wide variety of dies and complex combinations of materials and thicknesses, which in turn reduces production efficiency and increases costs.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for automatically calculating bending using a bending die library, comprising the following steps:

[0008] Receive bending characteristic data of the workpiece and capability data of the bending machine tool;

[0009] Based on the bending feature data, the effective bending length and theoretical inner radius of each bend are calculated, and geometric interference detection is performed, including side clearance detection, collision detection and springback space detection.

[0010] Candidate dies are retrieved from a pre-established parametric bending die library according to multi-level indexing rules based on V-shaped opening size, die height, and suitable thickness.

[0011] For each candidate mold, the springback correction amount is calculated based on the pre-stored bending compensation lookup table data of the mold-material-thickness three-dimensional index. When the lookup table data is not available, the regression prediction model trained using historical processing data is called to calculate the springback correction amount.

[0012] Calculate the bending force of the candidate mold under the corresponding bend according to the air bending force formula;

[0013] Candidate molds whose bending force exceeds the maximum bending force of the machine tool or whose geometric interference detection fails are rejected;

[0014] Among the remaining candidate mold combinations, a sequential optimization model with mold change cost is constructed based on mold change time cost and bending processing time. The optimal bending sequence and corresponding mold allocation are calculated using dynamic programming algorithm.

[0015] The output includes bending process data such as bending sequence, selected die number, bending force for each bend, and springback correction angle.

[0016] Preferably, the multi-level indexing rules include:

[0017] The first layer is a hash index based on the size of the V-shaped opening;

[0018] The second layer is a range tree index based on the mold height and the machine tool installation height;

[0019] The third layer is bitmask filtering based on the thickness range.

[0020] Preferably, the bending characteristic data includes bending angle, bending length, material type, material thickness, flange length, and target inner radius, and the machine tool capability data includes maximum bending force, maximum stroke, and number of installable molds;

[0021] The pre-stored bending compensation lookup data in the parametric bending die library is generated by fusing finite element simulation data and actual machining data, and is stored in a three-dimensional mesh. The three-dimensional indexes are material type, material thickness and bending angle, respectively.

[0022] Each die in the parametric bending die library includes V-shaped opening size, tip radius, die height, range of compatible materials, compatible thickness, and maximum load capacity.

[0023] Preferably, the regression prediction model is a gradient boosting regression model or a multilayer perceptron regression model. The model inputs include material yield strength, material thickness, V-shaped opening size, bending angle, effective bending length, and die height. The model output is the springback correction angle value.

[0024] Preferably, the formula for the air bending force is as follows:

[0025]

[0026] C is the die-material coefficient, σ is the material yield strength, t is the material thickness, L is the effective bending length, and V is the V-shaped opening size of the die;

[0027] C is obtained by calibrating the ratio of the measured bending force in the mold library to the calculated value by the formula, and is stored separately under different material and V-shaped opening size conditions.

[0028] Preferably, the sequential optimization model minimizes

[0029]

[0030] Let T be the objective function, where T bend (i) represents the processing time for the i-th bend, T change (j) represents the time required for the j-th mold change, with constraints including the machine tool's maximum bending force and maximum stroke.

[0031] Preferably, the geometric interference detection is performed by establishing a three-dimensional bending model of the workpiece and using a collision detection algorithm in the intermediate bending state to determine the interference between adjacent flanges and molds or machine tool components.

[0032] Preferably, when calling the regression prediction model, the prediction confidence value is also output. When the prediction confidence value is lower than a preset threshold, the fast finite element simulation module is triggered to recalculate the rebound correction amount.

[0033] Preferably, the bending process data is formatted before output to generate a process file containing machine tool control codes, a mold number table, and angle adjustment values ​​for each bend.

[0034] Preferably, after the actual processing is completed, the method collects the difference between the actual bending angle and the predicted bending angle fed back by the machine tool sensor, and writes the difference and the corresponding mold, material and thickness data into the mold library to update the pre-calculated bending compensation lookup data and regression prediction model.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] 1. This invention achieves sub-second retrieval of candidate sets that meet geometric and fitting conditions from a large-scale mold library by parameterized mold records and a three-layer index structure. This eliminates the manual search and inefficient traversal required for mold library retrieval, avoids the computational bottleneck caused by candidate explosion, significantly shortens mold matching time, improves the relevance of candidate molds, reduces operator time costs, and improves system response speed.

[0037] 2. This invention uses a pre-calculated mold-material-thickness compensation lookup table and lookup priority strategy to provide a high-confidence springback correction value for angle adjustment when the conditions are met, without the need for online simulation. This makes up for the contradiction between online calculation accuracy and real-time performance, avoids calling high-overhead simulations or complex models every time, which would cause delays, and achieves high-precision, low-latency compensation calculations under most common combinations, reducing the number of on-site trial folds and rework rate.

[0038] 3. This invention uses a hybrid quick calculation strategy of table lookup + regression model and a confidence triggering mechanism to quickly provide compensatory predictions even when table lookup is unavailable or outside the boundary. At the same time, it adopts an adaptive refinement strategy for uncertain predictions, which solves the problems of unreliable extrapolation of sparse table lookup and inaccuracy of single models on unseen data. It avoids rework due to uncertain predictions, improves the robustness and reliability of online predictions, and reduces on-site adjustments and quality risks caused by prediction errors.

[0039] 4. This invention generates a bending sequence and mold allocation that balances limited machine tool capacity and minimizes the number of mold changes through hierarchical sequence optimization with mold change costs and on-site closed-loop self-learning. It continuously improves prediction accuracy through on-site feedback, while coordinating mold change frequency, machine tool capacity constraints, and processing time. This avoids the overall efficiency decline caused by pursuing unidirectional indicators and solves the problem of long-term error accumulation due to static models. It reduces the number of mold changes and total processing time, improves equipment utilization, and as online closed-loop learning accumulates, the compensation prediction and sequence selection become more in line with the actual on-site situation, thereby reducing rework rate and improving long-term production stability. Detailed Implementation

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] One embodiment of the present invention provides a method for automatically calculating bending using a bending die library, comprising the following steps:

[0042] Receive bending characteristic data of the workpiece and capability data of the bending machine tool;

[0043] Based on the bending feature data, the effective bending length and theoretical inner radius of each bend are calculated, and geometric interference detection is performed, including side clearance detection, collision detection and springback space detection.

[0044] Candidate dies are retrieved from a pre-established parametric bending die library according to multi-level indexing rules based on V-shaped opening size, die height, and suitable thickness.

[0045] For each candidate mold, the springback correction amount is calculated based on the pre-stored bending compensation lookup table data of the mold-material-thickness three-dimensional index. When the lookup table data is not available, the regression prediction model trained using historical processing data is called to calculate the springback correction amount.

[0046] Calculate the bending force of the candidate mold under the corresponding bend according to the air bending force formula;

[0047] Candidate molds whose bending force exceeds the maximum bending force of the machine tool or whose geometric interference detection fails are rejected;

[0048] Among the remaining candidate mold combinations, a sequential optimization model with mold change cost is constructed based on mold change time cost and bending processing time. The optimal bending sequence and corresponding mold allocation are calculated using dynamic programming algorithm.

[0049] The output includes bending process data such as bending sequence, selected mold number, bending force for each bend, and springback correction angle;

[0050] The fully integrated process is as follows:

[0051] The system receives bending feature data (bending position, target bend angle, flange length, and bend length annotation for each bend) from CAD or unfolded drawings, along with material type and thickness information from the BOM; it also receives machine tool capability data (maximum bending force, maximum stroke, number of available die stations, and die change time); and creates an internal bending parameter object for each input bend (including bend angle θ, bend length L, thickness t, and desired inner radius R). target (e.g., edge markings), and perform initial screening of the bending sequence on the overall unfolded drawing of the workpiece (according to topological order or user-specified order);

[0052] For each bend, the effective bend length, theoretical inner radius, and intermediate workpiece state (represented as several discrete intermediate postures) are calculated using a parametric geometric model. Collision detection is performed on each intermediate posture to determine whether the side clearance and springback space are satisfied (if not satisfied, it is marked as "geometric failure" in the candidate generation).

[0053] A multi-level index is used to retrieve a set of candidate molds that meet the range of V-shaped opening, mold height, and thickness in the mold library. Each candidate mold is loaded with its metadata (V-shaped opening, tip radius, mold height, compatible material label, maximum load capacity, replacement time, etc.).

[0054] For candidate molds, the expected compensation value (springback correction angle Δθ and bending allowance) is first searched in the mold-material-thickness 3D index. If the table exists, it is directly read and Δθ is obtained according to the interpolation strategy. If the table does not match or the name has low confidence, the regression prediction model is called to estimate Δθ and the confidence index of the model output is obtained at the same time. The required bending force F is calculated according to the air bending force analytical formula combined with the mold coefficient C and the material yield strength, and then compared with the machine tool F. max Compare and eliminate infeasible candidates;

[0055] The candidate die set for each bend and the die change time are mapped into a sequence optimization problem with die change costs. A two-stage solution is adopted: first, the die is segmented and clustered based on die compatibility, and then dynamic programming or heuristic local search is run within each segment to obtain a near-optimal solution. The final bending sequence and die allocation for each bend are output, and the bending force, correction angle and die change time point for each step are generated.

[0056] The generated bending process data is formatted into a process package (including: bending sequence list, die number table, angle correction value and calculated bending force for each bend, and die change operation prompts); if it needs to be sent to the machine tool controller, the corresponding process file is generated according to the instruction format supported by the machine tool and the die clamping position information is attached.

[0057] One embodiment of the present invention: a method for automatic quick calculation of bending using a bending die library, with the following index structure design:

[0058] First-level index: Discretize and bucket based on the V-shaped opening size and build a hash index. The bucketing rules include both common multiples and custom ranges.

[0059] The second-level index is a range tree built based on the mold height and the machine tool installation height, which is used to quickly determine whether there is vertical interference or height mismatch of the mold on the machine tool;

[0060] The third-level index uses bitmasks to mark common thickness levels for the mold's applicable thickness range, making it easy to filter quickly using bitwise operations;

[0061] The system first queries the first-level hash index to obtain the initial candidate set based on the calculated recommended V opening range; then it performs interval tree filtering based on the mold height and the current machine tool installation height to exclude height non-compliant items; finally, it performs bitwise filtering based on the thickness bitmask to quickly remove molds with mismatched thickness; the remaining set is returned as the final candidate set for scoring and calculation.

[0062] The mold library supports adding, measuring, and deleting molds. Index updates are performed incrementally (only affected buckets, intervals, and bitmasks are rebuilt). For each change in mold parameters, the pre-calculated compensation table associated with that mold is triggered to undergo incremental recalculation or marked as requiring retesting.

[0063] One embodiment of the present invention: a method for automatically calculating bending using a bending die library;

[0064] Pre-calculation of table lookup generation and organization:

[0065] Offline, high-density sampling of typical molds and material combinations is performed using finite element simulation and factory measurement data to generate a compensation dataset for each mold ID. The data is organized in a three-dimensional network (material type × thickness × bending angle). The online system obtains compensation values ​​for query angles and thicknesses through multilinear interpolation or shape-preserving interpolation strategies and saves the hit confidence of the query.

[0066] Regression prediction model settings:

[0067] The regression model employs a lightweight structure (such as a gradient boosting regression tree or a small multilayer perceptron) to respond to online queries with low latency. The model input fields include material yield strength identifier, thickness t, V opening, bending angle θ, effective bending length L, and die height. The output is the springback correction angle Δθ and the model confidence score. The model training data consists of simulation data and historical actual processing data. The training process includes data cleaning, outlier removal, and cross-validation to avoid overfitting.

[0068] The logic for switching between table lookup and regression:

[0069] Prioritize using lookup table data: When a lookup table exists but the interpolation confidence level is below the threshold or the extrapolation rate in the lookup table exceeds the threshold, the system simultaneously calls the regression model to make a prediction and compares the differences between the two. If the confidence level of the regression model output is higher than that of the lookup table, the regression result is adopted and the query is included in the traceability log for subsequent offline correction. If the difference between the two exceeds the preset tolerance and both confidence levels are low, a fast finite element simulation is triggered to obtain a more accurate compensation value. The simulation results can be written back to the lookup table or labeled as measured data.

[0070] One embodiment of this invention provides a method for automatically calculating bending dies using a bending die library, applicable to sequence optimization and cost calculation with die change functionality:

[0071] Problem modeling:

[0072] Each bend is considered a work node, and each node has several candidate mold options. If different molds are selected between nodes and the molds are in different positions on the machine tool, a mold change action is generated. The objective function is the total processing time (the sum of bending times + the sum of mold change times), and the constraints include the maximum bending force of the machine tool, the maximum stroke, and the geometric constraints of a specific bend.

[0073] Solution strategy:

[0074] First layer: Mold compatibility clustering: Based on the compatibility of candidate molds (molds are the same or can be placed in the same clamping position, mold height is similar, etc.), the continuous bending segments are clustered to generate several compatible segments;

[0075] Second layer: Intra-segment optimization: Dynamic programming algorithm is used to solve the mold allocation and local sequence within each compatible segment. The DP state is indexed by the bending sequence number and the current mold state. The transition considers the mold changing action and processing time cost.

[0076] The third layer: Inter-segment adjustment: Greedy scheduling combined with local search (e.g., swapping the order of two bends or mold allocation) is used to make 2-opt or kopt local improvements between segments until the time budget or iteration threshold is met;

[0077] Special constraint handling:

[0078] If there is only one candidate die for a certain bend and the die conflicts with the die change in the adjacent steps, the optimizer will prioritize moving the adjacent bend to the compatible section to avoid repeated die changes without violating geometric constraints. If the die change time accounts for too high a percentage, the optimizer may suggest using a general die (a die that appears with a lower score in the candidate set) to reduce the number of die changes, while indicating the specific general die to be selected and suggesting the clamping position in the process package.

[0079] One embodiment of this invention provides: a method for automatically calculating bending dies using a bending die library, involving online closed-loop learning and on-site feedback updates.

[0080] On-site feedback collection:

[0081] During the machining execution phase, the system collects the actual angle, actual bending force curve, and mold change execution time for each bend from the machine tool's angle sensor (angle encoder), force sensor, and position sensor; these data are written to the feedback memory in a transactional manner after each bend is completed.

[0082] Feedback processing and update strategy:

[0083] The system calculates the error sequence between the actual angles collected and the system-predicted angles, and groups and statistically analyzes them by mold ID, material type, and thickness (e.g., recording average error and fluctuation index). When the error of a certain mold, material, or thickness combination exceeds a preset threshold, the system triggers one or more of three actions: updating the offset in the pre-calculated lookup table, marking the combination as needing to be re-measured and arranging it in the offline high-precision simulation queue, and using the data as an online fine-tuning training set for incremental training of the regression model.

[0084] Model security and version control:

[0085] The online fine-tuning of the regression model was completed in a low-risk isolated environment and validated on historical data through A / B testing. After successful validation, the production model was replaced. Both table lookups and model updates included version numbers and rollback mechanisms to ensure that if anomalies were introduced in the new version, the system could be quickly restored to the previous stable version.

[0086] One embodiment of this invention provides a method for automatic quick calculation of bending using a bending die library, applicable to compensation and die selection in segmented bending and special process situations:

[0087] Segmented bending scene recognition:

[0088] For long bends or bends where the bending force exceeds the machine tool's capacity, the system identifies the bend as "requiring segmented bending" or "requiring auxiliary fixtures" by detecting the ratio of the calculated bending force to the machine tool's Fmax and geometric constraints.

[0089] Segmented construction strategy:

[0090] If the bending is determined to be segmented, the system decomposes the original single bending into several continuous sub-bendings. The candidate molds for each sub-bending are provided by the mold library that is suitable for short bending or with auxiliary support. The system calculates a separate springback correction for each step after segmentation and takes into account the mold switching cost within the segment and the overall processing time during the sequential optimization stage to ensure that the mold changing arrangement under the segmented combination is feasible and the mold position can be realized.

[0091] Special molds and auxiliary tools description:

[0092] For situations requiring the use of auxiliary pads, supports, or multi-segment fixtures, the mold library records the assembly method, fixture position requirements, and installation time of such auxiliary tools. Sequence optimization takes into account the installation and disassembly time of these auxiliary tools to calculate the overall processing time. The process package clearly specifies the fixture installation position, recommended clamping force range, and cumulative angle requirements after segmentation for each segmented bending, which facilitates on-site implementation by operators.

[0093] Working principle: The system first obtains the unfolded diagram and bending features of the workpiece, as well as the machine tool capability parameters. For each bend, a three-dimensional geometric model is constructed on the discrete intermediate posture. Side clearance, collision and springback space detection are performed. The effective bending length, theoretical inner radius and edge and center position information of each bend are extracted as geometric constraint inputs for candidate mold matching and mechanical calculation.

[0094] Based on the parameterized mold library, multi-level indexes are used to quickly retrieve candidate molds. For each candidate mold, the discrete pre-calculated mold-material-thickness compensation lookup table is retrieved first to obtain the springback correction amount. If the lookup table is unavailable or the confidence level is low, a lightweight regression model trained with historical processing and simulation data is called to predict the springback correction. The required bending force is estimated based on the analytical bending force formula combined with the mold-material coefficient for feasibility screening.

[0095] The candidate mold allocation and bending sequence are modeled as a sequence optimization problem with mold change cost and machine tool constraints. A hierarchical solution is used to generate a bending sequence that minimizes the total processing time. The process package is output and actual angle and mechanical feedback are collected on site. The error is written back to the mold library and model. It supports table lookup offset update and incremental training of regression model to form an online closed-loop adaptive mechanism.

[0096] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A method for automatically calculating bending dies using a bending die library, characterized in that: Includes the following steps: Receive bending characteristic data of the workpiece and capability data of the bending machine tool; Based on the bending feature data, the effective bending length and theoretical inner radius of each bend are calculated, and geometric interference detection is performed, including side clearance detection, collision detection and springback space detection. Candidate dies are retrieved from a pre-established parametric bending die library according to multi-level indexing rules based on V-shaped opening size, die height, and suitable thickness. For each candidate mold, the springback correction amount is calculated based on the pre-stored bending compensation lookup table data of the mold-material-thickness three-dimensional index. When the lookup table data is not available, the regression prediction model trained using historical processing data is called to calculate the springback correction amount. Calculate the bending force of the candidate mold under the corresponding bend according to the air bending force formula; Candidate molds whose bending force exceeds the maximum bending force of the machine tool or whose geometric interference detection fails are rejected; Among the remaining candidate mold combinations, a sequential optimization model with mold change cost is constructed based on mold change time cost and bending processing time. The optimal bending sequence and corresponding mold allocation are calculated using dynamic programming algorithm. The output includes bending process data such as bending sequence, selected die number, bending force for each bend, and springback correction angle.

2. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The multi-level indexing rules include: The first layer is a hash index based on the size of the V-shaped opening; The second layer is a range tree index based on the mold height and the machine tool installation height; The third layer is bitmask filtering based on the thickness range.

3. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The bending characteristic data includes bending angle, bending length, material type, material thickness, flange length, and target inner radius; the machine tool capability data includes maximum bending force, maximum stroke, and the number of dies that can be installed. The pre-stored bending compensation lookup data in the parametric bending die library is generated by fusing finite element simulation data and actual machining data, and is stored in a three-dimensional mesh. The three-dimensional indexes are material type, material thickness and bending angle, respectively. Each die in the parametric bending die library includes V-shaped opening size, tip radius, die height, range of compatible materials, compatible thickness, and maximum load capacity.

4. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The regression prediction model is either a gradient boosting regression model or a multilayer perceptron regression model. The model inputs include material yield strength, material thickness, V-shaped opening size, bending angle, effective bending length, and die height. The model output is the springback correction angle value.

5. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The formula for air bending force is: C is the die-material coefficient, σ is the material yield strength, t is the material thickness, L is the effective bending length, and V is the V-shaped opening size of the die; C is obtained by calibrating the ratio of the measured bending force in the mold library to the calculated value by the formula, and is stored separately under different material and V-shaped opening size conditions.

6. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The sequential optimization model aims to minimize Let T be the objective function, where T bend (i) represents the processing time for the i-th bend, T change (j) represents the time required for the j-th mold change, with constraints including the machine tool's maximum bending force and maximum stroke.

7. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The geometric interference detection is achieved by establishing a three-dimensional bending model of the workpiece and using a collision detection algorithm in the intermediate bending state to determine the interference between adjacent flanges and molds or machine tool components.

8. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: When the regression prediction model is invoked, the prediction confidence value is also output. When the prediction confidence value is lower than the preset threshold, the fast finite element simulation module is triggered to recalculate the springback correction amount.

9. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The bending process data is formatted before output to generate a process file containing machine tool control codes, a mold number table, and angle adjustment values ​​for each bend.

10. The method for automatic quick calculation of bending using a bending die library according to claim 1, characterized in that: The method collects the difference between the actual bending angle and the predicted bending angle fed back by the machine tool sensor after the actual processing is completed, and writes the difference and the corresponding mold, material and thickness data into the mold library to update the pre-calculated bending compensation lookup data and regression prediction model.

Images