3D printing piece intelligent detection system with whole-process quality traceability

Abstract

This invention relates to the field of additive manufacturing quality control and data traceability technology, specifically to an intelligent inspection system for 3D printed parts with full-process quality traceability. The system includes: a data acquisition module for real-time acquisition of all-dimensional process parameters of the target printed part; a first processing module for generating a trust crack index characterizing the integrity of the data traceability chain; a second processing module for determining a traceability risk index characterizing the degree to which the system deviates from the safety boundary based on the trust crack index and a preset upper limit for the trust crack index; a risk decision module for outputting intervention actions based on the traceability risk index and all-dimensional process parameters via a preset deep reinforcement learning model; and a closed-loop control module for executing the intervention actions to adaptively correct the printing process of the target printed part. This invention achieves proactive avoidance and intelligent management of traceability risks, significantly improving the system's adaptive control level and the foresight of its decisions.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing quality control and data traceability technology, specifically to an intelligent inspection system for 3D printed parts with full-process quality traceability. Background Technology

[0002] With the widespread application of additive manufacturing technology in high-specification manufacturing fields such as medical devices, ensuring the reliability and compliance of product quality traceability throughout the entire process has become crucial. Currently, quality monitoring of the 3D printing process relies heavily on the independent monitoring of various process parameters. The system typically adopts a passive response mechanism based on fixed thresholds, triggering alarms or interventions when a parameter exceeds the preset range.

[0003] However, this traditional approach ignores the potential for data integrity damage during collection, transmission, and storage. In other words, the data traceability chain itself may be at risk of breakage. Existing technologies lack real-time, quantitative assessment methods for the health status of the data chain, and cannot perform proactive, closed-loop intelligent control based on the assessment results to ensure its resilience. Therefore, how to assess the integrity of the data traceability chain in the 3D printing process in real time and quantitatively, and accordingly conduct proactive closed-loop intervention to dynamically repair and maintain the resilience of the traceability network, has become a pressing technical problem to be solved in this field. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an intelligent inspection system for 3D printed parts with end-to-end quality traceability. Specifically, the technical solution of this invention includes:

[0005] The data acquisition module is used to collect all-dimensional process parameters of the target printed part in real time. The all-dimensional process parameters include the internal process parameters of the equipment, the internal logical timestamp of the equipment control system, and the physical timestamp when it is received by the data acquisition server.

[0006] The first processing module is used to generate a trust crack index that characterizes the integrity of the data traceability chain;

[0007] The second processing module is used to determine the source tracing risk index, which characterizes the degree to which the system deviates from the security boundary, based on the trust crack index and the preset upper limit of the trust crack index.

[0008] The risk decision-making module is used to output intervention actions based on the source risk index and full-dimensional process parameters through a preset deep reinforcement learning model.

[0009] The closed-loop control module is used to perform intervention actions to adaptively correct the printing process of the target print.

[0010] Preferably, the first processing module is specifically used for:

[0011] The normalized timestamp drift is determined based on the difference between the device's internal logical timestamp and the server's physical timestamp.

[0012] Determine the packet loss rate based on sequence number discontinuity detection;

[0013] A trust crack index is generated by combining the weighted sum of timestamp drift and packet loss rate.

[0014] Preferably, the second processing module is specifically used for:

[0015] Invoke the trust crack index generated by the first processing module;

[0016] The source tracing risk index was determined by normalizing the trust crack index.

[0017] The normalization process involves dividing the trust crack index by a preset upper limit for the trust crack index.

[0018] Preferred options also include:

[0019] The status classification module is used to classify the health status of the system's traceability network into safe status, early warning status, and dangerous status based on the traceability risk index.

[0020] Among them, the safety status corresponds to the traceability risk index being less than or equal to a preset first threshold;

[0021] The warning status corresponds to the source tracing risk index being greater than the first threshold and less than or equal to the preset second threshold.

[0022] A dangerous state corresponds to a traceability risk index that is greater than the second threshold.

[0023] Preferably, the risk decision-making module is specifically used for:

[0024] The full-dimensional process parameters and the source tracing risk index are used together to construct the state input of the deep reinforcement learning model;

[0025] A policy network based on a deep reinforcement learning model outputs the optimal intervention action in real time under the current input state.

[0026] Preferably, the deep reinforcement learning model is trained using a reward function, which includes:

[0027] A quality indicator used to measure the degree to which process parameters deviate from their optimal settings;

[0028] Efficiency metrics used to quantify the current production rate;

[0029] The core penalty item is used to punish an increase in the traceability risk index.

[0030] Preferably, the core penalty is determined by squaring the difference between the current traceability risk index and the previous traceability risk index.

[0031] Preferably, the closed-loop control module is specifically used for:

[0032] In response to the safe state determined by the state division module, the current printing process is maintained.

[0033] Preferably, the closed-loop control module is specifically used for:

[0034] In response to the warning status determined by the status division module, a first-level correction action is executed, which includes forcibly reducing the printing speed.

[0035] Preferably, the closed-loop control module is specifically used for:

[0036] In response to a dangerous state determined by the state division module, a secondary correction action is performed. The secondary correction action is to recalibrate the data flow by creating an anchor event in the data link through the execution of an emergency redundancy calibration procedure.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. By quantitatively integrating factors such as timestamp drift and data packet loss rate, this invention transforms the intangible data traceability chain integrity problem into a measurable and assessable comprehensive engineering indicator, realizing real-time and quantitative monitoring of the health status of the data chain and solving the technical gap of lacking corresponding assessment methods in the existing technology.

[0039] 2. This invention introduces a deep reinforcement learning model for risk decision-making and designs a comprehensive reward function that includes quality, efficiency, and risk change trends. This system can learn and execute forward-looking optimal intervention strategies, which changes the passive response mode of traditional technology based on fixed thresholds. It realizes the proactive avoidance and intelligent management of source-tracing risks, and significantly improves the adaptive control level and forward-looking decision-making of the system.

[0040] 3. This invention constructs a classification mechanism that maps continuous risk indices to discrete states such as safety, warning, and danger, and performs graded closed-loop correction actions accordingly. This refined management approach can accurately match the intervention intensity according to the severity of the risk, avoiding indiscriminate over-intervention or under-intervention, and minimizing unnecessary impact on normal production efficiency while ensuring traceability resilience.

[0041] 4. This invention addresses the dangerous situation where data links face severe risks of disruption and proposes an innovative data flow calibration method that creates anchor events by performing specific physical actions. This mechanism can forcibly establish a reliable trust benchmark in the midst of data chaos, enabling rapid resynchronization and recovery of the data link, thereby saving the entire production batch at minimal cost and enhancing the system's robustness and traceability resilience under extreme conditions. Attached Figure Description

[0042] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0043] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0045] Example 1:

[0046] Please see Figure 1 A fully traceable intelligent inspection system for 3D printed parts, including:

[0047] The data acquisition module is used to collect all-dimensional process parameters of the target printed part in real time. The all-dimensional process parameters include the internal process parameters of the equipment, the internal logical timestamp of the equipment control system, and the physical timestamp when it is received by the data acquisition server.

[0048] The first processing module is used to generate a trust crack index that characterizes the integrity of the data traceability chain;

[0049] The second processing module is used to determine the source tracing risk index, which characterizes the degree to which the system deviates from the security boundary, based on the trust crack index and the preset upper limit of the trust crack index.

[0050] The risk decision-making module is used to output intervention actions based on the source risk index and full-dimensional process parameters through a preset deep reinforcement learning model.

[0051] The closed-loop control module is used to perform intervention actions to adaptively correct the printing process of the target print.

[0052] This invention provides an intelligent inspection system for 3D printed parts with full-process quality traceability. Its purpose is to solve the problems in the prior art that lack real-time and quantitative assessment of the integrity of the 3D printing process data chain, and that cannot be based on this assessment to carry out proactive and closed-loop control to ensure the resilience of quality traceability.

[0053] In one implementation scenario, the system is applied to the additive manufacturing process of metal medical devices with high precision requirements. By constructing a complete technical closed loop from data acquisition and risk assessment to decision control, the system realizes real-time monitoring and adaptive adjustment of the health status of the printing process traceability network.

[0054] The data acquisition module aims to provide comprehensive and multi-dimensional data input for subsequent quality traceability and risk assessment. In this embodiment, the module is implemented through a multi-sensor network deployed inside and around the laser powder bed melting equipment.

[0055] The module collects real-time, multi-dimensional process parameters, including not only internal equipment parameters such as laser power, scanning speed, powder feeding rate, forming chamber temperature, and oxygen content, but also two key types of timestamp information: the internal logical timestamp from the equipment control system, which refers to the time stamp added to each data packet by the internal controller, reflecting the logical moment of the event; and the server physical timestamp when the data packet is received by the data acquisition server, which refers to the physical moment recorded by the server system when the data packet arrives at the external server. The comparison of these two types of timestamps is the basis for subsequent evaluation of data transmission reliability.

[0056] The first processing module aims to transform the intangible problem of data pollution into a measurable and assessable engineering indicator, namely, to generate a trust crack index that characterizes the integrity of the data traceability chain. This module receives the data stream transmitted from the data acquisition module and analyzes it in real time to quantify the integrity damage that may occur during the acquisition, transmission and storage of data.

[0057] The second processing module aims to transform the absolute physical quantity Trust Crack Index output by the first processing module into a standardized relative metric with an intuitive risk scale, namely, a source risk index that characterizes the degree to which the system deviates from the safety boundary. This module receives the Trust Crack Index and normalizes it based on a preset upper limit for the Trust Crack Index, thereby providing a unified and easily categorized input basis for subsequent risk decisions.

[0058] The risk decision-making module aims to intelligently output the optimal intervention action based on the current system state, so as to achieve proactive intervention and forward-looking control of traceability risks. The core of this module is a preset deep reinforcement learning model. It receives the traceability risk index and real-time full-dimensional process parameters as state input, and outputs intervention instructions aimed at restoring or ensuring the resilience of the data traceability network through internal policy network operations.

[0059] The purpose of the closed-loop control module is to transform the virtual instructions output by the risk decision module into actual operations in the physical world, forming a closed-loop control system that adaptively corrects the printing process of the target printed part. After receiving the intervention action instruction, the module will directly interact with the control system of the 3D printing equipment to execute the corresponding correction strategy, such as adjusting the printing speed or executing a specific calibration procedure.

[0060] Through the collaborative work of the above modules, this invention constructs a complete intelligent closed-loop system of perception-evaluation-decision-execution. It can not only evaluate the integrity and risks of the data traceability chain in real time and quantitatively, but also make forward-looking optimal decisions based on the evaluation results through a deep reinforcement learning model and automatically execute intervention actions. This dynamically maintains and repairs the resilience of the data traceability chain during the printing process, ensuring the reliability and compliance of the final product's full-process quality traceability and improving its applicability in high-specification manufacturing fields.

[0061] Example 2:

[0062] The first processing module is specifically used for:

[0063] The normalized timestamp drift is determined based on the difference between the device's internal logical timestamp and the server's physical timestamp.

[0064] Determine the packet loss rate based on sequence number discontinuity detection;

[0065] A trust crack index is generated by combining the weighted sum of timestamp drift and packet loss rate.

[0066] This embodiment is a specific implementation of the first processing module in Embodiment 1, aiming to accurately quantify the two core dimensions that constitute the trust crack index: time synchronization deviation and data packet loss;

[0067] The first processing module generates a trust gap index through the following steps. :

[0068] The normalized timestamp drift is determined based on the difference between the device's internal logical timestamp and the server's physical timestamp. Timestamp drift This refers to an indicator used to measure the accuracy and consistency of time recording during data stream transmission. Its function is to quantify the impact of timestamp inconsistencies caused by factors such as network latency and system load on data traceability. In this embodiment, it is calculated by averaging and normalizing the timestamp differences of all data packets within a specific time window, as shown in the formula: ;

[0069] in, : indicates the first The server's physical timestamp for each data packet is a timestamp type, and its source is obtained from real-time recordings by the data acquisition server. : Represents the summation operator; : A positive integer index of the data packet sequence;

[0070] : indicates the corresponding first The internal logical timestamp of each data packet is a timestamp type, and its source is recorded by the device control system when the data is generated;

[0071] : Represents the total number of data packets within the current time window. It is an integer and is obtained by counting the data streams entering the module in real time.

[0072] : Represents a reference time constant, which is of time type and is derived from pre-set process parameters such as theoretical data packet transmission interval; the function of this parameter is to perform dimensionless processing of the time difference to ensure the consistency of physical dimensions on both sides of the formula.

[0073] Determining packet loss rate based on sequence number discontinuity detection Packet loss rate This refers to an indicator used to measure whether data is lost during transmission, and its function is to quantify the degree of damage to the integrity of the data link. In this embodiment, the system assigns a consecutive sequence number to each data packet, and determines whether data loss has occurred by detecting whether the sequence numbers of the received data packets are consecutive. The calculation method is as follows: ;

[0074] in, : Represents the number of lost data packets detected by sequence number discontinuity. It is an integer type and is calculated by comparing the sequence number of the current data packet with that of the previous data packet.

[0075] : Indicates the total number of data packets that should theoretically be received. It is an integer and is derived from a preset value based on the device's sampling frequency and the current time window length.

[0076] A trust gap index is generated by combining a weighted sum of timestamp drift and packet loss rate. The purpose of this step is to integrate the damage from both the time and integrity dimensions into a single, comprehensive assessment index, calculated using the following formula: ;

[0077] in, : is a dimensionless trust crack index, and the higher the value, the worse the integrity of the tracing network;

[0078] and These are weighting factors corresponding to timestamp drift and data packet loss rate, respectively. They are floating-point numbers and are preset based on specific regulatory requirements and sensitivity analysis of the impact of different parameters on the final product quality; for example, in scenarios with extremely high time synchronization requirements, Assigned a higher value;

[0079] By independently quantifying the two key factors that cause data chain contamination—timestamp drift and data packet loss—and then fusing them through a weighted sum, the generated Trust Crack Index can more comprehensively and accurately reflect the true health of the data traceability chain. This avoids the one-sidedness of assessments based on a single indicator and provides a more reliable data foundation for subsequent risk assessments.

[0080] Example 3:

[0081] The second processing module is specifically used for:

[0082] Invoke the trust crack index generated by the first processing module;

[0083] The source tracing risk index was determined by normalizing the trust crack index.

[0084] The normalization process involves dividing the trust crack index by a preset upper limit for the trust crack index.

[0085] This embodiment is a concrete implementation of the second processing module in Embodiment 1. Its core technical motivation lies in transforming an absolute value with physical meaning but lacking an intuitive risk scale into a trust crack index. This is transformed into a relatively easy-to-classify and machine-determine traceability risk index. ;

[0086] The second processing module is implemented through the following steps:

[0087] Call the trust crack index generated by the first processing module This is the starting point of the processing flow of this module;

[0088] By normalizing the trust gap index, the source tracing risk index is determined. The normalization process here involves dividing the trust crack index by a preset upper limit for the trust crack index. .

[0089] Trust Crack Index Upper Limit This refers to the maximum level of data inconsistency that the system can tolerate, provided that the batch traceability chain is complete and acceptable to regulatory agencies. Its purpose is to provide a clear and acceptable benchmark for risk assessment. This value is pre-set by domain experts based on Good Manufacturing Practices (GMP) for pharmaceuticals or medical devices and extensive historical batch validation data; it is a traceability risk index. The calculation formula is as follows: ;

[0090] in, : is a dimensionless source risk index, which intuitively represents the degree to which the current system state deviates from the acceptable safety boundary; when When the value equals 1, it indicates that the system is exactly at the critical point of acceptable risk; the output of this index will directly serve as the state input for subsequent deep reinforcement learning models and the decision-making basis for closed-loop control systems.

[0091] By introducing Normalization is performed using this as a benchmark to make the traceability risk index It has a clear physical meaning and an intuitive risk scale; The value is directly related to the safety boundary, which provides a unified and standardized input for subsequent risk level classification and decision control, greatly simplifying the complexity of decision logic and improving the stability and interpretability of the system.

[0092] Example 4:

[0093] This system also includes:

[0094] The status classification module is used to classify the health status of the system's traceability network into safe status, early warning status, and dangerous status based on the traceability risk index.

[0095] Among them, the safety status corresponds to the traceability risk index being less than or equal to a preset first threshold;

[0096] The warning status corresponds to the source tracing risk index being greater than the first threshold and less than or equal to the preset second threshold.

[0097] A dangerous state corresponds to a traceability risk index that is greater than the second threshold.

[0098] This embodiment adds a state division module to the system in Embodiment 1. Its purpose is to transform the continuously changing source tracing risk index into discrete system health status levels with clear operational guidelines, providing clear triggering conditions for the closed-loop control module to execute the graded correction strategy.

[0099] The state classification module is based on the source tracing risk index. The system traceability network health status is divided into safe status, early warning status, and dangerous status;

[0100] This division is based on This is accomplished by comparing with two preset thresholds:

[0101] Safety status: corresponds to a traceability risk index that is less than or equal to a preset first threshold; in this embodiment, the first threshold is set to 1; when When the system is deemed to be in a secure state, it indicates that the current trust gap is within a pre-defined acceptable range and the data traceability chain is complete and reliable.

[0102] Warning status: corresponds to a source tracing risk index greater than a first threshold and less than or equal to a preset second threshold; in this embodiment, the second threshold is set to 1.5; when At this point, the system is identified as being in an early warning state, indicating that the trust crack has exceeded the normal range and there is a risk of breakage, requiring the initiation of preliminary intervention measures;

[0103] Dangerous state: corresponding to a traceability risk index greater than the second threshold; when At this point, the system is deemed to be in a dangerous state, indicating a serious trust breach that could lead to batch traceability failure at any time, necessitating stronger emergency measures.

[0104] The second threshold of 1.5 is set based on an empirical value obtained from the statistical analysis of historical data on the risk amplification effect. It is used to distinguish between general risks and high-risk risks that may lead to catastrophic consequences. This value is an adjustable parameter.

[0105] By introducing a state classification module, this invention transforms complex and continuous risk assessment results into three simple and clear state levels: safety, warning, and danger. This approach greatly improves the system's practicality and operability, enabling the subsequent closed-loop control module to accurately match and execute corresponding corrective actions based on different state levels. This achieves refined and hierarchical risk management, avoiding indiscriminate over-intervention or under-intervention.

[0106] Example 5:

[0107] The risk decision-making module is specifically used for:

[0108] The full-dimensional process parameters and the source tracing risk index are used together to construct the state input of the deep reinforcement learning model;

[0109] A policy network based on a deep reinforcement learning model outputs the optimal intervention action in real time under the current input state.

[0110] This embodiment is a concrete implementation of the risk decision-making module in Embodiment 1, aiming to illustrate how a deep reinforcement learning model can use system state information to generate the optimal intervention action;

[0111] The risk decision-making module operates in the following ways:

[0112] The full-dimensional process parameters and the source tracing risk index are jointly constructed as the state input of the deep reinforcement learning model. Status input It refers to At any given time, a complete set of information is provided to the DRL agent for decision-making; in this embodiment, it is a state vector, which is composed as follows: ;

[0113] in, : Represents a set of core physical process parameters, which is a vector type. The data is collected in real time by the data acquisition module, such as {temperature, power, scan speed}.

[0114] : Indicates that the second processing module is in The traceability risk index calculated at all times is a floating-point number, and its source is obtained by the second processing module in real time.

[0115] : Represents the discrete time step or current moment of the current system operation;

[0116] The technical consideration behind this design is that optimal intervention decisions must take into account not only the current level of risk. It is also necessary to consider the specific physical process state at the time the risk arose. This makes decision-making more context-aware.

[0117] A policy network based on a deep reinforcement learning model outputs the optimal intervention action in real time given the current input state. The policy network refers to the network formed within a DRL agent after extensive training, used to process state inputs. Mapping to intervention actions Neural networks; intervention actions It is the DRL agent from its preset action space Choose one or a set of operation instructions.

[0118] In this embodiment, the action space It includes a series of interventions designed to restore the resilience of the data traceability network, such as forcibly reducing the printing speed by 15% and executing a sensor data synchronization command; the agent learns a policy by maximizing the long-term cumulative reward through training, enabling it to adapt to a given state input. The next step is to select the intervention that can most effectively reduce future risks and balance production goals. ;

[0119] By using real-time process parameters and quantified risk indices as inputs to the DRL model, this invention transforms the decision-making process from a simple triggering based on a single threshold into a comprehensive assessment of the state of a complex system. Utilizing the powerful nonlinear fitting and sequential decision-making capabilities of the DRL model, the system can learn sophisticated control strategies that are difficult for human experts to formulate, thereby achieving more intelligent and forward-looking management of source-tracing risks and significantly improving the system's adaptive control level.

[0120] Example 6:

[0121] Deep reinforcement learning models are trained using reward functions, which include:

[0122] A quality indicator used to measure the degree to which process parameters deviate from their optimal settings;

[0123] Efficiency metrics used to quantify the current production rate;

[0124] The core penalty item used to punish an increase in the traceability risk index;

[0125] The core penalty is determined by squaring the difference between the current source risk index and the previous source risk index.

[0126] This embodiment is a concretization of the deep reinforcement learning model training method in Embodiment 5. Its core innovation lies in the design of the reward function, which guides the agent to learn a complex decision-making behavior that can delicately balance the three core manufacturing goals of quality, efficiency and compliance.

[0127] Deep reinforcement learning models use reward functions During training, this function quantifies the agent's state input. Intervention actions should be taken. The reward function is designed to prioritize compliance and traceability integrity in high-risk manufacturing environments, providing immediate feedback. The instantaneous reward value, defined as dimensionless, is calculated using the following formula: ;

[0128] in, , , These are dimensionless weighting coefficients representing quality, efficiency, and traceability costs, respectively. They are floating-point numbers and are derived from hyperparameters set by users based on enterprise needs, such as compliance priority or efficiency priority. They can also be optimized from decision data of human experts through inverse reinforcement learning.

[0129] The reward function comprises the following three key components:

[0130] Quality indicators used to measure the degree to which process parameters deviate from optimal set values Quality Indicators The purpose is to quantify the stability of the current printing process, and its value range is... The closer the value is to 1, the more stable the quality. In this embodiment, it is calculated using an exponential decay function of the normalized error based on multiple key parameters: ;

[0131] in, : This is an index of key physical process parameters, representing a preset set of key physical process parameters. ; Indicates temperature parameter; Indicates power parameters; This represents a pre-defined set of key physical process parameters; : Represents the natural constant An exponential function with base 0; : Represents the summation operator;

[0132] : is a parameter In time The real-time measurement value is a floating-point number, and its source is the data acquisition module that collects it in real time.

[0133] : This is its optimal process setting value, which is a floating-point number and is derived from the preset process specification.

[0134] : This is the maximum allowable fluctuation tolerance for this parameter. It is a floating-point number and its source is preset according to the process specifications. It is used to normalize errors.

[0135] Efficiency metrics used to quantify the current production rate Efficiency indicators Aimed at quantifying production efficiency, its value range is A higher value indicates higher efficiency; it is defined as the ratio of the current printing speed to the device's maximum theoretical printing speed. ;

[0136] in, : is the time of the printhead The actual operating speed is a floating-point number, which is obtained in real time from the device controller.

[0137] : This is the maximum safe operating speed set by the equipment. It is a floating-point number and its source is the equipment's factory parameters or safety regulations.

[0138] The core penalty item used to punish an increase in the traceability risk index. This penalty term aims to make the agent highly sensitive to the worsening trend of risk. In this embodiment, the core penalty term is determined by squaring the difference between the current source risk index and the previous source risk index. The underlying logic is that the rate of change of risk is more likely to trigger intervention than the absolute value of risk, and to impose a penalty far greater than linear growth on a sharp deterioration of risk, a quadratic form is adopted. The definition is as follows: ;

[0139] in, : Indicates the current time The traceability risk index; : Indicates the previous moment The traceability risk index; : This indicates a function that takes the maximum value among its internal parameters;

[0140] This formula ensures that only when the risk index rises... Only then will punishment be imposed, and the severity of the punishment increases with the square of the risk increase; this design incorporates the evaluation results of previous steps. Effectively transformed into the core behavioral driving force of DRL agents.

[0141] By designing a comprehensive reward function that incorporates three key elements—quality, efficiency, and risk trends—this invention can train a highly intelligent decision-making agent. This agent not only pursues high-quality and high-efficiency production, but more importantly, it learns to predict and proactively avoid any behavior that could worsen traceability risks by imposing a quadratic penalty on the increment of the risk index. This proactive decision-making model is far superior to the traditional passive response mechanism based on fixed thresholds, greatly enhancing the traceability resilience and forward-looking control capabilities of the entire system.

[0142] Example 7:

[0143] The closed-loop control module is specifically used for:

[0144] In response to the safe state determined by the state division module, the current printing process is maintained.

[0145] The closed-loop control module is specifically used for:

[0146] In response to the warning status determined by the status division module, a first-level correction action is executed, which includes forcibly reducing the printing speed;

[0147] The closed-loop control module is specifically used for:

[0148] In response to a dangerous state determined by the state division module, a secondary correction action is performed. The secondary correction action is to recalibrate the data flow by creating an anchor event in the data link through the execution of an emergency redundancy calibration procedure.

[0149] This embodiment is a detailed explanation of the specific execution logic of the closed-loop control module in Embodiment 4, demonstrating how the system executes a hierarchical, adaptive correction strategy based on the different health states determined by the state division module.

[0150] The closed-loop control module is responsible for receiving instructions from the risk decision module and translating them into physical control of the 3D printing equipment. Its specific behavior responds to different states determined by the state division module.

[0151] In response to the safe state determined by the state division module, the current printing process is maintained.

[0152] When the source tracing risk index When the system is in a safe state, the closed-loop control module will not perform any intervention actions; this indicates that the current data traceability chain is in good health, and the system will allow the printing process to continue according to the preset process parameters to maximize production efficiency; this ensures that the system will not make unnecessary interventions when there is no risk, avoids interference with normal production efficiency and product quality, and achieves the precision and necessity of control;

[0153] In response to the warning status determined by the status division module, a first-level correction action is executed, which includes forcibly reducing the printing speed.

[0154] when When the system enters an early warning state, the DRL agent tends to output a first-level corrective action to proactively mitigate further risk escalation. Upon receiving this instruction, the closed-loop control module will execute actions such as forcibly reducing the printing speed by 10%-30% or forcibly inserting redundant data synchronization instructions between layers. These operations, by slowing down the production process, provide more time for data transmission and verification, thereby proactively suppressing the trust gap index. The growth of the risk index aims to reduce the risk index. The risk has fallen back to a safe range. By implementing mild first-level corrective actions, the system can effectively intervene in the early stages of risk exposure, smoothing out data link fluctuations at a relatively low efficiency cost and preventing the risk from escalating into a more serious dangerous state, demonstrating the timeliness and preventative nature of the control.

[0155] In response to a dangerous state determined by the state division module, a secondary correction action is performed. The secondary correction action is to recalibrate the data flow by creating an anchor event in the data link through the execution of an emergency redundancy calibration procedure.

[0156] when When the system enters a dangerous state, the DRL agent's decision-making will shift to a more decisive secondary corrective action, the principle of which is to sacrifice the local to save the whole; the closed-loop control module will execute instructions such as an emergency redundant calibration procedure, which will actively perform an unnecessary physical action such as scraper movement or powder feeding, and forcibly record all sensor data throughout the entire cycle of the action.

[0157] An anchor event refers to a baseline event with unambiguous data and timestamps that is forcibly created by the system in a data chain that has become chaotic. Its core purpose is to provide an absolutely reliable reference point for subsequent data analysis, so as to recalibrate and verify the validity of the data flow before and after the anchor point.

[0158] This approach sacrifices short-term efficiency and localized quality in exchange for rapid recovery of the entire batch's traceability resilience, avoiding reaching the final failure boundary of the batch. When the data chain faces a severe risk of disruption, the secondary correction action provides an innovative forced reset mechanism. By proactively creating an anchor event, the system can rebuild the trust benchmark amidst chaos, achieving rapid resynchronization of the data stream, saving the entire production batch at minimal cost, and enhancing the system's robustness and recovery capabilities under extreme conditions.

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fully traceable intelligent inspection system for 3D printed parts, characterized in that: include: The data acquisition module is used to collect all-dimensional process parameters of the target printed part in real time. The all-dimensional process parameters include the internal process parameters of the equipment, the internal logical timestamp of the equipment control system, and the physical timestamp when it is received by the data acquisition server. The first processing module is used to generate a trust crack index that characterizes the integrity of the data traceability chain; The second processing module is used to determine the source tracing risk index, which characterizes the degree to which the system deviates from the security boundary, based on the trust crack index and the preset upper limit of the trust crack index. The risk decision-making module is used to output intervention actions based on the source risk index and full-dimensional process parameters through a preset deep reinforcement learning model. The closed-loop control module is used to perform intervention actions to adaptively correct the printing process of the target print.

2. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 1, characterized in that, The first processing module is specifically used for: The normalized timestamp drift is determined based on the difference between the device's internal logical timestamp and the server's physical timestamp. Determine the packet loss rate based on sequence number discontinuity detection; A trust crack index is generated by combining the weighted sum of timestamp drift and packet loss rate.

3. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 1, characterized in that, The second processing module is specifically used for: Invoke the trust crack index generated by the first processing module; The source tracing risk index was determined by normalizing the trust crack index. The normalization process involves dividing the trust crack index by a preset upper limit for the trust crack index.

4. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 1, characterized in that, Also includes: The status classification module is used to classify the health status of the system's traceability network into safe status, early warning status, and dangerous status based on the traceability risk index. Among them, the safety status corresponds to the traceability risk index being less than or equal to a preset first threshold; The warning status corresponds to the source tracing risk index being greater than the first threshold and less than or equal to the preset second threshold. A dangerous state corresponds to a traceability risk index that is greater than the second threshold.

5. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 1, characterized in that, The risk decision-making module is specifically used for: The full-dimensional process parameters and the source tracing risk index are used together to construct the state input of the deep reinforcement learning model; A policy network based on a deep reinforcement learning model outputs the optimal intervention action in real time under the current input state.

6. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 5, characterized in that, Deep reinforcement learning models are trained using reward functions, which include: A quality indicator used to measure the degree to which process parameters deviate from their optimal settings; Efficiency metrics used to quantify the current production rate; The core penalty item is used to punish an increase in the traceability risk index.

7. The intelligent inspection system for 3D printed parts with end-to-end quality traceability according to claim 6, characterized in that, The core penalty is determined by squaring the difference between the current source risk index and the previous source risk index.

8. The intelligent inspection system for 3D printed parts with end-to-end quality traceability as described in claim 4, characterized in that, The closed-loop control module is specifically used for: In response to the safe state determined by the state division module, the current printing process is maintained.

9. The intelligent inspection system for 3D printed parts with end-to-end quality traceability according to claim 4, characterized in that, The closed-loop control module is specifically used for: In response to the warning status determined by the status division module, a first-level correction action is executed, which includes forcibly reducing the printing speed.

10. The intelligent inspection system for 3D printed parts with end-to-end quality traceability according to claim 4, characterized in that, The closed-loop control module is specifically used for: In response to a dangerous state determined by the state division module, a secondary correction action is performed. The secondary correction action is to recalibrate the data flow by creating an anchor event in the data link through the execution of an emergency redundancy calibration procedure.

Images