Real-time monitoring method and system for injection molding production quality

Abstract

The invention relates to the technical field of injection molding quality monitoring, in particular to a real-time monitoring method and system for injection molding production quality. The method comprises the following steps: acquiring a standard pressure curve and a standard temperature curve of each production stage in each injection molding production cycle; calculating the local intensity of each sampling moment; obtaining an abnormal fluctuation coefficient of each sampling moment, and constructing a multi-dimensional coupling dynamic weight of each sampling moment; multiplying the abnormal fluctuation coefficient by the multi-dimensional coupling dynamic weight to obtain a weighted dynamic fluctuation curve; integrating the weighted dynamic fluctuation curve to obtain a form deviation index; and performing anomaly detection on the current production cycle based on the form deviation index. And the monitoring precision of injection molding production is improved.

Description

Technical Field

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

[0002] Injection molding is a widely used high-precision, high-volume polymer product manufacturing process in modern manufacturing. To ensure consistent and stable product quality, real-time monitoring of the injection molding process is crucial. Currently, one of the mainstream monitoring methods is to collect key process parameters of the injection molding machine during each production cycle, such as mold cavity pressure, injection speed, and screw position, and then use data-driven algorithms to analyze these parameters to identify abnormal production cycles that may lead to product defects.

[0003] Among numerous data-driven algorithms, the Isolation Forest algorithm has been applied in the field of industrial anomaly detection due to its advantages such as high computational efficiency and the absence of pre-labeled anomaly samples. This algorithm typically constructs a model and performs anomaly detection by extracting a set of discrete scalar values, such as peak pressure, holding time, and average temperature, from the feature vectors of process parameters for each production cycle.

[0004] However, this method has a limitation: it compresses a dynamic, continuous process into several static, isolated scalar features. This process loses a significant amount of information about the dynamic changes of process parameters over time, i.e., the overall morphological characteristics of the time series curve. In actual production, some product defects, such as slight shrinkage marks caused by unstable pressure maintenance during the holding phase, or flow marks caused by uneven injection speed, may not cause significant deviations in traditional scalar features such as pressure peaks or integral values, but will manifest as abrupt changes in the local slope or abnormal fluctuations in the shape of the pressure curve. Because existing technologies cannot detect such "morphological anomalies," they will determine the cycle to be normal when these isolated feature values ​​are still within the normal range, resulting in missed detections and the failure to promptly identify and isolate potential defects. Summary of the Invention

[0005] To address the problem that existing injection molding production monitoring methods ignore the dynamic morphological characteristics of curves, leading to false alarms and missed alarms, this application provides a real-time monitoring method and system for injection molding production quality.

[0006] Firstly, this application provides a method for real-time monitoring of injection molding production quality, employing the following technology: Obtain the standard pressure curves and standard temperature curves for each production stage in each injection molding production cycle; Calculate the local intensity at each sampling time, where the local intensity is the absolute value of the second derivative of the standard pressure curve at the corresponding sampling time; obtain the abnormal fluctuation coefficient at each sampling time, where the abnormal fluctuation coefficient is obtained based on the stage fluctuation response factor and the positive pressure gradient of the standard pressure curve; the stage fluctuation response factor characterizes the abnormal production behavior at the sampling time. Constructing a multidimensional coupled dynamic weight for each sampling moment includes: obtaining the time-sensitive weight for each sampling moment to characterize the importance of the sampling moment in different production stages; obtaining the morphological energy factor for each sampling moment, which is used to characterize the stability of the injection molding process; obtaining the stage coupling coefficient for each sampling moment; and fusing the time-sensitive weight, the morphological energy factor, and the stage coupling coefficient to obtain the multidimensional coupled dynamic weight. The abnormal fluctuation coefficient is multiplied by the multidimensional coupled dynamic weight to obtain the weighted dynamic fluctuation curve; the weighted dynamic fluctuation curve is integrated to obtain the morphological deviation index; and anomalies are detected in the current production cycle based on the morphological deviation index.

[0007] The beneficial effects are as follows: by constructing a complete monitoring link from local dynamic features, stage-differentiated response, multi-dimensional coupling weighting to morphological deviation characterization, the system can simultaneously identify multiple types of quality hazards such as local mutations, stage anomalies, and overall morphological deviations; it can utilize the dynamic change information of the entire pressure and temperature curve to perform more refined anomaly detection, reduce false alarms and missed alarms, and make the monitoring results closer to the real change law of the injection molding process.

[0008] Furthermore, the injection molding production cycle includes a filling stage, a holding pressure stage, a cooling stage, and a demolding stage.

[0009] Furthermore, the method for obtaining the stage fluctuation response factor is as follows: The ratio of the absolute value of the second derivative to the absolute value of the first derivative of the standard pressure curve is taken as the relative curvature. For each injection molding production cycle, the product of the median of the relative curvature and the median absolute deviation at all sampling times is obtained, and the sum of the median is used as the preset curvature threshold for the corresponding sampling time.

[0010] During the pressure holding and cooling stages, the stage fluctuation response factor is the absolute value of the second derivative of each standard pressure curve at the corresponding sampling time; during the filling and demolding stages, the stage fluctuation response factor is determined based on the relative curvature of the standard pressure curve and the duration exceeding the preset curvature threshold.

[0011] The beneficial effects are as follows: by introducing relative curvature, preset curvature threshold and duration constraint to construct stage fluctuation response factor, the algorithm can effectively distinguish between real abnormal fluctuations and normal stage disturbances; during the pressure holding and cooling stages, the algorithm maintains high sensitivity to any fluctuations; while during the filling and demolding stages, the response is triggered only when the curvature is abnormal and the duration is long enough, which improves the ability to suppress normal fluctuations of the curve and thus significantly reduces false alarms.

[0012] Furthermore, the formula for calculating the abnormal fluctuation coefficient is as follows:

[0013] In the formula, The abnormal fluctuation coefficient at sampling time t; Let be the stage fluctuation response factor at sampling time t. It is a function with maximum value. This is a cooling indicator variable. During the cooling phase, the value of the cooling indicator variable is 1, and during other phases, the value of the cooling indicator variable is 0. This is the normalization function.

[0014] The beneficial effects are as follows: By integrating the stage fluctuation response factor with the pressure monotonicity judgment during the cooling stage, the system can effectively identify key abnormal behaviors such as pressure rebound and localized rise during the cooling stage. This structure can automatically amplify real anomalies and suppress normal downward trends, making anomaly detection more consistent with the physical laws of cooling and improving the detection capabilities for problems such as uneven cooling and premature pressure relief.

[0015] Furthermore, the method for obtaining the time-sensitive weight is as follows: the first S-shaped function value and the second S-shaped function value are added together, wherein the input of the first S-shaped function value is the difference between the current sampling time and the starting point of the holding phase, and the input of the second S-shaped function value is the difference between the current sampling time and the ending point of the holding phase.

[0016] The beneficial effects are as follows: the time-sensitive weight uses two S-shaped functions to form a time window aligned with the holding pressure stage, so that anomalies in the holding pressure stage are automatically amplified in the overall detection results, while the impact of non-critical stages is naturally weakened; it can adapt to different cycle lengths and different product processes without the need for additional manual threshold setting, so that the algorithm can automatically focus on the most important process stages within the cycle.

[0017] Furthermore, the method for obtaining the morphological energy factor is as follows: the product of the absolute value of the first derivative and the absolute value of the second derivative of the standard pressure curve is divided by the ratio between the variance of the standard pressure curve within a preset time window and the sum of the values ​​of 1.

[0018] The beneficial effects are as follows: By integrating the first derivative, the second derivative, and the local variance, the morphological energy factor enables the system to distinguish between real changes with continuous trends and short-term noise disturbances; it can maintain a high response to strong changes with stable trends and automatically attenuate high-frequency random fluctuations, thereby improving the algorithm's ability to identify phenomena such as abnormal real morphology, unstable flow, or mechanical vibration.

[0019] Furthermore, the formula for calculating the stage coupling coefficient is as follows:

[0020] In the formula, The stage coupling coefficient at sampling time t; As an indicator variable for the pressure holding stage, As an indicator variable for the cooling phase, To maintain the coupling degree, This represents the cooling coupling degree. The pressure holding coupling degree is... Cooling coupling degree is ,in, The first derivative of the standard temperature curve at sampling time t; When the sampling time is in the holding pressure stage, the value of the holding pressure stage indicator variable is 1; otherwise, the value of the holding pressure stage indicator variable is 0. When the sampling time is in the cooling stage, the value of the cooling stage indicator variable is 1; otherwise, the value of the cooling stage indicator variable is 0.

[0021] Furthermore, the calculation formula for the multidimensional coupled dynamic weights is as follows:

[0022] In the formula, The multidimensional coupled dynamic weights at sampling time t The time-sensitivity weights at sampling time t To preset the first coupling coefficient, To preset the second coupling coefficient, The normalized morphological energy factor is the sampled time t. The normalized stage coupling coefficient is the sampling time t.

[0023] Furthermore, the anomaly detection of the current production cycle based on the morphological deviation index includes: combining the morphological deviation index with the pressure peak value and cycle duration of the standard pressure curve to form an enhanced feature vector; inputting the enhanced feature vector into a pre-trained isolated forest model to obtain an anomaly score; and performing anomaly detection of the current production cycle based on the anomaly score.

[0024] Secondly, this application provides a real-time monitoring system for injection molding production quality, employing the following technology: A real-time monitoring system for injection molding production quality includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the real-time monitoring method for injection molding production quality as described above.

[0025] The above-mentioned real-time monitoring method for injection molding production quality is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. Thus, a system is made based on the memory and processor for easy use.

[0026] This application has the following technical advantages: The real-time monitoring method for injection molding production quality proposed in this application achieves high-precision identification of key operating conditions in the injection molding process by constructing a multi-dimensional dynamic feature system. The method first uses local intensity to characterize the transient changes in the pressure curve, enabling the early detection of anomalies such as peaks and inflection points during the filling stage. Combined with stage fluctuation response factors, the method differentiates the fluctuation behavior of different process stages such as filling, holding pressure, and cooling, clearly distinguishing between normal stage disturbances and abnormal fluctuations. An abnormal fluctuation coefficient highlights potential quality hazards such as rebound and recovery during the cooling stage. Furthermore, a time-sensitive weight is introduced to strengthen the overall judgment of the holding pressure stage, the most critical stage for molding quality.

[0027] Meanwhile, the method utilizes morphological energy factors to reflect the stability of the local trend of the curve, and incorporates the relationship between pressure and temperature changes into a consistent evaluation framework through stage coupling coefficients. This allows the monitoring algorithm to automatically adjust its sensitivity according to the process status, avoiding false alarms. Furthermore, a multi-dimensional coupled dynamic weight is constructed to integrate energy changes, morphological changes, and stage coupling information, enabling the impact of fluctuations to be automatically amplified or suppressed according to different stages. The resulting morphological deviation index can reflect the degree of deviation of the overall curve shape from a complete cycle perspective, and can identify hidden quality anomalies that are difficult to detect by traditional peak-type indicators.

[0028] By incorporating features such as pressure peak value and cycle duration into the pre-trained model, the accuracy of anomaly identification is further improved. This method effectively identifies key defects such as slight fluctuations, uneven cooling, and unstable holding pressure, improving monitoring accuracy and robustness, significantly reducing false alarms and false negatives, and providing reliable assurance for the stability and molding quality of injection molding production. Attached Figure Description

[0029] Figure 1 This is a flowchart of the real-time monitoring method for injection molding production quality according to this application. Detailed Implementation

[0030] This application discloses a real-time monitoring method for injection molding production quality. The method acquires standard pressure and temperature curves for each production stage in each injection molding production cycle; calculates the local intensity at each sampling moment; acquires the abnormal fluctuation coefficient at each sampling moment and constructs a multi-dimensional coupled dynamic weight for each sampling moment; multiplies the abnormal fluctuation coefficient by the multi-dimensional coupled dynamic weight to obtain a weighted dynamic fluctuation curve; integrates the weighted dynamic fluctuation curve to obtain a morphological deviation index; and performs anomaly detection for the current production cycle based on the morphological deviation index. This improves the monitoring accuracy of injection molding production.

[0031] Reference Figure 1 The real-time monitoring method for injection molding production quality includes steps S1-S4.

[0032] Step S1: Obtain the standard pressure curve and standard temperature curve for each production stage in each injection molding production cycle.

[0033] For each injection molding production cycle, pressure sensor data and temperature sensor data within the mold cavity are synchronously collected at a preset high sampling frequency to obtain the original mold cavity pressure time series and temperature time series for each injection molding production cycle. In one embodiment of this application, the preset high sampling frequency is set to 1000Hz, but the implementer can choose other values ​​based on actual conditions. Furthermore, to eliminate the impact of slight variations in the total cycle time between different production batches, the amplitudes of the obtained pressure time series and temperature time series are normalized and standardized on the time axis. In one embodiment of this application, the normalization and standardization methods are maximum-minimum value normalization and Z-score standardization. Furthermore, the least squares method is used to fit the standardized pressure time series and temperature time series to obtain a standard pressure curve. and standard temperature curve.

[0034] Step S2: Calculate the local intensity at each sampling time, where the local intensity is the absolute value of the second derivative of the standard pressure curve at the corresponding sampling time; obtain the abnormal fluctuation coefficient at each sampling time, where the abnormal fluctuation coefficient is obtained based on the stage fluctuation response factor and the positive pressure gradient of the standard pressure curve; the stage fluctuation response factor characterizes the abnormal production behavior at the sampling time.

[0035] Based on the above steps, standard pressure curves for each injection molding production cycle were obtained. This was done to capture and quantify the pressure curves. To determine the degree of drastic change in local morphology at each time point, it is necessary to analyze the dynamic characteristics of the curve. Unlike focusing solely on the first derivative (slope) of the curve's numerical value, the second derivative of the curve can more sensitively reflect its concavity and convexity, as well as the acceleration of change. Therefore, for each sampling time point, the absolute value of the second derivative of the standard pressure curve at the sampling time point is calculated as the degree of local drastic change at each sampling time point.

[0036] It should be noted that, for each sampling moment, the value of the local severity is positively correlated with the curvature of the pressure curve at that sampling moment. In an ideal, stable injection molding production stage, pressure changes should be smooth, with the absolute value of its second derivative close to zero. If a sudden pressure change, fluctuation, or abrupt change in inflection point occurs at a certain sampling moment, it will significantly increase the value of the local morphological change at that sampling moment, thus effectively identifying local anomalies in the curve morphology.

[0037] Furthermore, in the injection molding process, the complete injection molding cycle consists of the filling stage, the holding pressure stage, the cooling stage, and the demolding stage. During the filling stage, the cavity pressure rises rapidly from zero, exhibiting a steep upward trend with a large slope, and there are some transient fluctuations before the flow front stabilizes. During the holding pressure stage, the cavity is almost completely filled, and the pressure curve reaches and maintains a stable high-pressure plateau range. Subsequently, as the melt gradually cools and its shrinkage capacity weakens, the pressure begins to decrease slowly. During the cooling stage, the holding pressure is released, and the melt in the cavity gradually changes from a liquid to a solid state, resulting in a gradual decrease in cavity pressure. During the demolding stage, the cavity pressure should theoretically be close to zero, but due to the ejection mechanism's action or the mechanical impact generated by mold parting, the sensor may detect instantaneous pressure spikes or brief vibration signals. Based on the above analysis, the degree of pressure change varies in different filling stages within the injection molding cycle. It should be noted that each stage of the injection molding process is based on manual settings. That is, during the filling stage, the machine needs to be manually controlled to fill the injection material. After the machine displays that the filling is complete, the machine needs to be manually set to start the holding pressure. The holding pressure duration and cooling duration are both manually set. Finally, during the demolding stage, the machine is manually controlled to demold.

[0038] Based on the above analysis, the mold cavity pressure variation patterns differ significantly across stages of the injection molding process. During the filling and demolding stages, transient fluctuations are more likely to occur in the pressure curves due to the influence of melt front advancement, gas expulsion, and ejection. However, these fluctuations are normal, and responding to them entirely would lead to false alarms. Furthermore, the pressure curves during the holding and cooling stages should remain stable; any fluctuations reflect potential molding anomalies. Therefore, differentiating the local dynamic fluctuations by stage, allowing the indicator to suppress normal fluctuations during the filling and demolding stages and enhance the response during the holding and cooling stages, better reflects the actual situation of the injection molding process, thus significantly improving the targeting and accuracy of fluctuation detection. Specifically, for different injection stages, a first case and a second case are constructed, where the first case includes: the holding stage, the cooling stage, or... The second scenario includes the filling stage and the demolding stage, which do not meet the above conditions.

[0039] Construction phase fluctuation response factor:

[0040] In the formula, Sampling time The phase fluctuation response factor; The local intensity at sampling time t, Let be the absolute value of the second derivative of the standard pressure curve at sampling time t. Let be the absolute value of the first derivative of the standard pressure curve at sampling time t; where, is the relative curvature, and is the ratio of the absolute value of the second derivative to the absolute value of the first derivative of the standard pressure curve at sampling time t; The relative curvature threshold, To continuously satisfy from sampling time t Duration To preset the minimum duration, in one embodiment of this application, the preset minimum duration is set to 5 seconds. Implementers may select other values ​​based on actual circumstances.

[0041] The method for obtaining the relative curvature threshold is as follows: for each stage of the injection molding production process at each sampling time, the median and median absolute deviation of the relative curvature at all sampling times in the stage are obtained, and the sum of the product of the preset sensitivity coefficient and the median absolute deviation and the median is used as the relative curvature threshold for each sampling time.

[0042] It should be noted that the stage fluctuation response factor adjusts the response mode to local dynamic fluctuations according to different stages of the injection molding process, so that the calculation of the stage fluctuation response factor is consistent with the actual situation; the response mode is dynamically adjusted according to the stage characteristics; in the formula of the stage fluctuation response factor, by combining the derivative relationship of the pressure curve at different stages with the duration constraint, the dynamic distinction between normal fluctuations and abnormal fluctuations is realized: in the holding and cooling stages, the pressure should remain stable, so the local intensity is directly used as the response value; while in the filling and demolding stages, the curve itself has normal transient fluctuations, and only when the ratio of the second derivative to the first derivative exceeds the preset relative curvature threshold and the state lasts for the minimum duration is it considered abnormal, thereby suppressing misjudgments caused by short-term disturbances or noise; the stage fluctuation response factor can sensitively capture slight unstable changes in the stages that should be stable, and can automatically filter normal disturbances in the stages that should fluctuate, realizing adaptive adjustment of the response intensity at different stages. Its beneficial effects are that it significantly improves the system's accuracy in identifying actual process anomalies, avoids false alarms and missed alarms caused by stage differences, and makes the calculation of local dynamic fluctuations more consistent with the real physical characteristics of the injection molding process.

[0043] Furthermore, during the cooling stage, the melt within the mold cavity should continuously contract as the temperature decreases, and the pressure curve should monotonically decrease. Local increases or rebounds often indicate problems such as premature pressure release, uneven cooling, or abnormal residual stress release. By introducing monotonicity judgment during the cooling stage and amplifying the pressure segments exhibiting upward trends, abnormal behavior during the cooling stage can be quickly identified without complex parameters. This supplementary judgment allows local dynamic fluctuations to reflect not only the geometric changes in the curve but also the physical rationality of the cooling process, thereby improving sensitivity to cooling quality anomalies; therefore, an abnormal fluctuation coefficient is constructed, calculated using the following formula:

[0044] In the formula, The abnormal fluctuation coefficient at sampling time t; Let be the stage fluctuation response factor at sampling time t. It is a function with maximum value. As a cooling indicator variable, the value of the cooling indicator variable is 1 when the control system detects that the pressure holding has ended and the cooling valve has opened; at other stages, the value of the cooling indicator variable is 0. As a normalization function, in one embodiment of this application, the normalization method selected is the Z-score normalization method. Implementers may select other normalization methods based on actual circumstances.

[0045] It should be noted that the abnormal fluctuation coefficient supplements the monotonicity judgment term of the cooling stage, using the first derivative of pressure. The positive or negative sign of the pressure is used to identify whether there is an abnormal rise in the cooling process. During the cooling stage of injection molding, the volume of melt in the mold cavity gradually shrinks as the temperature decreases, and the mold cavity pressure should show a monotonically decreasing trend. In actual production, due to factors such as uneven mold cooling, release of residual stress in the material, local backflow, or premature pressure relief, the mold cavity pressure curve may show a local rise or rebound, that is, the curve no longer shows a monotonically decreasing trend. Based on the above analysis, when the positive or negative sign of the pressure is detected... When this occurs, it indicates that the pressure curve has shown an abnormal upward trend, and the system automatically amplifies the fluctuation response value; if If the pressure change is consistent with normal cooling behavior, no correction is required.

[0046] At this point, the abnormal fluctuation coefficients at each sampling time are obtained.

[0047] Step S3: Construct multidimensional coupled dynamic weights for each sampling moment, including: obtaining time-sensitive weights for each sampling moment to characterize the importance of the sampling moment in different production stages; obtaining morphological energy factors for each sampling moment, which characterize the stability of the injection molding process; obtaining stage coupling coefficients for each sampling moment; fusing the time-sensitive weights, morphological energy factors, and stage coupling coefficients to obtain multidimensional coupled dynamic weights; and multiplying the abnormal fluctuation coefficients by the multidimensional coupled dynamic weights to obtain a weighted dynamic fluctuation curve.

[0048] In injection molding, the impact of process stability at different stages on the final product quality varies. Pressure fluctuations during the holding pressure stage are more likely to cause defects such as shrinkage marks or dimensional deviations than fluctuations at the end of injection or the initial cooling stage. Therefore, the abnormal fluctuation coefficient... Time-sensitivity weighting is applied to highlight morphological anomalies in key process stages. Based on the above analysis, the time-sensitivity weights for each sampling time are constructed, and the calculation formula is as follows:

[0049] In the formula, Sampling time Time-sensitivity weights; This indicates the starting point of the pressure holding phase. This indicates the end time of the pressure holding phase. A preset positive coefficient is used to control the steepness of the weighting function curve. In one embodiment of this application, the preset positive coefficient is 25, but the implementer can select other values ​​based on the actual situation. express function.

[0050] It should be noted that, and By analyzing the standard pressure curve The plateau phase characteristics are automatically obtained. The time-sensitive weights utilize the difference between two sigmoid functions to construct a weight for the holding pressure phase. A window function with a high intrinsic value and a value close to zero in other stages. Time-sensitive weights. The importance of the holding pressure phase is transformed into mathematical weights, at the sampling time... Time sensitivity weight during the pressure holding phase The value increases rapidly at the sampling time. Time sensitivity weight when leaving the holding pressure phase The value decreases rapidly.

[0051] Time-sensitivity weights only focus on the time position and cannot reflect the complex changes in the curve shape itself. In actual production, changes in the shape of the injection pressure curve reveal process stability more clearly than simple numerical changes, such as curve inflection points, local fluctuations, and trend deviations. If the dynamic geometric characteristics of the curve are not considered, the system will lose sensitivity to periods of "abnormal shape but normal numerical values" during real-time monitoring of injection molding production. Therefore, a shape energy factor is constructed, calculated using the following formula:

[0052] In the formula, The morphological energy factor at sampling time t; Let be the absolute value of the second derivative of the standard pressure curve at sampling time t. Let be the absolute value of the first derivative of the standard pressure curve at sampling time t. The variance of all amplitudes of the standard pressure curve with the sampling time t as the center point is given. In one embodiment of this application, the length of the preset window is 0.2, and the implementer may select other values ​​based on the actual situation.

[0053] It should be noted that when the first and second derivatives of the pressure curve are both large and have the same sign, it indicates that the pressure is in a phase of continuous and rapid rise or fall at that moment. Although this change is drastic, it is unidirectional and has a clear trend. Therefore, the local variance will not increase significantly, and the morphological energy factor value calculated by the formula is high, which can accurately reflect this strong change process with a stable trend. On the other hand, when the first and second derivatives have opposite signs or frequently alternate between positive and negative, it indicates that the pressure curve has undergone rapid rise and fall switching in a short period of time. The direction of the curve's concavity and convexity is constantly reversed, which is a manifestation of high-frequency fluctuations or mechanical oscillations. At this time, the local variance increases significantly, the denominator of the formula is amplified, and the morphological energy factor value is suppressed, thereby avoiding the system from over-responding to transient disturbances. When both the first and second derivatives are small and the curve changes gently, it indicates that the pressure change rate is low and the acceleration is small. The entire system is in a stable phase, and the corresponding local variance is also small. At this time, both the numerator and denominator are small, making the morphological energy factor approach zero. This design enables the algorithm to respond strongly only to real-world morphological changes with a sustained trend, while automatically reducing random noise or minor disturbances, thus accurately capturing stable and abnormal changes in the process.

[0054] The different stages of injection molding not only have a temporal sequence but also significant differences in process conditions. The holding pressure stage aims for pressure stability, while the cooling stage prioritizes temperature uniformity. If the time weight and morphological energy factor do not consider the differences between stages, the algorithm cannot reflect the variations in injection molding production conditions. This leads to the use of the same sensitivity for each stage, failing to adapt to the different physical characteristics of the holding pressure and cooling stages. Therefore, a stage coupling coefficient is constructed, and the calculation formula is as follows:

[0055] In the formula, The stage coupling coefficient at sampling time t; As an indicator variable for the pressure holding stage, As an indicator variable for the cooling phase, To maintain the coupling degree, This represents the cooling coupling degree. The pressure holding coupling degree is... Cooling coupling degree is ,in, Let be the first derivative of the standard temperature curve at sampling time t; where, when sampling time t is in the pressure holding stage, the value of the pressure holding stage indicator variable is 1, and otherwise it is 0; when sampling time t is in the cooling stage, the value of the cooling stage indicator variable is 1, and otherwise it is 0.

[0056] It should be noted that the stage coupling coefficient increases with decreasing pressure change rate during the holding pressure stage and increases with decreasing temperature change rate during the cooling stage, reflecting a preference for a stable state. This coefficient is high when the pressure curve is stable or the temperature decreases uniformly; it is low when pressure fluctuates or cooling is uneven. This design allows the system to automatically adjust its sensitivity according to the actual state of each process stage, amplifying the influence of stable processes and weakening the interference of unstable factors during the holding pressure and cooling stages, making the weighting function more consistent with the physical laws of the injection molding process.

[0057] Based on the above analysis, in order to ensure the consistency of dimensions in subsequent processes, the time-sensitive weights and stage coupling coefficients are normalized to construct a multidimensional coupled dynamic weight, the calculation formula of which is:

[0058] In the formula, The multidimensional coupled dynamic weights at sampling time t The time-sensitivity weights at sampling time t To preset the first coupling coefficient, To preset the second coupling coefficient, The normalized morphological energy factor is the sampled time t. The normalized stage coupling coefficient is the coefficient at sampling time t. In one embodiment of this application, the preset first coupling coefficient and the preset second coupling coefficient are 0.6 and 0.4, respectively. The implementer may select other values ​​based on the actual situation.

[0059] It should be noted that the dependent variable of the multidimensional coupled dynamic weight increases synchronously as the time window enters the critical stage, the morphological energy factor increases, and the stage coupling coefficient increases, thus exhibiting an adaptive amplification of morphological anomalies and process consistency changes during the critical period. This design enables the weight to not only be time-controlled but also comprehensively reflect pressure morphology and stage stability, thereby automatically enhancing abnormal responses during the pressure holding and cooling stages and automatically attenuating responses during non-critical stages, achieving high sensitivity, high stability, and physical interpretability of the monitoring results.

[0060] Thus, the multidimensional coupled dynamic weights at the sampling time are obtained, and these weights are applied to the abnormal fluctuation coefficients to obtain the weighted dynamic fluctuation curve. .

[0061] Step S4: Integrate the weighted dynamic fluctuation curve to obtain the morphological deviation index; perform anomaly detection for the current production cycle based on the morphological deviation index.

[0062] Based on the above steps, the curve morphology information for each injection molding production cycle is aggregated into a scalar feature that can be used by the Isolation Forest algorithm. This requires processing the weighted dynamic fluctuation curve. Integrate to obtain the morphological deviation index .

[0063] It should be noted that the morphological deviation index The value incorporates all local morphological fluctuations throughout the entire production cycle and amplifies the impact of fluctuations in key stages through a time-sensitive weighting function. The closer the pressure curve of a production cycle is to the ideal smooth curve, the lower the morphological deviation index. The smaller the value, the better; conversely, if the curve experiences any unexpected fluctuations, abrupt changes, or instability at critical stages, it will lead to a higher morphological deviation index. The value increases significantly.

[0064] Furthermore, the calculated morphological deviation index With peak pressure Cycle time Construct an enhanced feature vector This enhanced feature vector is then fed into a pre-trained isolated forest model for anomaly scoring. Because... Including both numerical and morphological information, the model can give higher anomaly scores to production cycles that only exhibit morphological abnormalities but have normal numerical characteristics, thereby effectively identifying and marking such potential defects and issuing alarms for unqualified injection molded parts.

[0065] In this application, the specific method for pre-training the isolated forest model is as follows: the isolated forest model is obtained through offline training on historical injection molding production data. The training data comes from multiple batches of sample periods collected by the injection molding equipment during the stable production phase. The raw data of each period are processed according to the methods described in steps S1 to S4 to obtain the corresponding morphological deviation index. Peak pressure With period time The three are then combined to form an enhanced feature vector. .

[0066] During the training phase, a preset number of feature vectors are randomly selected from the sample periods of historical production status as the training set. The random forest model is trained based on the training set, so that the trained model can automatically learn the distribution density and correlation between each feature, forming a benchmark model for subsequent anomaly detection.

[0067] This application also discloses a real-time monitoring system for injection molding production quality, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the real-time monitoring method for injection molding production quality according to this application is implemented.

[0068] 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.

[0069] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for real-time monitoring of the quality of injection-moulded products, characterised in that, The method comprises the steps of: obtaining standard pressure curves and standard temperature curves of each production stage in each injection molding production cycle; calculating the local intensity of each sampling time, the local intensity being the absolute value of the second derivative of the standard pressure curve at the corresponding sampling time; obtaining an abnormal fluctuation coefficient of each sampling time, the abnormal fluctuation coefficient being obtained based on the stage fluctuation response factor and the positive gradient of the standard pressure curve; the stage fluctuation response factor represents the abnormal behavior of the production at the sampling time; constructing a multi-dimensional coupling dynamic weight of each sampling time, comprising: obtaining a time sensitivity weight of each sampling time, which is used to represent the importance of the sampling time in different production stages; obtaining a morphology energy factor of each sampling time, which is used to represent the stability of the injection molding process; obtaining a stage coupling coefficient of each sampling time; and fusing the time sensitivity weight, the morphology energy factor and the stage coupling coefficient to obtain a multi-dimensional coupling dynamic weight; multiplying the abnormal fluctuation coefficient by the multi-dimensional coupling dynamic weight to obtain a weighted dynamic fluctuation curve; integrating the weighted dynamic fluctuation curve to obtain a morphology deviation index; and performing abnormal detection on the current production cycle based on the morphology deviation index.

2. The method for real-time monitoring of the quality of injection molding according to claim 1, characterized in that, The injection molding production cycle comprises a filling stage, a holding stage, a cooling stage and a demolding stage.

3. The method of real-time monitoring of injection molding production quality according to claim 2, characterized in that, The method for obtaining the stage fluctuation response factor is: taking the ratio of the absolute value of the second derivative to the absolute value of the first derivative of the standard pressure curve as the relative curvature; for each injection molding production cycle, obtaining the product of the median and the median absolute deviation of the relative curvatures of all sampling times, and the sum of the medians, as the preset curvature threshold value of the corresponding sampling time. In the holding stage and the cooling stage, the stage fluctuation response factor is the absolute value of the second derivative of each standard pressure curve at the corresponding sampling time; in the filling stage and the demolding stage, the stage fluctuation response factor is determined according to the relative curvature of the standard pressure curve and the duration of exceeding the preset curvature threshold value.

4. The method of claim 2, wherein the step of determining the quality of the injection molded product comprises the steps of: determining the quality of the injection molded product based on the at least one of the plurality of parameters. The calculation formula of the abnormal fluctuation coefficient is: In the formula, is an abnormal fluctuation coefficient at sampling time t; is a phase fluctuation response factor at sampling time t, is a maximum function, is a cooling indication variable, the value of the cooling indication variable is a numerical value 1 when in a cooling phase, and the value of the cooling indication variable is a numerical value 0 when in other phases; is a normalization function.

5. The method of claim 2, wherein the step of determining the quality of the injection molded product comprises the step of: The method for obtaining the time sensitivity weight is to add a first S-type function value and a second S-type function value, wherein the input of the first S-type function value is the difference between the current sampling time and the starting point of the holding stage, and the input of the second S-type function value is the difference between the current sampling time and the end point of the holding stage. ​ 6. The method of claim 1, wherein, The method for obtaining the morphology energy factor is to divide the product of the absolute value of the first derivative and the absolute value of the second derivative of the standard pressure curve by the ratio between the sum of the variances of the standard pressure curve within a preset time window and the sum of the values 1.

7. The method of claim 2, wherein the step of determining the quality of the injection molded product comprises the step of: The calculation formula of the stage coupling coefficient is: ​ wherein, is the phase coupling coefficient at the sampling instant t; is the hold phase indicator variable, is the cool phase indicator variable, is the hold coupling degree, is the cool coupling degree. The hold coupling degree is ; the cool coupling degree is wherein, is the first derivative of the standard temperature curve at the sampling instant t; when the sampling time is in the holding stage, the value of the holding stage indicator variable is 1, otherwise, the value of the holding stage indicator variable is 0; when the sampling time is in the cooling stage, the value of the cooling stage indicator variable is 1, otherwise, the value of the cooling stage indicator variable is 0.

8. The method of claim 1, wherein, The calculation formula of the multi-dimensional coupling dynamic weight is: In the formula, is a multi-dimensional coupling dynamic weight of the sampling time t, is a time sensitivity weight of the sampling time t, is a preset first coupling coefficient, is a preset second coupling coefficient, is a normalized morphological energy factor of the sampling time t, is a normalized stage coupling coefficient of the sampling time t.

9. The method of claim 1, wherein, The abnormality detection on the current production cycle based on the morphology deviation index comprises: combining the morphology deviation index with a pressure peak value and a cycle length of the standard pressure curve to form an enhanced feature vector; inputting the enhanced feature vector into a pre-trained isolation forest model to obtain an abnormality score; and performing abnormality detection on the current production cycle according to the abnormality score.

10. A system for real-time monitoring of the quality of injection-molded products, characterized in that The method comprises: A processor and a memory, wherein the memory stores computer program instructions which, when executed by the processor, implement the real-time monitoring method for injection molding production quality according to any one of claims 1-9.

Images