Discharging optimization method for anti-deformation cutting pieces of machine-washable western-style clothes

Abstract

The invention relates to the technical field of intelligent discharging of machine-washed western-style clothes, and particularly discloses a machine-washed western-style clothes anti-deformation cutting piece discharging optimization method, which comprises the following steps: in response to a discharging task request, collecting anisotropic mechanical parameters of machine-washed western-style clothes fabric, cutting equipment parameters and fabric batch difference data, carrying out standardization processing on the collected data to obtain a multi-dimensional state input vector; and inputting the multi-dimensional state input vector into a pre-trained deep Q network model, and generating an initial arrangement layout scheme by using the model according to a multi-target dynamic weight award function. According to the method, the deep Q network model containing the anisotropic mechanical parameters and the production environment data is constructed, so that the conflict that high utilization rate and low deformation risk of the machine-washed western-style clothes in the discharging stage are difficult to consider at the same time is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent pattern layout technology for machine-washable suits, and in particular to an optimized method for pattern layout of anti-deformation cut pieces for machine-washable suits. Background Technology

[0002] With the rise of machine-washable suits as a niche product, the market has placed extremely high demands on the dimensional stability and shape retention of ready-made suits after multiple washes. This has directly changed the priority logic of traditional garment nesting. In traditional suit production, the core objective of nesting is usually set solely to maximize fabric utilization. Algorithms often treat the fabric as an isotropic homogeneous material or simply follow warp constraints.

[0003] However, machine-washable suit fabrics typically exhibit significant anisotropic mechanical characteristics, such as substantial differences in warp and weft elastic modulus and Poisson's ratio, and fluctuations in shrinkage performance between different batches. If traditional pattern layout logic is followed, placing cut pieces in directions where the fabric's mechanical properties are unstable in pursuit of high utilization, or forcibly arranging them to meet stringent personalized customization requirements, such as matching stripes or checks, or adhering to specific texture directions, can easily lead to irreversible deformations such as twisting, shrinkage, or bulging during machine washing due to uneven stress release.

[0004] A more prominent problem is that there is an inherent contradiction between improving fabric utilization and ensuring garment deformation resistance. This contradiction, when combined with the introduction of variable customization requirements and unstable fabric batch differences, can easily lead to the nesting algorithm getting stuck in a multi-objective optimization deadlock.

[0005] Specifically, when algorithms attempt to meet the mechanical constraints of machine-washable deformation resistance, they often lead to a significant decrease in fabric utilization, resulting in unacceptable cost waste. Conversely, when algorithms attempt to improve utilization or meet customized geometric constraints by adjusting the position of cut pieces, they are prone to causing cut pieces to fall into high-deformation-risk zones or causing topological fragmentation of the remaining space, making it impossible to place the remaining standard components. Existing nesting methods lack an adaptive decision-making mechanism that can simultaneously quantify mechanical deformation risk and geometric nesting efficiency, and dynamically break this deadlock state based on batch fluctuations. Summary of the Invention

[0006] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the purpose of this invention is to propose an optimized method for the layout of anti-deformation cut pieces for machine-washable suits, thereby resolving the conflict between achieving high utilization and low deformation risk in the layout stage of machine-washable suits.

[0007] To achieve the above objectives, a first aspect of the present invention proposes a method for optimizing the layout of machine-washable suit anti-deformation cut pieces, applied in an intelligent layout calculation device, comprising:

[0008] In response to the material layout task request, the anisotropic mechanical parameters, cutting equipment parameters, and batch difference data of the machine-washed suit fabric are collected, and the collected data are standardized and processed into a multi-dimensional state input vector.

[0009] The multi-dimensional state input vector is input into a pre-trained deep Q-network model, and the model is used to generate an initial material layout scheme based on a multi-objective dynamic weight reward function.

[0010] Local conflict area detection is performed on the initial layout scheme. In response to the detection of conflict areas that do not meet the anisotropic deformation constraints, a local subdivision optimization process is initiated based on the equipment adaptation coefficient in the cutting equipment parameters and the batch difference coefficient in the fabric batch difference data to determine the target layout data stream.

[0011] Generate control instructions containing the target material feed data stream for the cutting equipment to perform the material feed operation;

[0012] The local subdivision optimization process includes: identifying areas in the initial fabric layout scheme where the predicted deformation exceeds a preset deformation threshold, and dynamically adjusting the fabric layout direction and allowance of the cut pieces within the area based on the shrinkage prediction index deviation of the current fabric batch.

[0013] To achieve the above objectives, a second aspect of the present invention provides a machine-washable suit anti-deformation cut piece layout optimization system, comprising:

[0014] The data acquisition and standardization processing module is used to respond to the material layout task request, collect the anisotropic mechanical parameters of machine-washed suit fabric, cutting equipment parameters, and fabric batch difference data, and standardize the collected data into a multi-dimensional state input vector.

[0015] The deep network decision module is used to input the multi-dimensional state input vector into a pre-trained deep Q-network model and use the model to generate an initial material layout scheme according to a multi-objective dynamic weight reward function.

[0016] The local subdivision optimization module is used to perform local conflict area detection on the initial fabric layout scheme. In response to the detection of conflict areas that do not meet the anisotropic deformation constraints, the local subdivision optimization process is started based on the equipment adaptation coefficient in the cutting equipment parameters and the batch difference coefficient in the fabric batch difference data to determine the target fabric layout data stream.

[0017] The instruction generation module is used to generate control instructions containing the target layout data stream for the cutting equipment to perform layout operations; wherein, when the local subdivision optimization module executes the local subdivision optimization process, it is specifically configured to: identify areas in the initial layout scheme where the predicted deformation exceeds a preset deformation threshold, and dynamically adjust the layout direction and allowance of the cut pieces in the area according to the shrinkage prediction index deviation of the current fabric batch.

[0018] To achieve the above objectives, a third aspect of the present invention provides a computing device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described method for optimizing the layout of anti-deformation cut pieces for machine-washable suits.

[0019] An embodiment of the present invention provides a method for optimizing the layout of machine-washable suit anti-deformation cut pieces. By constructing a deep Q-network model that includes anisotropic mechanical parameters and production environment data, the method resolves the conflict between balancing high utilization rate and low deformation risk in the layout stage of machine-washable suits.

[0020] Specifically: First, through a multi-objective dynamic weight reward mechanism, the model can prioritize learning the mechanical properties of the fabric in the early stages of training to ensure its resistance to deformation, and then adaptively optimize its utilization rate in the later stages, thereby maximizing economic benefits while ensuring machine washing stability. Second, to address the deadlock problem caused by customized needs and fabric batch fluctuations, this solution introduces a reward reshaping mechanism based on the washing stress potential energy field and a spatial aggregation strategy based on anisotropic sliding windows. This can actively guide the cut pieces to slide towards the mechanically most stable region or fine-tune and aggregate them in the low deformation direction, effectively eliminating the topological fragmentation phenomenon in the material layout process and ensuring the feasibility of the solution. Finally, through a closed-loop data feedback throughout the entire process, the system can continuously adapt to the shrinkage characteristics of different batches of fabric and equipment differences, realizing the continuous evolution and precise adaptation of the material layout strategy. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the optimized method for layout of machine-washable suit anti-deformation cutting pieces provided by the present invention.

[0022] Figure 2 This is a schematic diagram of the response surface of the equipment adaptation coefficient as a function of equipment precision and effective workbench area in the optimized layout method for anti-deformation cutting pieces of machine-washable suits provided by the present invention.

[0023] Figure 3 This is a dynamic trend diagram of the weight coefficients of each objective in the two-stage training strategy of the machine-washable suit anti-deformation cutting piece layout optimization method provided by the present invention, as a function of the number of iterations.

[0024] Figure 4This is a simulation diagram of deadlock state identification and parameter feature monitoring during model training in the machine-washable suit anti-deformation cutting piece layout optimization method provided by the present invention;

[0025] Figure 5 This is a schematic diagram of the washing stress potential energy field distribution (potential energy field guiding model) after introducing an anisotropic correction factor in the optimization method for anti-deformation cutting pieces of machine-washable suits provided by the present invention.

[0026] Figure 6 This is a simulation comparison of the model's ability to escape local extrema before and after the intervention of the reward reshaping mechanism in the machine-washable suit anti-deformation cutting piece layout optimization method provided by this invention;

[0027] Figure 7 This is a schematic diagram of the change in the aggregation degree index of the remaining space within the anisotropic sliding window and the optimization of the anchor point in the method for optimizing the layout of machine-washable suit anti-deformation cutting pieces provided by the present invention.

[0028] Figure 8 This is a schematic diagram illustrating the implementation of the machine-washable suit anti-deformation cutting piece layout optimization system provided by the present invention;

[0029] Figure 9 This is a schematic diagram of the structure of the computing device provided by the present invention. Detailed Implementation

[0030] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0031] The following description, with reference to the accompanying drawings, outlines an embodiment of the method, system, and computing device for optimizing the layout of machine-washable suit anti-deformation cut pieces.

[0032] Example 1:

[0033] This embodiment details a method for optimizing the layout of machine-washable suit fabric to prevent deformation. This method is specifically configured and applied to an intelligent layout computing device with high-performance parallel computing capabilities. This intelligent layout computing device can be an industrial-grade server containing a GPU cluster, or a distributed computing node deployed in the cloud. Its core capability lies in its ability to run complex deep reinforcement learning algorithms. For example... Figure 1 As shown, the method in this embodiment specifically includes the following steps:

[0034] S1: Acquisition of multi-source heterogeneous data and construction of standardized state vectors.

[0035] Specifically, when the production planning management system issues a new production instruction, the intelligent nesting calculation device first responds to the nesting task request and initiates the data acquisition process. This nesting task request includes not only the quantity of suits to be produced but also specific fabric codes, equipment specifications, and the customer's personalized customization ID. To achieve high-precision, deformation-resistant nesting, the system needs to collect data from three core dimensions: the anisotropic mechanical parameters of the machine-washable suit fabric, the parameters of the cutting equipment at the production end, and fabric batch variation data.

[0036] For example, the system acquires anisotropic mechanical parameters of machine-washable suit fabric by connecting to a precision fabric mechanical testing instrument. Unlike traditional pattern layout which only focuses on the length and width dimensions of the fabric, in this embodiment, the anisotropic mechanical parameters must include at least the warp modulus of elasticity, weft modulus of elasticity, warp Poisson's ratio, and weft Poisson's ratio. This is because the forces experienced by machine-washed suits in the washing drum are multidirectional and complex, and the deformation behaviors of the warp and weft directions are usually quite different. The warp modulus of elasticity characterizes the fabric's ability to resist tensile deformation along the warp direction, while the weft modulus of elasticity characterizes its resistance to deformation along the weft direction. The warp and weft Poisson's ratios describe the ratios of lateral shrinkage of the fabric under tension. After reading these physical quantities, the system uses them as the basic physical boundary conditions for assessing whether the cut pieces will undergo twisting deformation during washing.

[0037] It is important to note that collecting the parameters of the cutting equipment at the production end is crucial for subsequent layout design, as different equipment determines the physical limitations of the layout. In this embodiment, the parameters of the production-end cutting equipment include equipment type, cutting accuracy, worktable size, and maximum cutting width. The equipment type can be categorized as a laser cutting machine, a vibrating knife cutting machine, or a straight knife cutting machine, each with different safety clearances. Cutting accuracy determines the minimum allowable distance between cut pieces, while the worktable size and maximum cutting width define the physical boundaries of the layout diagram.

[0038] In particular, to quantify the impact of equipment performance on the material routing strategy, this embodiment introduces a key dimensionless parameter in the step of standardizing the collected data into a multi-dimensional state input vector: calculating the equipment adaptation coefficient of the production-end cutting equipment. This coefficient aims to balance the combined impact of equipment accuracy and worktable area on material discharge efficiency. The equipment adaptability coefficient... The calculation formula is as follows:

[0039] ;

[0040] In the formula, the equipment accuracy reference value is a preset normalization constant, for example, set to 0.1 mm, which is used as a reference anchor point for standard accuracy; the standard workbench area is also a preset normalization constant, for example, set to the typical area of ​​a standard single-person single-measurement cutting bed.

[0041] This formula shows that when the actual equipment accuracy value is smaller (i.e., higher accuracy) or the actual effective area of ​​the worktable is larger, the calculated equipment adaptability coefficient will be higher. A larger value indicates a stronger capacity for handling complex material discharges. The subsequent deep Q-network model is configured based on the device adaptation coefficient. The numerical range allows for dynamic adjustment of the spacing between cut pieces. For example, when... When the value is high, the model will learn to appropriately reduce the spacing between the cut pieces to improve utilization; conversely, it will increase the spacing to ensure cutting safety.

[0042] like Figure 2 The figure shows a three-dimensional response surface plot of the equipment adaptation coefficient and equipment parameters. This plot intuitively reveals the quantitative evaluation logic of the intelligent material feeding system for the performance of different production equipment through visualization.

[0043] Figure 2 The horizontal and vertical axes represent the actual equipment accuracy and the actual effective area of ​​the workbench, respectively. The vertical axis represents the calculated equipment adaptation coefficient. The color bar legend on the right side of the graph uses a gradient from cool to warm colors to map the change of the coefficient value from low to high.

[0044] from Figure 2 The waveform transformation trend shown clearly shows that as the actual equipment precision value decreases, that is, the equipment processing precision continuously improves, the surface exhibits a significant hyperbolic steep upward shape, indicating that the high precision characteristics have a decisive gain effect on improving the adaptation coefficient; at the same time, with the increase of the actual effective area of ​​the worktable, the surface shows a stable linear climbing trend along another dimension, reflecting the carrying advantage of the large-format worktable for complex material feeding tasks.

[0045] Figure 2The red peaks correspond to high-end intelligent cutting equipment with both extremely high precision and ultra-wide width, while the blue troughs correspond to basic equipment with limited performance. This non-linear surface distribution characteristic has clear technical guiding significance in the embodiment. The deep Q-network model dynamically adjusts the nesting strategy based on the coefficients fed back by this surface. When the input state falls in the red high-coefficient combination area, the model will automatically implement an aggressive strategy to reduce the spacing between cut pieces to maximize utilization. When it falls in the blue low-coefficient combination area, the model will automatically increase the reserved safety margin to prevent physical cutting errors. This strongly demonstrates the scientific validity and effectiveness of this coefficient calculation method in balancing the physical limits of hardware and the economic benefits of nesting.

[0046] Optionally, the collection of fabric batch variation data is intended to address the common batch instability problem in the textile industry. The fabric batch variation data includes the fabric batch variation coefficient. and the revised shrinkage prediction index Fabric batch variation coefficient This reflects the degree of fluctuation in the physical properties of the current batch of fabric relative to the standard sample. However, the accurate prediction system for shrinkage performance does not directly use the nominal shrinkage rate measured in the laboratory, but rather employs a modified index.

[0047] Specifically, the corrected shrinkage prediction index The acquisition process includes: first, obtaining the basic shrinkage prediction index of the fabric under standard testing conditions from the fabric database. Secondly, collect shrinkage data from sampling tests of the current batch of fabric and calculate the deviation of the shrinkage prediction index between the current batch and the standard batch. Finally, the deviations between the basic shrinkage prediction index and the shrinkage prediction index are summed to obtain the corrected shrinkage prediction index. This approach ensures that the data input to the model accurately reflects the actual shrinkage risk of the roll of fabric about to be cut, thus avoiding garment size discrepancies caused by batch data distortion.

[0048] Furthermore, when the material layout request includes personalized customization requirements, this embodiment also performs specific preprocessing steps. Before generating the initial material layout scheme, the system converts the personalized customization requirements into computer-recognizable material layout constraints using a preset mapping rule base. These constraints include at least material layout direction restrictions, allowance coefficients, and threshold values. For example, if a customer requires that "the front plaid pattern must be horizontally aligned," this will be converted into a hard constraint that "the front panel material layout direction is restricted to 0 degrees"; while if the customer selects a "loose fit and high-frequency machine washing," the system will generate a larger allowance coefficient.

[0049] It is worth noting that the deep Q-network model is specifically trained to execute a particular strategy when generating actions: prioritizing the placement of fabric pieces containing customized area identifiers in the geometric center region of the fabric. This is because the geometric center region of the fabric is typically least affected by edge effects and has the most stable mechanical properties, thus maximizing the shape retention of the customized areas.

[0050] Ultimately, the aforementioned collected mechanical parameters and equipment parameters (including calculated parameters) Batch data (including) The data, along with custom constraints, are processed through Z-score normalization or Min-Max normalization, and then concatenated and fused to form a multi-dimensional state input vector. This vector not only contains numerical information but also class information through one-hot encoding, forming the eyes of the Deep Q-Network (DQN) for perceiving the external environment.

[0051] S2: DQN decision-making and initial material feeding based on multi-objective dynamic weighted reward function.

[0052] Specifically, after data preparation is complete, the system inputs the multi-dimensional state input vector into a pre-trained deep Q-network model. This model's core architecture is based on deep reinforcement learning principles, generating the optimal strategy by evaluating the long-term benefits of different material placement actions in the current state. In this embodiment, the model is used to generate an initial material placement layout scheme based on a multi-objective dynamic weight reward function.

[0053] It is important to note that traditional material sorting algorithms often focus solely on utilization rate, while the innovation of this embodiment lies in designing a complex reward function that balances the interests of multiple parties. The multi-objective dynamic weighted reward function... Used to evaluate the quality of material handling actions during model training, it guides the model to learn how to make choices over hundreds or thousands of iterations. Its functional relationship can be expressed as:

[0054] ;

[0055] In the formula: This indicates a training iteration phase, meaning that the weights are not fixed but evolve as the training progresses. This is the expected deformation after machine washing, calculated by a physical simulation engine based on the current fabric feed angle and fabric mechanical parameters, representing the displacement of key parts of the garment after washing. The maximum allowable deformation is a red line set by the quality standard, for example, it cannot exceed 1.5 cm; the first item This means the smaller the deformation, the higher the reward.

[0056] The revised shrinkage prediction index obtained from the aforementioned calculations; The first metric allows for the maximum shrinkage prediction; similarly, the second metric rewards higher rewards for lower shrinkage risk.

[0057] The area already laid out is the total area of ​​all the cut pieces that have been placed. This refers to the total area of ​​the fabric. The aforementioned equipment compatibility coefficient is used here as a correction factor for utilization, meaning that high utilization obtained on high-performance equipment is more valuable. The score for the rationality of the layout is used to evaluate whether there are overlaps, boundary crossings, or other behaviors that violate the common sense of physics between the cut pieces. It is usually a binary or piecewise function. Custom requirement satisfaction rate, representing the proportion of all custom constraints that are met, ranging from 0 to 1; coefficient. to These are weight coefficients that change dynamically with each training iteration, determining the model's focus at different stages.

[0058] Regarding the training process of the model, to maximize the effectiveness of this complex reward function, the deep Q-network model employs a two-stage training strategy. The first stage is defined as the deformation-first training stage, in which the system forcibly sets... Its weight value is significantly higher than other weight coefficients, for example... Set it to 0.6, and the utilization weight Set to 0.1 only. The model's learning objective is extremely conservative, prioritizing the placement of cut pieces in the direction of optimal mechanical performance, even at the cost of significant fabric loss. This phase continues until the expected deformation after machine washing is achieved. The training continues until the percentage of samples with deformation values ​​less than or equal to the preset target value reaches a preset threshold. For example, if 95% of the samples are qualified, this is equivalent to the model first learning to produce qualified suits. The second stage is defined as the multi-objective optimization training stage. This stage begins when the model has mastered the basic rules of anti-deformation material layout. The system dynamically adjusts the ratio of the current training metric to the target metric. , and The value. For example, if the current deformation is well controlled, gradually reduce it. At the same time improve This encourages the model to explore methods that maintain deformation resistance while improving utilization. Training continues until the reward function is defined. If the fluctuation amplitude is less than the preset convergence threshold, it indicates that the model has found the Nash equilibrium point of the interests of all parties.

[0059] After receiving the input vector, the trained model will output a series of action sequences: which piece to select, how many degrees to rotate, and which coordinate to place it. The accumulation of these action sequences eventually forms the initial layout scheme.

[0060] like Figure 3 The figure shows the dynamic evolution curve of the multi-objective weights in the two-stage training. The figure intuitively shows the emphasis logic of the deep Q network model on each optimization objective in different training stages through time-series waveforms.

[0061] Figure 3 The horizontal axis represents the progress of the training iterations, the vertical axis represents the numerical value of each target weight coefficient, and the three curves of different colors and line types represent the dynamic change trajectory of deformation constraint weight, utilization weight, and shrinkage constraint weight, respectively.

[0062] from Figure 3 In the first stage area on the left, it can be clearly seen that the red solid line representing the deformation constraint weight always maintains a high value level, while the blue dotted line representing the utilization weight is suppressed to a low level. This huge difference in values ​​corresponds to the deformation priority training strategy described in the specific implementation, which aims to force the model to learn the anti-deformation rules at all costs in the early stage.

[0063] As the training process progresses... Figure 3 The stage boundary point shown by the dashed line marks the beginning of the second stage, where the curves undergo a significant convergence and transformation. The red solid line shows a smooth downward trend, while the blue dotted line rises synchronously. Both eventually converge to a stable state in the right region. This waveform reversal and balance vividly reveals the game process in which the model gradually releases the constraints on utilization after mastering the physical boundaries. It intuitively proves that the two-stage strategy can effectively guide the model to successfully converge to the Nash equilibrium point that balances material discharge efficiency and machine washing stability by dynamically adjusting the weight parameters, while ensuring the quality of the finished product.

[0064] S3: Although the initial nesting layout scheme generated by DQN is excellent on a macroscopic level, it may still have flaws in microscopic geometric matching or specific constraints. Therefore, this embodiment must perform local conflict region detection on the initial nesting layout scheme, a step similar to the review of drawings by a human nesting expert.

[0065] The system performs a full scan of the initial plan, identifying three main types of conflicts: equipment conflicts, batch conflicts, and customization conflicts. In response to the detection of conflict areas that do not meet anisotropic deformation constraints—a manifestation of batch conflicts (e.g., a piece of fabric is placed but falls within a region of abrupt changes in fabric warp and weft properties, resulting in excessive predicted deformation) or other types of conflicts, the system initiates a local subdivision optimization process based on the equipment adaptation coefficient in the cutting equipment parameters and the batch difference coefficient in the fabric batch difference data to determine the final target layout data stream.

[0066] The local subdivision optimization process is a fine-tuning process, specifically including: identifying areas in the initial fabric layout scheme where the predicted deformation exceeds a preset deformation threshold, i.e., the aforementioned high-deformation-risk areas. For these areas, the system will read the shrinkage prediction index deviation of the current fabric batch. If it is found that the shrinkage rate of the current batch is too high, the system will dynamically adjust the fabric layout direction of the cut pieces in the area (e.g., changing from bias cutting to straight cutting) and the allowance (e.g., adding 0.3 cm of allowance around the cut pieces).

[0067] Specifically, the step of initiating a local segmentation optimization process based on the device adaptation coefficient and the batch difference coefficient includes performing at least one of the following targeted operations:

[0068] The first scenario involves equipment conflicts. For example, in the initial plan, the length of the back panel of an oversized coat exceeds the effective length of the current cutting machine's worktable, or the cut pieces are too dense, causing the low-precision cutting machine to easily stick together. In this case, the system... Determine the equipment's limits. If it is confirmed that physical execution is not possible, the system will adjust the material arrangement (e.g., change from a single row to a segmented row) or disassemble large components at permissible joint locations to fit the worktable size.

[0069] In the second scenario, if batch conflicts exist and the batch difference coefficient is... The deformation exceeds the preset threshold. This means the current fabric is highly unstable and prone to deformation. DQN's initial approach may be biased towards a general strategy and not conservative enough. At this point, the optimization process will intervene, switching the cut pieces to a low-deformation direction. The low-deformation direction is usually the warp direction of the fabric. Simultaneously, a margin is increased, and this increase can be related to... Proportional to offset potential excessive shrinkage.

[0070] The third scenario involves customization conflicts. For example, in an initial design aimed for maximum utilization, a pocket flap requiring strict alignment might be slightly rotated by 2 degrees. While this is permissible in ready-to-wear garments, it's a fatal flaw in custom orders. In this case, the system prioritizes the layout constraints of the custom area. The system will force the cut piece back to its original position, precisely aligning it with the check pattern. This operation often encroaches on the space of surrounding standard cut pieces, leading to a decrease in overall utilization. This embodiment explicitly allows the utilization rate of non-custom areas to decrease within a preset range, for example, allowing a 3% reduction in total utilization rate, in exchange for perfectly fulfilling the customization requirements.

[0071] Through the aforementioned local optimizations, the initial solution was revised into a final solution that not only conforms to physical and mechanical constraints, but also adapts to specific equipment and meets personalized needs, namely the target material discharge data stream.

[0072] S4: After determining the target nesting data stream, the intelligent nesting calculation device converts it into a machine-readable language, such as G-code or HPGL format, and then generates control instructions containing the target nesting data stream. These instructions are sent to the production line for the cutting equipment to perform the nesting operation. Following the instructions, the cutting machine precisely cuts numerous garment pieces from the fabric.

[0073] It should also be noted that the method described in this embodiment does not stop at outputting instructions, but also includes data feedback steps for the entire process of material feeding, production and care, which enables the system to have the ability to evolve itself.

[0074] During the production phase, the system collects cutting utilization data from the cutting equipment during the execution of the target fabric layout data stream via an IoT interface, i.e., a comparison between the actual fabric length consumed and the theoretical length. In the subsequent finished product stage, especially after simulated home washing tests, the system collects actual deformation data of machine-washed suits during the care stage, which is typically obtained by comparing digital twin models of the garments before and after washing using a 3D scanner.

[0075] S5: The system feeds back the cutting utilization rate data and the actual deformation data to the intelligent nesting calculation device. This real-world data serves as error correction signals. Whenever the accumulated data reaches a preset update threshold, such as 1000 measured samples, the system initiates an incremental learning program, using the feedback data to perform incremental iterative training on the deep Q-network model.

[0076] This training is no longer starting from scratch, but rather a fine-tuning of existing model parameters. Simultaneously, the system updates the calculation parameters of the equipment adaptation coefficient (such as correcting the baseline accuracy parameters) or updates the fabric batch variation data based on deviation statistics, for example, correcting a supplier's shrinkage prediction baseline.

[0077] Through this closed-loop mechanism, as production tasks accumulate, the DQN model gains a deeper understanding of machine washing deformation and a more thorough exploration of equipment operation results, thereby enabling the generated material discharge scheme to achieve an increasingly perfect balance between deformation resistance and high utilization rate.

[0078] In summary, this embodiment achieves intelligent and deformation-resistant fabric layout for machine-washable suit pieces through standardized data input, reward function design based on both physical and economic logic, phased training strategies, and a sophisticated conflict handling mechanism. This solves the pain point of existing technologies that cannot simultaneously consider the anisotropic mechanical properties of fabrics and production efficiency.

[0079] Example 2:

[0080] This embodiment, based on Embodiment 1, further refines the technical details to address potential extreme situations during model training. Specifically, this embodiment focuses on the deadlock oscillation problem that a deep Q-network model may encounter during multi-objective optimization when faced with fabrics exhibiting extremely anisotropic mechanical characteristics and simultaneously subject to stringent personalized customization requirements. To overcome this technical bottleneck, this embodiment details a reward reshaping step to resolve deadlock oscillations during multi-objective optimization during model training, including the following:

[0081] Step 1: Identification mechanism for deadlock oscillation in multi-objective optimization.

[0082] Specifically, during the training of the deep Q-network model, the system does not iterate blindly, but instead introduces a real-time monitoring module. This module is configured to monitor the multi-objective dynamic weight reward function in real time. The numerical variation characteristics. In Example 1, the reward function has been defined. It is composed of multiple weighted components such as deformation, shrinkage, utilization rate, rationality, and customization satisfaction rate. However, in the actual calculation logic, when the physical properties of the fabric in the warp and weft directions are very different, for example, the warp elastic modulus is three times that of the weft, and the customer specifies that a certain key cut piece must be arranged along the weft direction to meet the pattern alignment, the model will face huge decision conflicts.

[0083] For example, this conflict often manifests as follows: the model attempts to place the fabric piece in a position with better mechanical properties, resulting in the custom constraints not being met, thus incurring a penalty for the custom weight; subsequently, the model attempts to meet the custom constraints, but is penalized again by the force weight due to excessive deformation. Within consecutive iteration cycles, the model repeatedly jumps between these two high-penalty states, causing the fabric layout scheme to remain unstable. To accurately capture this state, this embodiment sets a judgment logic: if within a preset consecutive iteration cycle, the reward function... The variance is lower than the preset convergence threshold, and the customized requirement satisfaction rate is... If the preset threshold is not met, the system determines that the model has entered a multi-objective deadlock state.

[0084] like Figure 4 The graph shown is a monitoring plot of the reward function variance and the customized satisfaction rate under deadlock conditions. This plot intuitively demonstrates the typical data characteristics of the model when it is trapped in multi-objective optimization deadlock through synchronous comparison of the dual-axis curves.

[0085] Figure 4 The middle horizontal axis represents the continuous training iteration cycle, the left vertical axis corresponds to the blue sliding variance curve of the reward function, and the right vertical axis corresponds to the red custom demand satisfaction rate curve.

[0086] As can be clearly observed from the evolution trend of the graph, as the iteration proceeds, the variance of the reward function shown by the blue solid line shows a trend of rapid decrease and eventually stabilizes in an extremely low value range, which is significantly lower than the preset convergence threshold shown by the blue dashed line in the figure. This indicates that the model has stopped exploring policies at a large scale at the numerical level and has entered a superficially stable convergence state.

[0087] However, in stark contrast, Figure 4 Although the customized demand satisfaction rate, indicated by the red dashed line, has increased slightly, it remains at a low level, far below the preset threshold shown by the red dashed line in the figure. This "scissors difference" phenomenon—where the reward function variance is extremely low while the key indicators fail to meet the standards—precisely describes the pseudo-convergence deadlock state defined in the specific implementation method. That is, to avoid the high penalties incurred from violating conflicting constraints, the model chooses to remain in a local, safe, and comfortable zone, rather than truly solving the complex material arrangement problem.

[0088] This monitoring chart not only verified the sensitivity of the deadlock identification mechanism, but also provided solid data for triggering the subsequent washing stress potential energy field guidance mechanism.

[0089] It should be explained in detail that the continuous iteration period here can be set as a sliding window during the training process, such as the past 1000 iterations. Reward function The variance is below the preset convergence threshold, meaning the total score fluctuates very little, and the model appears to have converged; however, the customized demand satisfaction rate... Failure to reach the preset threshold means that the key hard indicators have not actually been achieved. This is a pseudo-convergence or deadlock state, indicating that the model is trapped in a local minimum trap. The model's own random exploration strategy alone is no longer sufficient to escape this predicament, and external potential energy must be introduced to guide it.

[0090] Step 2: Reward reshaping strategy based on washing stress potential field.

[0091] In response to the multi-objective deadlock state, the system will automatically trigger a reward reshaping mechanism. The core idea of ​​this mechanism is to artificially introduce a virtual physical field, namely the washing stress potential energy field, into the state space of material discharge in order to break the deadlock.

[0092] In terms of specific operation, the system uses the original multi-objective dynamic weighted reward function. Superimposed washing stress potential energy guiding term At this point, the feedback received by the model is no longer the original... Instead, it is the revised reward function. .

[0093] Specifically, the revised reward function The expression is:

[0094] ;

[0095] In the formula, Defined as a preset guidance strength coefficient, this coefficient is a crucial hyperparameter that determines the intensity of external guidance intervention. If... If the value is too small, it may not generate enough thrust to push the model out of the deadlock region; if If the value is too large, the model may focus too much on potential energy guidance and ignore the original objectives such as utilization rate.

[0096] In practical applications, It is usually set to a dynamic value that increases with the duration of deadlock; that is, the longer the deadlock lasts, the greater the push exerted by the system, forcing the model to change its current strategy.

[0097] Step 3: Construction and physical significance of the washing stress potential energy guiding term.

[0098] It should also be noted that the washing stress potential energy guiding term It is not a simple numerical value; it is a complex geometric and mechanical function built on the physical properties of the fabric. Its core purpose is to transform the abstract concept of mechanical stability into geometric coordinates on the fabric layout plane.

[0099] Specifically, the washing stress potential energy guiding term The calculation formula is:

[0100] ;

[0101] In the formula: This represents the number of critical cut pieces in the current layout that do not meet the customization constraints.

[0102] Representing the The area weight of key fabric pieces. During machine washing, large fabric pieces, such as the front and back pieces, tend to accumulate greater washing stress, and their deformation has the greatest impact on the appearance of the garment. Therefore, introducing area weights into the formula means that the larger the area of ​​the fabric piece, the greater its mass in the potential energy field, and the system will preferentially drive these large components to move towards a safe area;

[0103] and Representing the The system obtains the centroid coordinates of each key piece of fabric in real time and the geometric position of these key pieces on the current layout. The coordinate system here is established on the plane of the fabric, usually with the lower left corner or geometric center of the fabric as the origin.

[0104] and The coordinates represent the minimum point of washing stress potential energy in the fabric coordinate system. In traditional fabric layout concepts, the fabric is considered uniform throughout. However, in the technical solution of this embodiment, due to the anisotropy of the fabric and the special mechanical analysis, a virtual safest point exists on the fabric. The minimum point of washing stress potential energy is the position of the mechanical stability extremum calculated by weighting the warp elastic modulus, weft elastic modulus, and shrinkage prediction index of the fabric. For example, if a piece of fabric has extremely strong warp deformation resistance but weak weft resistance, and the shrinkage rate shows a decreasing trend from the selvage to the center, then this... and This will result in the fabric falling at a point slightly off-center from the strong warp axis. This point can be understood as the eye of a storm; placing the cut piece here or near this point minimizes the probability of uncontrollable deformation during machine washing.

[0105] Represents the anisotropy correction factor, used to characterize the ratio of the warp to the weft mechanical attenuation gradient of a fabric.

[0106] Inside the square root in the above expression, Plus Multiply This actually defines a non-Euclidean distance metric, namely mechanical distance.

[0107] Specifically, if If the value is 1, then this is the standard Euclidean distance, which means that the fabric has the same properties in all directions and the potential energy field is distributed in a circular shape. However, in reality, machine-washable suit fabrics are usually anisotropic.

[0108] Assuming the fabric's risk of deformation along the Y-axis (weft direction) is twice that along the X-axis (warp direction), then The value of is set relatively large. This causes the displacement along the Y-axis to be amplified when calculating distance. In other words, the model will perceive that moving the piece along the Y-axis requires overcoming greater resistance, or that deviating from the optimal point along the Y-axis is more costly. Therefore, The presence of this feature stretches the originally circular equipotential lines into ellipses, with the major axis corresponding to the direction of weaker mechanical properties and the minor axis corresponding to the direction of stronger mechanical properties. This design forces the cut pieces to slide more along the spine of the fabric with better mechanical properties when moving, rather than blindly moving in a straight line.

[0109] like Figure 5 The image shows a contour plot of the washing stress potential energy field based on the anisotropic correction factor. The plot intuitively reveals the guiding logic of the virtual physical field in the reward reshaping mechanism through the distribution of color gradients and geometric shapes.

[0110] Figure 5 The horizontal and vertical axes represent the warp and weft physical dimensions of the fabric, respectively. The color bars on the right indicate the intensity level of the washing stress potential energy, gradually transitioning from the dark blue low potential energy area to the dark red high potential energy area.

[0111] Figure 5 The position marked with a white asterisk in the center is the point of lowest washing stress potential energy calculated by weighted mechanical parameters. This point represents the eye of the storm area where the fabric has the strongest resistance to deformation. Figure 5 The most significant feature is that the equipotential lines spreading around the lowest point do not present regular concentric circles, but rather exhibit a distinctly flattened elliptical shape. This asymmetric geometric waveform transformation directly corresponds to the physical effect of the anisotropy correction factor.

[0112] Specifically, the direction of the minor axis of the ellipse corresponds to the direction in which the potential energy increases sharply, indicating that the mechanical stability of this direction (such as the weft direction) is poor, and the cost of the fabric piece deviating from the optimal point along this direction is extremely high; while the direction of the major axis of the ellipse corresponds to the direction in which the potential energy increases gradually, indicating that this direction (such as the warp direction) has better deformation tolerance.

[0113] This anisotropic potential slope design generates a special sliding gravity when the model training gets stuck, forcing the cut pieces to fall back to the lowest point in the center along the long axis ridge with the least resistance. This physically avoids high-risk material layout areas and ensures the dimensional stability of the final solution in a machine washing environment.

[0114] Step 4: Bootstrapping and Deadlock Resolution.

[0115] In summary, the washing stress potential energy guiding term The calculated value is negatively correlated with the weighted distance of the set of fabric pieces that do not meet the custom constraints relative to the point of lowest washing stress potential energy. The negative sign before the formula clarifies this: the greater the distance, the larger the value within the square root; since it is a negative value, The smaller the distance, the greater the penalty; the closer the distance, The closer it is to zero, the greater the reward.

[0116] Specifically, when the deep Q-network model receives the corrected reward function Subsequently, its optimization objective shifted momentarily. The model, which had been struggling with the minor trade-off between utilization and custom rules, suddenly felt a powerful pull, a pull stemming from... This force doesn't demand that the model immediately solve all problems, but rather simply and directly requires the model to prioritize dragging those problematic key pieces to the area where the fabric's mechanical properties are most stable.

[0117] Under this strong guidance, the model temporarily abandons its meticulous attention to the utilization rate of edge areas, prioritizing the movement of the fabric pieces that do not meet the customized constraints towards the point of lowest washing stress potential energy. This movement is often accompanied by a restructuring of the fabric layout, breaking up the previously rigid and mutually restrictive fabric piece structure. When the critical fabric pieces finally fall into place... While some utilization might be sacrificed when the model is nearby, its resistance to deformation is physically guaranteed due to its advantageous position. At this point, the deformation penalty that previously caused the deadlock is significantly reduced, and the model regains positive reward feedback.

[0118] Once the model escapes its previous local minimum, the reward function... The variance begins to increase because a new region has been explored, or the customized demand satisfaction rate... Once the threshold is reached, the system determines that the multi-target deadlock state must be resolved. At this point, The value can gradually decay to zero, and the model reverts to the original reward function described in Example 1. The main phase is fine-tuning, which involves further optimizing utilization based on the new stable layout.

[0119] like Figure 6 The figure shows a comparison of the model convergence effect before and after the intervention of potential energy guidance. The figure intuitively confirms the decisive role of the reward reshaping mechanism in solving the deadlock problem of multi-objective optimization through two training curves with different trends.

[0120] Figure 6The horizontal axis represents the number of model training iterations, and the vertical axis represents the total model reward value, including multi-dimensional objectives such as deformation control and utilization. The blue dashed line represents the basic model training trajectory without the introduction of the washing stress potential energy guiding term. It can be seen that this curve oscillates violently in the low value range of around 0.45 throughout the process, indicating that the model is trapped in a local minimum deadlock state that it cannot escape because it cannot take into account both anisotropic constraints and customized requirements.

[0121] In stark contrast, the red solid line represents the training trajectory of the optimized model after the introduction of the potential energy guidance mechanism. This curve remained consistent with the basic model until the intervention point around the 600th iteration. However, after the intervention of the potential energy guidance term, the reward value exhibited an explosive growth trend, rapidly breaking through the original deadlock bottleneck and eventually converging stably in the high-scoring region around 0.92. This waveform transformation clearly demonstrates that the introduction of external physical field potential energy successfully broke the original deadlock, forcibly driving the material feeding strategy towards a more mechanically optimal global solution, thereby effectively achieving a dual improvement in deformation resistance and production utilization.

[0122] Through this reward reshaping mechanism based on potential energy field, this embodiment not only solves the mathematical convergence problem in multi-objective optimization, but more importantly, it deeply embeds the physical laws of textile materials science into the algorithm logic of artificial intelligence, so that the layout results not only look good on the data, but also meet the high-quality requirements of machine-washable suits in physical essence.

[0123] Example 3:

[0124] Building upon Examples 1 and 2, this embodiment further elaborates on a highly subtle yet fatal geometric problem that may arise during the material layout process. While prioritizing the execution of layout constraints in customized areas sacrifices some utilization rates to satisfy the customized rules, in actual production scenarios, this can easily lead to a topology deadlock phenomenon where the numerical requirements are met but the geometry is infeasible. This embodiment details a spatial aggregation optimization step for addressing the material layout deadlock problem caused by topological fragmentation, specifically including the following:

[0125] Step 1: After the intelligent nesting calculation device executes the nesting constraint strategy that prioritizes the customized area, although the nesting scheme logically meets the customer's strict requirements for stripes or specific texture directions, it often produces side effects in the physical layout.

[0126] This side effect manifests as follows: custom-made fabric pieces, forcibly anchored to specific locations on the fabric, such as the absolute center or a specific intersection of warp and weft yarns, act like nails, fragmenting the remaining fabric space into several unconnected or extremely elongated pieces. In this situation, although from a macroscopic statistical perspective, the area utilization rate of the non-customized area may only decrease by two percent, within the preset allowable reduction range, these remaining fragmented spaces cannot accommodate any standard garment piece with actual dimensions, such as a sleeve or facing. This state is the fabric layout deadlock caused by topological fragmentation as defined in this embodiment.

[0127] To enable accurate computer-aided identification of this state, this embodiment introduces a key geometric evaluation metric: the maximum inscribed rectangle size. After executing the priority-protection layout constraint strategy for customized areas, the system immediately initiates a background scanning program to calculate the maximum inscribed rectangle size of the non-customized blank areas in the current layout scheme. This typically requires calling the maximum empty rectangle algorithm in computational geometry to find the largest axis-aligned rectangle that can fit within the complex layout polygon boundaries.

[0128] For example, the decision logic is set as follows: if the calculated maximum inscribed rectangle size is less than the minimum bounding box size of the standard cut piece to be laid out, and the system simultaneously detects that the utilization rate of the non-customized area is within the preset reduction range, then the system formally determines that the current solution has a topological deadlock. Here, the minimum bounding box size of the standard cut piece to be laid out refers to the minimum rectangular space required by the smallest of the remaining unlaid components. If even the smallest component cannot be placed, then even if the total remaining area is large, it is meaningless waste for production. The rigor of this decision condition lies in its exclusion of failures caused by insufficient area itself, specifically locking out topological faults where the area is sufficient but the shape is incorrect.

[0129] Step 2: In response to the topological deadlock state determined by the system, the nesting engine will no longer attempt to forcibly insert the remaining fabric pieces. Instead, through a backtracking mechanism, it will initiate anisotropic sliding window optimization for the source of the breakage: the customized area fabric piece. The core innovation of this process lies in using the asymmetry of the fabric's physical properties to achieve geometrical continuity.

[0130] Specifically, the system first needs to revisit the physical parameters collected in Example 1, and based on the anisotropic mechanical parameters of the machine-washable suit fabric, determine the low-deformation principal axis direction of the customized area cut pieces. Machine-washable suit fabrics are typically woven from warp and weft yarns of different twists and densities, resulting in significant differences in their tensile deformation resistance in the warp and weft directions. Generally, the warp direction (along the selvage direction) has better dimensional stability than the weft direction due to higher yarn tension. By comparing the warp and weft elastic moduli, the system defines the axis with the larger modulus and lower deformation risk as the low-deformation principal axis direction. For example, if the warp elastic modulus is much greater than the weft elastic modulus, then the warp direction is the low-deformation principal axis direction. This means that if the cut piece is slightly translated along this direction, the risk of increased twisting deformation during machine washing is extremely low and controllable.

[0131] It is also important to note that after determining the direction of movement, the system constructs a linear sliding window along the low-deformation principal axis. This window does not extend infinitely but is subject to strict physical constraints. The length of the linear sliding window is determined by the allowable deformation tolerance of the custom patch and the tolerance of the custom rule. For example, if the custom rule requires the pocket to align with the front panel, then the step size of the sliding window must be an integer multiple of the pattern period; if the rule only applies to the chest area, then the sliding window can cover the entire chest area. Mathematically, this window is defined as a line segment, and the reference point of the custom patch is restricted to moving along this line segment. This design ensures that the optimization process is carried out within a safe zone and will not introduce new quality issues by solving topology problems.

[0132] Step 3: After establishing a safe operating space, the optimization algorithm begins executing its core actions, driving the customized area piece to traverse the linear sliding window. This process is similar to sliding building blocks along a track. The system gradually translates the customized piece along the low-deformation main axis in preset small steps, such as 1 millimeter or one yarn unit.

[0133] At each translation and temporary fixed position, the system needs to reassess the geometric quality of the remaining space. To this end, this embodiment proposes a novel evaluation dimension: calculating the corresponding remaining space aggregation index at each displacement position. Unlike traditional utilization indices that only focus on the sum of areas, this index focuses on the connectivity of areas.

[0134] For example, the remaining space aggregation index is defined as the ratio of the largest continuous area in the non-custom blank area to the total blank area. To ensure the unique accuracy of the definition, the largest continuous area here refers to the area of ​​the largest single connected area in the remaining fabric that can be traversed by a continuous path after deducting all currently arranged cut pieces; while the total blank area is the value of the current total fabric area minus the total area of ​​the arranged cut pieces. The physical meaning of this index is very clear: when the index approaches 1, it means that the remaining blank area is a complete large block, not fragmented, which is the ideal state and is very conducive to the subsequent placement of large components; when the index approaches 0, it means that the blank area is torn into countless isolated islands, which is the worst state. By calculating this ratio, the system can quantitatively perceive the degree of improvement of the overall spatial topology by each tiny movement.

[0135] Step 4: After completing the traversal calculation, the system will generate a curve showing the change in position and aggregation degree. Typically, this curve will exhibit distinct peaks and troughs. The peak position corresponds to the moment when the customized piece has moved to a certain avoidance point, causing the two previously blocked blank areas to merge into one due to the opening of the channel.

[0136] like Figure 7 The graph shows the change of the spatial aggregation degree index with the displacement of the sliding window. The graph visually demonstrates how the spatial aggregation optimization steps described in the embodiment solve the macroscopic topological deadlock problem through tiny geometric maneuvers by showing the evolution of peaks and troughs.

[0137] Figure 7 The horizontal axis represents the amount of physical displacement of the fixed piece along the main axis of low deformation of the fabric, and the vertical axis represents the residual spatial cohesion index used to evaluate the connectivity of the remaining fabric.

[0138] observe Figure 7 The blue solid line in the curve shows the change in the degree of aggregation. It can be seen that in the initial stage of displacement, the degree of aggregation index remains at a low level of around 0.35. This corresponds to the material discharge scheme being in a severely fragmented state, with the remaining space being cut into unusable islands.

[0139] However, as the cut piece gradually slides along the safety axis, the curve shows a significant upward trend, forming a steep peak when the displacement reaches approximately 35 mm. At this point, the aggregation index climbs above 0.9. This extreme position marked by the red dot is the optimal anchor point. Physically, it represents that the customized cut piece has moved to a critical avoidance position, instantly opening up and merging the two narrow blank areas that were originally blocked into a wide, continuous rectangular space.

[0140] This strategy of determining the layout coordinates by finding geometric peaks powerfully demonstrates that the present invention can transform an originally unusable topological deadlock structure into a highly valuable aggregate space by utilizing an anisotropic sliding window mechanism without compromising the mechanical stability of the fabric, thereby ensuring the smooth placement of subsequent standard large components.

[0141] Specifically, the system selects the position with the highest remaining space aggregation index as the final anchor point for the customized area pattern piece. This selection decision is based on global optimization considerations. Once the anchor point is determined, the customized pattern piece will be permanently locked at this position and will no longer participate in subsequent regular optimization. With the release of large continuous spaces, standard pattern pieces that were previously unable to be placed due to their excessive size, such as sleeves and backs, can now be smoothly placed into the blank area where the largest inscribed rectangle has been significantly expanded. Thus, the layout stagnation caused by spatial fragmentation is successfully resolved, thereby eliminating the topological deadlock state.

[0142] It's also important to note that this optimization strategy based on spatial aggregation is essentially a high-dimensional game between microscopic mechanical costs and macroscopic production feasibility. Although the cut pieces are displaced in the low-deformation direction, theoretically introducing a negligible deformation deviation, this deviation is far below the quality tolerance of machine-washed suits because the movement is strictly limited to a safety axis determined by anisotropic parameters. The benefit is that it saves a potentially unusable entire pattern layout, avoiding the huge costs of being forced to replace with larger fabrics or cancel custom orders due to a deadlock.

[0143] Through the technical solution described in this embodiment, the intelligent material layout system possesses advanced geometric intelligence. It is no longer a mechanical algorithm for filling rectangles, but an intelligent agent capable of understanding the physical properties of the fabric, sensing spatial connectivity, and proactively creating favorable layout conditions through maneuvering. This has irreplaceable industrial practical value for handling complex products such as high-end machine-washable suits, which are sensitive to both physical properties and have strict cost control requirements.

[0144] Example 4:

[0145] like Figure 8 As shown in the illustration, this embodiment details a machine-washable suit anti-deformation fabric layout optimization system. Based on a modular design architecture, this system aims to resolve the inherent contradiction between the utilization rate of machine-washable suit fabric and the garment's anti-deformation capability, as pointed out in the background art. Through the collaborative work of hardware and software, it achieves intelligent control over the layout process of anisotropic fabrics. Logically, this system corresponds one-to-one with the steps described in Embodiments 1 to 3, representing a concrete device-based embodiment of the aforementioned methods within a computer system.

[0146] The machine-washable suit anti-deformation cutting piece layout optimization system described in this embodiment comprises a core structure including a data acquisition and standardization processing module, a deep network decision-making module, a local subdivision optimization module, and an instruction generation module. These modules interact via a system bus to collaboratively complete the entire process from task reception to cutting instruction output.

[0147] First, the system is equipped with a data acquisition and standardization module. This module serves as the system's sensing front end, primarily responding to material layout requests by initiating a comprehensive scan of the production environment and material properties. As described in the background section, traditional material layout often overlooks the physical properties of the fabric, leading to severe deformation after machine washing. Therefore, this module is specifically designed to collect anisotropic mechanical parameters, cutting equipment parameters, and batch-to-batch variation data of machine-washed suit fabrics. The acquisition of anisotropic mechanical parameters quantifies the differences in the fabric's ability to resist stretching and twisting in the warp and weft directions, providing the physical basis for subsequent deformation resistance decisions. Simultaneously, this module standardizes the collected data into multi-dimensional state input vectors. This process involves normalizing data of different dimensions, such as fusing millimeter-level equipment precision data with dimensionless Poisson's ratio data to construct an environmental state descriptor that a deep learning model can understand, providing a precise input basis for subsequent intelligent decision-making.

[0148] Secondly, the system core incorporates a deep network decision-making module. This module integrates the deep reinforcement learning algorithm described in Example 1, used to input the multi-dimensional state input vector into a pre-trained deep Q-network model. The module is designed to break the deadlock between utilization and quality. By using the model to generate an initial material layout scheme based on a multi-objective dynamic weight reward function, this module can find the optimal solution from tens of thousands of material layout combinations that satisfies both mechanical constraints and cost-effectiveness.

[0149] In particular, during operation, this module can dynamically invoke the reward reshaping mechanism described in Example 2. When the model is detected to be trapped in a multi-objective optimization deadlock oscillation, the key cut pieces are forcibly driven to move to the mechanically most stable region by introducing a washing stress potential energy guiding term, thereby achieving priority protection of the ability to resist deformation at the system level.

[0150] Third, the system includes a local subdivision optimization module. This module primarily handles post-processing and fine-tuning of the design, acting as a quality guardian for the system. Its main function is to detect local conflict areas in the initial layout scheme. In actual operation, even if the macroscopic strategy of the deep network is correct, there may still be physically infeasible areas at the microscopic level. When this module detects a conflict area that does not meet the anisotropic deformation constraints, it immediately initiates a local subdivision optimization process based on the equipment adaptation coefficient in the cutting equipment parameters and the batch difference coefficient in the fabric batch difference data. This process is not a simple patching, but rather a systematic determination of the target layout data flow.

[0151] Specifically, when executing the local subdivision optimization process, the module is configured to identify areas in the initial fabric layout scheme where the predicted deformation exceeds a preset deformation threshold. For these high-risk areas, the module reads real-time fabric data and dynamically adjusts the fabric layout direction and allowance within the area based on the shrinkage prediction deviation of the current fabric batch. For example, if an abnormally high weft shrinkage rate is detected in a certain batch of fabric, the module will automatically rotate the fabric pieces originally planned to be laid out along the weft direction to the warp direction and increase the corresponding shrinkage allowance. Furthermore, this module integrates the spatial aggregation optimization logic described in Embodiment 3. When a topological deadlock is detected, it can construct an anisotropic sliding window to drive the fabric pieces to fine-tune in the low-deformation direction, thereby eliminating the fragmentation of remaining space caused by customization requirements and ensuring the executability of the fabric layout scheme.

[0152] Finally, the system includes an instruction generation module. This module serves as the interface between the system and the physical world, generating control instructions containing the target nesting data stream. These instructions, after encoding and conversion, can be directly recognized by the industrial cutting machine, enabling the cutting equipment to perform nesting operations. Through this module, the virtual nesting scheme calculated by the system is transformed into actual physical cutting actions, ensuring that each piece is precisely cut according to the optimized path and angle.

[0153] In summary, the machine-washable suit anti-deformation cut piece layout optimization system provided in this embodiment, through the close cooperation of the four core modules mentioned above, solidifies the complex layout optimization method into executable system logic. This system not only solves the deformation problem caused by fabric anisotropy mentioned in the background art, but also effectively addresses algorithmic deadlock caused by multi-objective conflicts, achieving the continuous evolution and precise adaptation of the layout strategy described in the beneficial effects, providing strong technical support for the high-quality production of machine-washable suits.

[0154] Example 5:

[0155] Corresponding to the above embodiments, the present invention also proposes a computing device.

[0156] like Figure 9 The diagram shows a structural schematic of a computing device according to the present invention. The computing device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the computing device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one type, and the structure of this computing device 100 does not constitute a limitation on the embodiments of the present invention.

[0157] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0158] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0159] The memory 103 stores a computer program corresponding to the optimized layout method for anti-deformation cutting pieces of a machine-washable suit according to the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0160] The computing device 100 includes, but is not limited to, mobile terminals such as laptops and tablets, as well as fixed terminals such as desktop computers. Figure 9 The computing device 100 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0161] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optimizing the layout of machine-washable suit fabric pieces to prevent deformation, characterized in that, Applications in intelligent material feeding calculation equipment include: In response to the material layout task request, the anisotropic mechanical parameters, cutting equipment parameters, and batch difference data of the machine-washed suit fabric are collected, and the collected data are standardized and processed into a multi-dimensional state input vector. The multi-dimensional state input vector is input into a pre-trained deep Q-network model, and the model is used to generate an initial material layout scheme based on a multi-objective dynamic weight reward function. Local conflict area detection is performed on the initial layout scheme. In response to the detection of conflict areas that do not meet the anisotropic deformation constraints, a local subdivision optimization process is initiated based on the equipment adaptation coefficient in the cutting equipment parameters and the batch difference coefficient in the fabric batch difference data to determine the target layout data stream. Generate control instructions containing the target material feed data stream for the cutting equipment to perform the material feed operation; The local subdivision optimization process includes: identifying areas in the initial fabric layout scheme where the predicted deformation exceeds a preset deformation threshold, and dynamically adjusting the fabric layout direction and allowance of the cut pieces within the area based on the shrinkage prediction index deviation of the current fabric batch.

2. The method according to claim 1, characterized in that, The cutting equipment parameters include equipment type, cutting accuracy, and worktable size; The step of standardizing the collected data into a multi-dimensional state input vector includes calculating the device adaptation coefficient of the cropping device. The calculation formula is as follows: ； Wherein, the equipment accuracy reference value and the standard worktable area are preset normalization constants; The deep Q-network model is configured to adapt to the device adaptation coefficient. The numerical range is used to dynamically adjust the spacing between cut pieces.

3. The method according to claim 1, characterized in that, The anisotropic mechanical parameters include the warp elastic modulus, weft elastic modulus, warp Poisson's ratio, and weft Poisson's ratio of the fabric. The fabric batch variation data includes the fabric batch variation coefficient. and the revised shrinkage prediction index ; The corrected shrinkage prediction index The acquisition process includes: obtaining basic shrinkage prediction indicators Calculate the deviation of the shrinkage prediction index between the current batch and the standard batch. The result is obtained by summing the deviation between the basic shrinkage prediction index and the shrinkage prediction index.

4. The method according to claim 1, characterized in that, The multi-objective dynamic weighted reward function Used to evaluate the quality of material feeding actions during the model training phase, its functional relationships include: ； in, Indicates the training iteration phase. The expected deformation after machine washing, To allow the maximum deformation, To correct the shrinkage prediction index, To allow for the maximum shrinkage prediction index, This represents the area where materials have already been laid out. This represents the total area of ​​the fabric. To score for the rationality of material arrangement, To improve the satisfaction rate of customized needs, to These are weight coefficients that change dynamically with each training iteration.

5. The method according to claim 4, characterized in that, The training process of the deep Q-network model adopts a two-stage training strategy: The first stage is the deformation-priority training stage, which is set as follows: The weight value is higher than other weight coefficients, until the expected deformation after machine washing. The percentage of samples with a target variable less than or equal to the preset threshold value reaches the preset threshold value. The second stage is the multi-objective optimization training stage, which dynamically adjusts the training metric based on the ratio of the current training metric to the target metric. , and The value of until the reward function The fluctuation amplitude is less than the preset convergence threshold.

6. The method according to claim 1, characterized in that, The process of performing local conflict area detection on the initial material layout scheme includes identifying equipment conflicts, batch conflicts, and customization conflicts. The local segmentation optimization process initiated based on the device compatibility coefficient and the batch difference coefficient includes at least one of the following operations: If there is a conflict between the equipment, adjust the material arrangement or disassemble large components to fit the workbench size. If there is a batch conflict and the batch difference coefficient is greater than the preset threshold, the cut piece will be switched to a low deformation direction and the allowance will be increased. If a customization conflict exists, the material layout constraint strategy that prioritizes the customized area will be implemented, and the utilization rate of the non-customized area will be allowed to decrease within a preset reduction range.

7. The method according to claim 1, characterized in that, The method also includes a data feedback step covering the entire process of material feeding, production, and nursing: Collect cutting utilization data of the cutting equipment during the execution of the target material layout data stream, and actual deformation data of machine-washed suits during the care stage; The cutting utilization rate data and the actual deformation data are fed back to the intelligent nesting calculation device; Whenever the accumulated data reaches a preset update threshold, the deep Q-network model is incrementally iteratively trained using the feedback data to update the calculation parameters of the device adaptation coefficient or the fabric batch difference data.

8. The method according to claim 1, characterized in that, The material sorting task request includes personalized customization requirement data; Before generating the initial material layout scheme, the method further includes: The personalized customization requirements data are transformed into material layout constraints through a preset mapping rule base; The layout constraints include at least layout direction restrictions, allowance coefficients, and index thresholds; when generating actions, the deep Q-network model prioritizes placing cut pieces with customized region identifiers in the geometric center region of the fabric.

9. The method according to claim 4, characterized in that, The method also includes a reward reshaping step, specifically comprising: Real-time monitoring of the multi-objective dynamic weighted reward function The numerical change characteristics of the reward function, if within a preset continuous iteration period, The variance is lower than the preset convergence threshold, and the customized requirement satisfaction rate is... If the preset threshold is not met, the model is determined to have entered a multi-objective deadlock state. In response to the multi-objective deadlock state, in the multi-objective dynamic weighted reward function Superimposed washing stress potential energy guiding term The revised reward function The expression is ; in, The washing stress potential energy guiding term is a preset guiding intensity coefficient. The calculated values ​​are negatively correlated with the weighted distance of the set of cut pieces that do not meet the custom constraints relative to the lowest point of the dry washing stress potential energy; The deep Q-network model receives the corrected reward function. Then, the cut pieces that do not meet the customized constraints are driven to move towards the point with the lowest washing stress potential energy until the multi-target deadlock state is released.

10. The method according to claim 6, characterized in that, The method further includes a spatial aggregation optimization step, specifically comprising: After implementing the priority protection of the customized area layout constraint strategy, the maximum inscribed rectangle size of the non-customized blank area in the current layout scheme is calculated. If the maximum inscribed rectangle size is less than the minimum bounding box size of the standard cut piece to be laid out, and the utilization rate of the non-customized area is within the preset reduction range, then a topological deadlock state is determined to exist. In response to the aforementioned topology deadlock state, an anisotropic sliding window optimization process is initiated: Based on the anisotropic mechanical parameters of the machine-washable suit fabric, the low-deformation principal axis direction of the custom area cut pieces is determined. A linear sliding window is constructed along the low-deformation main axis, and the customized area patch is driven to traverse the displacement within the linear sliding window; Calculate the corresponding remaining space aggregation index at each displacement position, and select the position with the largest remaining space aggregation index as the final anchor point of the customized area piece to release the topological deadlock state. The remaining space aggregation index is defined as the ratio of the largest continuous area in the non-customized blank area to the total blank area.

Images