A method for identifying material disturbance in a PE film preparation process

Abstract

The application relates to the technical field of plastic film preparation, in particular to a material disturbance identification method in a PE film preparation process, which comprises the following steps: constructing a multi-modal sensor matrix in a melt flow channel, synchronously collecting a first pressure signal, a second pressure signal and a complex dielectric spectrum sequence; constructing a baseline multi-dimensional feature vector; constructing a real-time feature vector on line; and judging a disturbance event under a double-level judgment criterion through a weighted Mahalanobis distance. The application can realize non-destructive and advanced disturbance identification, and improves the film preparation precision.

Description

Technical Field

[0001] This invention relates to the field of plastic film preparation technology, and in particular to a method for identifying material disturbances during the preparation of PE film. Background Technology

[0002] In modern continuous PE film production lines, the process from raw material melting, extrusion, and molding to subsequent traction and winding constitutes a complex, dynamic, and multivariate coupled flow. In traditional PE film manufacturing, to maintain process stability, engineers have developed a series of monitoring and control methods based on macroscopic process parameters. These methods primarily rely on temperature sensors, pressure sensors, and melt pump speed monitoring devices deployed along key nodes of the production line to collect real-time data on the temperature, pressure, and flow rate of the melt within the extruder barrel and die channels. However, with the increasing demand from downstream industries for ultra-thin, highly transparent, and extremely uniform thickness PE films, and the continuous improvement in production line operating speeds, the aforementioned technical solutions that only focus on macroscopic process parameters are gradually revealing their inherent limitations at the principle level. This is because the temperature and pressure monitored by these solutions are essentially statistical averages of the movement of countless micro-units within the melt flow field, representing a macroscopic characterization of the already formed overall flow state. They have shortcomings in addressing material disturbances caused by uneven local shear heating within the melt, fluctuations in the distribution of the dispersed phase of additives, or the presence of discrete high-viscosity points such as microgels within the raw material.

[0003] This disturbance does not directly manifest as a violent fluctuation in the overall pressure of the die head. In its initial stage, it is often hidden deep within the microstructure of the melt flow field, causing only slight disturbances in local streamlines or abrupt changes in viscosity at the submicroscopic scale. When this minute material disturbance occurs and develops only at the polymer chain level, macroscopic temperature and pressure sensors can hardly detect any abnormal signals. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying material disturbances during the preparation of PE films, in order to solve the technical problem that the control system of the existing PE film preparation process relies on macroscopic process parameters and cannot penetrate the surface of the melt flow field to identify material disturbances occurring at the microscale.

[0005] This invention provides a method for identifying material disturbances during the preparation of PE films, comprising: S1. Steps for establishing a reference signal array: In the melt flow channel of the extrusion system, a multimodal sensor matrix is ​​constructed, consisting of at least one type I pressure sensor, at least one type II high-frequency response pressure sensor, and a wideband dielectric spectrum sensor; the type I pressure sensor, the type II high-frequency response pressure sensor, and the wideband dielectric spectrum sensor are integrated in a common sensor mounting base, the sensor mounting base having a melt contact surface flush with the inner wall of the melt flow channel, and the physical center point of the three on the melt contact surface is located within a virtual circle with a diameter not greater than a predetermined threshold, the predetermined threshold being not greater than half of the hydraulic radius of the melt flow channel; S2. Establish a disturbance event marker reference signal: During the preset baseline process window period in which no material disturbance occurs, continuously acquire the first pressure signal, the second pressure signal, and the complex dielectric spectrum sequence, and process the complex dielectric spectrum sequence to construct a multidimensional baseline feature vector; S3. Online real-time disturbance identification: After the baseline process window period ends, a continuous online monitoring state is entered. In each discrete evaluation cycle, the first pressure signal, the second pressure signal, and the complex dielectric spectrum sequence are synchronously acquired and processed to construct a real-time multidimensional feature vector. The structural dimension of the real-time multidimensional feature vector is consistent with the multidimensional baseline feature vector. S4. Material Disturbance Determination: The real-time multidimensional feature vector is compared with the multidimensional baseline feature vector in a multidimensional feature space. When the comparison result meets the preset joint determination criterion, a material disturbance event is determined to have occurred in the current evaluation period. The joint determination criterion is based at least on a weighted Mahalanobis distance between the real-time multidimensional feature vector and the multidimensional baseline feature vector in the multidimensional feature space. In the calculation of the weighted Mahalanobis distance, the weighting coefficient determined by the high-frequency fluctuation energy spectral density of the second pressure signal in the evaluation period is used to weight and modulate the deviation of the corresponding dimension of the complex dielectric spectrum sequence.

[0006] In some embodiments, in S1, the baseline process window period is determined based on the output signals of the first type of pressure sensor and the second type of high-frequency response pressure sensor after the extrusion system is started: When the fluctuation amplitude of the first pressure signal is less than a predetermined first amplitude threshold for a continuous first preset duration, and at the same time, the root mean square value of the signal component of the second pressure signal after high-pass filtering is less than a predetermined second amplitude threshold within any continuous second preset duration, the continuous period that meets the conditions is automatically defined as the baseline process window period, and the establishment of the disturbance event marker reference signal is triggered.

[0007] In some embodiments, in S2, the processing of the complex dielectric spectrum sequence includes: based on the melt temperature measured in real time by a temperature sensor integrated in the sensor mounting base, performing a predetermined temperature compensation function on the real part and imaginary part of the acquired complex dielectric spectrum sequence respectively, so as to eliminate the influence of the small and slow drift of the melt body temperature on the dielectric spectrum.

[0008] In some embodiments, the broadband dielectric spectrum sensor is composed of an interdigital electrode array, which is directly fabricated on an insulating ceramic substrate, the insulating ceramic substrate forming part of the melt contact surface of the sensor mounting base; the finger width and finger spacing of the interdigital electrodes are set within a predetermined range of 10 micrometers to 100 micrometers, and the electrode material is a platinum-rhodium alloy. When the wideband dielectric spectrum sensor sweeps the frequency, the frequency range of its excitation electric field extends from 100 kHz to 10 MHz. Within this frequency range, no less than 20 discrete frequency points are selected in a logarithmically uniform distribution for point-by-point excitation and response measurement to obtain the complex dielectric spectrum sequence.

[0009] In some embodiments, in S3, constructing the multidimensional baseline feature vector includes an independent preprocessing procedure for the second pressure signal: The second pressure signal is passed through a high-pass filter whose cutoff frequency is located between the first and second frequency response bandwidths to obtain a high-frequency pressure fluctuation component. Within a time window of equal length to the evaluation period, the energy spectral density distribution of the high-frequency pressure fluctuation component is calculated, and its spectral energy values ​​in several specified characteristic frequency bands are extracted. These spectral energy values ​​are used to construct the weighting coefficients in the weighted Mahalanobis distance.

[0010] In some embodiments, the joint determination criterion is a two-layer determination criterion: The first layer is a threshold judgment layer, which requires that the weighted Mahalanobis distance must exceed a predetermined adaptive threshold determined by statistical analysis within the baseline process window. The adaptive threshold is set based on the statistical distribution characteristics of the multidimensional baseline feature vector calculated within the baseline process window, and is specifically set as a predetermined multiple of the root mean square value of the weighted Mahalanobis distance after the multidimensional baseline feature vector is centered. The second layer is a persistence verification layer, which requires that after the conditions of the first layer are met, the value of the weighted Mahalanobis distance must remain above a predetermined persistence threshold for the next N consecutive evaluation periods. The persistence threshold is lower than the adaptive threshold, and N is a pre-set positive integer.

[0011] In some embodiments, the second type of high-frequency response pressure sensor specifically adopts a piezoresistive pressure sensor based on microelectromechanical systems technology, with a pressure-sensing diaphragm thickness of less than 5 micrometers and an inherent frequency higher than 1 MHz, used to achieve non-distortion capture of flow noise and local pressure transients in the melt caused by microgel particles or local viscosity abrupt changes, with a frequency range of 10 kHz to hundreds of kHz; the first type of pressure sensor adopts a strain gauge pressure sensor.

[0012] In some embodiments, the sensor mounting base is a modular component that is sealed and embedded in the straight pipe section of the flow channel between the extruder die and the melt filter via a flange interface; all sensor signal cables are led out from inside the sensor mounting base through a sealed through-hole and connected to the high-speed synchronous data acquisition and processing unit.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The first type of pressure sensor provides a slowly varying reference standard, while the second type of high-frequency response pressure sensor is specifically designed for high-frequency, short-lived local mechanical anomalies in the melt flow field. However, the above alone is insufficient to achieve high-confidence identification under all operating conditions, as mechanical vibrations or external electromagnetic interference can also generate similar high frequencies. Therefore, a broadband dielectric spectrum sensor is introduced. When disturbances occur within the melt due to material composition or structural inhomogeneity, such as the appearance of microgels, agglomeration of additive particles, or local degradation of polymer chains, these microstructural changes will inevitably alter the local dipole moment motion and interfacial polarization effect, thereby causing characteristic changes in the dielectric spectrum within a specific frequency range. Because the physical mechanism of this change is completely different from mechanical fluctuations, it can form a strong cross-validation with the high-frequency pressure signal. 2. In the determination of material disturbance, the use of weighted Mahalanobis distance and dynamic weighted modulation of dielectric spectrum deviation establishes an asymmetric criterion with clear physical meaning. When the high-frequency energy spectral density of the second pressure signal is significantly enhanced in certain characteristic frequency bands, it means that there is indeed a high probability of mechanical disturbance source in the flow field at this time. Correspondingly, the joint determination criterion will automatically increase the attention to even small deviations in the characteristic dimensions of dielectric spectrum, that is, dynamically increase the contribution weight of these dimensions to the final distance calculation result. 3. This invention enables early warning of material disturbances to be pushed back from the quality defects after film forming to the stage when the melt rheological microstructure has just changed, providing a response window for proactive intervention in process parameters. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the material disturbance identification method during the PE film preparation process of the present invention; Figure 2 This is a system architecture diagram corresponding to the material disturbance identification method of the present invention. Detailed Implementation

[0016] The following will be based on embodiments of the present invention. Figures 1-2 The technical solutions in the embodiments of the present invention will be clearly and completely described together. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Application Overview In modern continuous PE film production lines, the process from raw material melting, extrusion, and molding to subsequent traction and winding constitutes a complex, dynamic, and multivariate coupled flow. The stability of this process directly affects the yield and production efficiency of the film products. Therefore, precise control of various process states during preparation has always been a core direction for technological research and engineering optimization in this field. Specifically, in traditional PE film preparation processes, to maintain process stability, engineers have developed a series of monitoring and control methods based on macroscopic process parameters. These methods mainly rely on temperature sensors, pressure sensors, and melt pump speed monitoring devices deployed along key nodes of the production line to collect real-time data on the temperature, pressure, and flow rate of the melt within the extruder barrel and die channel. The core control logic lies in ensuring the stability of these macroscopic physical quantities to indirectly maintain the stability of the melt delivery and molding process, thereby ensuring the uniformity of film thickness. For example, when slight fluctuations occur in the die pressure, the control system automatically adjusts the extruder screw speed or the output of the melt gear pump, using closed-loop feedback to restore the pressure value to a preset threshold range. This process control system, based on macroscopic parameters (such as pressure, temperature, and overall flow rate), can effectively suppress significant process drift caused by factors such as temperature drift and unstable material supply in early industrial production where precision requirements are relatively relaxed, providing a solid technical foundation for the large-scale production of PE film. However, under current conditions, many incompatibilities have emerged, hence this invention is proposed, detailed below.

[0018] Example refer to Figure 1The front-end sensing node is highly integrated within the melt flow channel of the extrusion system. The physical core of this sensing node is the sensor mounting base. This sensor mounting base has a precision-machined melt contact surface that is geometrically flush with the inner wall of the melt flow channel, thus embedding the sensor probe into the flow field while minimizing interference with the streamlines. Inside this sensor mounting base, three independent sensor elements are rigidly integrated, collectively forming a multimodal sensor matrix with functional hierarchical layering. Specifically, this matrix includes at least one type-1 pressure sensor, at least one type-2 high-frequency response pressure sensor, and a broadband dielectric spectrum sensor.

[0019] The first type of pressure sensor is configured with a first frequency response bandwidth. In this embodiment, the sensor specifically employs a traditional strain gauge pressure sensor, whose measurement accuracy is within one-thousandth of the full scale, but whose natural frequency is designed to be on the order of several kilohertz. This first frequency response bandwidth setting allows it to effectively filter out high-frequency flow noise, thereby focusing on capturing quasi-static pressure changes that characterize the macroscopic flow state of the melt, providing a pressure reference that evolves slowly over time, i.e., the first pressure signal. The second type of high-frequency response pressure sensor, which complements this design, is configured with a second frequency response bandwidth much higher than the first. As a preferred method for achieving high-frequency response characteristics, this sensor is a piezoresistive pressure sensor, whose core pressure-sensing element is a single-crystal silicon pressure-sensing diaphragm. The thickness of this diaphragm is reduced to less than 5 micrometers, thereby making the sensor's natural frequency higher than 1 megahertz. This characteristic ensures that it can capture with high fidelity the flow noise within the melt caused by events such as the rupture of microgel particles, local viscosity abrupt changes, or microscopic cavitation, with a frequency range covering 10 kilohertz or even hundreds of kilohertz. The output signal is recorded as the second pressure signal.

[0020] The working principle of the broadband dielectric spectrum sensor in the multimodal sensor matrix is ​​not single-point frequency measurement. Instead, within a predetermined time window, an alternating electric field including multiple discrete frequency points is actively applied to the melt flowing over its surface using a frequency sweeping method, while simultaneously and continuously measuring the complex response of the melt to this excitation, thereby obtaining a complex dielectric spectrum sequence that evolves over time. This complex dielectric spectrum sequence is a macroscopic manifestation of the orientation motion of polar polymer chain segments inside the melt, the frictional relaxation process between chain segments, and the interfacial polarization effect of dispersed phases such as additives under an external alternating electric field.

[0021] To ensure strict spatial synchronization and causal correlation analysis of the aforementioned physical signals, and to avoid signal interpretation ambiguity caused by differences in the flow field space due to different sampling locations, the relative physical positions of the first type of pressure sensor, the second type of high-frequency response pressure sensor, and the broadband dielectric spectrum sensor are defined. Their physical center points are all positioned within a virtual circle on the melt contact surface, with the diameter of this virtual circle not exceeding a preset threshold. This preset threshold is set to not exceed half the hydraulic radius of the melt flow channel, thus ensuring, in a fluid dynamics sense, that the signals are all collected from the same local flow field region.

[0022] Furthermore, the broadband dielectric spectrum sensor consists of an interdigitated electrode array, which is directly fabricated on an insulating ceramic substrate using thin-film deposition and photolithography. This insulating ceramic substrate itself constitutes the core region of the melt contact surface of the sensor mounting base. The finger width and finger spacing of the interdigitated electrodes are set within a preferred range of 10 to 100 micrometers, and the electrode material is a platinum-rhodium alloy with excellent resistance to high-temperature melt corrosion. During frequency sweep excitation, the electric field frequency is not fixed, but rather at least 20 discrete frequency points are selected in a logarithmically uniform distribution within a frequency span of 100 kHz to 10 MHz. Single-frequency excitation and complex impedance response measurements are performed sequentially, thereby constructing a complex dielectric spectrum sequence that can finely characterize the dielectric relaxation spectrum of the material.

[0023] To achieve the temperature drift compensation function inherent in this method, a temperature sensor, such as a high-precision thin-film platinum resistance thermometer, is also integrated inside the sensor mounting base to monitor the temperature of the melt body in this local area in real time and synchronously.

[0024] At the back-end processing level of signals and data, all sensor signal cables are led out from inside the sensor mounting base through a sealed through-hole with high temperature and high pressure resistance, and connected to a high-speed synchronous data acquisition and processing unit located in a non-explosion-proof safety area.

[0025] The core hardware of this processing unit consists of a high-speed analog-to-digital converter, a field-programmable gate array (FPGA), and a high-performance embedded processor, running dedicated signal processing firmware built on a real-time operating system. This firmware executes all signal processing flows, feature extraction, criterion calculation, and logic decision tasks, which will be described in detail later, in hard real-time, and ultimately outputs standard industrial digital signals, such as dry contact signals or a Modbus TCP message, for seamless integration into the distributed control system or programmable logic controller of the production line.

[0026] In practice, the sensor mounting base is designed as a replaceable modular component, sealed and embedded in the straight pipe section of the flow channel between the extruder die and the upstream melt filter via an industry-standard flange interface. The selection of this mounting location is based on detailed computational fluid dynamics simulations of the flow field. The simulation results show that the streamlines in this region are fully developed, the velocity profile is relatively stable, and it is located sufficiently far upstream of the final forming die, forming an ideal observation window for early detection and warning of various disturbances caused by the upstream extrusion system or the raw material itself.

[0027] The following combination Figure 2 The method for identifying material disturbances during the preparation of PE film according to the present invention is described in detail.

[0028] First, a multidimensional baseline reference is established to characterize the intrinsic features of the material under undisturbed conditions, i.e., the step of establishing a reference signal array. Then, a reference signal for marking disturbance events is established based on a determined baseline process window. Since the baseline cannot be acquired when production is disturbed, how to automatically lock the ideal, undisturbed baseline process window period becomes the primary basis for realizing the reference benchmark function of this method.

[0029] Furthermore, this method proposes a window discrimination mechanism based on the statistical characteristics of the sensor output signals. After the extrusion system has completed heating and begun stable extrusion, the first type of pressure sensor and the second type of high-frequency response pressure sensor begin to continuously output signals. The data processing unit continuously monitors these two signals. Specifically, during the continuous online process, the system continuously calculates the fluctuation amplitude of the first pressure signal within a sliding time window. When it is detected that the fluctuation amplitude is less than a pre-calibrated first amplitude threshold for a continuous first preset time period, it indicates that the zero-order state of the macroscopic melt transport has reached stability.

[0030] Simultaneously, the system performs high-pass filtering on the second pressure signal, retaining only its high-frequency fluctuation components, and calculates the root mean square (RMS) value of these high-frequency components over a continuous second preset time period. When the RMS value is consistently lower than a predetermined second amplitude threshold, it indicates from a mechanical fluctuation perspective that no abnormal micromechanical events have occurred inside the melt.

[0031] Only when the aforementioned macroscopic and microscopic conditions are simultaneously met, and this met state is maintained for a specified duration, will the system automatically define this continuous period of continuously met conditions as the baseline process window. For example, in a specific setting, the first preset duration is set to 120 seconds, and the first amplitude threshold is set to two-thousandths of the macroscopic pressure setting value under the current operating conditions; the second preset duration is set to 180 seconds, and the second amplitude threshold is calibrated to an extremely low statistical level of the signal amplitude after passing through a high-pass filter. Once this window is identified, the system will automatically trigger the establishment of a disturbance event marker reference signal.

[0032] During the formal establishment of the reference signal, the system performs statistical processing on all signals acquired within the baseline process window. For the first pressure signal, the system simply calculates the arithmetic mean of all sampled values ​​throughout the window, obtaining a macroscopic pressure mean in scalar form. For complex dielectric spectrum sequences with large data volumes and rich physical meaning, the processing is more nuanced. Because the temperature of the melt body exhibits tiny, slow drifts on the order of a few ten-thousandths of a degree Celsius even under steady-state conditions, this drift systematically affects the dielectric spectrum measurement results, causing the baseline data to include spurious changes from non-material perturbation factors.

[0033] To eliminate this systemic bias, this method actively performs a system drift removal operation when establishing the baseline reference signal. Specifically, this operation uses the melt temperature measured in real-time by an integrated temperature sensor at the corresponding moment to call a pre-calibrated temperature compensation function calibrated offline. This temperature compensation function analytically describes the relationship between the complex dielectric constant of a specific grade of PE melt and the minute changes in temperature under undisturbed conditions. The system substitutes the real and imaginary parts of the complex dielectric spectrum sequence acquired at each moment within the window period into this function, uniformly compensating to a common reference temperature, thereby obtaining a temperature-normalized complex dielectric spectrum sequence.

[0034] Next, the system performs multi-dimensional feature extraction on all complex dielectric spectrum sequences within the window period after temperature compensation to construct a multi-dimensional baseline feature vector. The constituent dimensions of this vector are clearly defined, including: the real part mean vector obtained by statistically analyzing all spectrum data within the window period, where each element of the vector corresponds to the average of the real part measurements at each swept frequency point; similarly constructed imaginary part mean vector; the loss tangent mean calculated based on the ratio of the imaginary part to the real part at several specified characteristic frequency points; and the previously calculated macroscopic pressure mean.

[0035] After the baseline process window ends and a multidimensional baseline feature vector is successfully established, the system seamlessly switches to continuous online monitoring, i.e., real-time online disturbance identification. During normal online monitoring, the system continuously identifies disturbances using discrete, equal-length evaluation periods as the basic time unit. In a preferred embodiment, the duration of this evaluation period is set to 500 milliseconds. This time interval is determined by balancing the response speed requirements for disturbance events with the data statistical length required for feature extraction. Within each 500-millisecond evaluation period, the system synchronously acquires a first pressure signal, a second pressure signal, and a complex dielectric spectrum sequence, and constructs a real-time multidimensional feature vector corresponding to the current evaluation period. The construction logic of this real-time feature vector is largely similar to that of its baseline version, but its unique feature is the introduction of an independent preprocessing procedure specifically designed for the second pressure signal. The purpose of this procedure is to quantify the intensity of the mechanical disturbance in the flow field at the current moment.

[0036] The preprocessing procedure for the second pressure signal relies on a high-pass filter. The cutoff frequency of the high-pass filter is set between the highest cutoff frequency of the first frequency response bandwidth and the lowest starting frequency of the second frequency response bandwidth, for example, at 5 kHz. Next, the high-frequency pressure fluctuation component data captured within a 500-millisecond evaluation period undergoes time-frequency transformation processing, such as using short-time Fourier transform or wavelet packet transform, to calculate the power spectral density distribution in the frequency domain, thereby obtaining the high-frequency fluctuation energy spectral density of the signal. The system does not use the sum of energy across the entire frequency band, but selectively extracts the spectral energy values ​​of this energy spectral density within several pre-defined characteristic frequency bands that are characteristic indicators of micromechanical perturbations. For example, the first characteristic frequency band might be set between 15 kHz and 25 kHz to capture typical acoustic radio frequency bands of microgel particle deformation and breakage; the second characteristic frequency band might be set between 60 kHz and 120 kHz to sense more subtle local viscosity abrupt changes.

[0037] Meanwhile, the processing of the first pressure signal and the complex dielectric spectrum sequence during this evaluation period strictly follows the same workflow architecture as when constructing the baseline. The same temperature compensation operation is performed on the complex dielectric spectrum data, and the real part mean, imaginary part mean, and loss tangent at a specified frequency point are extracted.

[0038] With real-time multidimensional feature vectors, multidimensional baseline feature vectors from historical baselines, and weight information parsed from high-frequency pressure signals, the process proceeds to material perturbation determination. Perturbation determination is not based on a simple distance metric, but rather executes a pre-defined two-layer determination criterion.

[0039] The first layer performs dynamically weighted modulation distance calculation in a multi-dimensional feature space and compares it with a statistically adaptive threshold. Here, the system calculates the weighted Mahalanobis distance between the real-time vector and the baseline vector. The final result of this distance is determined not only by the offset of the real-time vector relative to the baseline mean in each dimension, but more importantly, by the weighting coefficients from the second pressure signal that weight the deviation of the complex dielectric spectrum sequence in the corresponding dimension.

[0040] The specific mechanism is as follows: when calculating distance, for several dimensions involving the real and imaginary parts of the dielectric spectrum and the loss tangent, the square of the deviation is multiplied by a weighting factor. This weighting factor is a series of weighting coefficients determined based on the spectral energy values ​​extracted from the high-frequency pressure fluctuation components in the aforementioned independent preprocessing steps. The magnitude of this coefficient is positively correlated with the value of the energy spectral density within the specified characteristic frequency band. Physically, when the mechanical noise within the fluid is significantly enhanced, the joint decision criterion amplifies the sensitivity to even minute changes in the dielectric spectrum dimension, achieving a gain amplification in cross-validation.

[0041] After the weighted Mahalanobis distance is calculated, it is sent to the first-layer threshold judgment layer. This threshold is not a fixed constant, but an adaptive threshold predetermined through statistical analysis within the previous baseline process window. Specifically, it is determined as follows: all sample points obtained within the baseline process window are treated as instances of the normal state, and their weighted Mahalanobis distances relative to the centers of the multidimensional baseline feature vectors are calculated to obtain a distribution set of distance values; then, the root mean square value of all distance values ​​in this set is calculated; finally, this root mean square value is multiplied by a predetermined safety factor, such as a coefficient between 3 and 5, to obtain the adaptive threshold.

[0042] If the calculated real-time weighted Mahalanobis distance exceeds the adaptive threshold, the first-level condition of the decision criterion is temporarily satisfied. However, this does not immediately trigger a disturbance alarm signal, because isolated points exceeding the threshold are likely just instantaneous high-energy noise spikes. To confirm a disturbance event with real physical persistence, representing a change in material properties, the decision proceeds to the second level, the persistence verification layer. This layer requires that, starting the instant after the first-level condition is satisfied, the system enters a verification sequence with enhanced observation. In the following N consecutive evaluation cycles, the weighted Mahalanobis distance value calculated in each cycle, while not necessarily higher than the strict adaptive threshold, must all remain above a slightly lower, predetermined persistence threshold.

[0043] This persistence threshold is typically set to 60% to 80% of the adaptive threshold. Only when the persistence condition is met for N consecutive cycles will the system finally output a high-confidence judgment signal, confirming that a real material disturbance event has occurred at the current moment. N is a configurable positive integer, and its setting depends on the trade-off between the speed of disturbance identification and the resistance to interference. In a typical configuration, N is set to 5, corresponding to a verification time window of 2.5 seconds.

[0044] To more concretely demonstrate the technical superiority of the method of the present invention, specific examples and comparative examples are used below for comparison and explanation.

[0045] On a PE film production line, the extruder screw diameter is 90 mm, the raw material is low-density polyethylene, the melt temperature is set to 160 degrees Celsius, and the production line speed is maintained at 2 meters per second. According to this invention, an integrated multimodal sensor mounting base and the entire back-end processing system are installed on the straight pipe section of the flow channel between the extruder and the die. The finger width and finger spacing of the interdigital electrode array are both 50 micrometers. The frequency sweep range is set to 100 kHz to 10 MHz, with 30 frequency points selected in a logarithmic uniform distribution. The first frequency response bandwidth of the first type of pressure sensor is 0-1 kHz, and the second type of high-frequency response pressure sensor uses a piezoresistive sensor with a pressure-sensitive diaphragm thickness of 3 micrometers, and its second frequency response bandwidth is 10 Hz to 1 MHz. The establishment of the baseline process window is automatically completed as described above, forming a multidimensional baseline feature vector. During continuous operation, the system enters an online monitoring state. To simulate the challenging early-stage disturbances common in industrial practice—specifically, the dispersed flow field disturbances caused by trace amounts of high melt viscosity microgel particles in the raw material—we pulsedly injected a very small amount of moderately pre-crosslinked polyethylene gel particle suspension into the main flow channel at specific times via an upstream micro-injection system. The injection volume was precisely controlled to achieve a local gel particle concentration of 50 parts per million within the flow channel. This disturbance was extremely weak, causing almost no visible fluctuations on the macroscopic pressure gauge.

[0046] In an example of this invention, when the contaminated melt agglomerate flows past the sensor mounting base, the system captures the following event details: First, the first pressure signal shows almost no perceptible change. Second, the second high-frequency response pressure signal, in its high-pass filtered component, clearly exhibits a cluster of sudden high-frequency pressure transients with an amplitude approximately three times the background noise floor, whose energy is mainly concentrated in the characteristic frequency band of 20 kHz to 80 kHz. Simultaneously, the complex dielectric spectrum sequence captured by the broadband dielectric spectrum sensor shows a small, average shift of approximately 0.5% in both its real and imaginary parts relative to the baseline within the frequency range of 1 MHz to 5 MHz, while the loss tangent increases by approximately 1.2%. If this 0.5% shift in the dielectric spectrum is examined in isolation, it is hidden within the background noise of approximately 0.3% due to normal process fluctuations and is almost impossible to effectively identify by any fixed threshold method. However, in the joint determination step of the method of this invention, due to the significant enhancement of the high-frequency pressure signal, the dynamic weighting mechanism instantaneously amplifies the weight of the dielectric spectrum deviation in that frequency band. The calculated weighted Mahalanobis distance reached 6.2 times the baseline root mean square value, far exceeding the preset adaptive threshold of 4 times. Furthermore, the distance value remained above the persistence threshold for the next five consecutive evaluation periods, thus correctly triggering the system's warning signal for material disturbance. The total time from the arrival of the disturbed melt at the sensor to the system issuing the warning was less than one second.

[0047] As a comparative example, we used only the root mean square (RMS) value of the high-frequency signal from a type II high-frequency response pressure sensor for threshold judgment, simulating a monitoring logic based on a single high-frequency mechanical feature. The alarm threshold was set to five times the RMS value of the high-frequency pressure fluctuation component. In multiple experiments injecting the same level of gel particles, similar high-frequency mechanical artifacts were periodically generated due to electromagnetic interference in the industrial environment and the inherent mechanical vibration of the extruder. To avoid false alarms, the threshold had to be increased. When the threshold was set at five times, the RMS value of the pressure transient caused by a weak disturbance at a concentration of 50 parts per million was only about three times the background noise level, which was completely insufficient to trigger an alarm. Only when we increased the injection concentration to 200 parts per million, making the resulting mechanical disturbance strong enough to penetrate the high threshold, did the system issue an alarm. However, at this point, tiny fisheye defects visible to the naked eye could be detected on the film.

[0048] Table 1 below quantifies the key performance differences between this example and the comparative example.

[0049] Table 1 Comparison of Differences As can be seen from the data in Table 1 above, the present invention successfully reduces the lower limit of early microscopic material disturbance identification, and, while ensuring zero false alarms, shifts the early warning time window from after defect formation to the nascent stage when the material microstructure begins to change.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0051] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for identifying material disturbances during the preparation of PE film, characterized in that, include: S1. Steps for establishing a reference signal array: In the melt flow channel of the extrusion system, a multimodal sensor matrix is ​​constructed, consisting of at least one type I pressure sensor, at least one type II high-frequency response pressure sensor, and a wideband dielectric spectrum sensor; the type I pressure sensor, the type II high-frequency response pressure sensor, and the wideband dielectric spectrum sensor are integrated in a common sensor mounting base, the sensor mounting base having a melt contact surface flush with the inner wall of the melt flow channel, and the physical center point of the three on the melt contact surface is located within a virtual circle with a diameter not greater than a predetermined threshold, the predetermined threshold being not greater than half of the hydraulic radius of the melt flow channel; S2. Establish a disturbance event marker reference signal: During the preset baseline process window period in which no material disturbance occurs, continuously acquire the first pressure signal, the second pressure signal, and the complex dielectric spectrum sequence, and process the complex dielectric spectrum sequence to construct a multidimensional baseline feature vector; S3. Online real-time disturbance identification: After the baseline process window period ends, a continuous online monitoring state is entered. In each discrete evaluation cycle, the first pressure signal, the second pressure signal, and the complex dielectric spectrum sequence are synchronously acquired and processed to construct a real-time multidimensional feature vector. The structural dimension of the real-time multidimensional feature vector is consistent with the multidimensional baseline feature vector. S4. Material Disturbance Determination: The real-time multidimensional feature vector is compared with the multidimensional baseline feature vector in a multidimensional feature space. When the comparison result meets the preset joint determination criterion, a material disturbance event is determined to have occurred in the current evaluation period. The joint determination criterion is based at least on a weighted Mahalanobis distance between the real-time multidimensional feature vector and the multidimensional baseline feature vector in the multidimensional feature space. In the calculation of the weighted Mahalanobis distance, the weighting coefficient determined by the high-frequency fluctuation energy spectral density of the second pressure signal in the evaluation period is used to weight and modulate the deviation of the corresponding dimension of the complex dielectric spectrum sequence.

2. The method according to claim 1, characterized in that, In S1, the baseline process window period is determined based on the output signals of the first type of pressure sensor and the second type of high-frequency response pressure sensor after the extrusion system is started: When the fluctuation amplitude of the first pressure signal is less than a predetermined first amplitude threshold for a continuous first preset duration, and at the same time, the root mean square value of the signal component of the second pressure signal after high-pass filtering is less than a predetermined second amplitude threshold within any continuous second preset duration, the continuous period that meets the conditions is automatically defined as the baseline process window period, and the establishment of the disturbance event marker reference signal is triggered.

3. The method according to claim 1, characterized in that, In S2, the processing of the complex dielectric spectrum sequence includes: based on the melt temperature measured in real time by a temperature sensor integrated in the sensor mounting base, performing a predetermined temperature compensation function on the real and imaginary parts of the acquired complex dielectric spectrum sequence to eliminate the influence of the small, slow drift of the melt body temperature on the dielectric spectrum.

4. The method according to claim 1, characterized in that, The broadband dielectric spectrum sensor is composed of an interdigital electrode array, which is directly fabricated on an insulating ceramic substrate. The insulating ceramic substrate forms part of the melt contact surface of the sensor mounting base. The finger width and finger spacing of the interdigital electrodes are set within a predetermined range of 10 micrometers to 100 micrometers, and the electrode material is a platinum-rhodium alloy. When the wideband dielectric spectrum sensor sweeps the frequency, the frequency range of its excitation electric field extends from 100 kHz to 10 MHz. Within this frequency range, no less than 20 discrete frequency points are selected in a logarithmically uniform distribution for point-by-point excitation and response measurement to obtain the complex dielectric spectrum sequence.

5. The method according to claim 1, characterized in that, In S3, constructing the multidimensional baseline feature vector includes an independent preprocessing procedure for the second pressure signal: The second pressure signal is passed through a high-pass filter whose cutoff frequency is located between the first and second frequency response bandwidths to obtain a high-frequency pressure fluctuation component. Within a time window of equal length to the evaluation period, the energy spectral density distribution of the high-frequency pressure fluctuation component is calculated, and its spectral energy values ​​in several specified characteristic frequency bands are extracted. These spectral energy values ​​are used to construct the weighting coefficients in the weighted Mahalanobis distance.

6. The method according to claim 1, characterized in that, The joint determination criterion is a two-layer determination criterion: The first layer is a threshold judgment layer, which requires that the weighted Mahalanobis distance must exceed a predetermined adaptive threshold determined by statistical analysis within the baseline process window. The adaptive threshold is set based on the statistical distribution characteristics of the multidimensional baseline feature vector calculated within the baseline process window, and is specifically set as a predetermined multiple of the root mean square value of the weighted Mahalanobis distance after the multidimensional baseline feature vector is centered. The second layer is a persistence verification layer, which requires that after the conditions of the first layer are met, the value of the weighted Mahalanobis distance must remain above a predetermined persistence threshold for the next N consecutive evaluation periods. The persistence threshold is lower than the adaptive threshold, and N is a pre-set positive integer.

7. The method according to claim 1, characterized in that, The second type of high-frequency response pressure sensor specifically adopts a piezoresistive pressure sensor based on microelectromechanical systems technology. Its pressure-sensing diaphragm thickness is less than 5 micrometers and its inherent frequency is higher than 1 MHz. It is used to achieve non-distortion capture of flow noise and local pressure transients in the melt caused by microgel particles or local viscosity changes, with a frequency range of 10 kHz to hundreds of kHz. The first type of pressure sensor adopts a strain gauge pressure sensor.

8. The method according to claim 1, characterized in that, The sensor mounting base is a modular component, which is sealed and embedded in the straight pipe section of the flow channel between the extruder die and the melt filter through a flange interface; the signal cables of all sensors are led out from the inside of the sensor mounting base through a sealed through-piece and connected to the high-speed synchronous data acquisition and processing unit.

Images