Real-time monitoring method and system for plastic part production line

Abstract

The invention relates to the technical field of injection molding processes, in particular to a real-time monitoring method and system for a plastic part production line. The method comprises the following steps: acquiring and standardizing reference batch historical data of a plastic part production line, and constructing a reference process model containing a reference mass center and a reference form matrix; collecting and standardizing process parameter vectors of the current period in real time, and storing the process parameter vectors into a time sequence sliding window; calculating a current centroid and a current form matrix in the time sequence sliding window; and acquiring a current window form expansion factor and a current window position drift factor, and comparing the factors with a form expansion threshold and a position drift threshold respectively to complete real-time monitoring of the plastic part production line. According to the method, the reference process model is constructed, the sliding window and the form and position factors are combined, the system state is judged through double thresholds, the problem that time sequence perception of a traditional algorithm is insufficient is solved, the accuracy and practicability of real-time monitoring of a plastic part production line are improved, and early abnormity recognition and precise diagnosis are supported.

Description

Technical Field

[0001] This invention relates to the field of injection molding technology, and in particular to a method and system for real-time monitoring of plastic parts production lines. Background Technology

[0002] Injection molding of plastic parts is one of the core processes in modern manufacturing. The quality of plastic products, such as electrical appliance housings and precision components, depends heavily on the precise coordination of multiple process parameters, such as injection pressure, mold temperature, holding time, and cooling time. Ensuring that these multi-dimensional parameters are always maintained within the optimal process window during production is key to guaranteeing consistency and pass rate in mass production.

[0003] Currently, many production lines still rely on traditional monitoring methods, which involve setting fixed upper and lower thresholds for individual process parameters, such as ensuring that the mold temperature does not exceed a certain set value. This monitoring method can only detect extreme exceedances of a single parameter, but it cannot identify coordinated anomalies. Coordinated anomalies refer to situations where, although each parameter is within its allowable tolerance range when viewed individually, the combination of these parameters deviates from the optimal process setting, which can also lead to batch quality defects such as shrinkage and flash in the products.

[0004] To address the problem of detecting collaborative anomalies, existing technologies typically employ density-based unsupervised anomaly detection algorithms, such as the Local Anomaly Factor (LOF) algorithm. The LOF algorithm determines whether a data point deviates from the normal data cluster by comparing the local density of a data point (i.e., the local density of a multidimensional parameter combination of a production cycle) with the local density of its neighboring data points.

[0005] However, when applied to continuous, real-time time-series data streams such as injection molding production lines, traditional LOF algorithms lack the ability to perceive temporal context, focusing only on spatial neighborhood while ignoring temporal neighborhood. This leads to the failure to detect process drift-type anomalies. For example, when equipment wears down slowly, causing process parameters to deviate from the baseline window collaboratively and gradually over several hours, each new data point is still locally dense compared to its neighboring points that are also drifting, causing traditional LOF algorithms to consistently misclassify them as normal. Furthermore, traditional LOF algorithms cannot distinguish the type of anomaly. Whether it is an instantaneous spike anomaly caused by sudden equipment failure or a gradual drift anomaly caused by slow equipment wear, traditional algorithms can only categorize them as anomalies in general, failing to provide differentiated information for subsequent fault diagnosis and maintenance decisions, thus reducing the practicality of real-time monitoring systems for plastic parts production lines. Summary of the Invention

[0006] To address the technical problem that existing real-time monitoring technologies cannot effectively detect process drift anomalies due to the lack of temporal context, and also cannot distinguish anomaly types, this invention provides a real-time monitoring method and system for plastic parts production lines.

[0007] In a first aspect, the present invention provides a method for real-time monitoring of a plastic parts production line, employing the following technical solution: A real-time monitoring method for a plastic parts production line includes: acquiring historical data from multiple production cycles, including a benchmark batch of the plastic parts production line, and performing standardization processing; constructing a benchmark process model based on the standardized historical data, the benchmark process model including a benchmark centroid and a benchmark morphology matrix; acquiring process parameter vectors of the plastic parts production line in the current production cycle in real time, and performing standardization processing; storing the standardized process parameter vectors in a preset time-series sliding window; calculating the current centroid and current morphology matrix of all process parameter vectors within the time-series sliding window; and calculating the inverse of the current morphology matrix and the benchmark morphology matrix. A matrix is ​​used to determine the current window shape expansion factor, which characterizes the degree of data dispersion within a short-term time-series sliding window. Based on the current centroid, the reference centroid, and the inverse matrix of the reference shape matrix, a current window position drift factor is determined, which characterizes the degree of deviation of the short-term time-series sliding window centroid from the reference centroid. A shape expansion threshold and a position drift threshold are set. Based on the comparison results of the current window shape expansion factor and the shape expansion threshold, and the current window position drift factor and the position drift threshold, the state of the plastic parts production line is diagnosed as one of normal, instantaneous spike, gradual drift, or spike in drift, thus completing the real-time monitoring of the plastic parts production line.

[0008] The beneficial effects are as follows: By constructing a benchmark process model containing a benchmark centroid and a benchmark morphology matrix, the normal operating conditions of the plastic parts production line are modeled, accurately reflecting the statistical characteristics and covariance structure of the normal production process, providing a reliable benchmark reference for subsequent anomaly detection; through the time-series sliding window mechanism, dynamic management of real-time process parameter vectors is achieved, capturing changes in data characteristics within a short time window, providing a data foundation for time-series context awareness; by calculating the current window morphology expansion factor and position drift factor, quantitative characterization of different types of anomalies is achieved, with the morphology expansion factor reflecting changes in data dispersion and the position drift factor reflecting the degree of overall offset; through a dual judgment mechanism of setting morphology expansion threshold and position drift threshold, accurate classification of four states—normal, instantaneous spike, gradual drift, and spike in drift—is achieved, providing differentiated information support for subsequent fault diagnosis and maintenance decisions; this invention effectively solves the problem of the lack of time-series awareness capability in traditional LOF algorithms, improves the accuracy and practicality of real-time monitoring systems for plastic parts production lines, and provides reliable technical support for early identification and accurate diagnosis of process anomalies.

[0009] Furthermore, the construction of the benchmark process model includes: using the mean of all process parameter vectors of the standardized historical data as the benchmark centroid; and using the covariance matrix of all process parameter vectors of the standardized historical data as the benchmark morphological matrix.

[0010] The beneficial effects are as follows: by using the mean of standardized historical data as the reference centroid, the central trend under normal operating conditions of the plastic parts production line is accurately reflected; by using the covariance matrix as the reference morphology matrix, the correlation between various process parameters and the discrete characteristics of the data are effectively captured; the construction of the reference centroid ensures the accurate representation of the center position under normal operating conditions, providing a reliable reference point for the calculation of the position drift factor; and the construction of the reference morphology matrix retains the covariance information between parameters, providing an accurate shape reference for the calculation of the morphological expansion factor.

[0011] Furthermore, the standardization process employs Z-score standardization.

[0012] Furthermore, the current window shape expansion factor satisfies: In the formula, For the current production cycle The corresponding current window shape inflation factor, For the current production cycle The corresponding current form matrix, It is the inverse of the base form matrix. This is the trace operation of a matrix.

[0013] The beneficial effects are as follows: By calculating the trace of the product of the current morphological matrix and the inverse of the baseline morphological matrix, an evaluation index for the morphological expansion factor is constructed, which can effectively measure the change in the dispersion of the current window data relative to the baseline model. The matrix trace operation ensures that the morphological expansion factor is a scalar value, which facilitates subsequent threshold comparison and state diagnosis. When the current morphological matrix is ​​similar to the baseline morphological matrix, the product is close to the identity matrix, the trace value is close to the parameter dimension, and the morphological expansion factor is 1. When the dispersion of the current morphological matrix is ​​greater than that of the baseline, the trace value increases, and the morphological expansion factor is greater than 1. The morphological expansion factor can effectively identify the increase in data dispersion caused by equipment wear or process parameter drift, providing an important basis for the detection of gradual drift anomalies and improving the accuracy and sensitivity of anomaly detection.

[0014] Furthermore, the current window position drift factor satisfies: In the formula, For the current production cycle The corresponding current window position drift factor, For the current production cycle The corresponding current centroid, Using the reference centroid, It is the inverse of the base form matrix. This is a vector transpose operation.

[0015] The beneficial effects are as follows: By constructing a position drift factor using the Mahalanobis distance formula, a scientific assessment of the degree to which the current window centroid deviates from the baseline centroid is achieved, fully considering the covariance structure of the baseline distribution and ensuring the accuracy and rationality of the distance calculation; the introduction of the inverse matrix of the baseline morphology matrix allows the distance calculation to consider the correlation between various process parameters and the variability of different parameters. When the centroid deviation direction is consistent with the direction of high variability, the distance is relatively small, and when the deviation direction is consistent with the direction of low variability, the distance is relatively large; the position drift factor can effectively identify the overall position change caused by equipment drift or systematic deviation, providing an important basis for distinguishing between instantaneous spikes and gradual drift anomalies, and improving the accuracy and reliability of anomaly type identification.

[0016] Furthermore, the conditions for diagnosing the instantaneous spike include: the current window shape expansion factor is greater than the shape expansion threshold, and the current window position drift factor is less than or equal to the position drift threshold.

[0017] Furthermore, the conditions for diagnosing the gradual drift include: the current window shape expansion factor is not greater than the shape expansion threshold, and the current window position drift factor is greater than the position drift threshold.

[0018] Furthermore, the conditions for diagnosing a spike in the drift include: the current window shape expansion factor is greater than the shape expansion threshold, and the current window position drift factor is greater than the position drift threshold.

[0019] Furthermore, it also includes adaptively updating the baseline process model, including: after the gradual drift anomaly continues for a set number of cycles, triggering manual confirmation; if it is confirmed as benign drift, extracting the process parameter vectors of the most recent multiple cycles as new historical data, and reconstructing the baseline process model.

[0020] The beneficial effects are as follows: By setting a number of cycles to trigger a manual confirmation mechanism, intelligent identification and handling of continuous gradual drift are achieved, which can ensure system stability while responding promptly to possible process improvements or equipment adjustments; the manual confirmation process ensures the accuracy and reliability of the baseline model update, avoiding erroneous updates caused by real faults. The baseline model is only rebuilt after a benign drift is confirmed, ensuring the accuracy of the monitoring system; by extracting the process parameter vectors from the most recent multiple cycles as new historical data, adaptive updates of the baseline model are achieved, which can adapt to the normal evolution of the process and maintain the timeliness and accuracy of the baseline model.

[0021] Secondly, the present invention provides a real-time monitoring system for a plastic parts production line, which adopts the following technical solution: A real-time monitoring system for a plastic parts production line includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned real-time monitoring method for a plastic parts production line is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned real-time monitoring method for plastic parts production lines and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of terminal equipment based on the memory and processor, making it convenient to use.

[0023] The present invention has the following technical effects: (1) Breaking through the limitations of traditional LOF algorithms that only focus on spatial neighborhood and ignore temporal context, this algorithm achieves effective identification of slow drift anomalies through a time-series sliding window and position drift factor. Based on historical steady-state data, a benchmark process model is constructed, and the deviation between the centroid of the current window and the benchmark centroid is calculated in real time. When equipment wear and other factors cause process parameters to deviate from the benchmark in a coordinated and slow manner, the position drift factor will continue to accumulate and increase. Even if a single data point and the adjacent drift data remain locally dense, it can be accurately determined as a gradual drift through the temporal characteristics of continuous centroid deviation, thus solving the problem of missed detection of process drift anomalies by traditional algorithms.

[0024] (2) Abandoning the traditional algorithm's mode of only judging anomalies and not classifying them, the system achieves accurate diagnosis of various abnormal states through a two-dimensional combination of morphological expansion factor and positional drift factor. The classification diagnosis directly provides differentiated guidance for operation and maintenance. For example, instantaneous spikes require emergency troubleshooting of sudden faults, and gradual drifts can be scheduled for shutdown maintenance. This solves the problem that the traditional method has only single abnormal information and cannot support accurate decision-making, thus improving the practicality of the real-time monitoring system for plastic parts.

[0025] (3) By constructing a historical benchmark model and standardizing data, the monitoring benchmark is made to fit the actual steady-state working conditions of the production line, avoiding the misjudgment of normal process fluctuations, such as parameter fluctuations caused by slight differences in raw material batches, as abnormalities; at the same time, the comparison between the current morphology matrix and the benchmark morphology matrix can more accurately assess the degree of data dispersion, filter out false fluctuations caused by measurement noise, ensure the reliability of anomaly judgment, and solve the problems of traditional algorithms being easily interfered with and having a high false alarm rate.

[0026] (4) The time-series sliding window updates the process parameter vector of the current production cycle in real time. By calculating the current centroid and current morphology matrix in real time and comparing them with the benchmark, it can achieve rapid response to anomalies. Whether it is the instantaneous spike of sudden equipment damage or the gradual drift of slow equipment wear, it can be captured and diagnosed in time, avoiding the expansion of faults due to monitoring lag, such as batch non-conforming products and serious equipment damage caused by drift accumulation, thus meeting the continuous and high real-time monitoring requirements of plastic parts production lines. Attached Figure Description

[0027] Figure 1 This is a flowchart of a real-time monitoring method for a plastic parts production line according to an embodiment of the present invention.

[0028] Figure 2 This is a timing diagram of the original process parameters in a real-time monitoring method for a plastic parts production line according to an embodiment of the present invention.

[0029] Figure 3 This is a timing diagram of dual-index diagnosis in a real-time monitoring method for a plastic parts production line according to an embodiment of the present invention. Detailed Implementation

[0030] 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, not all, of the embodiments of the present invention. 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.

[0031] This invention discloses a real-time monitoring method for a plastic parts production line, referring to... Figure 1 This includes steps S001-S004: S001: Offline construction of baseline process model.

[0032] In the initial stage of deployment in this embodiment, it is necessary to first establish a mathematical model of a baseline process window.

[0033] (1) Data Acquisition and Preprocessing: Historical production data is retrieved, and production line managers or quality inspectors select at least one production cycle data that is confirmed as a benchmark batch. The production cycle data of the benchmark batch represents the combination of process parameters when the production quality is optimal. It is assumed that each production cycle is monitored. One process parameter, for example, The parameters are: injection pressure, mold temperature, and cooling time. Therefore, the data for one production cycle is one... dimensional vector , such as 3D vectors Injection pressure mold temperature Cooldown time The process parameter vectors of all benchmark batches constitute the historical benchmark set. Because the dimensions and numerical ranges of different process parameters vary greatly (e.g., pressure is measured in MPa, temperature in °C, and time in seconds), normalization is necessary. This embodiment uses Z-score normalization for calculation. Each dimension mean and standard deviation ,in, The mean vector is calculated. Standard deviation vector All subsequent data will be standardized using the aforementioned statistical vectors.

[0034] (2) Constructing a benchmark process model: Calculating and storing the standardized process model. The core statistical features constitute the Base Craftsmanship Model (BCM). , It contains two core components: the reference centroid. and the baseline morphological matrix , yes All data points A mean vector, according to the definition of Z-score standardization, the mean of the standardized data is the zero vector, i.e. ,in, This is a vector transpose operation. It represents the center position of the reference process window, i.e., the reference centroid; yes of The covariance matrix describes the shape and size of the baseline process window, i.e., the baseline shape matrix. Its diagonal elements are the variances of each parameter individually, which are standardized to 1. The off-diagonal elements are the pairwise covariances between parameters, such as the synergistic relationship between pressure and temperature. (Offline storage) and normalization parameters .

[0035] (3) Initialize timing sliding window: Initialize a short-term timing sliding window that is first-in-first-out. The capacity is preset to Exemplary One cycle.

[0036] By building a baseline process model offline, a gold standard is provided for real-time monitoring of subsequent plastic parts production lines, establishing a benchmark center and normal fluctuation pattern for evaluating process status.

[0037] S002: Real-time calculation of time-series sliding window state characteristics.

[0038] During the real-time monitoring phase of the plastic parts production line, the following operations are performed for each new production cycle: (1) Data update: collecting production cycle The original data vector Exemplary At that time, the data was collected. Use the data stored in S1 and right Perform the same Z-score standardization to obtain The calculation process is as follows: ; then, deposit ,like If 100 cycles have been stored, remove the oldest data. ,Right now .

[0039] (2) Calculate the statistics within the window: in the window At least fill After each data point, the current centroid of the time-series sliding window is calculated in real time. and the current form matrix , yes middle The mean vector of the data points reflects the recent Average process status over one cycle; yes middle The covariance matrix of the data points reflects the recent The process fluctuation pattern of each cycle.

[0040] (3) Calculate the morphological expansion factor : Used to evaluate the current time-series sliding window Relative to the reference process window The degree of looseness, a stable process window, and its internal data form It should be consistent with the baseline morphological matrix To maintain consistency, an intrusion at a momentary spike will cause it Dramatic expansion, i.e., a dramatic increase in variance, is calculated as follows: ; In the formula, For the current production cycle The corresponding current window shape inflation factor, For the current production cycle The corresponding current form matrix, It is the inverse of the base form matrix. The trace operation is the sum of the elements on the main diagonal. The above formula calculates the generalized variance ratio of the current time-series sliding window shape relative to the baseline shape.

[0041] Calculation example: When the production line is stable, see Figure 2 0-200 cycles and Very close, assuming close identity matrix , ,therefore, , such as in In the example See Figure 3 The red curve remained at a low level in the early stage; When a transient spike occurs, see Figure 2 The area marked by the red dashed line represents an extreme anomaly, such as... Breaking into the window ,lead to The variance increases dramatically, such as , It will rise instantly, much greater than See Figure 3 Red curve It surges instantly during peak hours; When gradual drift occurs, see Figure 2 After 200 cycles, all points within the window move closely together, maintaining a compact internal shape. The changes are minor, therefore, Staying low, still Nearby, see Figure 3 The red curve remained at a low level during the drift phase after 200 cycles.

[0042] (4) Calculate the position drift factor : The metric is used to evaluate the current time series sliding window. center of mass Offset from the reference process window centroid The degree of a gradual drift process, the shape inside the window. Maintain stability, but center of gravity The displacement will continue, and the calculation process is as follows: ; In the formula, For the current production cycle The corresponding current window position drift factor, For the current production cycle The corresponding current centroid, Using the reference centroid, It is the inverse of the base form matrix. This is a vector transpose operation. It is the centroid of the current time-series sliding window and the reference centroid. The difference vector is calculated as follows: arrive The square of the Mahalanobis distance is a measure of the data's shape, specifically its covariance. The normalized distance.

[0043] Calculation example: When the production line is stable, see Figure 2 0-200 cycles Very close ,Right now ,therefore, It will be close to 0, see Figure 3 The orange curve remained at a low level in the early stage; When a transient spike occurs, see Figure 2 Although the point marked by the red dashed line is an anomaly. It's extreme, but yes The average of the data points is used to smooth out outliers, thus the centroid is reduced. The changes are very small. Keep it low, see Figure 3 The orange curve shows only slight fluctuations at its peak. When gradual drift occurs, see Figure 2 After 200 cycles, Inside the window The data points move as a whole, such as It continued to move to Clearly far away ,like , It will continue to rise, see Figure 3 The orange curve continued to climb steadily after 200 cycles.

[0044] By calculating the shape indicators of the current window in real time and location indicators Two decoupled feature quantities were obtained that can distinguish different abnormality types, thus providing a basis for subsequent diagnostic steps.

[0045] S003: Anomaly diagnosis of plastic parts production line based on dual-index decoupling.

[0046] This step is obtained through real-time calculation. and Two indicators enable the diagnosis of anomalies in plastic parts production lines.

[0047] (1) Threshold setting: Set the morphological expansion threshold. The morphological expansion threshold can be based on data dimensions. Configure it, for example. This allows for 50% morphological expansion; a positional drift threshold is set. The position drift threshold can be based on The statistical characteristics can be set, for example, by setting the statistical characteristics. Set as Data points to The 99.9th percentile of the squared Mahalanobis distance ensures statistical robustness, such as... ;exist Figure 3 middle, Corresponding to the red dashed line, i.e., the peak threshold, The corresponding orange dashed line represents the drift threshold.

[0048] (2) Anomaly diagnosis logic: at each point in time, i.e. the current production cycle The system according to and The comparison with the corresponding thresholds is performed according to the following logic for diagnosis, see [link to relevant documentation]. Figure 3 : The criteria for diagnosing a transient spike are: and ,like and The diagnostic logic is: the average position of the window is normal, that is... Normal, but the internal structure suddenly becomes loose, that is... The value is too high, indicating a sudden anomaly that entered the window. See [link / reference]. Figure 3 The moment when the red curve instantly crosses the threshold, while the orange curve remains below the threshold; The criteria for diagnosing gradual drift are: and , and The diagnostic logic is: the internal shape of the window is compact, that is... Normal, but the average position has deviated significantly from the historical benchmark, that is... A higher value corresponds to a slow, coordinated movement of the entire process, i.e., gradual drift. See [link / reference needed]. Figure 3 The orange curve crosses the threshold, while the red curve remains below the threshold. The criteria for diagnosing a spike in drift are: and ,like and The diagnostic logic is as follows: the average position of the window has deviated, and its internal shape has become loose, which indicates that the production line has been subjected to another instantaneous peak impact while it was already in a state of drift.

[0049] The conditions for a normal diagnosis are: and ,like and The diagnostic logic is: the shape and position of the window are both within the range allowed by the historical baseline.

[0050] By combining the two decoupled indicators of morphology and location, the system achieves accurate classification and diagnosis of anomalies with different physical causes, such as instantaneous spikes and gradual drifts, thus solving the problem that traditional algorithms cannot distinguish anomaly types.

[0051] S004: Post-processing and adaptive update.

[0052] (1) Alarm distribution: The system will push different types of alarms diagnosed to the production line management system or the field Andon system to guide relevant personnel to carry out differentiated handling: For example, instantaneous spikes may indicate that the sensor or actuator is experiencing occasional failure; while gradual drift indicates that the equipment may have slow wear and requires planned maintenance.

[0053] (2) Adaptive benchmark model: This step is to solve the problem of a large number of false alarms caused by changes in the static benchmark when the process changes, such as changing raw materials or molds; Trigger: When the gradual drift alarm continues After one cycle, for example, The system automatically pauses the alarm and sends a process window drift confirmation request to the production line administrator or quality inspector.

[0054] Confirmation: The operator checks the current batch, such as the most recent one. Product quality for each cycle.

[0055] Decision: Malicious drift, operator confirmation of non-compliance, such as shrinkage or flash, the system locks the current value. If no update is made and alarms continue, the production line must be shut down for maintenance; if the drift is benign and the operator confirms that it is acceptable, it means that the current drifted process window is a new and acceptable baseline process window, such as the overall shift of process parameters caused by changing to a new batch of raw materials or a new mold.

[0056] Update: The system triggered a baseline model update, automatically extracting the most recent data. One cycle, for example, qualified data as new Completely re-execute all steps in S001 to generate a new... .

[0057] Recovery: The system uses a new... Continue monitoring, due to the current window At this point, we are approaching a new benchmark. , Restore to The alarm will then be automatically deactivated, and the system will have successfully adapted to the new process.

[0058] By establishing a closed loop of alarm distribution and adaptive updates, differentiated guidance is provided for the operation and maintenance of plastic parts production lines. At the same time, the real-time monitoring system can proactively learn and adapt to positive process changes, ensuring the long-term availability and accuracy of the monitoring system in dynamic production environments.

[0059] This invention also discloses a real-time monitoring system for a plastic parts production line, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a real-time monitoring method for a plastic parts production line according to the present invention is implemented.

[0060] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0061] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of a plastic parts production line, characterized in that, include: Historical data from multiple production cycles, including a baseline batch used as a plastic parts production line, is acquired and standardized. A baseline process model is constructed based on the standardized historical data. The baseline process model includes a baseline centroid and a baseline morphology matrix. The process parameter vector of the plastic parts production line in the current production cycle is acquired in real time, and standardized. The standardized process parameter vector is then stored in a preset time-series sliding window. Calculate the current centroid and current morphology matrix of all process parameter vectors within the time-series sliding window; Based on the inverse matrix of the current morphology matrix and the reference morphology matrix, the current window morphology inflation factor, which characterizes the degree of data dispersion within a short-term time-series sliding window, is determined. Based on the current centroid, the reference centroid, and the inverse matrix of the reference morphological matrix, the current window position drift factor, which characterizes the degree to which the centroid of the short-term time-series sliding window deviates from the reference centroid, is determined. Set a shape expansion threshold and a position drift threshold. Based on the comparison results of the current window shape expansion factor and shape expansion threshold and the current window position drift factor and position drift threshold, diagnose the state of the plastic parts production line as one of the following: normal, instantaneous spike, gradual drift, or spike in drift, and complete the real-time monitoring of the plastic parts production line.

2. The real-time monitoring method for a plastic parts production line according to claim 1, characterized in that, The construction of the baseline process model includes: The mean of all process parameter vectors of the standardized historical data is used as the reference centroid. The covariance matrix of all process parameter vectors of the standardized historical data is used as the baseline morphological matrix.

3. The real-time monitoring method for a plastic parts production line according to claim 1, characterized in that, The standardization process uses Z-score standardization.

4. The real-time monitoring method for a plastic parts production line according to claim 1, characterized in that, The current window shape expansion factor satisfies: ； In the formula, For the current production cycle The corresponding current window shape inflation factor, For the current production cycle The corresponding current form matrix, It is the inverse of the base form matrix. This is the trace operation of a matrix.

5. The real-time monitoring method for a plastic parts production line according to claim 1, characterized in that, The current window position drift factor satisfies: ； In the formula, For the current production cycle The corresponding current window position drift factor, For the current production cycle The corresponding current centroid, Using the reference centroid, It is the inverse of the base form matrix. This is a vector transpose operation.

6. A real-time monitoring method for a plastic parts production line according to claim 1, 4, or 5, characterized in that, The criteria for diagnosing the transient spike include: The current window shape expansion factor is greater than the shape expansion threshold, and the current window position drift factor is less than or equal to the position drift threshold.

7. A real-time monitoring method for a plastic parts production line according to claim 1, 4, or 5, characterized in that, The conditions for diagnosing the gradual drift include: The current window shape expansion factor is not greater than the shape expansion threshold, and the current window position drift factor is greater than the position drift threshold.

8. A real-time monitoring method for a plastic parts production line according to claim 1, 4, or 5, characterized in that, The criteria for diagnosing a spike in the drift include: The current window shape expansion factor is greater than the shape expansion threshold, and the current window position drift factor is greater than the position drift threshold.

9. The real-time monitoring method for a plastic parts production line according to claim 1, characterized in that, It also includes adaptively updating the baseline process model, including: After the gradual drift anomaly continues for a set number of cycles, manual confirmation is triggered. If it is confirmed to be a benign drift, the process parameter vectors of the most recent multiple cycles are extracted as new historical data, and the baseline process model is reconstructed.

10. A real-time monitoring system for a plastic parts production line, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a real-time monitoring method for a plastic parts production line according to any one of claims 1-9.

Images