A production quality control system for mold products

Abstract

This invention discloses a production quality control system for mold products, belonging to the field of production control technology. Specifically, it includes: establishing a dynamically updated quality status benchmark library and constructing a directed network reflecting the propagation relationship of quality characteristics between processes to achieve real-time monitoring of the production process; calculating the state offset of each process node, collecting process entropy and result entropy data when an anomaly is detected, and accurately distinguishing between material characteristic anomalies and tool status anomalies by analyzing the cooperative jump characteristics of the dual entropy sequence and the time-series pattern of entropy change sensitivity ratio, and generating corresponding process parameter adjustment instructions or tool maintenance instructions. This invention overcomes the limitations of traditional quality control systems in identifying the root causes of anomalies, achieving accurate diagnosis and classification of quality problems, and significantly improving the scientific nature and effectiveness of production quality management.

Description

Technical Field

[0001] This invention relates to the field of production control technology, and specifically to a production quality control system for mold products. Background Technology

[0002] The metal mold manufacturing process involves a complex manufacturing system with multiple steps and parameters, and its quality control level directly affects the enterprise's production efficiency and resource utilization. In the production management system of modern manufacturing enterprises, how to achieve accurate production decisions through effective quality data analysis and process monitoring has become a key issue for improving operational efficiency.

[0003] Currently, manufacturing enterprises have generally established quality management systems based on statistical process control (SPC). These systems monitor key process parameters and inspect product quality indicators to achieve basic management of production status. Some advanced enterprises have begun to introduce digital quality dashboards and quality traceability systems, recording parameter data during the production process and enabling reverse tracking of quality issues. These management tools provide the necessary data foundation for enterprise quality control and play a positive role in standardizing work processes and recording quality information.

[0004] However, when quality fluctuations occur, while the management system can identify anomalies, it cannot accurately determine whether process parameters need adjustment or equipment maintenance needs to be arranged. This lack of decision-making basis directly affects the accuracy of production scheduling. Therefore, there is an urgent need to develop a quality control system that can intelligently adapt to production status and accurately guide management decisions. By dynamically optimizing monitoring resources and clarifying the direction of anomaly handling, it can provide more effective technical support for enterprise production management. Summary of the Invention

[0005] The purpose of this invention is to provide a production quality control system for molded products, solving the following technical problems:

[0006] When quality fluctuations occur, although the management system can identify the anomalies, it cannot accurately determine whether process parameters need to be adjusted or equipment maintenance needs to be arranged. This lack of decision-making basis directly affects the accuracy of production scheduling.

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

[0008] A production quality control system for molded products includes:

[0009] The data acquisition module is used to establish a quality status benchmark library based on multi-source quality data collected during the mold production process.

[0010] The network construction module is used to mine process relationships in the quality status benchmark library and build a quality feature propagation network.

[0011] The data verification module is used to monitor the state offset of each process node in the quality feature propagation network in real time. When the state offset of any process node exceeds the preset offset threshold, the process entropy and result entropy corresponding to that process node are obtained.

[0012] The data analysis module is used to construct real-time change trajectories of process entropy and result entropy based on a sliding time window, synchronously monitor the coordinated jump events of process entropy sequence and result entropy sequence according to the real-time change trajectory, and generate an entropy change sensitivity ratio time series sequence by performing point-by-point ratio calculation between process entropy change rate sequence and result entropy change rate sequence.

[0013] The result generation module is used to determine the abnormal material properties and generate process parameter adjustment instructions when a coordinated jump event exists; when no coordinated jump event exists and the entropy change sensitivity ratio time series continuously exceeds the upper limit of the preset benchmark range, it is determined that the tool status is abnormal and tool maintenance instructions are generated.

[0014] As a further aspect of the present invention: the specific process of constructing the quality feature propagation network in the network construction module is as follows;

[0015] Extract the time series of process parameters and product quality parameters of each process node in the continuous production cycle from the quality status benchmark library. Divide the time series of process parameters of adjacent process nodes into sliding window segments. Calculate the covariance between the change trend of process parameters of the preceding process and the change trend of process parameters of the subsequent process to obtain the process parameter covariance matrix.

[0016] Simultaneously, the covariance between the changing trends of process parameters in the preceding process and the changing trends of product quality parameters in the subsequent process is calculated to obtain the quality parameter covariance matrix; the process parameter covariance matrix and the quality parameter covariance matrix are weighted and fused to obtain the process correlation strength matrix;

[0017] A directed graph network is constructed based on the process association strength matrix. Each process node is used as a vertex of the directed graph. When the element value in the process association strength matrix exceeds the association threshold, a directed edge is established between the corresponding process nodes. The direction of the directed edge is from the previous process to the subsequent process. Weight coefficients are assigned to the directed edges according to the element values ​​in the process association strength matrix to generate a quality feature propagation network.

[0018] As a further aspect of the present invention: the specific calculation process of the state offset in the data verification module is as follows:

[0019] Establish a dynamic parameter benchmark library based on a sliding time window. The mean and standard deviation of process parameters and the mean and standard deviation of product quality parameters at each process node are dynamically updated by continuously collecting recent production data. The real-time collected process parameter values ​​are compared with the corresponding dynamic process parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the process parameter offset coefficient.

[0020] The real-time collected product quality parameter values ​​are compared with the corresponding dynamic product quality parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the quality parameter offset coefficient. The process parameter offset coefficient and the quality parameter offset coefficient are fused according to a preset weight to obtain the initial state offset. The initial state offset is weighted and corrected by combining the weight coefficient of the node in the quality feature propagation network to obtain the state offset.

[0021] As a further aspect of the present invention: the specific process for calculating the process entropy and result entropy corresponding to the process node in the data verification module is as follows:

[0022] The process parameter sequence of this process node within a preset monitoring period is collected. The process parameter sequence includes at least one of equipment vibration signal, spindle power signal, and temperature signal. Time-frequency domain feature extraction is performed on the process parameter sequence to obtain the probability distribution of the feature parameters. Based on the probability distribution, the information entropy value of the process parameter is calculated as the process entropy. At the same time, the quality parameter sequence of the output of this process is obtained within the monitoring period. The quality parameter sequence includes at least one of product size data, shape tolerance data, and surface feature data. A deviation distribution model of the quality parameter value relative to the standard value is established, and the information entropy value of the deviation distribution is calculated as the result entropy.

[0023] As a further aspect of the present invention: the specific process of monitoring the coordinated jump event of the process entropy sequence and the result entropy sequence in the data analysis module is as follows:

[0024] Based on the process entropy sequence and result entropy sequence collected within a preset sliding time window, the statistical mean and standard deviation of the process entropy sequence and the statistical mean and standard deviation of the result entropy sequence are calculated respectively; the process entropy fluctuation threshold is obtained by adding a preset multiple of the standard deviation to the statistical mean of the process entropy sequence and the result entropy fluctuation threshold is obtained by adding a preset multiple of the standard deviation to the statistical mean of the result entropy sequence.

[0025] The instantaneous change in the process entropy sequence is calculated in real time. When the instantaneous change exceeds the process entropy fluctuation threshold, the current time point is recorded as the starting point of the process entropy jump. The instantaneous change in the result entropy sequence is calculated synchronously. When the instantaneous change exceeds the result entropy fluctuation threshold, the current time point is recorded as the starting point of the result entropy jump. The time difference between the starting point of the process entropy jump and the starting point of the result entropy jump is calculated. When the time difference is less than the preset synchronization time tolerance, it is determined that a coordinated jump event has occurred between the process entropy sequence and the result entropy sequence.

[0026] As a further aspect of the present invention: the specific process for obtaining the preset benchmark range in the result generation module is as follows:

[0027] Monitor the state offset of the process node. When the state offset does not exceed the preset offset threshold, collect the corresponding entropy change sensitivity ratio data to form a benchmark dataset. Calculate the statistical characteristic parameters of the benchmark dataset and determine the upper and lower limits of the benchmark range based on the statistical characteristic parameters. During continuous operation, when the state offset of the process node remains within the preset offset threshold, incorporate the real-time collected entropy change sensitivity ratio data into the benchmark dataset and recalculate the benchmark range. When the state offset of the process node exceeds the preset offset threshold, pause the update process of the benchmark dataset for that process node.

[0028] As a further aspect of the present invention: in the data acquisition module, the quality status benchmark library is constructed by collecting time-series data of process parameters of each process node within the benchmark production cycle and quality parameter detection data of the corresponding products. The process parameter feature distribution is formed based on statistical feature parameters and probability density functions extracted from the time-series data. The product quality feature distribution is realized by establishing a quality feature model based on statistical process control. This model includes specification center values, control limits, and distribution morphology features. The statistical parameters of each feature distribution are dynamically updated by continuously collecting qualified production batch data in a sliding time window manner.

[0029] As a further aspect of the present invention, the result generation module further includes a condition where, if there is no coordinated jump event but the entropy change sensitivity ratio time series does not continuously exceed the upper limit of the preset benchmark range, it is determined to be a composite anomaly and process parameter adjustment instructions and tool maintenance instructions are generated simultaneously.

[0030] The beneficial effects of this invention are:

[0031] 1) By establishing a dynamic quality benchmark library and feature propagation network, this invention achieves precise perception of production quality status and early warning of anomalies. Through constructing a dynamic benchmark library integrating multi-source quality data, it effectively tracks gradual changes such as raw material characteristic fluctuations and natural tool wear, overcoming the limitations of fixed threshold monitoring. The system dynamically updates the benchmark distribution of process parameters and quality characteristics based on a sliding time window, ensuring that quality evaluation standards remain synchronized with the current production conditions. Furthermore, the quality feature propagation network constructed through process correlation mining clearly presents the propagation path and impact range of quality anomalies between processes. This combination of dynamic benchmark management and networked monitoring significantly improves the system's sensitivity to detecting minute quality deviations, making changes in the quality status of the production process perceptible and quantifiable, providing a reliable basis for implementing preventative quality control.

[0032] 2) Based on multi-dimensional entropy analysis and collaborative jump identification, this invention achieves accurate diagnosis and classification guidance for the root causes of anomalies. By introducing a dual monitoring mechanism of process entropy and result entropy, and utilizing the essential differences in information entropy change patterns among different anomaly types, it achieves accurate differentiation of the root causes of anomalies. When material property anomalies occur, the process entropy and result entropy will undergo collaborative jumps because the changes in material properties simultaneously affect processing stability and product quality. However, in cases of tool state anomalies, the gradual change in process entropy and the stability of result entropy decouple, manifesting as a continuous anomaly in the entropy change sensitivity ratio. This diagnostic method based on entropy change characteristic patterns overcomes the shortcomings of traditional systems that can only detect problems but cannot pinpoint their root causes. It provides production managers with clear directions for handling issues, effectively avoids resource waste caused by misjudgments, and significantly improves the accuracy and efficiency of anomaly handling. Attached Figure Description

[0033] The invention will now be further described with reference to the accompanying drawings.

[0034] Figure 1 This is a schematic diagram of a production quality control system for mold products according to the present invention. Detailed Implementation

[0035] 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.

[0036] Please see Figure 1 As shown, the present invention is a production quality control system for mold products, comprising:

[0037] The data acquisition module is used to establish a quality status benchmark library based on multi-source quality data collected during the mold production process.

[0038] A multi-source sensor network is deployed at each process node of the mold production line, including vibration acceleration sensors (sampling frequency ≥10kHz) installed on the machine tool spindle, pressure sensors (accuracy ±0.1%FS) installed in the hydraulic system, and thermocouples (accuracy ±0.5℃) installed in the temperature control system to synchronously collect equipment status parameters. A coordinate measuring machine (accuracy ±1.5μm) and a surface roughness meter (measuring range 0.1-10μm) are arranged at the inspection station to obtain product quality parameters. All sensor data is transmitted to a central server via industrial Ethernet to establish a quality status benchmark database containing timestamps, equipment numbers, and process identifiers.

[0039] The network construction module is used to mine process relationships in the quality status benchmark library and build a quality feature propagation network.

[0040] The process parameter time series (such as spindle vibration RMS value, hydraulic pressure, and temperature deviation) and product quality parameter time series (such as dimensional tolerance, form and position error, and surface roughness) of each process in a continuous production cycle are extracted. A sliding window (window length of 30 production cycles, step size of 1 cycle) is used to calculate the covariance matrix between parameters of adjacent processes. Specifically, the covariance between the change trends of process parameters in the preceding process and the change trends of process parameters in the subsequent process is calculated to form the process parameter covariance matrix. At the same time, the covariance between the change trends of process parameters in the preceding process and the change trends of product quality parameters in the subsequent process is calculated to form the quality parameter covariance matrix. The two matrices are then weighted and fused with a weight of 0.6:0.4 to obtain the process correlation strength matrix. A directed graph network is constructed based on this matrix. When the matrix element value exceeds the correlation threshold (30% of the maximum value of the matrix), directed edges are established between nodes, and edge weight coefficients are assigned according to the element values ​​to generate a quality feature propagation network.

[0041] The data verification module is used to monitor the state offset of each process node in the quality feature propagation network in real time. When the state offset of any process node exceeds the preset offset threshold, the process entropy and result entropy corresponding to that process node are obtained.

[0042] A dynamic parameter benchmark library based on a sliding time window (window length 50 sampling points) is established, continuously updating the benchmark mean and standard deviation of process parameters and product quality parameters for each process node. The real-time collected process parameter values ​​are compared with the corresponding dynamic benchmark mean, and the deviation of the process parameter from the dynamic benchmark is calculated as the process parameter offset coefficient; the quality parameter offset coefficient is calculated simultaneously. The two offset coefficients are fused with a weight of 0.7:0.3 to obtain the initial state offset, which is then weighted and corrected by the node's weight coefficient in the network (the harmonic mean of the node's in-degree and out-degree) to obtain the final state offset. When the state offset of any node exceeds a preset offset threshold (2.5 times the benchmark standard deviation), the calculation of the corresponding process entropy and result entropy is triggered.

[0043] The data analysis module is used to construct real-time change trajectories of process entropy and result entropy based on a sliding time window, synchronously monitor the coordinated jump events of process entropy sequence and result entropy sequence according to the real-time change trajectory, and generate an entropy change sensitivity ratio time series sequence by performing point-by-point ratio calculation between process entropy change rate sequence and result entropy change rate sequence.

[0044] Time-frequency domain features are extracted from the process parameter sequences (vibration signals, pressure signals, and temperature signals), and the information entropy of their probability distribution is calculated as the process entropy. A deviation distribution model relative to the standard value is established for the mass parameter sequences, and its information entropy is calculated as the result entropy. Cooperative jump events of the dual-entropy sequences are monitored: the statistical mean and standard deviation of the process entropy sequence and the result entropy sequence within the window are calculated separately, and the mean plus twice the standard deviation is used as the fluctuation threshold. When the instantaneous change in process entropy exceeds its fluctuation threshold, the jump start point is recorded, and the jump start point of the result entropy is detected synchronously. If the time difference between the two start points is less than the synchronization time tolerance (3 times the sampling interval), a cooperative jump event is determined to have occurred. Simultaneously, the point-by-point ratio of the process entropy change rate sequence to the result entropy change rate sequence is calculated. When the absolute value of the result entropy change rate is less than the set significance threshold (1.5 times the standard deviation of the historical stable period change rate), a baseline value of 1 is output; otherwise, the actual ratio is output, generating an entropy change sensitivity ratio time series sequence.

[0045] The result generation module is used to determine the abnormal material properties and generate process parameter adjustment instructions when a coordinated jump event exists; when no coordinated jump event exists and the entropy change sensitivity ratio time series continuously exceeds the upper limit of the preset benchmark range, it is determined that the tool status is abnormal and tool maintenance instructions are generated.

[0046] When a coordinated jump event occurs, it is determined to be an abnormal material property, and an adjustment instruction (including injection speed compensation and holding pressure correction value) is sent to the process parameter management system. When no coordinated jump event occurs, but the entropy change sensitivity ratio time series continuously exceeds the upper limit of the preset benchmark range (taken as the 95th percentile of the benchmark dataset), it is determined to be an abnormal tool status, and a tool replacement or regrinding instruction is sent to the equipment maintenance system. All instructions are accompanied by an anomaly confidence assessment (calculated based on the offset magnitude and duration), providing a quantitative basis for production decisions.

[0047] It is worth noting that when material properties become abnormal, the change in material properties simultaneously affects the stability of the processing and the consistency of the final product's quality. This synchronicity causes process entropy (an information entropy indicator reflecting the orderliness of the production process) and result entropy (an information entropy indicator reflecting the consistency of product quality) to exhibit a coordinated jump over time. Specifically, changes in material properties immediately disrupt the adaptability of the original process window, leading to a decrease in processing stability (a sudden increase in process entropy) and directly causing a deterioration in product quality consistency (a sudden increase in result entropy). This pattern of simultaneous jumps in dual entropy becomes a typical characteristic of abnormal material properties. Tool wear-related anomalies exhibit a different entropy change pattern. The gradual deterioration of the tool condition initially leads to a gradual decrease in processing stability, manifested as a continuous increase in process entropy. However, due to the equipment's adaptive adjustment mechanism and quality compensation, product quality may still remain within acceptable limits in the early stages of wear, keeping the result entropy relatively stable. This decoupling phenomenon between decreased process orderliness and product quality stability is specifically quantified as a persistently high entropy change sensitivity ratio (the dynamic ratio of the rate of change of process entropy to the rate of change of result entropy). When this ratio continues to exceed the normal baseline range, it indicates that the quality loss corresponding to the increase in unit process entropy is significantly increased, which is a characteristic signal of abnormal tool condition.

[0048] In a preferred embodiment of the present invention, the specific process of constructing the quality feature propagation network in the network construction module is as follows:

[0049] Extract the time series of process parameters and product quality parameters of each process node in the continuous production cycle from the quality status benchmark library. Divide the time series of process parameters of adjacent process nodes into sliding window segments. Calculate the covariance between the change trend of process parameters of the preceding process and the change trend of process parameters of the subsequent process to obtain the process parameter covariance matrix.

[0050] Simultaneously, the covariance between the changing trends of process parameters in the preceding process and the changing trends of product quality parameters in the subsequent process is calculated to obtain the quality parameter covariance matrix; the process parameter covariance matrix and the quality parameter covariance matrix are weighted and fused to obtain the process correlation strength matrix;

[0051] A directed graph network is constructed based on the process association strength matrix. Each process node is used as a vertex of the directed graph. When the element value in the process association strength matrix exceeds the association threshold, a directed edge is established between the corresponding process nodes. The direction of the directed edge is from the previous process to the subsequent process. Weight coefficients are assigned to the directed edges according to the element values ​​in the process association strength matrix to generate a quality feature propagation network.

[0052] First, process parameter time series and product quality parameter time series for each process node within a continuous production cycle are extracted from the quality status benchmark library. These time series reflect the dynamic characteristics of each stage in the production process. A sliding window is used to segment the process parameter time series of adjacent process nodes; the sliding window method captures local features of parameter changes. Next, the covariance between the changing trends of process parameters in preceding and subsequent processes is calculated. Covariance calculation quantifies the degree of synergy between the changes in parameters of two processes; positive values ​​indicate changes in the same direction, negative values ​​indicate changes in opposite directions, and the absolute value reflects the strength of the correlation, thus obtaining the process parameter covariance matrix. Simultaneously, the covariance between the changing trends of process parameters in preceding and subsequent processes and the changing trends of product quality parameters in subsequent processes is calculated, reflecting the production fluctuations of the preceding process. The impact on product quality in subsequent processes is assessed to obtain the quality parameter covariance matrix. The two covariance matrices are then weighted and fused. The process parameter covariance matrix primarily reflects the equipment linkage between processes, while the quality parameter covariance matrix reflects the transmission path of quality characteristics. By assigning different weight coefficients, a balance between the two types of relationships can be achieved, ultimately yielding the process association strength matrix. A directed graph network is constructed based on this matrix, with each process node as a vertex. When the element value in the matrix exceeds a set association threshold, a directed edge is established between the corresponding process nodes, pointing from the preceding process to the subsequent process. This directional design aligns with the actual production process flow. Based on the matrix element values, corresponding weight coefficients are assigned to the directed edges. The magnitude of the weight values ​​directly reflects the strength of the association between processes, ultimately generating a complete quality characteristic propagation network.

[0053] By calculating the covariance between process parameters, implicit correlations in the production process can be quantitatively identified. These correlations often reflect mechanical transmission, thermal coupling, or inheritance of quality characteristics between equipment. Sliding window analysis captures changes in these correlations under dynamic production conditions, making the constructed network time-varying and adaptable. Setting the direction of directed edges from preceding processes to subsequent processes perfectly aligns with the propagation direction of quality characteristics in the actual production process, ensuring a clear physical meaning for the network structure. By setting correlation thresholds, weak, accidental correlations can be filtered out, retaining strong, significantly influential paths, making the network structure clearer and more reliable. After establishing such a quality characteristic propagation network, the system can accurately track the propagation path of quality anomalies between processes, identify key process nodes affecting quality, and provide an important topological foundation for subsequent state offset calculations and anomaly diagnosis. This network-based analysis method overcomes the limitations of traditional single-point quality monitoring, enabling the system to understand the formation process of production quality from a global perspective, laying a solid foundation for precise quality control.

[0054] In another preferred embodiment of the present invention, the specific calculation process of the state offset in the data verification module is as follows:

[0055] Establish a dynamic parameter benchmark library based on a sliding time window. The mean and standard deviation of process parameters and the mean and standard deviation of product quality parameters at each process node are dynamically updated by continuously collecting recent production data. The real-time collected process parameter values ​​are compared with the corresponding dynamic process parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the process parameter offset coefficient.

[0056] The real-time collected product quality parameter values ​​are compared with the corresponding dynamic product quality parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the quality parameter offset coefficient. The process parameter offset coefficient and the quality parameter offset coefficient are fused according to a preset weight to obtain the initial state offset. The initial state offset is weighted and corrected by combining the weight coefficient of the node in the quality feature propagation network to obtain the state offset.

[0057] In a preferred embodiment of the present invention, the specific process for calculating the process entropy and result entropy corresponding to the process node in the data verification module is as follows:

[0058] The process parameter sequence of this process node within a preset monitoring period is collected. The process parameter sequence includes at least one of equipment vibration signal, spindle power signal, and temperature signal. Time-frequency domain feature extraction is performed on the process parameter sequence to obtain the probability distribution of the feature parameters. Based on the probability distribution, the information entropy value of the process parameter is calculated as the process entropy. At the same time, the quality parameter sequence of the output of this process is obtained within the monitoring period. The quality parameter sequence includes at least one of product size data, shape tolerance data, and surface feature data. A deviation distribution model of the quality parameter value relative to the standard value is established, and the information entropy value of the deviation distribution is calculated as the result entropy.

[0059] In another preferred embodiment of the present invention, the specific process of monitoring the coordinated jump event of the process entropy sequence and the result entropy sequence in the data analysis module is as follows:

[0060] Based on the process entropy sequence and result entropy sequence collected within a preset sliding time window, the statistical mean and standard deviation of the process entropy sequence and the statistical mean and standard deviation of the result entropy sequence are calculated respectively; the process entropy fluctuation threshold is obtained by adding a preset multiple of the standard deviation to the statistical mean of the process entropy sequence and the result entropy fluctuation threshold is obtained by adding a preset multiple of the standard deviation to the statistical mean of the result entropy sequence.

[0061] The instantaneous change in the process entropy sequence is calculated in real time. When the instantaneous change exceeds the process entropy fluctuation threshold, the current time point is recorded as the starting point of the process entropy jump. The instantaneous change in the result entropy sequence is calculated synchronously. When the instantaneous change exceeds the result entropy fluctuation threshold, the current time point is recorded as the starting point of the result entropy jump. The time difference between the starting point of the process entropy jump and the starting point of the result entropy jump is calculated. When the time difference is less than the preset synchronization time tolerance, it is determined that a coordinated jump event has occurred between the process entropy sequence and the result entropy sequence.

[0062] First, process entropy sequences and result entropy sequences are continuously collected based on a sliding time window of preset length. This sliding window approach dynamically tracks the latest state changes in the production process. Next, the statistical mean and standard deviation of the process entropy sequence and the result entropy sequence within the window are calculated respectively. A dynamic benchmark for the current production state is established through the calculation of statistical characteristics. Then, a preset multiple of the standard deviation of the process entropy sequence is added to the statistical mean of the process entropy sequence to determine the process entropy fluctuation threshold. The result entropy fluctuation threshold is determined using the same method. This threshold setting based on dynamic statistical characteristics can adapt to the normal fluctuation range in the production process. During the real-time monitoring phase, the process entropy sequence is calculated... The difference between adjacent sampling points yields the instantaneous change. When this change exceeds the process entropy fluctuation threshold, it indicates a significant abrupt change in the orderliness of the production process. At this point, the current time point is recorded as the starting point of the process entropy jump. Simultaneously, the instantaneous change of the result entropy sequence is calculated using the same method. When it exceeds the result entropy fluctuation threshold, it indicates a significant abnormality in product quality consistency, and this is recorded as the starting point of the result entropy jump. Finally, the time difference between the two jump starting points is calculated. When this difference is less than the preset synchronization time tolerance, it is determined that a coordinated jump event has occurred between the process entropy sequence and the result entropy sequence. This time synchronization judgment can effectively capture the causal relationship between the two types of entropy value changes.

[0063] The synchronicity analysis of dual-entropy sequences is used to distinguish the root causes of anomalies. When material properties become abnormal, the change in material properties simultaneously affects the stability of the processing and the consistency of product quality. This synchronicity causes process entropy and result entropy to exhibit a coordinated jump characteristic over time. However, in cases of gradual anomalies such as tool wear, the change in process entropy often precedes the change in result entropy, and there is a lack of strict temporal synchronicity between the two. By setting reasonable time tolerance parameters, it is possible to capture truly coordinated anomaly events while filtering out random fluctuations. The advantage of this monitoring mechanism is that it can deeply analyze the anomaly propagation pattern at the temporal characteristic level, providing a key basis for subsequent anomaly classification and diagnosis. This enables accurate differentiation between material property anomalies and tool condition anomalies, avoiding the misjudgment problems easily caused by traditional single-threshold judgment methods, and ultimately improving the decision-making accuracy and reliability of the quality control system.

[0064] In another preferred embodiment of the present invention, the specific process for obtaining the preset benchmark range in the result generation module is as follows:

[0065] Monitor the state offset of the process node. When the state offset does not exceed the preset offset threshold, collect the corresponding entropy change sensitivity ratio data to form a benchmark dataset. Calculate the statistical characteristic parameters of the benchmark dataset and determine the upper and lower limits of the benchmark range based on the statistical characteristic parameters. During continuous operation, when the state offset of the process node remains within the preset offset threshold, incorporate the real-time collected entropy change sensitivity ratio data into the benchmark dataset and recalculate the benchmark range. When the state offset of the process node exceeds the preset offset threshold, pause the update process of the benchmark dataset for that process node.

[0066] First, the state offset of the process node is continuously monitored. When the state offset does not exceed the preset offset threshold, it indicates that the current production process is in a relatively stable state. At this time, the corresponding entropy change sensitivity ratio data is collected and included in the benchmark dataset. This step ensures that the benchmark data all come from normal production conditions. Next, statistical analysis is performed on the benchmark dataset to calculate its statistical characteristic parameters, including the central tendency and dispersion of the data distribution. Based on these statistical characteristic parameters, the upper and lower limits of the benchmark range are determined. For example, the normal fluctuation range can be defined by calculating specific percentiles or the mean plus or minus a certain number of standard deviations of the dataset. During the continuous operation of the system, when the state offset of the process node always remains within the preset offset threshold, it indicates that the production process is continuously stable. At this time, newly collected entropy change sensitivity ratio data is continuously included in the benchmark dataset and the benchmark range is recalculated. This dynamic update mechanism allows the benchmark range to adapt to the slow changes in production conditions. Conversely, when the state offset exceeds the preset offset threshold, it indicates that the production process is abnormal. At this time, the update process of the benchmark dataset for that process node is immediately suspended to prevent abnormal data from contaminating the benchmark dataset and to ensure the purity and representativeness of the benchmark range.

[0067] The purity of the benchmark dataset is ensured by continuously monitoring state offsets. New data is only added to the benchmark set when the production process is normal, thus avoiding contamination of the benchmark range by abnormal data. Dynamically updating the benchmark range allows the system to adapt to slow drifts caused by equipment aging and environmental changes during production, maintaining the accuracy of diagnostic criteria. When an anomaly is detected, benchmark dataset updates are immediately paused, effectively preventing data under abnormal conditions from interfering with the benchmark standards and ensuring the reliability of subsequent diagnoses. The advantage of this mechanism lies in establishing an adaptive dynamic benchmark, avoiding the shortcomings of fixed thresholds that cannot adapt to changes in production conditions, and preventing benchmark distortion caused by data contamination. This provides a reliable basis for accurately distinguishing tool state anomalies, ultimately improving the adaptability and diagnostic accuracy of the entire quality control system in complex production environments.

[0068] In another preferred embodiment of the present invention, in the data acquisition module, the quality status benchmark library is constructed by collecting time-series data of process parameters of each process node within a benchmark production cycle and quality parameter detection data of the corresponding products. The process parameter feature distribution is formed based on statistical feature parameters and probability density functions extracted from the time-series data, and the product quality feature distribution is realized by establishing a quality feature model based on statistical process control. This model includes specification center values, control limits and distribution morphology features, and dynamically updates the statistical parameters of each feature distribution by continuously collecting qualified production batch data in a sliding time window manner.

[0069] In another preferred embodiment of the present invention, the result generation module further includes determining a composite anomaly and simultaneously generating process parameter adjustment instructions and tool maintenance instructions when there is no coordinated jump event but the entropy change sensitivity ratio time series does not continuously exceed the upper limit of the preset benchmark range.

[0070] Different types of anomalies in the production process exhibit distinguishable characteristics in their information entropy change patterns. When material properties become abnormal, the change in material properties simultaneously affects the stability of the processing process and the consistency of the final product's quality. This synchronicity causes process entropy and result entropy to exhibit a time-dependent, coordinated jump characteristic. However, in cases of tool condition anomalies, the gradual change in process entropy decouples from the stability of result entropy, manifesting as a persistent anomaly in the entropy change sensitivity ratio. When no typical coordinated jump event is detected, and the entropy change sensitivity ratio does not consistently exceed the upper limit of the baseline range, it indicates a complex situation in the production process that differs from both simple material anomalies and typical tool anomalies. This situation often stems from the combined effect of slight fluctuations in material properties and early tool wear, or from certain transitional states that have not yet developed into typical anomalies. Although the changes in process entropy and result entropy are correlated, they do not reach the level of strictly synchronized jumps; simultaneously, although the entropy change sensitivity ratio fluctuates, it has not yet formed a stable pattern of continuous over-limit. This unique combination of entropy change characteristics reveals the combined effect of multiple minor anomalies in the production system, requiring comprehensive measures to address.

[0071] By identifying this transitional state between typical material anomalies and tooling anomalies, the system can detect potential problems earlier and take preventative measures before the anomaly fully develops. Simultaneously, the identification of complex anomalies avoids incomplete handling issues that may result from simple categorization, ensuring the comprehensiveness and effectiveness of the response plan. This design enables the quality control system to handle complex production situations, adapting to real-world scenarios where multiple factors interact, significantly improving the accuracy and reliability of anomaly diagnosis, and providing more comprehensive technical support for maintaining the stable operation of the production system.

[0072] 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. A production quality control system for molded products, characterized in that, include: The data acquisition module is used to establish a quality status benchmark library based on multi-source quality data collected during the mold production process. The network construction module is used to mine process relationships in the quality status benchmark library and build a quality feature propagation network. The data verification module is used to monitor the state offset of each process node in the quality feature propagation network in real time. When the state offset of any process node exceeds the preset offset threshold, the process entropy and result entropy corresponding to that process node are obtained. The data analysis module is used to construct real-time change trajectories of process entropy and result entropy based on a sliding time window, synchronously monitor the coordinated jump events of process entropy sequence and result entropy sequence according to the real-time change trajectory, and generate an entropy change sensitivity ratio time series sequence by performing point-by-point ratio calculation between process entropy change rate sequence and result entropy change rate sequence. The specific process of monitoring the coordinated jump event of the process entropy sequence and the result entropy sequence is as follows: Based on the process entropy sequence and result entropy sequence collected within a preset sliding time window, the statistical mean and standard deviation of the process entropy sequence and the statistical mean and standard deviation of the result entropy sequence are calculated respectively. The statistical mean of the process entropy sequence plus a preset multiple of its standard deviation is used as the process entropy fluctuation threshold, and the statistical mean of the result entropy sequence plus a preset multiple of its standard deviation is used as the result entropy fluctuation threshold. The instantaneous change in the process entropy sequence is calculated in real time. When the instantaneous change exceeds the process entropy fluctuation threshold, the current time point is recorded as the starting point of the process entropy jump. The instantaneous change in the result entropy sequence is calculated synchronously. When the instantaneous change exceeds the result entropy fluctuation threshold, the current time point is recorded as the result entropy jump start point. The time difference between the starting point of the process entropy jump and the starting point of the result entropy jump is calculated. When the time difference is less than the preset synchronization time tolerance, it is determined that a coordinated jump event has occurred between the process entropy sequence and the result entropy sequence. The result generation module is used to determine the abnormal material properties and generate process parameter adjustment instructions when a coordinated jump event occurs. When there is no collaborative jump event and the entropy change sensitivity ratio time series continuously exceeds the upper limit of the preset benchmark range, the tool state is determined to be abnormal and a tool maintenance instruction is generated.

2. The production quality control system for molded products according to claim 1, characterized in that, The specific process of constructing the quality feature propagation network in the network construction module is as follows: Extract the time series of process parameters and product quality parameters of each process node in the continuous production cycle from the quality status benchmark library. Divide the time series of process parameters of adjacent process nodes into sliding window segments. Calculate the covariance between the change trend of process parameters of the preceding process and the change trend of process parameters of the subsequent process to obtain the process parameter covariance matrix. Simultaneously, the covariance between the changing trends of process parameters in the preceding process and the changing trends of product quality parameters in the subsequent process is calculated to obtain the quality parameter covariance matrix. The process parameter covariance matrix and the quality parameter covariance matrix are weighted and fused to obtain the process correlation strength matrix; A directed graph network is constructed based on the process association strength matrix. Each process node is used as a vertex of the directed graph. When the element value in the process association strength matrix exceeds the association threshold, a directed edge is established between the corresponding process nodes. The direction of the directed edge is from the previous process to the subsequent process. Weight coefficients are assigned to the directed edges according to the element values ​​in the process association strength matrix to generate a quality feature propagation network.

3. The production quality control system for molded products according to claim 1, characterized in that, The specific calculation process for the state offset in the data verification module is as follows: Establish a dynamic parameter benchmark library based on a sliding time window, and dynamically update the benchmark mean and standard deviation of process parameters and product quality parameters for each process node by continuously collecting recent production data; The real-time collected process parameter values ​​are compared with the corresponding dynamic process parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the process parameter offset coefficient. The real-time collected product quality parameter values ​​are compared with the corresponding dynamic product quality parameter benchmark mean, and the standard deviation multiple of the deviation from the dynamic benchmark is calculated as the quality parameter offset coefficient. The initial state offset is obtained by fusing the process parameter offset coefficient and the quality parameter offset coefficient according to a preset weight. By combining the weight coefficient of the node in the quality feature propagation network, the initial state offset is weighted and corrected to obtain the state offset.

4. The production quality control system for molded products according to claim 1, characterized in that, In the data verification module, the specific process for calculating the process entropy and result entropy corresponding to the process node is as follows: The process parameter sequence of this process node within a preset monitoring period is collected. The process parameter sequence includes at least one of equipment vibration signal, spindle power signal, and temperature signal. Time-frequency domain feature extraction is performed on the process parameter sequence to obtain the probability distribution of the feature parameters. Based on the probability distribution, the information entropy value of the process parameter is calculated as the process entropy. At the same time, the quality parameter sequence of the output of this process is obtained within the monitoring period. The quality parameter sequence includes at least one of product size data, shape tolerance data, and surface feature data. A deviation distribution model of the quality parameter value relative to the standard value is established, and the information entropy value of the deviation distribution is calculated as the result entropy.

5. A production quality control system for molded products according to claim 3, characterized in that, The specific process for obtaining the preset benchmark range in the result generation module is as follows: Monitor the state offset of the process node. When the state offset does not exceed the preset offset threshold, collect the corresponding entropy change sensitivity ratio data to form a benchmark dataset. Calculate the statistical characteristic parameters of the benchmark dataset and determine the upper and lower limits of the benchmark range based on the statistical characteristic parameters. During continuous operation, if the state offset of the process node does not exceed the preset offset threshold, the entropy change sensitivity ratio data collected in real time will be included in the benchmark dataset and the benchmark range will be recalculated; if the state offset of the process node exceeds the preset offset threshold, the update process of the benchmark dataset of the process node will be paused.

6. The production quality control system for molded products according to claim 1, characterized in that, In the data acquisition module, the quality status benchmark library is constructed by collecting time-series data of process parameters of each process node within the benchmark production cycle and quality parameter detection data of the corresponding products. The process parameter feature distribution is formed based on the statistical feature parameters and probability density function extracted from the time-series data. The product quality feature distribution is realized by establishing a quality feature model based on statistical process control. This model includes specification center value, control limits and distribution morphology features, and dynamically updates the statistical parameters of each feature distribution by continuously collecting qualified production batch data in a sliding time window manner.

7. The production quality control system for molded products according to claim 1, characterized in that, The result generation module also includes a condition where, if there is no coordinated jump event and the entropy change sensitivity ratio time series does not continuously exceed the upper limit of the preset benchmark range, it is determined to be a composite anomaly and process parameter adjustment instructions and tool maintenance instructions are generated simultaneously.

Images