A real-time monitoring method for injection molding process based on multi-sensor data fusion

Abstract

The present application relates to the technical field of industrial process dynamic characteristic monitoring, and discloses a kind of injection molding process real-time monitoring method based on multi-sensor data fusion, comprising: firstly, the dynamic response characteristics of drive source itself are identified and compensated to obtain corrected process output signal;Further, process input signal and the corrected process output signal are cooperatively processed to generate process gain and time constant capable of overall representing process dynamic transfer characteristics, and process stability is judged accordingly, the present application separates the disturbance source of drive system performance fluctuation from the monitoring signal, realizes the direct quantitative characterization of the dynamic transfer process connecting input and output itself, converts a complex industrial process stability problem into the monitoring of process gain and time constant two-dimensional coordinate point, and improves the reliability of monitoring.

Description

TECHNICAL FIELD

[0001] The application relates to a real-time monitoring method for an injection molding process based on multi-sensor data fusion, and belongs to the technical field of dynamic characteristic monitoring of industrial processes. BACKGROUND

[0002] At present, in industrial production such as injection molding, a commonly used online monitoring method is to configure sensors at the driving end of equipment and the load end of a process respectively, to synchronously acquire independent state time series data, such as process input quantities such as injection pressure or speed at the driving end, and process output quantities such as cavity pressure or temperature at the load end, and then analyze the shape or value of the data curves through a data processing system to determine whether an abnormality exists in the production process.

[0003] The technical method of observing the driving end and the load end as independent information sources has high dependence on the stability of production conditions for the accuracy of monitoring results; when material batches, equipment working conditions or environmental temperatures fluctuate, small disturbances occurring in the force and energy transmission path often do not produce characteristics that can be directly identified on a single input or output data curve, but the cumulative effect has already affected the product quality, and this method has not actually directly quantified the dynamic transmission process connecting the input and the output.

[0004] Specifically, the prior art has the following main problems: 1. Due to the lack of a unified index that can comprehensively evaluate the state of the complete transmission path from the driving force to the cavity pressure response, when the process drifts, it is difficult for technicians to distinguish whether the root cause of the problem is the change in the performance of the driving system or the change in the state of the melt in the flow channel; 2. The judgment of the monitoring system on the process abnormality relies on the direct observation of the shape of the input or output data curve, so it cannot effectively identify potential process fluctuations that do not cause changes in the shape of the curve, but have caused changes in the dynamic response relationship between input and output; Such a fusion process only at the data level, without decoupling the disturbance source at the physical model level, is also reflected in the prior art; For example, the Chinese invention patent with publication number CN120469321A discloses an intelligent factory monitoring method and system based on multi-sensor fusion, which synchronizes and fuses the device state and environmental state data, but its essence is still to correlate and analyze the driving end and the load end as parallel information sources, without separating the dynamic response characteristics of the driving system itself as a disturbance source to be identified and compensated from the monitoring signal, which leads to the monitoring result being easily confused by equipment working condition fluctuations. When an anomaly is detected, it is still difficult to fundamentally clarify whether the anomaly is caused by changes in the transmission process itself or only by drift in the performance of the driving unit. Therefore, how to obtain a dynamic transmission characteristic that can quantitatively represent the dynamic transmission characteristic from the control input to the process output in a physical process in real time and as a whole, and use it as a direct basis for judging the process stability, becomes a technical problem to be solved by the present application. SUMMARY

[0005] The present application provides an injection molding process real-time monitoring method based on multi-sensor data fusion, which mainly aims to solve the problem that there is a lack of a means to directly quantify the dynamic transmission characteristic from the driving source to the cavity in the prior art, making it difficult to effectively judge the process stability.

[0006] To achieve the above-mentioned purpose, the present application provides an injection molding process real-time monitoring method based on multi-sensor data fusion, comprising the following steps:

[0007] Step a, obtaining the instruction signal for the driving source and the actual action data of the driving source responding to the instruction signal; based on the instruction signal and the actual action data, using online system identification to determine a transfer function model representing the driving source itself, and generating a set of driving end dynamic parameters including driving end gain and driving end time constant;

[0008] Step b, synchronously obtaining input time series data representing the state of the driving source and output time series data representing the state of the melt in the cavity during the injection molding cycle;

[0009] Step c, based on the driving end dynamic parameters generated in step a, performing reverse compensation operation on the output time series data obtained in step b to generate corrected output time series data;

[0010] Step d, applying the input time series data and the corrected output time series data as a set of inputs to a first-order plus delay process model for online system identification to generate a set of process gain parameters and time constant parameters ;

[0011] Step e, arranging the process gain parameters and time constant parameters as a two-dimensional coordinate point in a two-dimensional process health space constructed with the process gain as the first orthogonal axis and the time constant as the second orthogonal axis, and determining whether the two-dimensional coordinate point falls within a preset reference area in the two-dimensional process health space to determine the stability of the injection molding process.

[0012] Preferably, the preset reference area is an elliptical area with a preset standard process gain and a preset standard time constant as the center, and a preset process gain allowable variation range and a preset time constant allowable variation range as the semi-axes; the step of determining whether the two-dimensional coordinate point falls within the preset reference area is implemented by executing the following determination rule: if , then determining that the injection molding process is stable.

[0013] Preferably, the instruction signal obtained in step a is a perturbation signal injected before the formal injection of the injection molding cycle; the method further comprises: extracting the echo delay and echo energy as response features from the response of the perturbation signal to the output time series data; and comparing the response features with a preset reference range to pre-judge the initial state of the injection molding process.

[0014] Preferably, the method further comprises: performing high-pass filtering on the output time series data in parallel to obtain a high-frequency noise signal; calculating the energy of the high-frequency noise signal in a preset frequency band; and based on whether the energy exceeds a preset energy threshold, determining whether there is a risk of microscopic defects caused by melt degradation or gas precipitation in the injection molded part.

[0015] Preferably, the method further comprises: after step d, generating a residual signal between the actual value of the corrected output time series data and the predicted value of the first-order plus delay process model; extracting the energy, skewness and zero-crossing rate of the residual signal as time domain features; and based on the matching result of the time domain features with a preset fault knowledge base, identifying different types of physical fault sources of the injection molding process.

[0016] Preferably, the method further comprises, before step b of each injection cycle, acquiring baseline output values of the sensors from which the output time series data is derived, in a machine logic state where the mold cavity is in communication with atmosphere and stress-free, defined by the ejection completion signal from the machine ejector pin sensor and the mold open-in-position signal from the mold position sensor being true simultaneously; and in step b, zero-point correcting the output time series data based on the baseline output values.

[0017] Preferably, the input time series data is selected from one of injection cylinder hydraulic pressure data, servo motor drive current data and servo motor drive torque data; and the output time series data is mold cavity pressure data.

[0018] Preferably, the method further comprises automatically adjusting one or more control set values of the injection molding machine based on the process gain parameter and the time constant parameter deviation from the boundary of the preset reference region.

[0019] Preferably, the inverse compensation operation performed in step c is a deconvolution operation performed on the output time series data based on the drive end gain and the drive end time constant.

[0020] Preferably, the step of identifying different types of physical fault sources comprises: when the process gain parameter decreases and the energy of the residual signal increases in the late filling stage, identifying the fault source as a gate blockage; when the process gain parameter and the time constant parameter are stable but the zero-crossing rate of the residual signal increases, identifying the fault source as a heating system performance fluctuation.

[0021] Compared with the prior art, the beneficial effects of the present application are:

[0022] 1. The time series data of the drive source state and the melt state in the mold cavity are collectively used as inputs to identify a dynamic process model, not to accurately reproduce the physical details of the injection molding process, but to extract process gain and time constant that can directly represent the characteristics of the conversion process from driving force to effective filling pressure; by constructing a two-dimensional coordinate point with these two parameters and determining its position in the preset two-dimensional healthy region, the stability of a complex industrial process with multiple physical fields and multiple variables coupled with each other is converted into a drift monitoring problem of a two-dimensional coordinate point, which provides an overall, visual and quantitative representation for monitoring and judging the health status of a transfer process that is originally fuzzy and affected by multiple factors.

[0023] 2. Before the injection molding cycle begins, a perturbation signal is first generated by the control drive source, and the actual action data of the drive source executing this perturbation signal and the output response data in the mold cavity are acquired simultaneously. Then, based on the perturbation signal and the actual action data, the dynamic response characteristics of the drive source itself are determined, and the output response data is corrected using these characteristics. Finally, based on the corrected output response data, the initial state of the injection molding process is predicted. This processing method, through online identification and compensation of the state of the measuring tool itself, separates the source of interference, the performance fluctuation of the drive system, from the process monitoring signal. This avoids confusion caused by the machine's own performance drift when judging the initial state of the process, thereby achieving separate characterization of the equipment state and the process state.

[0024] 3. While utilizing the low-frequency main signal of the mold cavity pressure data to identify the dynamic characteristics of the process, a parallel data processing channel is also used to filter and extract the high-frequency noise signal of the pressure data. By calculating the energy of this high-frequency noise signal within a specific frequency band, the risk of micro-defects in the injection molded part can be determined. This method utilizes the different physical information carried by different frequency components in the same sensor signal, enabling the monitoring of process stability reflected by changes in overall flow resistance and material quality caused by micro-events such as melt degradation or gas release within a single monitoring method, without modifying any measurement and control hardware configuration. Risk is monitored in parallel across different dimensions. After determining the parameters of the process model, a residual signal is generated between the actual values ​​of the output time series data and the predicted values ​​of the process model. Based on the time-domain characteristics of this residual signal, specific failure modes of the injection molding process are identified. This step utilizes information components that cannot be explained by the simplified model. Based on the judgment of the overall stability of the process through the process model parameters, further analysis of the model residuals provides a basis for distinguishing different sources of physical failures. This enables the monitoring system to provide a directional failure mode identification result while outputting the judgment of process instability. Attached Figure Description

[0025] Fig. 1 This is a schematic diagram of the functional architecture and data flow of the monitoring method of the present invention;

[0026] Fig. 2 This is a diagram showing the identification results of the dynamic parameters of the drive end of the present invention changing with the operating conditions.

[0027] Fig. 3 This is a timing diagram of the automatic adjustment decision-making process of the closed-loop feedback control of the present invention. Detailed Implementation

[0028] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0029] The application relates to a real-time monitoring method for an injection molding process based on multi-sensor data fusion, which comprises a driving end dynamic characteristic online identification unit, a process transfer characteristic online identification unit and a process stability determination unit. The function of the driving end dynamic characteristic online identification unit is to quantize the performance state of the injection molding equipment driving system in real time, and to correct the melt state signal in the mold cavity collected subsequently by taking the state as a dynamic reference, so that the performance fluctuation of the driving system is separated from the monitoring of the injection molding process. Then, the process transfer characteristic online identification unit processes the mold cavity response signal corrected by the driving system performance and the input signal of the driving source to generate the process gain and time constant which can represent the dynamic transfer characteristic of the physical process from the driving force application to the mold cavity pressure establishment. Finally, the process stability determination unit compares the process parameters as a two-dimensional coordinate point with a preset two-dimensional health region to realize real-time monitoring of the stability of the injection molding process. In a specific application scenario, the method is applied to solve the quality problem of uneven internal stress of products caused by performance drift of the material batch fluctuation or the oil temperature change of the equipment hydraulic system after a long time of running in the injection molding production process of precise optical lenses. In this scenario, monitoring the injection pressure of the driving end or the mold cavity pressure of the load end cannot effectively identify the problems caused by the change of the transfer process. The method is configured to perform the following steps. In the initial stage of each injection molding cycle, in order to evaluate the instantaneous performance of the driving system in the current cycle, the system first performs step a, that is, acquires an instruction signal for driving the injection molding machine screw to perform the injection action, and the actual action data fed back by the injection molding machine screw encoder representing the actual motion state of the screw. The instruction signal is a standardized perturbation signal with an amplitude of 5% of the set injection pressure and a duration of 50 ms injected by the control system before the formal injection. The system synchronously collects the time sequence of the instruction signal and the time sequence of the actual action data at a sampling frequency of 1 kHz, and performs online system identification based on the input and output data to determine a first-order plus delay model representing the transfer function of the driving source, and generates a set of driving end dynamic parameters including the driving end gain and the driving end time constant This step aims to calibrate the dynamic characteristics of the measurement tool online.

[0030] Before applying the method to a specific production task, a model parameter initialization calibration procedure needs to be performed, which first acquires a set of input and output data under stable working conditions, and calculates the Akaike Information Criterion (AIC) value of the first-order and second-order process models respectively, and selects the model order corresponding to the lower AIC value for subsequent identification. Then, after determining the model order, the forgetting factor λ in the recursive least squares algorithm used for real-time parameter calculation is set, and its initial value can be set to 0.98. When the identified process parameters [K, T] appear excessive oscillation under stable working conditions, the value of λ is increased to 0.99 to enhance the smoothness, otherwise when its response to process changes is sluggish, it is decreased to 0.95 to improve the sensitivity. In addition, for the perturbation signal used to identify the driving end dynamic characteristics, the determination process of its amplitude and duration includes: first, gradually increase the signal amplitude from 1% of the set pressure until the signal-to-noise ratio of the actual motion response data of the driving source exceeds 10, to determine the lower limit of the amplitude, and then adjust the signal duration, so that the response curve of the driving source can return to steady state after the signal ends and does not overlap with the curve of the formal injection phase, thereby obtaining a set of signal parameters that do not affect the normal process and can fully stimulate the system dynamics. Then, in the filling stage of the same injection cycle, the system performs step b to synchronously acquire input time series data representing the state of the driving source and output time series data representing the state of the melt in the mold cavity at the same sampling frequency. In this embodiment, the input time series data is the hydraulic data of the injection cylinder of the injection molding machine, which is measured by the pressure sensor provided by the device. The output time series data is the mold cavity pressure data measured by the piezoelectric pressure sensor installed at the far end of the mold cavity.

[0031] To eliminate the interference of the performance drift of the driving system itself on the process monitoring results, the system then performs step c, that is, based on the driving end dynamic parameters [K, T] generated in step a, which can reflect the actual response capability of the driving system in the current cycle, the system calculates the process parameters [K, T] in step d, which can reflect the actual response capability of the melt in the mold cavity in the current cycle. Step b] Performs a reverse compensation operation on the original output time series data obtained in step b. This reverse compensation operation is specifically a deconvolution operation, which calculates, from the cavity pressure response excited by the actual non-ideal driving action, the standard pressure response that a theoretically delay-free and gain-free driving action should elicit, thereby generating corrected output time series data. This data characterizes the melt transfer characteristics within the mold's flow channel. Step c] The reverse compensation operation performed on the output time series data based on the dynamic parameters of the driving end is specifically implemented as a Wiener deconvolution method performed in the frequency domain to suppress the amplification of noise in the original measurement signal during the calculation process. The execution steps are as follows: the output time series data obtained in step b and the impulse response sequence corresponding to the driver transfer function identified in step a are transformed to the frequency domain by Fast Fourier Transform (FFT). At the same time, the noise power spectral density is estimated based on the sensor background noise data collected by the device in the non-operating state. A Wiener filter is constructed based on the ratio of the signal power spectral density to the noise power spectral density. This filter acts as a frequency domain multiplier on the spectrum of the output data to attenuate the frequency components dominated by noise. Finally, the filtered spectrum is transformed back to the time domain by Inverse Fast Fourier Transform (IFFT), thereby generating corrected output time series data that eliminates the dynamic influence of the driver and has a controlled noise level.

[0032] Subsequently, to transform the fluid dynamics process into quantifiable dynamic characteristic indicators, the system executes step d, which involves treating the input time series data obtained in step b and the corrected output time series data generated in step c as an input-output pair and applying them to a pre-defined first-order plus-delay process model for online system identification. This identification process also employs a recursive least squares algorithm, which determines the two core parameters of the process model in real time by fitting the dynamic relationship between the input and corrected output data: the process gain parameter. With time constant parameter Among them, process gain It measures the steady-state transfer efficiency of how many units of mold cavity pressure can be converted from a unit injection pressure, while the time constant... This characterizes the sluggishness of the cavity pressure response reaching a new steady state; finally, to determine the process stability, the system executes step e, which takes the process gain parameter generated in step d. and time constant parameters As a two-dimensional coordinate point, it is arranged in a two-dimensional process health space constructed by taking the process gain as the first orthogonal axis and the time constant as the second orthogonal axis; the system compares the two-dimensional coordinate point with a preset reference area to determine the stability of the current injection molding process; the preset reference area is determined by statistical analysis of the production cycle data of 100 excellent products, and the shape is an elliptical area with the standard process gain =0.85 and the standard time constant =150ms as the center, the process gain allowable variation range =0.05 and the time constant allowable variation range =20ms as the half axis; the specific judgment rule is that if the inequality is established, it is determined that the current injection molding process is stable, otherwise the system will output an instability alarm.

[0033] The method claimed in the application can also be configured as a closed-loop control system to automatically compensate and adjust the dynamic transfer characteristics of the process; in this configuration, when step e determines that the [ ] coordinate point deviates from the preset reference area, the system will further calculate the normalized distance vector of the coordinate point from the boundary of the elliptical area; the control law module built in the system will associate the components of the distance vector with the control set value of one or more injection molding machines; for example, when the system monitors that the process gain is lower than the lower boundary of the reference area for three consecutive periods, and the deviation vector mainly points to the negative direction of the axis, the control law module will determine that the process transfer efficiency is continuously decreasing; accordingly, the controller will automatically generate an adjustment instruction to gradually increase the injection pressure set value in increments of 0.5 MPa until the [ ] coordinate point returns to the inside of the preset reference area; for another example, when it is monitored that the time constant is continuously increasing, indicating that the melt response is slowing down, the controller will then increase the melt temperature set value in increments of 2 The method of the invention can further comprise several parallel or pre-posed auxiliary technical steps without conflict; for example, to achieve the monitoring of the melt state, the system can perform a high-pass filtering process on the output time series data in parallel to obtain high-frequency noise signals with frequencies above 1 kHz; further, by calculating the signal energy of the high-frequency noise signals in the frequency band of 1 kHz to 5 kHz, and judging whether the energy exceeds an energy threshold calibrated by experiment, to identify whether there is a risk of internal defects of the injection molded part caused by the release of trace gas due to melt degradation; again, to achieve the identification of specific failure modes, after step d, the system can also generate a residual signal between the actual value of the corrected output time series data and the predicted value of the first-order plus delay process model; by extracting the time domain features such as energy, skewness and zero-crossing rate of the residual signal, and performing pattern matching with a fault knowledge base that stores typical residual characteristics corresponding to different physical faults, to give an identification result about the specific physical fault source while judging the process instability; again, to eliminate the influence of baseline drift of the sensor itself due to long-term thermal cycling on the measurement, before step b of each injection molding cycle, the system is also configured with a dynamic zero-point calibration step, which identifies the machine state where the mold cavity is in communication with the atmosphere and in a stress-free state by defining a logic gate that is true at the same time when the ejection completion signal is sent by the machine's ejector pin sensor and the mold opening in place signal is sent by the mold position sensor; and within this time window, the output of the pressure sensor is obtained as the dynamic baseline output value of the current cycle, and then the baseline value is subtracted from the original measured value when collecting data in step b, to achieve the zero-point correction of the output time series data cycle by cycle.

[0034] In a continuous injection molding process for producing high-precision polycarbonate lenses for automotive applications, surface flow marks and internal filling defects started to occur intermittently after the production line had been running stably for a while. The traditional monitoring system, which only monitored the hydraulic pressure of the injection cylinder at the drive end and the melt pressure in the mold cavity at the load end, did not issue an alarm because the hydraulic pressure curve of the injection cylinder and the pressure curve of the previous good products were identical in shape and value according to the sensor data at the drive end, and the control system executed according to the preset instructions. Although the peak value of the mold cavity pressure decreased, this phenomenon could be caused by various reasons such as changes in material viscosity or fluctuations in the temperature of the heating system, making it difficult for technicians to immediately locate the root cause of the problem. When the method claimed in the present application was used to monitor the production process, the system first performed the online identification of the dynamic characteristics at the drive end in the initial stage of the failure occurrence period by analyzing the command and execution data of the perturbation signal to determine the dynamic parameters of the drive end in the current period ]. The results showed that the parameters did not change compared to the previous stable period, ruling out the possibility that the fault originated from the performance drift of the hydraulic system of the injection molding machine. Subsequently, the system simultaneously acquired the input time series data of the injection cylinder hydraulic pressure and the output time series data of the mold cavity pressure during the filling stage, and used the obtained ] parameters to perform reverse compensation operation on the output time series data to generate corrected output time series data. This step provided a signal basis for subsequent process characteristic identification by online identification and compensation of the state of the measurement tool itself, resolving the dilemma of being unable to distinguish whether the problem originated from the drive system or the process.

[0035] Further, the system applied the input time series data and the corrected output time series data as a set of inputs to a first-order plus delay process model for online system identification and calculated the process gain parameter and the time constant parameter of the current period. The calculation results showed that the value of the process gain parameter decreased from the baseline value of 0.85 to 0.70, while the time constant parameter remained around 150 ms without significant change. When the system plotted this two-dimensional coordinate point [0.70, 150] in the two-dimensional process health space, the point had deviated from the preset baseline area centered at =0.85, =150]. The decrease in the process gain indicated a loss in the transfer efficiency from the driving force to the mold cavity pressure, i.e., an increase in the flow resistance in the transfer path, while the time constant The stability of the melt itself is indicated by the response characteristics of the melt itself; this set of parameters provides the skilled person with a basis for determining the source of the fault, i.e. the problem is not caused by a global change in the material viscosity, but by a local blockage in the flow channel; the skilled person then inspects the nozzle area of the injection molding machine and finds and removes a semi-solidified material block that has formed due to a local temperature drop at the front end of the barrel; in this scenario, the focus of the monitoring is shifted from observing isolated input or output state quantities to quantifying the system parameters that characterize the dynamic transfer relationship between the two; by transforming a multi-factor coupled process problem into a process gain and time constant monitoring, a process disturbance that is otherwise difficult to identify by conventional monitoring methods is clearly identified and located.

[0036] Example 2: In order to objectively verify the effectiveness of the method claimed in the present application in distinguishing between stable and unstable process states, a set of control tests were designed and performed; the purpose of the tests was to quantify the changes in the process gain parameters and time constant parameters calculated by the method claimed in the present application under stable production conditions and under unstable conditions artificially introduced by simulating process faults, and to compare them with the monitoring method that only observes a single sensor signal; the tests were performed on an electric injection molding machine equipped with a closed-loop control system; the platform was configured with a data acquisition system capable of collecting servo motor drive torque data as input time series data and piezoelectric pressure sensor output as output time series data at a synchronous frequency of 1 kHz, where the measurement accuracy of the pressure sensor was ±0.5%; the tests were divided into two groups, each consisting of 10 consecutive injection molding cycles: the first group was the control group, which used a set of standard process parameters known to be able to produce qualified products stably; the second group was the test group, which, based on the process parameters of the control group, only lowered the heating temperature setting value of the nozzle area of the injection molding machine by 15 This operation was used to simulate the fault scenario of increased melt flow resistance caused by nozzle blockage or local solidification of the material; except for this parameter, all other process settings, including equipment, mold, raw materials, and other process settings, were kept consistent between the two groups; during the test, the data of each cycle of the two groups were processed using the method claimed in the present application, and the calculated results of the process gain parameters and time constant parameters were recorded; at the same time, as a comparison, the peak value of the servo motor drive torque at the drive end was also recorded; the key parameters and calculation results of the test are shown in Table 1.

[0037] Table 1: Comparison of process parameter calculation results under different conditions.

[0038]

[0039] Referring to Table 1, the data show that, in the 10 stable production cycles of the control group, the process gain... The mean is 0.85, the fluctuation range is less than ±0.02, and the time constant is... The mean is 150ms, and the fluctuation range is less than ±3ms. All coordinate points fall within the preset two-dimensional process health space, and the state is judged to be stable. In contrast, in the 10 cycles of the experimental group, although the peak value of the driving torque representing the process input did not change compared with the control group, and remained at about 85.2%, the process gain... The value showed a systematic decrease, with its average value dropping to 0.72, deviating from the baseline value of the steady state. The mechanism of this phenomenon is that the peak drive torque only reflects that the servo motor executed the same control command, while the process gain... As a parameter characterizing the transmission properties of a system, its decrease objectively quantifies the physical fact that the increased melt flow resistance due to the drop in nozzle temperature leads to a reduction in the efficiency of converting driving force into effective mold cavity pressure. Experimental results confirm that the method claimed in this invention, by identifying process gain parameters… and time constant parameters This method can identify process instability caused by changes in process transfer characteristics. Compared with monitoring methods that only observe a single input or output signal, this method provides a direct and reliable quantitative basis for judging process stability.

[0040] To further verify the necessity of the specific technical step of online identification and reverse compensation of the dynamic characteristics of the driving end in the present invention, a comparative example 1 is set up.

[0041] Comparative Example 1: This comparative example aims to verify the limitations of conventional technical approaches in distinguishing between equipment status fluctuations and process fluctuations when online identification and compensation of the dynamic characteristics of the drive source itself are lacking. The experiment used the same electric injection molding machine, mold, raw materials, data acquisition system, and basic process settings as Example 2. The experiment was also divided into two groups, each undergoing 10 consecutive injection cycles. The first group served as the control group, with operating conditions and data identical to the control group in Example 2. The second group was the experimental group for this comparative example. Based on the process parameters of the control group, no process settings were changed; instead, a stable heat source was introduced into the hydraulic system's oil circuit, raising the operating temperature of the hydraulic oil in the drive system by 15°C compared to standard operating conditions. , which is used to simulate the driving performance drift caused by the change of the equipment itself (not the melt process) in the long-period production; the only difference between the processing method of the comparative example and the method claimed in the present application is that the original input time series data (servo motor driving torque data) and the original output time series data (mold cavity pressure data) collected are directly used for online system identification of the first-order plus delay process model to calculate the process gain parameter and the time constant parameter , i.e. steps a (identifying the driving end dynamic parameters and ) and step c (performing reverse compensation operation on the output time series data based on the driving end dynamic parameters) in the present application scheme are omitted, the key parameters and the calculation results of the test are shown in Table 2.

[0042] Table 2: Comparison table of process parameter calculation results using conventional technical path under different working conditions.

[0043]

[0044] Referring to Table 2, the data shows that in the 10 cycles of the test group, although the driving torque peak value of the process input does not change compared with the control group, and the physical process of the melt flow does not change, the average of the calculated process gain systematically decreases from 0.85 to 0.76, while the average of the time constant increases from 150 ms to 166 ms, and accordingly, the system determines all test group cycles as unstable; the test results confirm that due to the lack of online identification and compensation of the dynamic characteristics of the driving source itself, the conventional technical path cannot separate the performance drift (slower response, lower efficiency) of the driving system due to the increase of oil temperature from the real process state, and mistakenly attributes the performance decline of the driving system to the deterioration of the process transfer characteristics, thereby issuing an unstable false alarm.

[0045] Example 3: This example combines Figs. 1 to 3 to explain a kind of injection molding process real-time monitoring method based on multi-sensor data fusion, as shown in Fig. 1 , the method starts from the multi-source sensor signal input link, i.e. obtaining the original time series data of the driving end and the load end, and in a parallel dynamic zero point calibration module, the signal is corrected for baseline drift cycle by cycle, the calibrated signal enters the driving end dynamic characteristic online identification unit to quantify the dynamic response characteristics of the driving system itself, and generates a set of driving end dynamic parameters , which is used in the output signal inverse compensation unit to separate the disturbance of the drive system performance fluctuation on the cavity response signal, then the compensated output signal and the input signal are sent into the process transfer characteristic online identification unit to quantify the dynamic transfer process from the drive to the load and generate the core process gain and time constant , and a residual signal, based on the two core parameters, the process stability judgment unit judges whether the process is stable by monitoring the coordinate drift in the preset two-dimensional process health space, if stable, the process ends, if unstable, an instability alarm is output, at the same time, the method also includes three parallel functional modules: a microscopic defect risk monitoring module for processing high-frequency noise signals to judge material quality, a specific physical fault identification module which diagnoses based on the residual signal generated by the aforementioned process identification unit and a preset fault knowledge base, and outputs a fault diagnosis report, and a closed-loop feedback control module which can automatically adjust the injection molding machine setting value based on the deviation of the process parameters .

[0046] As shown in Fig. 2 , the figure compares four working conditions, namely the standard working condition, temperature -15 , pressure +5% and speed -10%, and presents the values of the drive end gain and the drive end time constant identified under the four working conditions, from the figure, it can be seen that compared with the standard working condition, when the machine operating parameters change, the two parameters representing the performance of the drive system also change, which confirms that the drive system is not a constant measurement reference,

[0047] As shown in Fig. 3 , the process involves the interaction of six logical units of the judgment module, the control law module, the distance calculator, the control system, the injection molding machine and the verification module, when the judgment module monitors the process instability, it will send the instability parameter deviation information to the control law module, the control law module will then call the distance calculator to calculate the deviation vector, and generate a control adjustment strategy according to the analyzed deviation direction and amplitude, for example, when the process gain is judged to be too low, the module will suggest to increase the injection pressure, or when the time constant is judged to be too large, it will suggest to increase the barrel temperature, these adjustment suggestions are sent to the control system, which converts them into specific setting value adjustment instructions for the injection molding machine, after the injection molding machine executes the new parameters, the verification module will monitor the adjustment effect, if the parameters return to the stable region, it is confirmed that the adjustment is successful and the normal production continues, if the parameters still deviate, it enters the next adjustment cycle.

[0048] Example 4: This example aims to provide a standardized engineering calibration procedure for building the fault knowledge base of the core components of the technical solution for identifying specific physical fault sources based on residual signals, and to clarify the operation mechanism of its online diagnosis; in a production environment where the monitoring method of the present application has been deployed, when the system judges that the current production process has deviated from the stable state by monitoring the changes in process gain parameters and time constant parameters , in order to locate the fault source from numerous potential causes, this example discloses a method for establishing the correlation between different physical fault modes and their model residual time domain signature; first, the residual signal between the actual value of the corrected output time series data of each cycle and the predicted value of the first-order plus delay process model needs to be collected and analyzed under the condition that the production equipment is in a stable running state and 50 consecutive qualified products are produced; by calculating the time domain features of the 50 stable cycle residual signals, i.e. residual energy, skewness and zero-crossing rate, a baseline feature vector representing the healthy state is established; during this calibration process, the baseline range of residual energy is determined to be 0.01 to 0.05 , the baseline range of skewness is -0.1 to 0.1, and the baseline range of zero-crossing rate is 150 Hz to 200 Hz; subsequently, a first type of physical fault is artificially and controllably introduced on the production equipment, i.e. by setting an obstacle occupying 20% of the cross-sectional area at the mold gate to simulate the gate clogging condition; under this condition, the system is continuously operated for 10 injection cycles, and the residual signal of each cycle is analyzed; the analysis results show that under this fault mode, the process gain has a moderate decrease, at the same time, the residual energy has a systematic increase in the late filling stage, with an average value of , which exceeds the upper limit of the baseline range, while the skewness and zero-crossing rate have no significant change; the system accordingly takes the combination of moderate decrease in process gain and increase in residual energy as a specific residual signature, correlates it with the physical fault of gate clogging, and stores it as the first rule in the fault knowledge base.

[0049] Then, the gate obstacle is removed, and after the system returns to the stable baseline state, a second type of physical fault is introduced, i.e. by superimposing a temperature disturbance signal with a period of 30 seconds and an amplitude of ±5 in the control loop of a heating ring of the injection molding machine to simulate the performance fluctuation condition of the heating system; under this condition, the system is also continuously operated for 10 cycles and the residual signal is analyzed; the analysis results show that under this fault mode, the process gain and time constant the mean value fluctuates within the benchmark range, but its cycle-to-cycle stability deteriorates, and the zero-crossing rate of the residual signal increases, with a mean value of 450 Hz, which is more than twice the upper limit of the benchmark range; accordingly, the system will and the time-domain feature of the combination of stable mean value, increased variance, and elevated residual zero-crossing rate, as another specific residual signature, is associated with physical fault-induced process performance fluctuation, and is stored as the second rule; by repeating this process, multiple known physical fault modes can be associated with their corresponding specific residual signatures, thereby constructing a fault knowledge base containing multiple fault mode rules; in subsequent online real-time monitoring, as soon as the system determines that the process is unstable, it will immediately start the diagnosis process: the system first calculates the residual signal of the current unstable cycle and extracts its time-domain feature signature; then, it performs pattern matching of this signature with each rule stored in the fault knowledge base, by calculating the Euclidean distance between it and each pre-stored signature, to find the closest match; finally, the system outputs a combined information containing the preliminary diagnosis opinion, which includes the unstable state and the matched fault mode, such as process instability, the value decreases and the residual signature matches the sprue blockage mode, thereby narrowing the scope of fault diagnosis and improving repair efficiency.

[0050] Embodiment 5: This embodiment aims to provide a standardized engineering procedure for the dynamic zero-point calibration step to eliminate the influence of sensor baseline drift, and the calibration step to establish the pre-set benchmark area in the two-dimensional process health space; in a long-period continuous production application, to deal with the situation that the zero-point output of the cavity pressure sensor will slowly drift due to repeated thermal cycles and material aging, which may be incorrectly interpreted by the system as a change in process gain The method claimed in the present application is configured with a set of cycle-by-cycle dynamic zero-point calibration procedures when deployed; the procedure is set to trigger the data acquisition system to obtain the baseline output value of the cavity pressure sensor at the logical moment when the mold cavity is in a stress-free state and in communication with the atmosphere, i.e., when the ejection completion signal from the machine's ejector pin sensor and the mold position sensor's open mold in place signal are both true, at a specific stage of each injection cycle; the baseline output value is defined by the system as the dynamic zero-point bias value exclusive to the next injection cycle, and in the data acquisition of the next cycle, the bias value is subtracted from all the raw output time series data collected, thereby generating a set of zero-point corrected data for subsequent system identification calculations; this procedure uses the inherent logic state of the equipment to build a self-calibration loop for the core sensor that runs continuously without human intervention.

[0051] On the basis of the calibration procedure ensuring the long-term validity of the input data, for a specific production task to establish an objective and statistically significant stable process benchmark, the system will perform a two-dimensional process health space calibration procedure after the first deployment or replacement of the production task; the procedure requires that the process engineer first debug the production equipment to a stable state that can continuously produce qualified products, then the system is switched to learning mode, in which mode the system will continuously collect and calculate the values of the process gain parameters and time constant parameters of the next 100 production cycles; after the collection is completed, the system automatically calculates the mean value and and the standard deviation and of the 100 pairs of parameters; further, the system sets the center point coordinates of the preset benchmark region as the calculated mean value , and sets the lengths of the two semi-axes of the elliptical region and as three times the standard deviation, that is and ; in this way, a confidence region covering the normal production fluctuations is automatically generated, providing a traceable and standardized basis for subsequent real-time monitoring.

[0052] Example 6: This embodiment is a specific application of the technical solution in an application scenario involving closed-loop control. In the injection molding process of producing a precision catheter for medical use, due to the slight viscosity difference between batches of high molecular materials used, the process gain parameter may deviate slightly in the initial period of replacing a new batch of materials, which may affect the dimensional consistency of the product. To address this condition, the method claimed in the present application is configured in an automatic closed-loop control mode; after replacing a new batch of materials, the system monitors that the mean value of the process gain decreases from the standard value 0.85 to 0.81, which value has not exceeded the hard shutdown threshold, but has deviated from the center of the preset benchmark region, indicating that the fluidity of the current batch of materials has decreased slightly compared to the standard batch; the system immediately starts the automatic adjustment procedure, according to the deviation of the value ( ), and according to a preset linear control model, that is, there is a relationship between the injection pressure adjustment amount and the process gain deviation , where is the control coefficient calibrated through previous experiments, for example MPa; the system calculates the required pressure adjustment to be 0.4 MPa accordingly.

[0053] Subsequently, the system controller automatically adjusts the original injection pressure setting of the injection molding machine from 80.0 MPa to 80.4 MPa; in the following production cycle, the system continues to monitor the pair of parameters and observes that the value of the process gain returns to the vicinity of 0.84, re-entering the central range of the preset reference region; in this way, the system automatically compensates for the process shift introduced by the fluctuation in the raw material batch, using direct quantification of the process transfer characteristics, thereby maintaining the stability of the production process and the consistency of the product quality without human intervention.

[0054] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0055] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for real-time monitoring of an injection molding process based on multi-sensor data fusion, characterized in that, The method comprises the following steps: Step a, obtaining an instruction signal for the driving source and actual action data of the driving source in response to the instruction signal; based on the instruction signal and the actual action data, using online system identification to determine a transfer function model for characterizing the driving source itself, and generating a set of driving end dynamic parameters including driving end gain and driving end time constant; Step b, synchronously obtaining input time series data representing the state of the driving source and output time series data representing the state of the melt in the mold cavity during the injection molding cycle; Step c, based on the driving end dynamic parameters generated in step a, performing reverse compensation operation on the output time series data obtained in step b to generate corrected output time series data; Step d. Apply the input time series data and the corrected output time series data as a set of inputs to a first order plus delay process model for online system identification to generate a set of process gain parameters and time constant parameters ; Step e, the process gain parameter and the time constant parameter as a two-dimensional coordinate point, is disposed in a two-dimensional process health space constructed with the process gain as a first orthogonal axis and the time constant as a second orthogonal axis, and it is determined whether the two-dimensional coordinate point falls within a preset reference region in the two-dimensional process health space to determine the stability of the injection molding process.

2. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, characterized in that, The preset reference region is an elliptical region with a preset process gain and a preset standard time constant as the center, the preset process gain allows a range of variation and the preset time constant allows a range of variation The preset reference region is an elliptical region with a preset process gain If the two-dimensional coordinate point falls within the preset reference region, the injection molding process is determined to be stable.

3. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The instruction signal obtained in step a is a perturbation signal injected before the official injection of the injection molding cycle; the method further comprises: extracting echo delay and echo energy as response features from the response of the output time series data to the perturbation signal; and comparing the response features with a preset reference range.

4. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The method further comprises: performing high-pass filtering on the output time series data in parallel to obtain a high-frequency noise signal; calculating the energy of the high-frequency noise signal in a preset frequency band; and based on whether the energy exceeds a preset energy threshold, determining whether the injection part has a risk of micro defects caused by melt degradation or gas precipitation.

5. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The method further comprises: after step d, generating a residual signal between the actual value of the corrected output time series data and the predicted value of the first-order plus delay process model; extracting the energy, skewness and zero-crossing rate of the residual signal as time domain features; and based on the matching results of the time domain features with a preset fault knowledge base, identifying different types of physical fault sources of the injection molding process.

6. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The method further comprises, before step b of each injection molding cycle: in a machine logic state in which the mold cavity is in communication with the atmosphere and in a stress-free state when the ejection completion signal is sent by the machine ejector sensor and the mold opening signal is sent by the mold position sensor, obtaining the baseline output value of the sensor from which the output time series data is derived; and in step b, zero-point correcting the output time series data based on the baseline output value.

7. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The input time series data is selected from one of injection cylinder hydraulic data, servo motor driving current data and servo motor driving torque data; the output time series data is mold cavity pressure data.

8. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The method further includes automatically adjusting one or more control settings of the injection molding machine based on the process gain parameter and the time constant parameter and a deviation of the boundary of the preset reference area, 9. The method for real-time monitoring of injection molding process based on multi-sensor data fusion according to claim 1, wherein, The reverse compensation operation performed in step c is a deconvolution operation performed on the output time series data based on the driving end gain and the driving end time constant. The reverse compensation operation performed in step c is a deconvolution operation performed on the output time series data based on the driving end gain and the driving end time constant.

10. The method of claim 5, wherein the method is based on multi-sensor data fusion for real-time monitoring of an injection molding process. The step of identifying different types of physical fault sources includes: when the process gain parameter decreases and the energy of the residual signal increases at the late filling stage, the fault source is identified as a gate blockage; when the process gain parameter and the mean value of the time constant parameter are stable but the zero-crossing rate of the residual signal increases, the fault source is identified as a heating system performance fluctuation.

Images