A method for dynamic adjustment of thin film tension based on multi-sensor data fusion

Abstract

This invention relates to the field of automatic control, and more particularly to a method for dynamic adjustment of film tension based on multi-sensor data fusion. The method includes: setting initial Kalman filter parameters using equipment rated parameters and acquiring real-time data from sensors in the production line; analyzing the energy characteristics and rate of change of the film tension residual data to obtain a film tension abnormality impact intensity factor; evaluating the temporal correlation of the film tension residual data to obtain a film tension abnormality event confirmation factor; evaluating the adaptive measurement noise covariance using the film tension abnormality impact intensity factor and the film tension abnormality event confirmation factor, and updating the film tension estimate based on the adaptive measurement noise covariance; and obtaining the film winding / unwinding adjustment amount through closed-loop feedback of the optimal film tension estimate, thereby improving the accuracy and stability of the control system.

Description

Technical Field

[0001] This invention relates to the field of automatic control, and in particular to a method for dynamic adjustment of thin film tension based on multi-sensor data fusion. Background Technology

[0002] In modern industrial manufacturing, thin-film materials are widely used as functional carriers in high-precision fields such as flexible displays, lithium-ion battery motors, high-end optical components, and integrated circuit packaging. In the roll-to-roll production of these thin-film materials, processes such as coating, sputtering, etching, or lamination require precise and stable control of the tension of the thin-film substrate, which is one of the core technologies for ensuring product yield and performance. Excessive fluctuations or uneven distribution of tension can directly lead to permanent tensile deformation, wrinkles, and fractures in the thin film, or cause fatal defects such as uneven thickness and cracking of the functional coating.

[0003] To achieve high-precision tension control, existing technologies typically employ closed-loop feedback control systems. These systems directly measure the real-time tension of the film using tension sensors installed at key locations on the production line, while simultaneously using encoders mounted on the take-up and unwind motors to acquire indirect information such as linear speed and roll diameter. This multi-sensor data from different physical sources provides a comprehensive understanding of the system's state. To fuse this noisy and diverse data and extract the tension value closest to the true physical state, advanced control systems commonly employ the Kalman filter algorithm. As an optimal linear state estimation algorithm, the Kalman filter effectively filters out Gaussian white noise commonly found in sensor measurements, providing a smooth and accurate tension estimate to the backend PID (Proportional-Integral-Derivative) controller. The PID controller then adjusts the torque or speed of the take-up and unwind motors, thus forming a complete high-performance tension control closed loop. This method performs exceptionally well in handling random noise under stable operating conditions and is currently an important technical means to improve the accuracy of film tension control.

[0004] However, the optimal performance of the standard Kalman filter algorithm used in existing technologies relies strictly on a core theoretical premise: both the process noise and measurement noise of the system must follow a Gaussian distribution, and their statistical characteristics must be known and fixed. In practical applications, especially in precision manufacturing scenarios such as high-speed sputtering coating of high-end optical thin films, there exists a typical non-Gaussian disturbance source that severely impacts product quality. Specifically, minute physical defects, such as crystal points, gel points, or localized abrupt changes in thickness, inevitably occur in the thin film substrate during manufacturing. When these micro-defects pass through a highly sensitive tension detection roller at speeds on the order of meters per second, they instantaneously generate a violent pulse-like impact on the tension sensor with an amplitude far exceeding the normal noise level and a very short duration. This signal is statistically a typical non-Gaussian outlier, violating the theoretical assumptions of the standard Kalman filter. If the filter's measurement noise parameter is set too high to suppress the pulse impact, thus reducing the confidence level in the measured value, the system's response to real, minute tension changes will become sluggish under normal operating conditions, sacrificing the system's dynamic performance. Conversely, if the measured value is trusted to ensure dynamic response, the filter will incorrectly identify the spurious pulse signal as a real, dramatic tension change and quickly transmit this erroneous estimate to the PID controller. Upon receiving this erroneous spike signal, the controller will immediately execute an incorrect compensation action, which itself will impose a destructive secondary tension fluctuation on the film, ultimately causing wrinkles to form near the defect point. Therefore, how to identify and effectively suppress this sudden non-Gaussian pulse measurement noise caused by micro-defects in the substrate without sacrificing the system's normal dynamic performance, and avoid the control system being misled by it to produce destructive adjustment actions, is a pressing technical problem that needs to be solved in the current field. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method for dynamic adjustment of thin film tension based on multi-sensor data fusion, in order to solve the problem of wrinkles in the thin film near the defect point caused by sudden non-Gaussian pulse measurement noise caused by micro-defects in the substrate.

[0006] This invention provides a method for dynamic adjustment of thin film tension based on multi-sensor data fusion, the method comprising the following steps:

[0007] Step S1: Set the initial Kalman filter parameters using the equipment's rated parameters and collect real-time data using sensors in the production line;

[0008] Step S2: By analyzing the energy characteristics and rate of change of the film tension residual data, the abnormal impact intensity factor of film tension is obtained;

[0009] Step S3: Obtain the confirmation factor for abnormal film tension events by performing a time-series correlation assessment on the film tension residual data;

[0010] Step S4: Perform adaptive measurement noise covariance evaluation using the film tension abnormality impact intensity factor and the film tension abnormality event confirmation factor, and update the film tension estimate based on the adaptive measurement noise covariance.

[0011] Step S5: Obtain the film winding and unwinding adjustment amount through closed-loop feedback of the optimal film tension estimate.

[0012] Preferably, the step of setting the initial Kalman filter parameters through the equipment's rated parameters and acquiring real-time data through sensors in the production line includes:

[0013] Initialize the Kalman filter basic model, define the state vector in the Kalman filter basic model as the core component of the film tension, determine the state transition matrix and control input matrix describing the dynamic behavior of the system, and set the basic measurement noise covariance and process noise covariance characterizing the noise level of the system under normal operating conditions.

[0014] The sensor sampling frequency is set, and in each control cycle, the actual measured value of the film tension is collected in real time by the sensors deployed on the production line. The control input command applied to the take-up and unwinding servo motor is obtained through the central motion controller.

[0015] Preferably, the step of obtaining the abnormal impact intensity factor of film tension by analyzing the energy characteristics and rate of change of the film tension residual data includes:

[0016] The impact factor of the film tension is obtained by performing micro-defect impact characteristic analysis on the residual data of the film tension; the steepness factor of the film tension is obtained by performing impact rate analysis on the residual data of the film tension; and the abnormal impact intensity factor of the film tension is obtained by fusing and mapping the impact factor and the steepness factor of the film tension.

[0017] Preferably, the step of obtaining the impact factor of the film tension by performing micro-defect impact characteristic analysis on the residual data of the film tension, and obtaining the steepness factor of the film tension by performing impact rate analysis on the residual data of the film tension, includes:

[0018] The residual data of the film tension at the target time is calculated by combining the actual measured value of the film tension at the target time with the state prediction value obtained by the Kalman filter prediction step at the target time; the residual covariance data in the Kalman filter process at the target time is obtained; the square of the film tension residual data is used as the numerator, the residual covariance data is used as the denominator, and the resulting fraction is used as the impact factor of the film tension.

[0019] The square of the result of subtracting the film tension residual data at the target time from the film tension residual data at the time before the target time is used as the numerator, and the result of multiplying the constant 2 with the residual covariance data is used as the denominator. The resulting fraction is used as the steepness factor of the film tension.

[0020] Preferably, the step of obtaining the abnormal impact intensity factor of film tension by fusing and mapping the impact factor and the steepness factor of film tension includes:

[0021] The result of adding the square of the impact factor of the film tension to the square of the steepness factor of the film tension is used as the first impact intensity assessment; the square root of the first impact intensity assessment is mapped by a power function with the natural constant as the base, and the result of subtracting the constant 1 from the obtained mapping result is used as the abnormal impact intensity factor of the film tension at the target time.

[0022] Preferably, the step of obtaining the confirmation factor for abnormal film tension events by performing a time-series correlation assessment on the film tension residual data includes:

[0023] By performing time-series correlation evaluation on the residual data of film tension, the cumulative factor of residual time-series correlation of film tension is obtained; by performing gating mapping processing on the cumulative factor of residual time-series correlation of film tension, the confirmation factor of abnormal events of film tension is obtained.

[0024] Preferably, the step of obtaining the cumulative factor of residual temporal correlation of film tension by performing time-series correlation evaluation on the film tension residual data includes:

[0025] Set a forgetting factor for the exponential moving average; multiply the film tension residual data at the target time with the film tension residual data at the time before the target time, and divide by the calculated result of the residual covariance data at the target time as the time series correlation assessment of film tension at the target time; use the film tension time series correlation assessment of the first time in the time series data formed by the residual data of film tension as the residual time series correlation accumulation factor of the first time.

[0026] The forgetting factor is used as the weight for the time-series correlation assessment of film tension at the target time. The constant 1 is subtracted from the forgetting factor as the weight for the cumulative factor of residual time-series correlation of the time before the target time. The result of the weighted sum of the time-series correlation assessment of film tension at the target time and the cumulative factor of residual time-series correlation of the time before the target time is used as the cumulative factor of residual time-series correlation of the target time.

[0027] Preferably, the step of obtaining the film tension anomaly event confirmation factor by gating and mapping the residual temporal correlation accumulation factor of film tension includes:

[0028] Set the cumulative activation threshold for relevance and the sigmoid function transformation scaling factor to determine the authenticity of an event;

[0029] The result of subtracting the cumulative correlation factor of the residual time series at the target time from the cumulative correlation activation threshold and dividing by the calculated result of the sigmoid function transformation scaling factor is used as the abnormal event evaluation factor at the target time.

[0030] The abnormal event evaluation factor at the target time is mapped using the tanh function to obtain the first mapping result. The first mapping result is added to a constant 1 and then multiplied by half to obtain the calculated result as the film tension abnormal event confirmation factor at the target time.

[0031] Preferably, the step of evaluating the adaptive measurement noise covariance using the abnormal film tension impact intensity factor and the abnormal film tension event confirmation factor, and updating the film tension estimate based on the adaptive measurement noise covariance, includes:

[0032] The result of multiplying the abnormal impact intensity factor of the film tension at the target time and the confirmation factor of the abnormal film tension event at the target time and adding it to a constant 1 is used as a weight to weight the basic measurement noise covariance. The corresponding weighted basic measurement noise covariance is then used as the adaptive measurement noise covariance.

[0033] The adaptive measurement noise covariance replaces the basic measurement noise covariance in the Kalman filter, and the Kalman gain is calculated and the optimal estimate of the system state is updated according to the Kalman filter process, as well as the error covariance is updated, thereby updating the film tension estimate.

[0034] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0035] This invention constructs a thin film tension anomaly impact intensity factor based on the residual of thin film tension measurement. By fusing the instantaneous energy and rate of change characteristics of the residual, it effectively quantifies the instantaneous anomaly intensity caused by physical disturbances such as micro-defects. This is significantly more sensitive and has stronger anti-interference capabilities than the single judgment method of residual amplitude in existing methods. Furthermore, this invention constructs a thin film tension anomaly event confirmation factor. By calculating the cumulative temporal correlation between residuals at consecutive time points, it dynamically judges whether the abnormal signal has physical temporal continuity. This mechanism can effectively suppress isolated spike misjudgments caused by occasional events such as electromagnetic interference and servo device switching, significantly reducing the false trigger rate and improving the accuracy of the filter's response to real events. Further, this invention controls the release degree of the impact intensity factor through the event confirmation factor, jointly generates an adaptive measurement noise covariance, and feeds it back to the Kalman filter in real time to dynamically adjust the confidence level of the measured value. This dual-factor gating mechanism enables the system to maintain high dynamic response characteristics under normal operating conditions and has the ability to quickly suppress abnormal interference, achieving a compatibility improvement in response speed and robustness. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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.

[0037] Figure 1 This is a flowchart of a method for dynamically adjusting thin film tension based on multi-sensor data fusion, provided in Embodiment 1 of the present invention. Detailed Implementation

[0038] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0039] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0040] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0041] See Figure 1 This is a flowchart of a method for dynamically adjusting thin film tension based on multi-sensor data fusion, as provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include:

[0042] Step S1: Initial Kalman filter parameters are set using the equipment's rated parameters, and real-time data is collected using sensors in the production line.

[0043] This invention is an optimization method implemented within a standard Kalman filter system framework for thin-film tension control. First, a basic Kalman filter model needs to be established and initialized based on the specific characteristics of the production line. The establishment of this model is a conventional technique in the field. Specifically, it includes: defining the state vector, with thin-film tension as the core component in the basic Kalman filter model; determining the state transition matrix and control input matrix describing the system's dynamic behavior; and setting the basic measurement noise covariance and process noise covariance characterizing the system's noise level under normal operating conditions. These matrices and parameters are set according to the technical manual provided by the equipment manufacturer. Simultaneously, the initial values ​​of the state vector and the initial error covariance are set. The initial values ​​of the state vector are set to the initial physical quantities measured by calibration equipment at the start of production. The initial error covariance, reflecting the assessment of the reliability of the initial state values, is set as a diagonal matrix with large diagonal elements, indicating that the initial state has high uncertainty and ensuring that the filter can converge quickly through subsequent measurements. In this embodiment, the initial error covariance is set as a diagonal matrix with 100 diagonal elements and units of N / m (corresponding to...). (The variance estimate), the diagonal elements can be adjusted according to the actual scenario, and are not required.

[0044] After the basic model is established and initialized, the system enters real-time operation. Subsequent steps of this invention are executed within discrete control cycles, defined by the system's sampling frequency. In this embodiment, the sampling frequency is set to 1000Hz. In each control cycle, the Kalman filter system collects and digitizes raw data in real time using sensors deployed on the production line for subsequent use. The collected data includes: the actual measured value of the film tension at the current moment, obtained through a high-sensitivity tension detection roller based on strain gauge technology installed at key locations on the production line; and the control input command applied to the unwinding and take-up servo motors at the current moment, obtained through the central motion controller.

[0045] At this point, the initial Kalman filter parameters have been set using the equipment's rated parameters, and real-time data acquisition has been completed using sensors in the production line.

[0046] Step S2: The impact intensity factor is obtained by analyzing the energy characteristics and rate of change of the film tension residual data.

[0047] The problem addressed by this invention stems from physical micro-defects present on the thin film substrate. When these micro-defects pass through the tension detection roller, the resulting disturbance in the film tension data collected by the sensor exhibits two physical characteristics distinct from normal Gaussian noise: first, the instantaneous energy carried by this disturbance signal far exceeds the energy statistical range of normal noise; second, the energy of this disturbance signal shows a sudden increase, meaning the rate of energy change is also much greater than the random fluctuations of normal noise. The standard Kalman filter algorithm used in existing technologies can theoretically only handle noise with energy distribution conforming to Gaussian properties. It cannot effectively handle pulse impacts that simultaneously possess both high instantaneous energy and high energy change rate characteristics, and may even be misled by them. Therefore, in order to effectively identify and quantify this specific pulse impact, the first step of this invention is not simply to determine whether the residual film tension, i.e., the difference between the measured film tension value and the predicted film tension value, is too large. Instead, it is necessary to construct a metric that can simultaneously capture the two energy characteristics mentioned above. Only when a signal simultaneously exhibits extremely high instantaneous energy and extremely rapid energy change rate can there be sufficient basis to preliminarily determine that it originates from the physical impact of micro-defects.

[0048] Based on this logic, the present invention first obtains the impact factor of film tension by performing micro-defect impact characteristic analysis on the residual data of film tension; then, it obtains the steepness factor of film tension by performing impact rate analysis on the residual data of film tension; finally, it obtains the abnormal impact intensity factor of film tension by fusing and mapping the impact factor and the steepness factor of film tension.

[0049] Specifically, firstly, the residual data of the film tension at the target time is calculated by combining the actual measured value of the film tension at the target time with the state prediction value obtained through the Kalman filter prediction step; the residual covariance data during the Kalman filter process at the target time is obtained; the square of the film tension residual data is used as the numerator, the residual covariance data as the denominator, and the resulting fraction is used as the impact factor of the film tension; then, the square of the result of subtracting the film tension residual data at the target time from the film tension residual data at the previous time is used as the numerator, and the result of multiplying the constant 2 with the residual covariance data is used as the denominator, and the resulting fraction is used as the steepness factor of the film tension. Finally, the result of adding the square of the impact factor of the film tension to the square of the steepness factor of the film tension is used as the first impact intensity assessment; the square root of the first impact intensity assessment is mapped using a power function with the natural constant as the base, and the result of subtracting the corresponding mapping result from the constant 1 is used as the abnormal impact intensity factor of the film tension at the target time.

[0050] In one implementation, it is assumed that in the first... The film tension residual data at each time point are: ; in the The covariance of the film tension residuals at each time step is Then in the first The formula for calculating the abnormal impact intensity factor of membrane tension at a given moment is:

[0051]

[0052] in, Indicates the first Abnormal impact intensity factor of membrane tension at a given moment; Denotes the natural constant e; Indicates the first Thin film tension residual data at each moment; Indicates the first Covariance of film tension residuals at each moment; Indicates the first Thin film tension residual data at each time point.

[0053] It should be noted that in the formula for calculating the abnormal impact intensity factor of membrane tension, The impact factor, representing the film tension, is expressed as the instantaneous energy of the film tension measurement data represented by the square of the residual, and is then compared with the expected variance. By making comparisons, the standardization of residual energy was achieved. This term directly responds to the high-energy characteristics of micro-defect impact. When the residual energy far exceeds its normal expectation, this term will increase significantly, thereby capturing the amplitude information of the impact. The steepness factor, representing the film tension, is used to capture the suddenness of impacts. It is calculated by squared the difference between the residuals at two consecutive moments, representing the rate of change energy of the film tension measurement data, and is similarly expressed through the expected variance. Standardization is performed to quantify the steepness of thin film tension measurement data. When the measurement data undergoes a sharp jump, the steepness factor increases dramatically, thus capturing the impact rate information. Finally, [the following is used]... The norm form integrates the impact factor and steepness factor of film tension, and uses this comprehensive metric as a power of the natural exponent. Subtracting a constant 1 gives the abnormal impact intensity factor of film tension. When the system is under normal operating conditions and the value of the comprehensive metric is close to 0, the value of the abnormal impact intensity factor of film tension is also close to 0 and has no effect on the system. However, once the impact of micro-defects causes the comprehensive metric to increase, the amplification effect of the exponential function will cause the value of the abnormal impact intensity factor of film tension to grow exponentially, generating a sufficiently strong adjustment signal. The final value of the abnormal impact intensity factor of film tension then becomes a non-negative potential adjustment range that can quantify the severity of micro-defect impacts.

[0054] Thus, by analyzing the energy characteristics and rate of change of the film tension residual data, the impact intensity factor was obtained.

[0055] Step S3: Obtain the confirmation factor for abnormal film tension events by performing a time-series correlation assessment on the film tension residual data.

[0056] In step S2, the impact intensity factor has been obtained. However, this factor can only assess the severity of the impact from a single time slice. It cannot distinguish the source of the disturbance. In complex industrial environments, besides real pulse impacts caused by micro-defects in the thin film, there may also be isolated or transient electromagnetic interference signals caused by equipment start-up and shutdown, or power grid fluctuations. If all severe impacts are suppressed indiscriminately, erroneous over-adjustments to these non-physical pseudo-signals will occur. A real impact originating from a physical micro-defect will necessarily exhibit temporal continuity in the thin film tension measurement data; that is, the thin film tension residual data will show significant temporal correlation within a short period before and after the impact. In contrast, the signal values ​​before and after an isolated, random electromagnetic interference spike are statistically independent and do not possess this correlation. Therefore, after initially detecting a severe impact, this invention introduces a confirmation mechanism. By analyzing the intrinsic temporal correlation of the thin film tension residual data, it determines whether the impact is a physically continuous real event. Only after the authenticity of the impact event is confirmed should the system authorize the execution of the corresponding suppression adjustment. Therefore, a confirmation mechanism is needed to generate a gating signal that determines whether the adjustment action should be executed by quantifying the temporal correlation of the film tension residual data. In this step, firstly, the residual temporal correlation accumulation factor of the film tension is obtained by performing a temporal correlation evaluation on the film tension residual data; then, the film tension abnormal event confirmation factor is obtained by performing gating mapping processing on the residual temporal correlation accumulation factor of the film tension.

[0057] Specifically, firstly, a forgetting factor is set for the exponential moving average; the film tension residual data at the target time is multiplied by the film tension residual data at the time before the target time, and the result is divided by the residual covariance data at the target time as the film tension time-series correlation assessment at the target time; the film tension time-series correlation assessment at the first time in the time-series data formed by the film tension residual data is used as the residual time-series correlation accumulation factor at the first time; the forgetting factor is used as the weight of the film tension time-series correlation assessment at the target time, and the constant 1 is subtracted from the forgetting factor as the weight of the residual time-series correlation accumulation factor at the time before the target time; the weighted sum of the film tension time-series correlation assessment at the target time and the residual time-series correlation accumulation factor at the time before the target time is used as the residual time-series correlation accumulation factor at the target time.

[0058] In one embodiment, the first The formula for calculating the cumulative factor of residual temporal correlation at each time point is:

[0059]

[0060] in, Indicates the first Cumulative factor of residual temporal correlation at each time point; Indicates the first Cumulative factor of residual temporal correlation at each time point; This represents the forgetting factor used for exponential moving average. In this embodiment, in order to effectively capture short-term pulse events lasting for several sampling periods, the value of the forgetting factor is set to 0.1. This forgetting factor can be adjusted according to the actual scenario and is not required. Indicates the first Thin film tension residual data at each moment; Indicates the first Thin film tension residual data at each moment; Indicates the first Covariance of film tension residuals at each time step.

[0061] It should be noted that, in order to quantify the temporal correlation of thin film tension residual data, this invention introduces a cumulative residual temporal correlation. The core of its calculation formula is That is, the product of the residual at the current moment and the residual at the previous moment. For a real pulse impact caused by a physical micro-defect, the sign of the residual has a high probability of remaining consistent during the occurrence and duration of the impact, which will lead to... The residual time-series correlation cumulative factor remains consistently positive, causing it to steadily accumulate and increase through an exponential moving average. For normal Gaussian white noise, the residual values ​​at different times are statistically uncorrelated. The signs of the thin film tension residual data will alternate between positive and negative, making... The expected value is 0, and the residual time-series correlation cumulative factor will fluctuate slightly around the value of 0. Divide by residual covariance Normalization can eliminate the influence of the film tension measurement data itself, making it a dimensionless statistical correlation measure. This makes the value of the residual time-series correlation cumulative factor an effective indicator for measuring whether there has been a sustained physical impact in the near future.

[0062] After obtaining the residual temporal correlation accumulation factor at the target time, a correlation accumulation activation threshold and a sigmoid function transformation scaling factor are set to determine the authenticity of the event. The result of subtracting the residual temporal correlation accumulation factor at the target time from the correlation accumulation activation threshold and dividing by the sigmoid function transformation scaling factor is used as the abnormal event evaluation factor at the target time. The abnormal event evaluation factor at the target time is mapped using the tanh function to obtain the first mapping result. The first mapping result is added to a constant 1 and multiplied by half to obtain the calculated result as the film tension abnormal event confirmation factor at the target time.

[0063] In one embodiment, the first The formula for calculating the confirmation factor of the abnormal film tension event at a given time is:

[0064]

[0065] in, Indicates the first Confirmation factor for abnormal film tension events at any given moment; Indicates the first Cumulative factor of residual temporal correlation at each time point; This represents the cumulative activation threshold for determining the authenticity of an event; This represents the scaling factor for the transformation of the sigmoid function; This represents the tanh function.

[0066] It should be noted that the determination of the correlation cumulative activation threshold used to determine the authenticity of the event in the formula for calculating the confirmation factor of abnormal thin film tension events follows a standard engineering calibration method. This is achieved through offline statistical analysis of a large amount of historical sensor data collected under normal and stable operating conditions on a specific production line. The basic principle is to maximize the suppression of random noise while ensuring sufficient sensitivity to real physical impact events, thereby achieving an extremely low false trigger rate. In this embodiment, the application scenario is a magnetron sputtering production line for optical thin films processing PET substrates, with a film running line speed of 1.5 m / s and a tension sensor sampling frequency of 1000 Hz. Under this specific operating condition, the threshold is determined using the aforementioned statistical calibration method. This threshold is set under normal operating conditions. The mean of the value distribution is added with a safety margin sufficient to cover more than 99.9% of normal fluctuations. In this embodiment, the specific value is set as follows: This value can be adjusted according to different working conditions and is not required. Regarding the S-shaped function transformation scaling factor... This value controls the smoothness of the gating signal transition from 0 to 1. This value determines the decisiveness of the confirmation process. In this embodiment of the invention, to achieve a relatively fast and clear gating switch, the specific value is set to... For the obtained residual time-series correlation accumulation factor, the role of the thin film tension anomaly confirmation factor is to transform the residual time-series correlation accumulation metric into a standardized gating signal. This invention uses the hyperbolic tangent function tanh to construct an S-shaped transformation function. Much smaller than the preset threshold When the system is in normal operating condition or encounters an isolated noise point, the input to the tanh function is a large negative number, and its output value approaches negative one, ultimately causing the value of the film tension anomaly event confirmation factor to approach 0, representing gating closed; when a real pulse impact causes Continuous accumulation and significantly exceeding the threshold At this point, the input to the tanh function becomes a large positive number, and its output value approaches 1, ultimately causing the value of the film tension anomaly event confirmation factor to approach 1, representing that the gating is fully open. This is achieved by embedding the tanh function into... In the structure, change its output range from Mapped to This range allows the final value of the film tension event confirmation factor to become a gated signal between 0 and 1, which can accurately quantify the confidence level of the authenticity of the impact event.

[0067] Step S4: Perform adaptive measurement noise covariance evaluation using the film tension abnormality impact intensity factor and the film tension abnormality event confirmation factor, and update the film tension estimate based on the adaptive measurement noise covariance.

[0068] After obtaining the abnormal impact intensity factor and the abnormal event confirmation factor of the film tension at the target time, this step continues to construct an adaptive measurement noise covariance and uses this adaptive measurement noise covariance to complete the update process of the standard Kalman filter. Specifically, the calculation result of multiplying the abnormal impact intensity factor and the abnormal event confirmation factor of the film tension at the target time and adding it to the constant 1 is used as a weight to weight the basic measurement noise covariance, and the calculation result of the corresponding weighted basic measurement noise covariance is used as the adaptive measurement noise covariance.

[0069] It should be noted that the gating modulation mechanism is implemented by determining the extent to which the adjustment range calculated by the abnormal film tension impact intensity factor can be actually executed by the film tension abnormal event confirmation factor. Only when the film tension abnormal event confirmation factor confirms the authenticity of the event (the value of the film tension abnormal event confirmation factor approaches 1) is the adjustment effect of the film tension abnormal impact intensity factor fully released. Conversely, if the film tension abnormal event confirmation factor determines that the current impact is isolated noise (the value of the film tension abnormal event confirmation factor approaches 0), the adjustment effect of the film tension abnormal impact intensity factor will be effectively suppressed, thereby avoiding unnecessary disturbances to the system.

[0070] After obtaining the adaptive measurement noise covariance, the adaptive measurement noise covariance is used to replace the basic measurement noise covariance in the Kalman filter. The Kalman gain is calculated according to the Kalman filter process, and the optimal estimate of the system state is updated, as well as the error covariance is updated, thereby updating the film tension estimate.

[0071] Step S5: Obtain the film winding and unwinding adjustment amount through closed-loop feedback of the optimal film tension estimate.

[0072] In step S2, the proposed dual-factor gating control mechanism obtains a tension estimate that accurately reflects the true physical state of the thin film. Non-Gaussian impulse noise interference caused by micro-defects is effectively suppressed in this tension estimate. The purpose of this step is to utilize the optimized estimate to achieve closed-loop feedback control of the thin film tension, thereby ultimately solving the technical problem mentioned in the background art. The system sends the tension estimate output from step S2 as a closed-loop feedback quantity to the PID controller. The PID controller compares this feedback quantity with a preset tension target value and calculates the deviation between the two. Subsequently, the PID controller calculates the control adjustment quantity based on the internally set proportional, integral, and derivative parameters for this deviation. This control adjustment quantity is then encapsulated into a digitized target torque command and sent to the servo driver controlling the take-up and unwinding motors. Upon receiving the command, the servo driver controls and adjusts the output torque of the servo motor in real time according to the target torque command. Because the tension estimate input to the PID controller reflects the true physical state, avoiding the misleading effect of erroneous spike signals, the adjustment output by the PID controller will no longer produce destructive compensation actions. This ensures that the actual tension of the film is stably maintained near the target value, guaranteeing uniform and stable tension throughout the entire production process, even in the presence of micro-defects in the film substrate.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for dynamic adjustment of thin film tension based on multi-sensor data fusion, characterized in that, The method for dynamic adjustment of thin film tension based on multi-sensor data fusion includes: Step S1: Set the initial Kalman filter parameters using the equipment's rated parameters and collect real-time data using sensors in the production line; Step S2: By analyzing the energy characteristics and rate of change of the film tension residual data, the abnormal impact intensity factor of film tension is obtained; Step S3: Obtain the confirmation factor for abnormal film tension events by performing a time-series correlation assessment on the film tension residual data; Step S4: Perform adaptive measurement noise covariance evaluation using the film tension abnormality impact intensity factor and the film tension abnormality event confirmation factor, and update the film tension estimate based on the adaptive measurement noise covariance. Step S5: Obtain the film winding and unwinding adjustment amount through closed-loop feedback of the optimal film tension estimate; The method of obtaining the abnormal impact intensity factor of film tension by analyzing the energy characteristics and rate of change of film tension residual data includes: obtaining the impact factor of film tension by performing micro-defect impact characteristic analysis on film tension residual data; obtaining the steepness factor of film tension by performing impact rate analysis on film tension residual data; and obtaining the abnormal impact intensity factor of film tension by fusing and mapping the impact factor and the steepness factor of film tension. The step of obtaining a confirmation factor for abnormal film tension events by evaluating the temporal correlation of film tension residual data includes: obtaining a cumulative factor for the temporal correlation of film tension residuals by evaluating the temporal correlation of film tension residuals; and obtaining a confirmation factor for abnormal film tension events by performing gating mapping processing on the cumulative factor for the temporal correlation of film tension residuals. The adaptive measurement noise covariance evaluation based on the abnormal impact intensity factor and the abnormal event confirmation factor of the film tension, and the updating of the film tension estimate based on the adaptive measurement noise covariance, includes: multiplying the abnormal impact intensity factor of the film tension at the target time by the abnormal event confirmation factor of the film tension at the target time and adding it to a constant 1 as a weight to weight the basic measurement noise covariance; using the calculated result of the corresponding weighted basic measurement noise covariance as the adaptive measurement noise covariance; replacing the basic measurement noise covariance in the Kalman filter with the adaptive measurement noise covariance; and completing the calculation of the Kalman gain and updating the optimal estimate of the system state and the error covariance according to the Kalman filter process, thereby updating the film tension estimate.

2. The method for dynamic adjustment of thin film tension based on multi-sensor data fusion according to claim 1, characterized in that, The process of setting the initial Kalman filter parameters using the equipment's rated parameters and acquiring real-time data from sensors on the production line includes: Initialize the Kalman filter basic model, define the state vector in the Kalman filter basic model as the core component of the film tension, determine the state transition matrix and control input matrix describing the dynamic behavior of the system, and set the basic measurement noise covariance and process noise covariance characterizing the noise level of the system under normal operating conditions. The sensor sampling frequency is set, and in each control cycle, the actual measured value of the film tension is collected in real time by the sensors deployed on the production line. The control input command applied to the take-up and unwinding servo motor is obtained through the central motion controller.

3. The method for dynamic adjustment of thin film tension based on multi-sensor data fusion according to claim 1, characterized in that, The impact factor of the film tension is obtained by performing micro-defect impact characteristic analysis on the residual data of the film tension. By performing impact rate analysis on the residual data of film tension, the steepness factor of film tension is obtained, including: The residual data of the film tension at the target time is calculated by combining the actual measured value of the film tension at the target time with the state prediction value obtained by the Kalman filter prediction step at the target time; the residual covariance data in the Kalman filter process at the target time is obtained; the square of the film tension residual data is used as the numerator, the residual covariance data is used as the denominator, and the resulting fraction is used as the impact factor of the film tension. The square of the result of subtracting the film tension residual data at the target time from the film tension residual data at the time before the target time is used as the numerator, and the result of multiplying the constant 2 with the residual covariance data is used as the denominator. The resulting fraction is used as the steepness factor of the film tension.

4. The method for dynamic adjustment of thin film tension based on multi-sensor data fusion according to claim 1, characterized in that, The method of obtaining the abnormal impact intensity factor of film tension by fusing and mapping the impact factor and steepness factor of film tension includes: The result of adding the square of the impact factor of the film tension to the square of the steepness factor of the film tension is used as the first impact intensity assessment; the square root of the first impact intensity assessment is mapped by a power function with the natural constant as the base, and the result of subtracting the constant 1 from the obtained mapping result is used as the abnormal impact intensity factor of the film tension at the target time.

5. The method for dynamic adjustment of thin film tension based on multi-sensor data fusion according to claim 1, characterized in that, The step of obtaining the cumulative factor of the residual temporal correlation of film tension by performing a time-series correlation evaluation on the film tension residual data includes: Set a forgetting factor for the exponential moving average; multiply the film tension residual data at the target time with the film tension residual data at the time before the target time, and divide by the calculated result of the residual covariance data at the target time as the time series correlation assessment of film tension at the target time; use the film tension time series correlation assessment of the first time in the time series data formed by the residual data of film tension as the residual time series correlation accumulation factor of the first time. The forgetting factor is used as the weight for the time-series correlation assessment of film tension at the target time. The constant 1 is subtracted from the forgetting factor as the weight for the cumulative factor of residual time-series correlation of the time before the target time. The result of the weighted sum of the time-series correlation assessment of film tension at the target time and the cumulative factor of residual time-series correlation of the time before the target time is used as the cumulative factor of residual time-series correlation of the target time.

6. The method for dynamic adjustment of thin film tension based on multi-sensor data fusion according to claim 1, characterized in that, The step of obtaining a film tension anomaly event confirmation factor by gating and mapping the residual temporal correlation accumulation factor of film tension includes: Set the cumulative activation threshold for relevance and the sigmoid function transformation scaling factor to determine the authenticity of an event; The result of subtracting the cumulative correlation factor of the residual time series at the target time from the cumulative correlation activation threshold and dividing by the calculated result of the sigmoid function transformation scaling factor is used as the abnormal event evaluation factor at the target time. The abnormal event evaluation factor at the target time is mapped using the tanh function to obtain the first mapping result. The first mapping result is added to a constant 1 and then multiplied by half to obtain the calculated result as the film tension abnormal event confirmation factor at the target time.

Images