A method and system for optimizing the digital cutting and layout of ship steel plates

Abstract

This invention discloses a digital cutting and layout optimization method and system for ship steel plates. The method includes: acquiring morphological data of the steel plate to be cut and generating a processing area model; based on the processing area model and the specification data of the target part, using a deep reinforcement learning-genetic hybrid algorithm as the solution strategy to output an optimal layout scheme; generating a CNC cutting path and corresponding CNC code according to the optimal layout scheme, and performing kerf gap constraint and thermal deformation compensation on the CNC cutting path; performing positioning calibration on the steel plate to be cut based on the deviation between the ideal coordinates of the reference marker points in the processing area and their real coordinates projected onto the steel plate to be cut; executing the cutting command, evaluating the cutting result, and feeding back deviation data to adjust the optimization strategy parameters. This invention achieves maximum utilization of steel plates, reduces costs, and improves cutting quality while ensuring process feasibility. It is suitable for shipbuilding scenarios with complex parts, diverse components, and high cutting precision requirements.

Description

Technical Field

[0001] This invention relates to CNC cutting technology in the field of shipbuilding, and in particular, to a method and system for optimizing the digital cutting and layout of ship steel plates. Background Technology

[0002] In shipbuilding, most hull parts are produced by cutting plates of the required specifications from large steel plates and then welding / bolting them together. This raises the question of how to arrange target plates of different shapes and sizes on large steel plates. Existing shipbuilding steel plate cutting and layout technologies have the following main shortcomings:

[0003] 1. Low material utilization: Traditional typesetting mainly relies on manual experience or simple heuristic algorithms to arrange parts on steel plates, which makes it difficult to make full use of the steel plate material. After the cutting operation, a lot of scattered scraps are left. Especially when dealing with irregularly shaped plates, large areas of irregular shapes remain on the steel plate, resulting in a lot of material waste and increasing production costs.

[0004] 2. Insufficient algorithm optimization: Commonly used automatic layout algorithms include heuristic rules, genetic algorithms, simulated annealing, etc., but these algorithms are prone to getting stuck in local optima for complex layout problems (which are NP-hard problems) and cannot achieve the highest material utilization rate. In addition, these algorithms usually output the solution at once, lack self-learning ability, and cannot continuously improve the layout strategy based on the feedback of the cutting results.

[0005] 3. Unoptimized cutting path: Existing CNC cutting paths are usually generated directly based on the layout results without sufficient optimization of the cutting sequence, resulting in a long idle path and reduced cutting efficiency. In addition, there is a lack of compensation mechanism for the thermal deformation of the steel plate during the cutting process. The steel plate will expand and deform after being cut at high temperature. If no compensation is made, it will affect the cutting accuracy and cause deviation in the actual size of the parts.

[0006] 4. Positioning accuracy issues: When steel plates are placed on the cutting platform, there are usually position and angle deviations. If only manual positioning or single camera vision positioning is used, the cutting path may not match the actual position of the plate, affecting cutting quality and production safety. Existing technologies lack high-precision automatic calibration methods, which leads to the need to increase the gap between parts or retain excess edges, further hindering the realization of zero-excess cutting.

[0007] 5. Lack of feedback mechanism: Traditional cutting systems rarely analyze and provide feedback on the results of each cut, so they cannot adjust subsequent layout strategies based on the actual amount of leftover material. For example, if it is found that the actual leftover material is more than expected after a cut, the system will not "learn" this deviation, which may lead to similar problems recurring in subsequent layouts.

[0008] 6. Low system integration: Different stages of steel plate cutting (such as scanning measurement, layout calculation, CNC cutting, calibration, and feedback) are often completed by different equipment and software, and the data interfaces are not unified, which affects production efficiency. For example, the shape of the plate obtained by scanning needs to be manually imported into the layout software, while the cutting code is generated by another system. There is a lack of unified communication protocols and data formats to achieve real-time collaboration between the modules.

[0009] In view of the above problems, the industry urgently needs a new technical solution that can automatically acquire the shape of steel plates, intelligently optimize the layout of parts to maximize material utilization, generate optimized cutting paths, calibrate the actual cutting position, monitor and provide feedback on the cutting results, and form a closed-loop self-learning mechanism with the optimization algorithm to continuously improve the efficiency and accuracy of layout and cutting. Summary of the Invention

[0010] The purpose of this invention is to provide a digital cutting and layout optimization technology for ship steel plates to overcome the shortcomings of existing ship steel plate nesting and CNC cutting technologies.

[0011] To achieve the above objectives, this invention discloses a method for optimizing the digital cutting and layout of ship steel plates, comprising:

[0012] Step 1: Collect the morphological data of the steel plate to be cut and generate a processing area model;

[0013] Step 2: Based on the processing area model and the specification data of the target part, a deep reinforcement learning-genetic hybrid algorithm is used as the solution strategy to output the optimal layout scheme;

[0014] Step 3: Generate CNC cutting path and corresponding CNC code according to the preferred layout scheme, and perform kerf gap constraint and thermal deformation compensation on the CNC cutting path;

[0015] Step 4: Based on the deviation between the ideal coordinates of the reference marker points within the processing area and their actual coordinates projected onto the steel plate to be cut, perform positioning calibration on the steel plate to be cut;

[0016] Step 5: Execute the cutting command, evaluate the cutting results, and provide feedback on deviation data to adjust and optimize strategy parameters.

[0017] Preferably, step 1 includes:

[0018] The steel plate to be cut is a whole plate or a leftover plate from the previous cut. The outline coordinate point cloud of the steel plate is obtained by a laser scanner or machine vision, and a processing area model is generated, as shown in the following expression:

[0019] ;

[0020] in, These are the coordinates of the processing area boundary.

[0021] Preferably, step 2 includes:

[0022] Input the processing area model and the specification data of the target part, and use a deep reinforcement learning agent and a genetic algorithm to explore the optimal layout of the target part on the steel plate, that is, all the layout parts do not overlap and the unused space in the processing area is minimized, as follows:

[0023] The reinforcement learning agent arranges the parts one by one in the processing area to form a sequence of actions. N represents the total number of parts that need to be laid out, and the blank processing area with no parts to be placed is taken as the initial state. In action The upcoming After each part is placed in the machining area, its new status is recorded. ;

[0024] In the early stages of the action, parts are arranged randomly or based on heuristic rules. After each arrangement, a reward is calculated based on the material utilization rate and the conflict of parts. A positive reward is given if the material utilization rate increases or there is no conflict, and a negative reward is given if the material utilization rate decreases or a conflict occurs. The reinforcement learning agent adjusts its strategy accordingly to try different layout methods.

[0025] A genetic algorithm is used to update the policy network parameters of feasible layout schemes given by the reinforcement learning agent. After several iterations, crossover and mutation operations are performed on several of the better layout schemes to generate new layout schemes. After completing a preset number of iterations of training and evaluation, a set of optimized part layout parameters is output as the preferred layout scheme. The component layout parameters include the rotation angles of each component. And the amount of translation in the horizontal direction Vertical translation .

[0026] Preferably, the preferred typesetting scheme The condition for zero residual is satisfied, and the expression is as follows:

[0027] ;

[0028] in, Let be the region where the t-th part is located after rotation and translation operations, and Area be a function to calculate the area of ​​the region.

[0029] It should be noted that the definition of zero margin in this invention does not mean that there is no margin at all, but rather that all parts are placed in the steel plate processing area. Inside, the parts do not overlap, and the unused areas are only the extremely fine mesh skeleton formed by laser cutting. The area of ​​leftover material is almost zero. The specific judgment threshold is set according to the actual production situation.

[0030] Preferably, step 3 includes:

[0031] The position of each part in the processing area is determined according to the preferred layout scheme. The outline of each part is divided into multiple line segments according to nodes. All line segments are arranged in sequence to generate a CNC cutting path. The planning of the CNC cutting path must meet the following requirements:

[0032] ① For any two adjacent parts, maintain ,in, Let be the minimum distance between the contour edges of part i and part j. This is the minimum clearance allowed by the process.

[0033] ② The cutting path length is corrected and compensated based on the thermal expansion effect of steel, i.e. ,in, To compensate for the path length, is the coefficient of linear expansion of steel. This represents the original length of the cutting path before compensation. To increase the temperature of the cutting area;

[0034] Generate the corresponding CNC code (such as G code or other NC code) based on the CNC cutting path, and upload the code file to the CNC cutting machine controller.

[0035] Preferably, step 3 uses the 2-Opt algorithm to optimize the cutting order. Based on the initial path scheme, the cutting order of local paths is repeatedly swapped to reduce the length of empty paths until the path can no longer be optimized or the optimization increment is lower than the threshold.

[0036] It should be noted that by introducing the 2-Opt algorithm, the idle time of the cutting head moving between the contours of the parts is minimized, thereby improving cutting efficiency and reducing energy consumption.

[0037] Preferably, step 4 includes:

[0038] Select at least three reference marker points within the processing area and project them onto the surface of the steel plate to be cut, then record the ideal coordinates of the reference marker points. ;

[0039] Machine vision is used to acquire images of the steel plate to be cut and to identify the true coordinates of the projection points on the steel plate. ;

[0040] Based on the coordinate differences, the rigid body transformation parameters of each projection point relative to the ideal position are calculated, as shown in the following expressions:

[0041] ;

[0042] in, For rotation angle, This represents the horizontal translation. This represents the vertical translation amount;

[0043] The least squares method is used to fit the solution of the rigid body transformation to obtain estimates of the translation and rotation angles. The positioning error is calculated and compared with a preset accuracy threshold, as shown in the following expression:

[0044] ;

[0045] If all the above equations are true, the calibration is considered successful. The data is uploaded to the CNC cutting machine controller, which triggers the cutting execution command and performs coordinate transformation compensation on the cutting path to be executed. Conversely, if at least one of the above formulas is not true, the calibration is deemed to have failed, and an alarm is triggered to request adjustment of the steel plate placement position or addition of more reference markers to recalculate the positioning error until the accuracy requirements are met.

[0046] Preferably, step 5 includes:

[0047] The CNC cutting machine performs cutting operations according to the generated CNC code. The cutting head cuts out the outline of the part segment by segment from the edge of the steel plate or the pre-pierced part. It moves along the shortest path to the starting point of the next part outline. After the operation is completed, all parts are removed and the remaining material is left on the cutting machine table.

[0048] The morphological data of the scrap material is acquired using a laser scanner or machine vision. The actual scrap material is compared with the theoretical scrap material expected in the optimized layout scheme, and the overlap index between the two is calculated. The expression is as follows:

[0049] ;

[0050] Where IoU is the overlap index, and Area is a function for calculating the area of ​​a region. for For theoretical surplus ;

[0051] When the IoU is less than the preset overlap threshold, the difference area between the actual surplus material and the theoretical surplus material is rasterized and discretized to obtain a difference raster image, as shown in the following expression:

[0052] ;

[0053] in, For set difference operations, For discretization operations;

[0054] The differential raster image is input into the neural network of the reinforcement learning agent for training, so that the algorithm can identify situations where differential regions are not effectively utilized in the scheme, and adaptively adjust the subsequent typesetting strategy parameters accordingly.

[0055] Preferably, the function expression for calculating the reward in step 2 is as follows:

[0056] ;

[0057] in, Let be the reward value for the t-th action. Let be the increment of material utilization rate after the t-th action compared to the previous step. This is a conflict indicator function. It returns 1 if a newly placed part overlaps with an already placed part, and 0 otherwise. This represents the current cumulative cutting time. For the maximum allowable cutting time, These are the corresponding weighting coefficients.

[0058] It should be noted that this invention guides the reinforcement learning agent to optimize decisions towards the goal of efficient layout with zero redundancy by designing a specific reward function. This reward function is an evaluation index that comprehensively considers material utilization, collision conflicts, and cutting efficiency. The material utilization rate is the ratio of the total area of ​​the layoutd parts to the area of ​​the steel plate. This indicates the proportion of the current cumulative cutting path length or cutting time to the maximum allowable value, reflecting the cutting efficiency. The reward function is determined through practical experience to adjust the severity of rewards and penalties for different state changes. Through this reward function design, the agent can balance improving material utilization, avoiding component conflicts, and reducing cutting complexity during typesetting, gradually learning better typesetting strategies during training.

[0059] Preferably, the deep reinforcement learning-genetic hybrid algorithm includes reinforcement learning based on policy gradient update and global search using a genetic algorithm, and the expression for the policy gradient update is as follows:

[0060] ;

[0061] in, For policy network parameters, For policy function Regarding parameters gradient, For the policy network in state Select action The probability distribution function, The probability distribution function with respect to the parameters The logarithmic derivative, For strategy Mathematical expectation of the next action-state distribution The advantage function, used to evaluate the merits of a particular action relative to the average level of that state, is expressed as follows:

[0062] ;

[0063] in, As a discount factor, and The value network is for the current state. Next state The estimated state value.

[0064] It should be noted that the reinforcement learning part of the hybrid algorithm in this invention adopts an Actor-Critic architecture, utilizing the action probability distribution output by the policy network and the state value estimated by the value network to guide policy improvement. The policy network parameters can be adjusted using the aforementioned policy gradient update formula. Updated to improve long-term cumulative rewards.

[0065] Preferably, the genetic algorithm is based on the current policy network parameter population and generates new policy individuals through selection, crossover, and mutation operations, as shown in the following expression:

[0066] ;

[0067] in, These are the parameters of the new policy network generated after the genetic operation. and Let be the parameters of the i-th and j-th policy individuals selected from the current policy population, respectively. The cross ratio coefficient and , This is a Gaussian noise variation term.

[0068] It should be noted that the present invention achieves perturbation optimization of policy network parameters through the above-mentioned genetic update, so that the policy network parameters can be further diversified on the basis of reinforcement learning gradient update, so as to escape the local optima that reinforcement learning may get stuck in and increase the probability of finding a better global solution.

[0069] This invention also discloses a digital cutting and layout optimization system for ship steel plates, comprising:

[0070] Plate morphology acquisition module: Collects morphology data of the steel plate to be cut and generates a processing area model;

[0071] Intelligent layout optimization module: Based on the processing area model and the specification data of the target part, it adopts a deep reinforcement learning-genetic hybrid algorithm as the solution strategy and outputs the optimal layout scheme;

[0072] Cutting path generation module: Based on the preferred layout scheme The corresponding CNC code is used to constrain the kerf gap and compensate for thermal deformation in the CNC cutting path;

[0073] Projection positioning calibration module: Based on the deviation between the ideal coordinates of the reference mark points in the processing area and their actual coordinates projected onto the steel plate to be cut, the module performs positioning calibration on the steel plate to be cut.

[0074] Feedback self-learning module: Executes cutting instructions, evaluates cutting results, and provides feedback on deviation data to adjust and optimize strategy parameters.

[0075] In the above system:

[0076] The plate morphology acquisition module collects the morphology data of the steel plate to be cut and outputs a digital representation of the steel plate area (such as a polygonal outline or raster image), which serves as the basic input data for subsequent layout optimization.

[0077] The intelligent typesetting optimization module uses a deep reinforcement learning-genetic hybrid algorithm as its core. Through repeated trials by reinforcement learning agents and global search by genetic algorithms, it continuously improves the typesetting scheme and finally outputs the optimal or near-optimal typesetting result that meets the conditions.

[0078] When generating a cutting path corresponding to the preferred layout scheme, the cutting path generation module constrains the kerf gap by setting a lower limit to prevent adjacent parts from sticking together due to excessive proximity, thus ensuring cutting quality. It also compensates and corrects the cutting path length to eliminate the thermal expansion effect on the steel plate caused by the cutting operation, ensuring accurate actual cutting dimensions.

[0079] The projection positioning calibration module calibrates the actual pose of the steel plate before cutting, aligning the actual coordinate system of the steel plate with the ideal coordinate system in the preferred layout scheme. This ensures accurate matching between the CNC cutting path and the part contour, avoiding cut deviations that could cause omissions or additional material waste.

[0080] The feedback self-learning module forms a closed-loop optimization that adaptively optimizes the strategy parameters through analysis and feedback of the actual cutting effect; the overlap ratio index IoU involved is essentially the actual residual material area. With theoretical surplus area The Intersection over Union (IoU) value ranges from 0 to 1. The closer it is to 1, the better the actual leftover material matches the expected leftover material. The closer it is to 0, the greater the deviation between the current cutting and the expected material. The algorithm strategy needs to be adjusted to improve it. Through such a cyclical self-learning process, the system can continuously evolve and optimize the strategy to reduce subsequent layout errors and improve material utilization and cutting accuracy.

[0081] Preferably, in the system, the data communication between modules adopts the following standard protocol and data format:

[0082] The inter-module communication protocol adopts the OPC-UA (Open Platform Communications Unified Architecture) industrial communication standard, with each module (scanning device, computer, CNC machine tool, etc.) exchanging data as an OPC-UA client / server;

[0083] The data transmission format adopts a custom JSON-NC format, which uses JSON (JavaScript Object Notation) to represent CNC-related data (such as sheet geometry information, layout result coordinates, cutting order and parameters, etc.), thus making the data have good readability and universal parsing.

[0084] The system's feedback data (such as the utilization rate of each cut, IoU results, algorithm adjustment parameters, etc.) is uploaded to the monitoring terminal or cloud platform in real time via the MQTT (Message Queuing Telemetry Transport) protocol to enable remote monitoring and big data analysis of the production process.

[0085] It should be noted that by adopting the aforementioned standard protocols and data formats, this invention enables the system to be reliably, conveniently, and smoothly integrated into existing digital shipbuilding production lines, achieving transparent information transmission and sharing, and improving the system's scalability and maintainability.

[0086] Furthermore, for occasional but high-risk low-probability anomalies in actual production (such as local warping of the sheet metal due to internal stress release during cutting, temporary failure of projected marks due to molten slag obstruction, path error accumulation due to instantaneous nozzle offset, and deviation of the thermal expansion coefficient from the calibration value caused by abnormal local temperature rise), this invention also proposes an online closed-loop fallback solution that runs through the execution-detection-correction process, namely, a robust compensation and update method combining deformation field estimation and local path replanning. Specifically:

[0087] During the cutting process, online vision / laser displacement sensors are used to intermittently / sliding window sample the uncut area. First, rigid body registration (ICP / G-ICP) is used to estimate the local attitude drift. Then, the registration residuals are fitted with thin plate spline (TPS) or bilinear displacement field to obtain non-rigid deformation. For path curves that have not yet been executed Perform online mapping correction using the following expression:

[0088] ;

[0089] For the corrected path curve The minimum clearance and thermal compensation constraints of the process are checked in real time. If it is found that the local minimum clearance approaches the threshold due to warping or the size deviation exceeds the limit after thermal compensation, local dynamic replanning is triggered for the affected path segment (only the order and start and end points are rearranged among several contours to be cut). The existing 2-Opt / LKH neighborhood can be reused and only the disturbed subgraph is solved to ensure the shortest downtime.

[0090] Simultaneously, an online recursive estimation of the temperature compensation coefficient is introduced to improve the robustness of thermal compensation: based on the current sensing temperature. With the length / time of the cut already performed As the independent variable in the regression, the actual profile deviation after cutting is used as the observation, and the linear expansion coefficient is updated using the RLS / Kalman method. :

[0091] ;

[0092] in, For the deviation of the profile dimensions measured online, For adaptive gain;

[0093] The system is configured with multiple trigger thresholds: when the positioning error or temperature rise deviation exceeds the first threshold, only coordinate correction is performed; when it exceeds the second threshold, local replanning is performed; when it exceeds the third threshold, a cooling / pause and recalibration process is inserted (which can automatically add / replace reference markers), and abnormal segments are marked and sent back to the feedback self-learning module to improve the prior adaptability of subsequent typesetting strategies under similar deformation patterns.

[0094] This fallback solution ensures the feasibility of maintaining dimensional and clearance constraints under abnormal disturbances, significantly reducing the risk of rework and scrap. At the same time, it incorporates the deviation of compensation parameters into the online estimation, avoiding the accumulation of errors in long batch processing.

[0095] Beneficial effects

[0096] In summary, this invention provides a digital cutting and layout optimization method and system for ship steel plates. Through the above technical solutions, the material utilization rate and production efficiency of ship steel plate cutting are significantly improved. Specific beneficial effects include, but are not limited to:

[0097] 1. Maximize material utilization and reduce costs: By optimizing the layout through intelligent algorithms, the utilization rate of steel plates is brought close to 100% (only the cutting gap loss remains). Compared with traditional layout methods, it significantly reduces the generation of scrap steel after cutting, achieves "zero surplus" material cutting, and greatly reduces material waste and production costs.

[0098] 2. Improve the intelligence level of typesetting: Introduce an optimization strategy that integrates deep reinforcement learning and genetic algorithms, enabling the system to autonomously explore the best part layout scheme and automatically generate high-quality typesetting schemes without manual intervention. At the same time, a feedback self-learning mechanism is further introduced, allowing the system to adjust the optimization strategy from each cutting feedback, becoming more intelligent with use and continuously adapting to different steel plate shapes and part combinations, thereby improving typesetting quality.

[0099] 3. Improve cutting accuracy and quality: By combining projection and vision, the actual position and posture of the steel plate are accurately calibrated before cutting. Automatic calibration is achieved within the accuracy threshold, eliminating the need for tedious manual adjustments. This ensures that the cutting path is precisely aligned with the steel plate, reducing cutting defects caused by steel plate placement deviations. It guarantees safe implementation even in extremely compact "zero margin" layouts. Combined with kerf gap constraints and thermal deformation compensation, it ensures accurate part dimensions and good edge quality. At the same time, the 2-Opt algorithm is used to optimize the cutting sequence when planning the cutting path to reduce idle travel, making the cutting machine's movement path more economical and efficient, reducing machine idle time and waiting, and improving cutting operation efficiency.

[0100] 4. Achieve system integration and digitalization: Each functional module communicates through standard OPC-UA and interacts with data in JSON format, realizing an integrated digital twin system. The information flow of each link, including scanning, optimization, cutting, and feedback, is seamlessly connected, avoiding delays and errors caused by manual data transmission. Real-time data upload via MQTT enables managers to understand the production status in real time, realizing real-time monitoring and optimization of the intelligent manufacturing process.

[0101] 5. High versatility and adaptability: This system can be applied to whole plates and scrap plates of various shapes. Regardless of the size or shape of the plate, it can achieve automatic processing and optimization. At the same time, the modular design and standard interface facilitate the connection with different cutting equipment and production management systems, which has good versatility. Through continuous self-learning, the system can adapt to the differences caused by the characteristics of different batches of materials and environmental factors, and has strong robustness.

[0102] Therefore, the technical solution of this invention has significant practical value in the field of shipbuilding, which can significantly reduce material waste, reduce labor intensity and human error, improve production efficiency and intelligence level, and is in line with the development direction of green shipbuilding and lean production. Attached Figure Description

[0103] Figure 1This is a flowchart illustrating the digital cutting and layout optimization method for ship steel plates in Embodiment 2 of the present invention. Detailed Implementation

[0104] For the purpose of describing the system structure and workflow of the present invention in detail below with reference to the accompanying drawings and specific embodiments, it should be understood that the following embodiments are intended to aid in understanding the present invention, but do not constitute a limitation on the scope of protection of the present invention.

[0105] Example 1

[0106] This embodiment provides a digital cutting and layout optimization system for ship steel plates, including:

[0107] Plate morphology acquisition module: Collects morphology data of the steel plate to be cut and generates a processing area model;

[0108] Intelligent layout optimization module: Based on the processing area model and the specification data of the target part, it adopts a deep reinforcement learning-genetic hybrid algorithm as the solution strategy and outputs the optimal layout scheme;

[0109] Cutting path generation module: Generates CNC cutting path and corresponding CNC code according to the preferred layout scheme, and performs kerf gap constraint and thermal deformation compensation on the CNC cutting path;

[0110] Projection positioning calibration module: Based on the deviation between the ideal coordinates of the reference mark points in the processing area and their actual coordinates projected onto the steel plate to be cut, the module performs positioning calibration on the steel plate to be cut.

[0111] Feedback self-learning module: Executes cutting instructions, evaluates cutting results, and provides feedback on deviation data to adjust and optimize strategy parameters.

[0112] The plate morphology acquisition module includes a laser scanner or industrial camera installed on the CNC cutting machine. The intelligent layout optimization module and the feedback self-learning module run on an industrial computer or server. The cutting path generation module is integrated into the control system of the CNC cutting machine. The projection positioning calibration module includes a digital projector and an industrial camera. The entire system is connected via Ethernet or wireless network. Each functional module exchanges data according to the OPC-UA protocol to form a platform for overall collaborative work.

[0113] Example 2

[0114] See Figure 1 This embodiment provides a method for optimizing the digital cutting and layout of ship steel plates based on the system in Embodiment 1, including:

[0115] I. Collecting Steel Plate Morphology Data

[0116] First, the steel plate to be cut is placed on the CNC cutting machine. If it is a whole plate, it can be scanned directly. If it is a leftover irregularly shaped scrap plate from the previous cut, it is also placed on the machine. The plate shape acquisition module is activated, and the laser scanner moves along the surface of the steel plate to scan and measure the edges, collecting a large number of edge point coordinates. Alternatively, the outline image of the steel plate can be acquired through the camera above. After image processing, the edges are extracted. The plate shape acquisition module processes the collected edge data to obtain the steel plate processing area. The digital model (represented by the coordinates of polygon vertices) is expressed as follows:

[0117] ;

[0118] in, These are the coordinates of the processing area boundary.

[0119] In this embodiment, the scanning results show that the outline of the scrap plate is an irregular polygon with an approximate length of 2.5 meters and a width of 1.2 meters. At this time, the outline can be divided into several line segments and recorded, and the coordinate system can be calibrated for use by subsequent modules.

[0120] II. Optimize Parts Layout Scheme

[0121] Input the processing area model and the target parts list (including the quantity and specifications of the parts to be cut) into the intelligent layout optimization module. This module uses a deep reinforcement learning agent and a genetic algorithm to solve for the optimal layout, as follows:

[0122] The intelligent typesetting optimization module uses a Markov decision process to model the typesetting problem, and its states and actions can be defined as follows: State This indicates information about the currently placed parts and remaining space on the steel plates (including the layout of occupied space, the shape characteristics of the remaining available areas, etc.), with the blank processing area without any parts yet serving as the initial state. The parts are arranged one by one in the processing area to form a sequence of actions. N is the total number of parts to be typed, and the action is... This indicates the selection of a part to be placed and the determination of its placement method on the steel plate, in the action. The upcoming After each part is placed in the machining area, its new status is recorded. By designing reasonable state and action spaces, reinforcement learning agents can effectively explore the placement order and posture combinations of different parts and correspond them to specific board layouts.

[0123] In the early stages of the action, parts are arranged randomly or based on heuristic rules. For example, the largest part is placed in a corner of the steel plate first, and then smaller parts are placed one by one. The reward is calculated by simulating the result after each placement. Positive reward is given if the material utilization rate increases or there is no conflict, and negative reward is given if the material utilization rate decreases or there is a conflict. The reinforcement learning agent adjusts its strategy accordingly and keeps trying different layout methods.

[0124] Specifically, the function expression for the reward is as follows:

[0125] ;

[0126] in, Let be the reward value for the t-th action. Let be the increment of material utilization rate after the t-th action compared to the previous step. This is a conflict indicator function. It returns 1 if a newly placed part overlaps with an already placed part, and 0 otherwise. This represents the current cumulative cutting time. For the maximum allowable cutting time, These are the corresponding weight coefficients;

[0127] Meanwhile, the genetic algorithm maintains a set of possible layouts (as a population of policy network parameters), updates the policy network parameters for a set of feasible layouts given by the reinforcement learning agent, and after several iterations, performs crossover and mutation operations on several of the better layouts to generate new candidate schemes.

[0128] Specifically, the deep reinforcement learning-genetic hybrid algorithm includes reinforcement learning based on policy gradient update and global search using a genetic algorithm. The expression for the policy gradient update is as follows:

[0129] ;

[0130] in, For policy network parameters, For policy function Regarding parameters gradient, For the policy network in state Select action The probability distribution function, The probability distribution function with respect to the parameters The logarithmic derivative, For strategy Mathematical expectation of the next action-state distribution The advantage function, used to evaluate the merits of a particular action relative to the average level of that state, is expressed as follows:

[0131] ;

[0132] in, As a discount factor, and The value network is for the current state. Next state The estimated state value;

[0133] The genetic algorithm, based on the current policy network parameter population, generates new policy individuals through operations such as selection, crossover, and mutation, as shown in the following expression:

[0134] ;

[0135] in, These are the parameters of the new policy network generated after the genetic operation. and Let be the parameters of the i-th and j-th policy individuals selected from the current policy population, respectively. The cross ratio coefficient and , The Gaussian noise variation term (a small random vector with zero mean);

[0136] After completing a preset number of iterations of training and evaluation, the intelligent layout optimization module outputs a set of optimized part layout parameters as the optimal layout scheme. The component layout parameters include the rotation angles of each component. And the amount of translation in the horizontal direction Vertical translation ;

[0137] The preferred typesetting scheme The condition for zero residual is satisfied, and the expression is as follows:

[0138] ;

[0139] in, Let be the region where the t-th part is located after rotation and translation operations, and Area be a function to calculate the area of ​​the region.

[0140] In this embodiment, the target parts list includes 10 parts with known specifications, and a preferred layout scheme is used. To place the largest trapezoidal plate in the upper left area of ​​the scrap plate by rotating it 15°, and the second largest arc-shaped plate in the lower right corner by rotating it 33°, with the remaining small parts cleverly interspersed in the gaps, this solution has a compact and reasonable layout. All parts are placed within the scrap plate area without overlapping each other, achieving a material utilization rate of over 99%. The unused plate space is only the extremely fine grid skeleton formed by laser cutting.

[0141] III. Cutting Path Planning and Code Generation

[0142] The cutting path generation module generates the cutting path based on the preferred layout scheme. Determine the position of each part in the processing area, divide the outline of each part into multiple line segments according to the vertices, and arrange all line segments in a certain order to generate the cutting path;

[0143] To improve cutting efficiency, this module uses the 2-Opt algorithm to optimize the cutting order. That is, based on the initial path scheme, the cutting order of local paths is repeatedly swapped to reduce the length of empty paths until the path can no longer be optimized or the optimization increment is lower than the threshold.

[0144] In this embodiment, the initial cutting sequence is to cut the steel plate piece by piece from the upper left to the lower right. After optimization by the 2-Opt algorithm, it is adjusted to cut the central parts of the steel plate first and then the edge parts, which reduces the movement distance of the cutting head in the non-working area.

[0145] Insert inlet / outlet leads, common-edge cutting, and other process paths (if necessary) into the determined cutting path, and the cutting path planning must meet the following conditions:

[0146] ① For any two adjacent parts, maintain ,in, Let be the minimum distance between the contour edges of part i and part j. This is the minimum clearance allowed by the process.

[0147] ② The cutting path length is corrected and compensated based on the thermal expansion effect of steel, i.e. ,in, To compensate for the path length, is the coefficient of linear expansion of steel. This represents the original length of the cutting path before compensation. To increase the temperature of the cutting area;

[0148] In this embodiment, the minimum allowable gap in the process is 5mm, while the closest distance between adjacent parts is controlled to be 5.2mm or more. The thermal compensation is calculated based on the coefficient of thermal expansion of the steel and the cutting parameters: the coefficient of linear expansion of the steel... Local temperature rise during plasma cutting ℃, for each length The straight cutting segment, estimate Therefore, when generating CNC code, each segment of the cut straight line is lengthened proportionally (curves are approximated by segmentation).

[0149] For parts with long outlines on steel plates, the cutting path can also adopt a segmented intermittent cutting method, that is, cut a part first and then pause, wait for the material to cool down and then continue cutting, so as to reduce cumulative deformation.

[0150] The corresponding CNC code is generated based on the cutting path. The cutting path generation module exports the corresponding code file, which contains cutting instructions for all part contours, tool movement paths, and control commands such as speed and piercing. This code file is then uploaded to the CNC cutting machine controller for execution.

[0151] IV. Cutting Positioning Calibration

[0152] Before the cutting machine begins actual operation, the projection positioning calibration module is activated to accurately calibrate the position of the steel plate. First, the digital projector selects three reference marker points in the processing area and projects them onto the surface of the steel plate to be cut (e.g., in the processing area). A crosshair is projected at the lower left, upper right, and center positions. The design coordinates of these points in the steel plate coordinate system are known, denoted as . Then, an industrial camera fixed to the machine tool is used to photograph the steel plate and its projected markings. A visual algorithm is then used to identify and calculate the true coordinates of the projected points on the steel plate, denoted as... ;

[0153] Based on the coordinate differences of corresponding points in each group, the rigid body transformation parameters of each projection point relative to the ideal position are calculated, as shown in the following expressions:

[0154] ;

[0155] in, For rotation angle, This represents the horizontal translation. This represents the vertical translation amount;

[0156] The least squares method is used to fit the solution of the rigid body transformation to obtain estimates of the translation and rotation angles. This describes the offset of the steel plate relative to the ideal position, calculates the positioning error, and compares it with a preset accuracy threshold. The expression is as follows:

[0157] ;

[0158] If all the above equations are true, the calibration is considered successful. The data is uploaded to the CNC cutting machine controller, which triggers the cutting execution command and performs coordinate transformation compensation on the cutting path to be executed. Conversely, if at least one of the above formulas is not true, the calibration is deemed to have failed, and an alarm is triggered to request adjustment of the steel plate placement position or addition of more reference markers to recalculate the positioning error until the accuracy requirements are met.

[0159] In this embodiment, it was found that the actual position of the steel plate was offset to the right from the ideal position. It shifted downwards. And rotated clockwise None of the values ​​exceeded the accuracy threshold, so the calibration was passed. The projection positioning calibration module then sent the offset parameters to the CNC cutting machine controller, which used these parameters to perform coordinate transformation compensation on the cutting path to be executed, in order to ensure that the actual cutting trajectory was aligned with the steel plate.

[0160] V. Perform the cutting

[0161] After completing the positioning calibration, the CNC cutting machine begins to execute the cutting operation according to the generated CNC code;

[0162] In this embodiment, the CNC cutting machine uses a plasma cutting head to cut the outline of the parts segment by segment, starting from the edge of the steel plate or the pre-perforated area. Since the layout and path sequence of the parts in the layout scheme have been optimized, the cutting process is smooth. The cutting head completes the cutting of each part outline in turn and moves to the starting point of the next part by moving the shortest path. The cutting of the entire plate is completed in about 15 minutes. During this period, the process control module of the cutting machine adjusts the cutting size appropriately according to the thermal compensation command, so that the actual size of the cut parts is highly consistent with the drawing requirements. After the cutting is completed, all 10 parts are successfully separated from the plate. The operator removes the parts, and the remaining material has a mesh-like fine skeleton structure, which is left on the table for recycling or further processing.

[0163] VI. Results Feedback and Self-Learning

[0164] After cutting, the feedback self-learning module is activated to evaluate the cutting results and collect data. The morphological data of the remaining material is acquired using a laser scanner or machine vision. The actual remaining material is compared with the theoretical remaining material expected in the optimized layout scheme, and the overlap index between the two is calculated. The expression is as follows:

[0165] ;

[0166] Where IoU is the overlap index, and Area is a function for calculating the area of ​​a region. for For theoretical surplus ;

[0167] When the IoU is less than the preset overlap threshold, the difference area between the actual surplus material and the theoretical surplus material is rasterized and discretized to obtain a difference raster image, as shown in the following expression:

[0168] ;

[0169] in, For set difference operations, For discretization operations;

[0170] The differential raster image is input into the neural network of the reinforcement learning agent for training, so that the algorithm can identify situations where differential regions are not effectively utilized in the scheme, and adaptively adjust the subsequent typesetting strategy parameters accordingly.

[0171] In this embodiment, actual scanning revealed that some areas of the leftover material were slightly larger than theoretically predicted. For example, a small strip of sheet metal that was originally planned to be cut between two parts did not completely detach due to slight deformation caused by thermal stress, forming an actual leftover material fragment. The feedback module analyzed this difference: The intersection area of ​​the surplus material area and the theoretical surplus material area is 200 cm². 2 The union area is 220 cm². 2 Then calculate Although most areas met expectations, the IoU was below the preset threshold of 95%, indicating that there is still considerable room for improvement.

[0172] The feedback self-learning module extracts this discrepancy region and rasterizes it into an image. Then, it updates the optimization algorithm's strategy based on the discrepancy type: the discrepancy raster image is input into the neural network of the reinforcement learning agent for training, enabling it to recognize situations where such narrow gap regions may not have been effectively utilized in previous solutions. The next time a similar steel plate and part arrangement requirement is encountered, the agent will tend to adjust the part spacing or cutting order to avoid this problem. At the same time, the system records key data of this cutting (utilization rate 99%, IoU approximately 0.91, cutting time approximately 15 minutes, etc.) and publishes it to the monitoring server via MQTT for engineers to view and analyze.

[0173] After the above feedback learning, when the system moves on to the next piece of material for layout and cutting, the reinforcement learning agent has updated the strategy parameters based on the previous experience, and the initial population of the genetic algorithm also includes the improved layout scheme factor, which makes the starting point of this optimization better. After multiple iterations, the self-learning effect of the system gradually becomes apparent: for subsequent similar types of leftover material, the success rate of the layout scheme increases, and the average IoU after cutting gradually increases (from 91% to 95%, 98%... and even 100%), indicating that the actual leftover material almost completely meets the expected scheme, the material utilization rate is also maintained at an extremely high level, and there are no longer any uncut small fragments around the parts. Through such continuous evolution, the system of this invention optimizes itself in continuous practice, achieving the ideal "zero leftover material" cutting effect and highly stable cutting quality.

[0174] It should be noted that the above embodiments of the present invention are specific application schemes provided to facilitate understanding of the present invention, but the present invention is not limited thereto. For those skilled in the art, various equivalent substitutions and improvements can be made to the specific implementation of the system without departing from the principles of the present invention. For example, different forms of reinforcement learning algorithms (such as Deep Q-Network (DQN), PPO, etc.), more complex evolutionary algorithms, or other types of scanning and projection devices can be used. These variations should all be considered within the scope of protection of the present invention. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for optimizing the digital cutting and layout of ship steel plates, characterized in that, include: Step 1: Collect the morphological data of the steel plate to be cut and generate a processing area model; Step 2: Based on the processing area model and the specification data of the target part, a deep reinforcement learning-genetic hybrid algorithm is used as the solution strategy to output the optimal layout scheme; include: Input the processing area model and the specification data of the target part, and use a deep reinforcement learning agent and a genetic algorithm to explore the optimal layout of the target part on the steel plate, that is, all the layout parts do not overlap and the unused space in the processing area is minimized, as follows: The reinforcement learning agent arranges the parts one by one in the processing area to form a sequence of actions. N represents the total number of parts that need to be laid out, and the blank processing area with no parts to be placed is taken as the initial state. In action The upcoming After each part is placed in the machining area, its new status is recorded. ; In the early stages of the action, parts are arranged randomly or based on heuristic rules. After each arrangement, a reward is calculated based on the material utilization rate and the conflict of parts. A positive reward is given if the material utilization rate increases or there is no conflict, and a negative reward is given if the material utilization rate decreases or a conflict occurs. The reinforcement learning agent adjusts its strategy accordingly to try different layout methods. A genetic algorithm is used to update the policy network parameters of feasible layout schemes given by the reinforcement learning agent. After several iterations, crossover and mutation operations are performed on several of the better layout schemes to generate new layout schemes. After completing a preset number of iterations of training and evaluation, a set of optimized part layout parameters is output as the preferred layout scheme. The component layout parameters include the rotation angles of each component. And the amount of translation in the horizontal direction Vertical translation ; Step 3: Generate CNC cutting paths and corresponding CNC codes based on the preferred layout scheme, and perform kerf gap constraint and thermal deformation compensation on the CNC cutting paths; including: The position of each part in the processing area is determined according to the preferred layout scheme. The outline of each part is divided into multiple line segments according to nodes. All line segments are arranged in sequence to generate a CNC cutting path. The planning of the CNC cutting path must meet the following requirements: ① For any two adjacent parts, maintain ,in, Let be the minimum distance between the contour edges of part i and part j. This is the minimum clearance allowed by the process. ② The cutting path length is corrected and compensated based on the thermal expansion effect of steel, i.e. ,in, To compensate for the path length, is the coefficient of linear expansion of steel. This represents the original length of the cutting path before compensation. To increase the temperature of the cutting area; Generate the corresponding CNC code based on the CNC cutting path, and upload the code file to the CNC cutting machine controller; Step 4: Based on the deviation between the ideal coordinates of the reference marker points within the processing area and their actual coordinates projected onto the steel plate to be cut, perform positioning calibration on the steel plate to be cut; Step 5: Execute the cutting command, evaluate the cutting results, and provide feedback on deviation data to adjust and optimize strategy parameters.

2. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 1, characterized in that, Step 1 includes: The steel plate to be cut is a whole plate or a leftover plate from the previous cut. The outline coordinate point cloud of the steel plate is obtained by a laser scanner or machine vision, and a processing area model is generated, as shown in the following expression: ； in, These are the coordinates of the processing area boundary.

3. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 1, characterized in that, Step 4 includes: Select at least three reference marker points within the processing area and project them onto the surface of the steel plate to be cut, then record the ideal coordinates of the reference marker points. ; Machine vision is used to acquire images of the steel plate to be cut and to identify the true coordinates of the projection points on the steel plate. ; Based on the coordinate differences, the rigid body transformation parameters of each projection point relative to the ideal position are calculated, as shown in the following expressions: ； in, For rotation angle, This represents the horizontal translation. This represents the vertical translation amount; The least squares method is used to fit the solution of the rigid body transformation to obtain estimates of the translation and rotation angles. , , The positioning error is calculated and compared with a preset accuracy threshold, as shown in the following expression: ； ； If all the above equations are true, the calibration is considered successful. , , The data is uploaded to the CNC cutting machine controller, which triggers the cutting execution command and performs coordinate transformation compensation on the cutting path to be executed. Conversely, if at least one of the above formulas is not true, the calibration is deemed to have failed, and an alarm is triggered to request adjustment of the steel plate placement position or addition of more reference markers to recalculate the positioning error until the accuracy requirements are met.

4. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 1, characterized in that, Step 5 includes: The CNC cutting machine performs cutting operations according to the generated CNC code. The cutting head cuts out the outline of the part segment by segment from the edge of the steel plate or the pre-pierced part. It moves along the shortest path to the starting point of the next part outline. After the operation is completed, all parts are removed and the remaining material is left on the cutting machine table. The morphological data of the scrap material is acquired using a laser scanner or machine vision. The actual scrap material is compared with the theoretical scrap material expected in the optimized layout scheme, and the overlap index between the two is calculated. The expression is as follows: ； Where IoU is the overlap index, and Area is a function for calculating the area of ​​a region. for For theoretical surplus ; When the IoU is less than the preset overlap threshold, the difference area between the actual surplus material and the theoretical surplus material is rasterized and discretized to obtain a difference raster image, as shown in the following expression: ； in, For set difference operations, For discretization operations; The differential raster image is input into the neural network of the reinforcement learning agent for training, so that the algorithm can identify situations where differential regions are not effectively utilized in the scheme, and adaptively adjust the subsequent typesetting strategy parameters accordingly.

5. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 1, characterized in that, The function expression for calculating the reward in step 2 is as follows: ； in, Let be the reward value for the t-th action. Let be the increment of material utilization rate after the t-th action compared to the previous step. This is a conflict indicator function. It returns 1 if a newly placed part overlaps with an already placed part, and 0 otherwise. This represents the current cumulative cutting time. For the maximum allowable cutting time, These are the corresponding weighting coefficients.

6. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 1, characterized in that, The deep reinforcement learning-genetic hybrid algorithm includes reinforcement learning based on policy gradient update and global search using a genetic algorithm. The expression for the policy gradient update is as follows: ； in, For policy network parameters, For policy function Regarding parameters gradient, For the policy network in state Select action The probability distribution function, The probability distribution function with respect to the parameters The logarithmic derivative, For strategy Mathematical expectation of the next action-state distribution The advantage function, used to evaluate the merits of a particular action relative to the average level of that state, is expressed as follows: ； in, As a discount factor, and The value network is for the current state. Next state The estimated state value.

7. The method for optimizing the digital cutting and layout of ship steel plates as described in claim 6, characterized in that, The genetic algorithm, based on the current policy network parameter population, generates new policy individuals through selection, crossover, and mutation operations, as shown in the following expression: ； in, These are the parameters of the new policy network generated after the genetic operation. and Let be the parameters of the i-th and j-th policy individuals selected from the current policy population, respectively. The cross ratio coefficient and , This is a Gaussian noise variation term.

8. A digital cutting and layout optimization system for ship steel plates, characterized in that, include: Plate morphology acquisition module: Collects morphology data of the steel plate to be cut and generates a processing area model; Intelligent layout optimization module: Based on the processing area model and the specification data of the target part, it adopts a deep reinforcement learning-genetic hybrid algorithm as the solution strategy and outputs the optimal layout scheme; Cutting path generation module: Generates CNC cutting path and corresponding CNC code according to the preferred layout scheme, and performs kerf gap constraint and thermal deformation compensation on the CNC cutting path; Projection positioning calibration module: Based on the deviation between the ideal coordinates of the reference mark points in the processing area and their actual coordinates projected onto the steel plate to be cut, the module performs positioning calibration on the steel plate to be cut. Feedback self-learning module: Executes cutting instructions, evaluates cutting results, and provides feedback on deviation data to adjust and optimize strategy parameters; To achieve the digital cutting and layout optimization method for ship steel plates as described in any one of claims 1-7.

Images