A metal 3D printing industrial control system

Abstract

This invention relates to the field of metal 3D printing and industrial automation control technology, specifically to a metal 3D printing industrial control system, comprising: a data acquisition module for acquiring multimodal time-series data of the target processing area and external environmental interference data, and sending the multimodal time-series data and the external environmental interference data to a risk quantification assessment module; a risk quantification assessment module for sending a safety margin to a control strategy generation module based on the multimodal time-series data and the external environmental interference data; a control strategy generation module for acquiring a preset safety margin threshold and comparing the safety margin with the safety margin threshold; and an instruction execution closed-loop module for acquiring the processing scheduling instruction and injecting the processing scheduling instruction into the underlying CNC execution flow. This invention effectively avoids the risk of logic inversion under extremely dangerous conditions.

Description

Technical Field

[0001] This invention relates to the field of metal 3D printing and industrial automation control technology, specifically to a metal 3D printing industrial control system. Background Technology

[0002] In the current metal 3D printing manufacturing environment, especially under extreme manufacturing throughput conditions, due to continuous high-frequency energy injection and uneven local cooling in the processing area, residual thermal stress will continuously accumulate inside the component and is difficult to release on its own.

[0003] To ensure safety during the processing, existing solutions generally employ control logic based on conventional static rules and static single thresholds for risk assessment and processing scheduling. While this approach provides basic system assurance under normal stable operating conditions, it lacks cascading perception of multidimensional external environmental disturbances and internal thermodynamic characteristics. Furthermore, it struggles to address the nonlinear thermal hysteresis effect that occurs after localized overheating of metallic materials, resulting in sluggish risk assessment and severe dynamic mismatch. This control logic is highly susceptible to causing the system to make incorrect judgments and continue executing basic processing commands under extremely dangerous conditions, ultimately leading to thermodynamic runaway, causing macroscopic mechanical warping and deformation of components, and triggering severe system-level interference failures such as powder-spreading scraper collisions.

[0004] Therefore, how to achieve in-situ dynamic sensing of the risk of underlying thermodynamic mismatch under complex operating conditions, and effectively prevent system-level interference disasters caused by thermodynamic runaway under extreme throughput, has become an urgent technical problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide an industrial control system for metal 3D printing, which solves the following technical problems:

[0006] It solves the system-level interference failure caused by thermodynamic runaway under extreme manufacturing throughput, effectively alleviates the continuous accumulation of thermal stress and eliminates the risk of excessive local heat accumulation, thereby avoiding severe powder spreading scraper collision events triggered by macroscopic mechanical warping deformation.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An industrial control system for metal 3D printing includes: a data acquisition module, a risk quantification and assessment module, a control strategy generation module, and an instruction execution closed-loop module;

[0009] The data acquisition module is used to acquire multimodal time-series data of the target processing area and external environmental interference data during the layer-by-layer processing, and send the multimodal time-series data and external environmental interference data to the risk quantification and assessment module.

[0010] The risk quantification assessment module is used to calculate the global potential stress entropy based on the multimodal time series data and the external environmental interference data through a preset risk quantification model, and to predict the safety margin of the distance-triggered physical interference failure event based on the global potential stress entropy, and send the safety margin to the control strategy generation module.

[0011] The control strategy generation module is used to obtain a preset safety margin threshold and compare the safety margin with the safety margin threshold.

[0012] In response to the safety margin being less than the safety margin threshold, the control strategy generation module is configured to generate a processing scheduling instruction containing a state recovery compensation instruction sequence; or, in response to the safety margin being greater than or equal to the safety margin threshold, the control strategy generation module is configured to generate a processing scheduling instruction containing a basic processing efficiency instruction sequence.

[0013] The instruction execution closed-loop module is used to acquire the machining scheduling instruction and inject the machining scheduling instruction into the execution flow of the underlying CNC system used to drive the laser energy source and / or motion actuator, so as to dynamically adjust the physical machining state.

[0014] Preferably, the multimodal time-series data includes the time-series characteristics of temperature distribution and the visual feedback characteristics of morphology of the target processing area; the external environmental interference data includes the power grid frequency band fluctuation characteristics and the turbulence characteristics of the protective airflow.

[0015] Preferably, the risk quantification assessment module is further configured to: extract transient thermal gradient features from the multimodal time-series data and retrieve the residual stress genetic sequence passed from the historical printing layer to the current layer; extract high-frequency noise features within a preset sensitive frequency band from the external environmental interference data; and input the transient thermal gradient features, the residual stress genetic sequence, and the high-frequency noise features into the risk quantification model to output the global potential stress entropy.

[0016] Preferably, the risk quantification assessment module is further configured to: obtain a preset critical parameter for the material yield strength; calculate the difference between the global potential stress entropy characterizing the equivalent mechanical stress and the critical parameter for the material yield strength; and determine the safety margin of the distance-triggered physical interference failure event based on the difference.

[0017] Preferably, the control strategy generation module is further configured to: obtain the initial motion rate parameter and initial energy output parameter of the current machining layer from the pre-programmed instructions in the underlying CNC system when generating the state recovery compensation instruction sequence; reduce the initial motion rate parameter to generate the target motion rate parameter; and generate the state recovery compensation instruction sequence based on the target motion rate parameter.

[0018] Preferably, the control strategy generation module is further configured to: generate an idle mobile heat dissipation path with no energy output when the state recovery compensation instruction sequence is generated; and insert the idle mobile heat dissipation path into the state recovery compensation instruction sequence.

[0019] Preferably, the control strategy generation module is further configured to: when generating the state recovery compensation instruction sequence, obtain the current slice file corresponding to the current processing task, and obtain the initial local scan vector strategy from the current slice file; change the initial local scan vector strategy to generate a target local scan vector strategy; and integrate the target local scan vector strategy into the state recovery compensation instruction sequence.

[0020] Preferably, the control strategy generation module is further configured to: obtain a preset equipment throughput index and a preset hardware energy consumption index when generating the basic processing efficiency instruction sequence; and generate the basic processing efficiency instruction sequence based on the equipment throughput index and the hardware energy consumption index using a preset optimization algorithm.

[0021] Preferably, the instruction execution closed-loop module is further configured to: obtain the current instruction running state of the execution flow of the underlying CNC system; and smoothly connect the machining scheduling instruction with the current instruction running state without interrupting the continuous machining process of the execution flow of the underlying CNC system.

[0022] Preferably, physical interference failure events include macroscopic mechanical warping deformation caused by local thermal stress release and powder spreading scraper collision events triggered by the macroscopic mechanical warping deformation.

[0023] The beneficial effects of this invention are:

[0024] 1. Achieving in-situ dynamic perception of underlying thermodynamic runaway risk: This system integrates multimodal time-series data and external environmental disturbance data to calculate the global potential stress entropy and safety margin. This multi-dimensional cascade perception breaks through the limitations of conventional static rules and realizes in-situ dynamic construction and assessment of underlying thermodynamic mismatch risk, effectively avoiding the risk of logical inversion under extremely dangerous conditions.

[0025] 2. Accurately quantifying the safety margin to overcome the kinetic mismatch defect: The system calculates the safety margin based on the difference between the global potential stress entropy and the critical parameter of the material's yield strength. This overcomes the kinetic mismatch defect of the traditional static threshold when facing the nonlinear thermal hysteresis effect of metallic materials, and accurately assesses the remaining buffer space between the current accumulated thermal stress and the material's mechanical collapse boundary.

[0026] 3. Multi-dimensional dynamic compensation and active intervention in thermodynamic runaway: When the safety margin falls below the threshold, the system actively generates a state recovery compensation command sequence; by dynamically reducing the motion rate, inserting an unloaded moving heat dissipation path, and reconstructing the local scan vector, the system actively suppresses heat injection and extends heat dissipation time without cutting off the energy source, thus accurately eliminating the risk of excessive local heat accumulation.

[0027] 4. Ensure seamless and safe switching of control commands under extreme scheduling. The instruction execution closed-loop module smoothly splices the machining scheduling command with the current command running state without interrupting the continuous machining process of the underlying CNC execution flow. This mechanism avoids the severe vibration of the machine tool caused by sudden changes in commands during extreme risk avoidance scheduling, and realizes the smooth switching of control strategies under complex working conditions.

[0028] 5. Fundamentally avoid system-level interference disasters and collision accidents. Through the above dynamic perception and control strategies, the system effectively avoids macroscopic mechanical warping deformation caused by local thermal stress release. This prevents the powder spreading scraper collision event triggered by warping deformation from the physical source, completely blocks system-level interference disasters, and ensures production safety under extreme throughput. Attached Figure Description

[0029] The invention will now be further described with reference to the accompanying drawings.

[0030] Figure 1 This is a schematic diagram of a module of an industrial control system for metal 3D printing provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 This embodiment provides an industrial control system for metal 3D printing, which aims to solve the system-level interference disaster caused by thermodynamic runaway under extreme manufacturing throughput. The system includes a data acquisition module, a risk quantification and assessment module, a control strategy generation module, and an instruction execution closed-loop module.

[0033] The data acquisition module acquires multimodal time-series data and external environmental interference data of the target processing area, and sends the multimodal time-series data and external environmental interference data to the risk quantification and assessment module. During the processing, the data acquisition module continuously performs high-frequency sampling of the underlying physical environment to capture extremely small changes in operating conditions. Within a set millisecond time window, it calculates the data change rate of adjacent sampling points and removes abnormal spikes with change rates exceeding physical limits, thereby effectively suppressing the interference of sensor noise on subsequent control logic.

[0034] The risk quantification assessment module receives the cleaned data, calculates the global potential stress entropy based on multimodal time series data and external environmental interference data through a preset risk quantification model, and predicts the safety margin of the distance to trigger physical interference failure events based on the global potential stress entropy. The safety margin is then sent to the control strategy generation module. Here, the global potential stress entropy represents the total residual thermal stress accumulated inside the component due to uneven local cooling that has not yet been released. This design, which transforms the cross-influence of multi-dimensional data into a single quantitative indicator, breaks through the limitations of conventional static rules and realizes the in-situ construction and perception of the underlying thermodynamic mismatch risk.

[0035] The control strategy generation module obtains a preset safety margin threshold and compares the safety margin with the safety margin threshold. In response to the safety margin being less than the safety margin threshold, the control strategy generation module generates a state recovery compensation instruction sequence as a processing scheduling instruction to force the equipment into an active cooling and stress release state. Conversely, in response to the safety margin being greater than or equal to the safety margin threshold, the control strategy generation module generates a basic processing efficiency instruction sequence as a processing scheduling instruction.

[0036] Since the safety margin is designed to be negatively correlated with stress entropy, that is, the higher the stress entropy, the smaller the safety margin value, when the safety margin falls below the safety margin threshold, it means that the system is on the verge of an extremely dangerous state of interference and collision, thus accurately triggering the forced cooling logic and avoiding the risk of logic reversal where the system is judged to be safe and continues to execute basic processing instructions in an extremely dangerous state.

[0037] The instruction execution closed-loop module acquires the machining scheduling instructions and injects them into the underlying CNC execution flow. The upper half of the display on the industrial control terminal uses a dynamic heat map to indicate the current global potential stress entropy level and safety margin value, while the lower half displays differentiated instructions based on the currently generated machining scheduling instructions. This allows operators to manually intervene or trigger an emergency isolation process with one click when necessary. This embodiment demonstrates the adaptability of the system cascading mechanism under harsh working conditions and verifies the robustness of this technical solution under different interference risks.

[0038] In a preferred embodiment of the present invention, the metal 3D printing industrial control system is further defined. The multimodal time-series data acquired by the data acquisition module includes the time-series characteristics of temperature distribution and the visual feedback characteristics of morphology of the target processing area. The external environmental interference data includes the power grid frequency band fluctuation characteristics and the turbulence characteristics of the protective airflow.

[0039] The temporal characteristics of temperature distribution are specifically manifested as the absolute temperature gradient and cooling rate of the center of the molten pool and the heat-affected zone within a continuous time slice, while the visual feedback characteristics of morphology are manifested as the geometric aspect ratio of the molten pool and the flow field fluctuation pattern of the liquid metal on the surface. These characteristics directly reflect the microscopic thermodynamic equilibrium state between the current laser energy injection and the heat dissipation of the underlying material.

[0040] External environmental interference data reflects the asymmetric disturbances of the complex factory environment on the printing process. Among them, the power grid frequency band fluctuation characteristics are manifested as small voltage dips and frequency shifts caused by the start-up and shutdown of heavy machinery in the factory. These shifts directly lead to transient jitters in the laser output power. The protective airflow turbulence characteristics are manifested as local nonlinear eddies generated by the unstable air pressure in the argon circulation system in the printing chamber. These eddies instantly change the convective heat transfer coefficient of the target processing area.

[0041] This design, which cascades monitoring of external environmental disturbances and internal thermodynamic characteristics, is similar to the stringent control logic for micro-environment vibration and aerosol concentration in precision semiconductor lithography. Although these two disturbances do not directly cause shutdowns, they induce high-frequency and severe overcompensation adjustments in the underlying closed-loop control circuit, injecting chaotic thermal stress into the printed component. By introducing grid frequency fluctuation characteristics and protective airflow turbulence characteristics, the system achieves an unexpected anti-interference synergy effect, significantly reducing the latent defect rate caused by external dynamic mismatch.

[0042] In a preferred embodiment of the present invention, the risk quantification assessment module is further refined. The risk quantification assessment module extracts transient thermal gradient features and residual stress genetic sequence of the pre-processed layer from multimodal time series data, and extracts high-frequency noise features from external environmental interference data. The transient thermal gradient features, residual stress genetic sequence and high-frequency noise features are used as joint feature vectors and input into the risk quantification model to output the global potential stress entropy in a multidimensional nonlinear fusion manner.

[0043] The process of extracting transient thermal gradient features involves calculating the ratio of the difference in local temperature peaks between adjacent scanning trajectories to their spatial distance. This feature directly reflects the intensity of microscopic tensile and compressive stresses generated inside the material due to rapid heating and cooling. The process of extracting high-frequency noise features involves converting the joint time-domain signal of power grid fluctuations and airflow turbulence into a frequency-domain signal through fast Fourier transform, and extracting the energy integral value within a specific sensitive frequency band.

[0044] The process of extracting the genetic sequence of residual stress from the previous processing layer is as follows: the system retrieves the cumulative value of the transient thermal gradient of the historical printed layer directly below the current processing position, and performs exponential reduction according to the preset interlayer thermal conduction attenuation coefficient to obtain the genetic characteristic value of residual stress transmitted to the current layer.

[0045] When the risk quantification model integrates the above features to output the global potential stress entropy, the feature weights inside are dynamically and adaptively adjusted according to the current interference intensity. The system calculates the deviation ratio between the energy value of the current high-frequency noise feature and the energy benchmark value of the historical stable period, and uses this ratio as a dynamic gain coefficient to directly apply to the initial weights of the transient thermal gradient feature.

[0046] Among them, the historical stable period energy benchmark value is determined by continuously collecting external environmental interference data within a set time window when the system is in a silent state with no equipment load and no laser energy output. The frequency domain energy integral value of each sampling period is extracted by the same fast Fourier transform, and these integral values ​​are processed by time-dimensional moving average. This historical stable period energy benchmark value purely reflects the current basic environmental noise of the workshop, providing a reliable reference anchor point for dynamic weighting; for example, when the external environment is stable, the deviation ratio is 1, and the initial weight remains unchanged.

[0047] Once the airflow turbulence intensifies and causes the high-frequency noise energy value to rise rapidly to 1.5 times the historical stable period energy benchmark value, the system will immediately modify and amplify the initial weight of the transient thermal gradient feature by 1.5 times to obtain the dynamically adjusted target weight. Through this specific data flow and dynamic weighting logic, the model will force the allocation of higher computational weights to the potential stress accumulation caused by the thermal gradient when external disturbances intensify. Finally, the fused feature value will be transformed into a quantified global potential stress entropy value through a nonlinear mapping function.

[0048] To accurately reflect the rapid deterioration of stress accumulation as it approaches the material limit, the system abandons simple linear superposition and instead employs a logarithmic activation function with soft saturation characteristics as the aforementioned nonlinear mapping function. The input to this function is the sum of three dynamically weighted features: transient thermal gradient characteristics, high-frequency noise characteristics, and residual weighted genetic feature values. The logic is as follows:

[0049] When the eigenvalues ​​and eigenvalues ​​are in the lower range, the function output increases slowly and almost linearly with the eigenvalues ​​and eigenvalues, representing the stable accumulation of initial stress.

[0050] When the features and exceed the set critical inflection point, the global potential stress entropy of the function output will show a step-like surge; in the specific numerical calculation process, the system adds a constant one to the input features and uses it as the true number of the logarithmic operation to ensure that the true number is always greater than one and the logarithm is always positive, and extracts the natural logarithm from the true number.

[0051] Based on this, the system introduces a dynamic scaling factor multiplied by the natural logarithm result to obtain the final global potential stress entropy. The selection of this dynamic scaling factor is strictly controlled by the relative magnitude of the feature sum and the critical inflection point: when the feature sum is less than or equal to the critical inflection point, the dynamic scaling factor is fixed as the basic characterization constant, making the output smooth; once the feature sum is greater than the critical inflection point, the system calculates the difference between the feature sum and the critical inflection point, multiplies the difference by a preset excitation factor, and then adds it to the basic characterization constant as a new dynamic scaling factor.

[0052] The eigenvalues ​​corresponding to the critical inflection point are set to 0.8; when the eigenvalues ​​are 0.5, the argument after adding the constant is 1.5. After extracting the natural logarithm and combining it with the basic characterization constant, the value of which is set to approximately 2.96, the output global potential stress entropy is 1.2.

[0053] When external disturbances cause the feature value to climb to 0.85 and break through the inflection point, the dynamic scaling factor is rapidly increased, causing the output global potential stress entropy to be amplified sharply to 4.5, directly breaking through the subsequent safety margin threshold. The system selects this logarithmic activation function with soft saturation characteristics here, abandoning the double-ended saturation characteristics of the Sigmoid function in the conventional model, because the Sigmoid function will suppress the change of features in the high value region, causing the model to become passive in dangerous states. The step mapping design in this embodiment perfectly matches the physical fact of the nonlinear thermal hysteresis effect and uncontrolled thermal expansion of metal materials after local overheating.

[0054] This dynamic weight allocation mechanism and explicit nonlinear mapping design significantly improve the model's risk perception sensitivity in complex interference environments, and verify the robustness of the feature cascade mechanism.

[0055] In a preferred embodiment of the present invention, the subsequent calculation logic of the risk quantification assessment module is described. The risk quantification assessment module obtains the preset material yield strength critical parameter, calculates the difference between the global potential stress entropy and the material yield strength critical parameter, and determines the safety margin of the distance-triggered physical interference failure event based on the difference.

[0056] The critical parameter for the yield strength of the material here is not a single static constant, but a dynamic physical benchmark that is strongly correlated with the current printing material properties, the current processing layer height, and the ambient preheating temperature. To achieve this multi-dimensional dynamic solution, the system abandons the predictive network that lacks interpretability and instead constructs a multi-dimensional lookup table architecture based on real physical test calibration. The system pre-stores the mechanical property calibration matrix for different metal powder materials in the control terminal. During the processing, the system locks a specific calibration sub-table according to the current printing material properties, extracts the current processing layer height and ambient preheating temperature as search coordinates in real time, and performs node matching in the calibration sub-table.

[0057] When the search coordinates fail to accurately hit the preset node, the system extracts the four nearest reference nodes in space and calculates the current accurate parameter value through a bilinear interpolation algorithm. This parameter represents the maximum thermal stress limit that the component can accommodate without macroscopic irreversible warping deformation. This dynamic reference design, which combines lookup tables and interpolation based on real physical test data, not only ensures the determinism and millisecond-level response of the calculation process, but also overcomes the dynamic mismatch defect of traditional static thresholds when facing the thermal hysteresis effect of materials.

[0058] The risk quantification assessment module executes the difference calculation logic to assess the remaining buffer space between the current accumulated systemic thermal stress and the material mechanical collapse boundary. The system obtains the difference characterizing the remaining bearing capacity by subtracting the global potential stress entropy from the material yield strength critical parameter, and divides the difference by the material yield strength critical parameter, thereby normalizing the deviation of the physical quantity into a safety margin in the form of a positive percentage.

[0059] When the dynamically determined critical parameter of material yield strength is 400 MPa, if the currently calculated global potential stress entropy is 320 MPa, then the difference between the critical parameter of material yield strength and the global potential stress entropy is 80 MPa, and the safety margin calculated after normalization is 0.2.

[0060] If the global potential stress entropy climbs to a more dangerous 380 MPa, the difference between the two will narrow to 20 MPa, and the safety margin will correspondingly narrow to 0.05. In algebraic logic, 0.05 is less than 0.2, which means that the more dangerous the system is, the smaller the calculated safety margin value will be, so as to accurately trigger subsequent emergency recovery compensation intervention when the safety margin is less than the safety margin threshold.

[0061] The system establishes a clear safety margin threshold for this safety margin. When the safety margin is greater than or equal to the safety margin threshold, the system maintains the generation of basic processing efficiency instructions to maintain the production rhythm. When the safety margin is less than the safety margin threshold, it indicates that macroscopic mechanical warping may occur at any time, and extreme control strategies must be triggered immediately. This solution effectively ensures that the system can perform deterministic binary state switching when facing extreme dangers by clearly defining a single threshold, avoiding decision delays caused by complex multi-level judgments and demonstrating good process adaptability.

[0062] In a preferred embodiment of the present invention, this embodiment provides a detailed explanation of the specific actions of the control strategy generation module when generating the state recovery compensation instruction sequence. When generating the state recovery compensation instruction sequence, the control strategy generation module obtains the initial motion rate parameter and the initial energy output parameter of the current processing layer, reduces the initial motion rate parameter to generate the target motion rate parameter, and generates the state recovery compensation instruction sequence based on the target motion rate parameter.

[0063] When the system determines that the safety margin has fallen below the safety margin threshold, the control strategy generation module initiates the regression safety control logic. It extracts the pre-programmed initial motion rate parameters (i.e., galvanometer scanning speed) and initial energy output parameters (i.e., laser power) from the underlying CNC system. In order to actively suppress heat injection and extend local heat dissipation time without interrupting processing, the system forcibly reduces the initial motion rate parameters. The reduction in rate is not a fixed value, but is dynamically calculated based on the urgency of the current safety margin deviating from the safety margin threshold.

[0064] The system sets a basic deceleration ratio and calculates an additional deceleration ratio based on the deviation of the safety margin and the preset compensation slope. Finally, the target motion rate parameter is obtained by subtracting the sum of these two ratios from the initial motion rate parameter.

[0065] The base deceleration ratio is set to 0.1, and the compensation slope is 2.0. When the safety margin deviates from the threshold by 0.05, the calculated additional deceleration ratio is 0.1. After the two are added together, the total deceleration ratio reaches 0.2. At this time, if the initial motion speed parameter is 1000 mm per second, the system will force it to be reduced by 200 mm per second, and finally output 800 mm per second as the target motion speed parameter.

[0066] The system encapsulates the calculated target motion rate parameters into a state recovery compensation command sequence, forcing the CNC system to perform a deceleration action in subsequent machining.

[0067] This solution effectively alleviates the continuous accumulation of thermal stress without cutting off the energy source, demonstrating precise intervention in the thermal hysteresis effect.

[0068] In a preferred embodiment of the present invention, the heat dissipation path planning logic of the control strategy generation module is extended. When generating a state recovery compensation instruction sequence, the control strategy generation module generates an idle mobile heat dissipation path with no energy output and inserts the idle mobile heat dissipation path into the state recovery compensation instruction sequence.

[0069] In order to gain more global heat dissipation time on the edge of extreme thermodynamic runaway, the control strategy generation module further introduces a spatial dimension cooling strategy on the basis of speed reduction. The system plans one or more unloaded moving heat dissipation paths with no energy output. During the execution of the path, the control command forcibly shuts down the energy output of the laser, but keeps the optical galvanometer or mechanical gantry moving back and forth above the printed high temperature area according to a specific trajectory.

[0070] This scheduling logic, which inserts the idle moving heat dissipation path into the state recovery compensation instruction sequence, forces the underlying CNC system to perform an idle running action without energy injection after executing a high-intensity laser melting instruction.

[0071] This maintains the continuity of the CNC execution flow at the instruction level, avoids mechanical shocks caused by sudden stops and starts, and utilizes this physical window period to allow the extreme thermal gradient inside the component to be conducted and dissipated to the underlying substrate, generating unexpected system thermal stress resistance and significantly improving the robustness of the control strategy.

[0072] In a preferred embodiment of the present invention, the local vector reconstruction mechanism of the control strategy generation module is described. When generating a state recovery compensation instruction sequence, the control strategy generation module obtains an initial local scan vector strategy, changes the initial local scan vector strategy to generate a target local scan vector strategy, and integrates the target local scan vector strategy into the state recovery compensation instruction sequence.

[0073] The control strategy generation module not only intervenes in the time and speed dimensions, but also reconstructs the topology of spatial heat distribution. The system obtains the initial local scan vector strategy in the current slice file. This strategy usually manifests as long-distance continuous parallel unidirectional scan lines in pursuit of the ultimate processing efficiency. However, under high thermal stress conditions, such long continuous vectors can cause severe heat accumulation and superposition effects in a single direction.

[0074] The system modifies the initial local scan vector strategy through an algorithm, forcibly truncating it into multiple short vectors, and then rearranges them using a spatial discretization allocation algorithm based on thermal isolation constraints to generate the target local scan vector strategy. The system obtains the boundary contour of the current high-risk area and divides the area into multiple independent grid matrices with a fixed step size equal to the preset thermally affected zone radius.

[0075] To strictly prevent heat accumulation between adjacent grids, the system sets a hard constraint condition: the spatial Euclidean distance between the scanning grid points in two adjacent processing time slices must be strictly greater than twice the radius of the heat-affected zone.

[0076] Based on this spatial distance constraint, the system traverses and matches all truncated short vectors, and selects the sequence that satisfies the distance maximization principle as the final jump trajectory. For example, the command sequence will control the laser beam to immediately jump to the far end of the diagonal after melting a specific grid, thus abandoning the traditional nearby continuous scanning rule.

[0077] Considering the stringent requirements of industrial control systems for millisecond-level response, the system did not adopt the full permutation traversal with factorial computational complexity, but instead introduced a greedy selection strategy based on spatial hash index.

[0078] Specifically, the system maps the center coordinates of all candidate grids to a one-dimensional hash table. In each processing time slice, the system constructs a tabu search region with the current grid as the center and twice the radius of the heat-affected zone as the radius. Candidate grids falling into this region are directly blocked in the hash table. The node farthest from the current grid is selected from the remaining compliant grids as the next processing point. This matching logic, which combines spatial dimensionality reduction and greedy algorithm, compresses the time complexity of the algorithm from exponential to linear under the physical constraint of ensuring absolute heat isolation, thus breaking through the real-time bottleneck of complex spatial topology reconstruction in the underlying CNC execution flow.

[0079] This design, which integrates a highly dispersed target local scanning vector strategy into a state recovery compensation instruction sequence, severs the continuous transmission path of the macroscopic thermal stress chain at the physical source, eliminates the risk of excessive local heat accumulation, and demonstrates an innovative consideration of in-situ construction of microscopic thermal equilibrium.

[0080] In a preferred embodiment of the present invention, this embodiment describes the optimization logic of the control strategy generation module in a safe state. When generating a basic processing efficiency instruction sequence, the control strategy generation module obtains a preset device throughput index and a preset hardware energy consumption index. Based on the device throughput index and the hardware energy consumption index, it generates a basic processing efficiency instruction sequence through a preset optimization algorithm.

[0081] When the risk quantification assessment module determines that the current system is in a thermodynamically safe state, that is, the safety margin is greater than or equal to the safety margin threshold, the system enters the normal high-efficiency production mode. The control strategy generation module extracts the equipment throughput index, which is expressed as the metal powder melting volume requirement per unit hour, and the hardware energy consumption index, which is expressed as the upper limit of the combined power consumption of the laser and the cooling water chiller.

[0082] The acquisition of these two preset indicators is supported by rigorous business logic: the equipment throughput indicator is a dynamic benchmark calculated by dividing the total volume of the components to be printed issued by the upper-level manufacturing execution system by the remaining delivery time required by the order, which ensures that the single-layer processing efficiency can strictly match the overall production delivery time.

[0083] The hardware energy consumption index is determined by reading the maximum safe continuous power of each core component of the equipment as specified by the factory and multiplying it by the dynamic derating factor generated by the current power grid load state of the factory. For example, during the peak period of power consumption in the workshop, the derating factor is set to 0.8 to avoid local power grid overload, thereby ensuring the absolute safety and compliance of the optimization boundary from the physical source.

[0084] The system employs a gradient optimization algorithm based on a dynamic penalty function to find the optimal solution for the current single-layer printing efficiency in a multi-dimensional solution space under the boundary constraints of the two key indicators mentioned above. Under the premise of not exceeding the hardware energy consumption index, it maximizes the laser scanning speed and toner feeding to maximize the equipment throughput index. The system chooses a gradient optimization algorithm based on a dynamic penalty function instead of traditional heuristic algorithms such as genetic algorithms or particle swarm optimization. The core motivation for this choice is that the machining instruction flow of the underlying CNC system requires a high degree of determinism and millisecond-level response latency.

[0085] Heuristic algorithms exhibit uncontrollable randomness in convergence time within a multidimensional solution space, while model predictive control can utilize physical mechanisms to anticipate and predict the model's output of a safe control action with defined boundaries within a fixed computation period. Specifically, the structure of the cost function within the system is finely decomposed into a dynamic weighted combination of efficiency penalty terms and energy consumption penalty terms.

[0086] Among them, the predicted throughput and predicted energy consumption rely on the built-in physical mechanism prospective model; the system calculates the effective heat input per unit time based on the laser power and scanning speed in the current solution system, combined with the fixed photothermal absorptivity of the current material, and infers the volumetric flow rate of the transient molten pool as the predicted throughput.

[0087] The specific parameter flow and deduction logic is as follows: The system multiplies the laser power in the current solution system by the fixed photothermal absorptivity of the current material to obtain the effective absorption power actually intercepted in the target processing area; at the same time, the system extracts the latent heat of melting phase transition of the current metal powder and the sensible heat required to rise from the ambient preheating temperature to the liquidus temperature from the underlying material database, and adds the two to obtain the critical volumetric enthalpy required for the complete melting of a unit volume of material.

[0088] The system divides the aforementioned effective absorbed power by the critical volumetric enthalpy to calculate the theoretical maximum volumetric melting rate.

[0089] The system introduces a molten pool morphology transformation matrix based on Gaussian photothermal distribution characteristics, which maps the theoretical melting rate to initial dynamic melt width and melt depth parameters;

[0090] To better fit the actual physical boundaries, the system further extracts the current scanning speed and, in combination with the preset laser spot radius and initial dynamic melting depth parameters, calculates the tailing stretching ratio of the molten pool. Based on this tailing stretching ratio, the corresponding geometric correction coefficient is matched in a preset correction lookup table.

[0091] By multiplying the theoretical maximum volumetric melting rate by this geometric correction factor, the volumetric flow rate of the transient molten pool can be inferred as the predicted throughput.

[0092] For the power consumption assessment of the cooling water chiller, the system pre-established a heat load-energy consumption mapping matrix in the control terminal. The system extracts the laser power in the current solution set as the real-time heat load input to the matrix, and dynamically matches the steady-state power consumption required by the water chiller under the heat load through a linear interpolation algorithm.

[0093] Meanwhile, the system divides the laser power by the laser's rated electro-optical conversion efficiency and adds the calculated steady-state power consumption of the cooling water chiller under the corresponding heat load to calculate the overall predicted energy consumption.

[0094] The efficiency penalty is calculated as the absolute value of the difference between the device throughput index and the predicted throughput under the current preset parameters. The energy consumption penalty is defined as the ratio of the current predicted energy consumption to the hardware energy consumption index. During the optimization iteration process, the system uses laser power and scanning speed as a two-dimensional solution system. When the predicted energy consumption approaches the hardware energy consumption index, the energy consumption penalty weight corresponding to the energy consumption penalty is forcibly amplified by the system in a stepwise manner.

[0095] The energy consumption penalty weight is set to 1.0 under the daily benchmark. When the predicted energy consumption reaches 0.9 times the hardware energy consumption index, that is, when the approximation ratio reaches the safety red line of 0.9, the system will instantly step up the energy consumption penalty weight to 50.0 in the next calculation cycle. This huge asymmetric penalty mechanism forces the evaluation value of the cost function to increase sharply at the boundary edge, thereby generating a strong physical backlash in the gradient descent direction of finding the optimal solution. This forces the solution system to quickly deviate from the high energy consumption region and converge to the robust balance region that takes into account moderate speed and low power in order to obtain the power and speed matching combination that minimizes the overall processing cost. This is the optimal solution for the current single-layer printing efficiency and is compiled into the basic processing efficiency instruction sequence.

[0096] This design, which translates abstract multi-objective optimization into explicit penalty constraints and weighted step mechanisms, eliminates the risk of uncontrollable random trial and error in the process of finding the optimal solution. It ensures that the equipment can operate at near-physical limits under stable conditions without collision risks, effectively overcoming the dynamic mismatch between safety and efficiency in traditional systems.

[0097] In a preferred embodiment of the present invention, the smooth transition mechanism of the instruction execution closed-loop module is described in detail. The instruction execution closed-loop module obtains the current instruction running state of the underlying CNC execution flow and smoothly splices the processing scheduling instruction with the current instruction running state without interrupting the continuous processing of the underlying CNC execution flow.

[0098] The instruction execution closed-loop module acts as a bridge connecting the upper-level intelligent decision-making and the lower-level rigid physical execution mechanism. It polls the look-ahead buffer of the lower-level CNC system in real time to obtain the current instruction running status of the lower-level CNC execution flow, including the line number of the currently executed code, the instantaneous position coordinates, and the motion acceleration vector of each axis.

[0099] To avoid severe machine tool vibration or stepper motor step loss caused by sudden command changes, the system executes smooth splicing logic when injecting machining scheduling commands. The system extracts the position vector and velocity vector of the current tool path endpoint and uses them as the initial boundary conditions of the newly injected command sequence. A transition curve command that conforms to the machine tool dynamics limit is automatically inserted between the two.

[0100] In terms of specific curve selection, the system specifically adopts quintic B-spline curves instead of cubic polynomial curves commonly used in traditional industrial CNC. The underlying motivation is that quintic B-spline curves can mathematically guarantee the absolute continuity of acceleration and jerk, i.e., the rate of change of acceleration.

[0101] When facing extreme deceleration or avoidance scheduling, if the acceleration is discontinuous, the transient command step will directly cause the servo motor torque to generate a high-frequency sudden impact, which will excite the physical resonance frequency band of the machine tool robot arm or guide rail. In the simulation calculation, the system presets the maximum acceleration limit of the machine tool spindle modification hardware to be 5000mm / s³. If rigid connection interpolation is directly performed based on the increased speed vector and the deceleration target required by the machining scheduling command, the calculated instantaneous acceleration may be as high as 7500. At this time, the system will actively lengthen the spatial physical length of the fifth B-spline transition curve, exchanging time for smoothness, and forcibly limit and flatten the peak acceleration within the safe envelope of 4500.

[0102] During this lengthening process, the system does not blindly increase the length, but executes strict reverse derivation logic; the system takes the maximum jerk that the machine tool guide rail hardware can withstand as a rigid boundary condition, and combines the initial velocity vector with the target deceleration difference to calculate the minimum time window required to complete the smooth transition of the speed. The minimum time window is multiplied by the current average running speed to obtain the minimum physical length of space required.

[0103] The system dynamically redistributes the control point positions of the quintic B-spline curve based on the minimum spatial physical length; the system achieves seamless switching of control strategies without interrupting the continuous machining process of the underlying CNC execution flow, avoiding stress fatigue of the mechanical structure, and verifying the robustness of the smooth splicing mechanism under extreme working conditions.

[0104] In a preferred embodiment of the present invention, the physical interference failure event is specifically defined, including macroscopic mechanical warping deformation caused by local thermal stress release and powder spreading scraper collision event triggered by macroscopic mechanical warping deformation.

[0105] Physical interference failure events define the absolute failure boundary that this industrial control system attempts to avoid. In the continuous printing process of ultra-large components, if the global potential stress entropy is not released in time, the accumulated thermal stress at the bottom layer will eventually break through the yield strength of the material. This micro-mechanical imbalance manifests in the physical form as an upward macro-mechanical warping deformation of the component edge or overhanging structure, i.e., an irreversible bulge in the vertical direction.

[0106] A rigid spreading blade will directly impact the warped area during high-speed horizontal movement, thus triggering a severe spreading blade collision event.

[0107] In order to achieve adaptive optimization and disturbance resistance of the control system throughout its entire life cycle, when a slight scraper friction interference occurs on-site, approaching the critical point of a collision event, the system will automatically package the multimodal time series data, external disturbance characteristics and corresponding control command stream within a specified time window before that moment into an engineering file and send it back to the cloud server, and recalibrate the baseline of the material yield strength critical parameter in the risk quantification model in an offline state.

[0108] The system continuously refines its perception accuracy of extreme thermodynamic nonlinear changes, ensuring that the equipment can maintain a dynamic safety balance between thermodynamics and control commands during uninterrupted high-risk operation for weeks, demonstrating extremely high industrial applicability.

[0109] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An industrial control system for metal 3D printing, characterized in that, include: The module includes a data acquisition module, a risk quantification and assessment module, a control strategy generation module, and an instruction execution closed-loop module. The data acquisition module is used to acquire multimodal time-series data of the target processing area and external environmental interference data during the layer-by-layer processing, and send the multimodal time-series data and external environmental interference data to the risk quantification and assessment module. The risk quantification assessment module is used to calculate the global potential stress entropy based on the multimodal time series data and the external environmental interference data through a preset risk quantification model, and to predict the safety margin of the distance-triggered physical interference failure event based on the global potential stress entropy, and send the safety margin to the control strategy generation module. The control strategy generation module is used to obtain a preset safety margin threshold and compare the safety margin with the safety margin threshold. In response to the safety margin being less than the safety margin threshold, the control strategy generation module is configured to generate a processing scheduling instruction containing a state recovery compensation instruction sequence; or, in response to the safety margin being greater than or equal to the safety margin threshold, the control strategy generation module is configured to generate a processing scheduling instruction containing a basic processing efficiency instruction sequence. The instruction execution closed-loop module is used to acquire the machining scheduling instruction and inject the machining scheduling instruction into the execution flow of the underlying CNC system used to drive the laser energy source and / or motion actuator, so as to dynamically adjust the physical machining state.

2. The industrial control system for metal 3D printing according to claim 1, characterized in that, The multimodal time-series data includes the time-series characteristics of temperature distribution and the visual feedback characteristics of shape in the target processing area; the external environmental interference data includes the power grid frequency band fluctuation characteristics and the turbulence characteristics of the protective airflow.

3. The industrial control system for metal 3D printing according to claim 1, characterized in that, The risk quantification assessment module is also used to: extract transient thermal gradient features from the multimodal time series data and retrieve the residual stress genetic sequence passed from the historical printing layer to the current layer; extract high-frequency noise features within a preset sensitive frequency band from the external environmental interference data; and input the transient thermal gradient features, the residual stress genetic sequence, and the high-frequency noise features into the risk quantification model to output the global potential stress entropy.

4. The metal 3D printing industrial control system according to claim 3, characterized in that, The risk quantification assessment module is also used to: obtain a preset critical parameter for material yield strength; and calculate the difference between the global potential stress entropy characterizing equivalent mechanical stress and the critical parameter for material yield strength. Based on the difference, the safety margin for the distance-triggered physical interference failure event is determined.

5. The industrial control system for metal 3D printing according to claim 1, characterized in that, The control strategy generation module is further configured to: when generating the state recovery compensation instruction sequence, obtain the initial motion rate parameter and initial energy output parameter of the current machining layer from the pre-arranged instructions in the underlying CNC system; reduce the initial motion rate parameter to generate the target motion rate parameter; and generate the state recovery compensation instruction sequence based on the target motion rate parameter.

6. The metal 3D printing industrial control system according to claim 5, characterized in that, The control strategy generation module is further configured to: generate an idle mobile heat dissipation path with no energy output when the state recovery compensation instruction sequence is generated; and insert the idle mobile heat dissipation path into the state recovery compensation instruction sequence.

7. The metal 3D printing industrial control system according to claim 6, characterized in that, The control strategy generation module is further configured to: when generating the state recovery compensation instruction sequence, obtain the current slice file corresponding to the current processing task, and obtain the initial local scan vector strategy from the current slice file; change the initial local scan vector strategy to generate a target local scan vector strategy; and integrate the target local scan vector strategy into the state recovery compensation instruction sequence.

8. The industrial control system for metal 3D printing according to claim 1, characterized in that, The control strategy generation module is further configured to: obtain a preset equipment throughput index and a preset hardware energy consumption index when generating the basic processing efficiency instruction sequence; and generate the basic processing efficiency instruction sequence based on the equipment throughput index and the hardware energy consumption index using a preset optimization algorithm.

9. The industrial control system for metal 3D printing according to claim 1, characterized in that, The instruction execution closed-loop module is also used to: obtain the current instruction running status of the execution flow of the underlying CNC system; and smoothly splice the machining scheduling instruction with the current instruction running status without interrupting the continuous machining process of the execution flow of the underlying CNC system.

10. The industrial control system for metal 3D printing according to claim 1, characterized in that, The physical interference failure events include macroscopic mechanical warping deformation caused by local thermal stress release and powder spreading scraper collision events triggered by the macroscopic mechanical warping deformation.

Images