A method and system for failure prediction of an integrated membrane product processing apparatus

Abstract

The application relates to the technical field of fault prediction, and provides a fault prediction method and system for an integrated membrane product processing device.The method comprises the following steps: acquiring a main material mechanism and an auxiliary material mechanism of the integrated membrane product processing device; collecting material parameters corresponding to the main material mechanism and the auxiliary material mechanism respectively; defining a multimodal synchronism feature item; collecting a multimodal synchronism feature data set of an asynchronous transfer process by using the multimodal synchronism feature item; calculating a variation coefficient of the multimodal synchronism feature data set based on a short-term window sliding mean algorithm, and converting the variation coefficient into a fault prediction probability; and triggering a fault signal to remind when the fault prediction probability is greater than a preset threshold value.The application solves the technical problem that it is difficult to accurately predict faults caused by multimodal synchronism disorder in the asynchronous transfer process of the integrated membrane product processing device, and achieves the technical effect of realizing early fault warning, improving equipment operation stability and production efficiency by means of multimodal synchronism feature analysis and variation coefficient calculation.

Description

Technical Field

[0001] This application relates to the field of fault prediction technology, specifically to a fault prediction method and system for integrated membrane product processing equipment. Background Technology

[0002] Traditional production processes typically involve multiple independent steps, such as leveling, asynchronous transfer, and die-cutting, each performed by separate equipment. This results in complex production line layouts, large floor space requirements, low inter-process coordination efficiency, and difficulties in fault diagnosis. Therefore, integrating the main material conveying mechanism, auxiliary material conveying mechanism, and related processing equipment into a single unit significantly improves production efficiency and space utilization. However, while this integrated equipment enhances efficiency, the coupled operation of multiple mechanisms presents new challenges: during asynchronous transfer, the main material conveying mechanism and auxiliary material transfer mechanism are prone to complex fault modes such as misalignment, abnormal tension fluctuations, and inconsistent step distances due to differences in material parameters (e.g., differences in tension response characteristics caused by different material types, and bonding pressure fluctuations caused by thickness variations) and control delays. Traditional fault diagnosis methods based on single-parameter threshold monitoring cannot capture the dynamic coupling relationships between multi-modal parameters, leading to delayed fault warnings or high false alarm rates. Therefore, there is an urgent need to develop an intelligent fault prediction method and system for integrated film product processing equipment to proactively warn of faults before they occur, ensuring the stable operation of integrated production equipment. Summary of the Invention

[0003] This application provides a fault prediction method and system for integrated film product processing equipment, aiming to solve the technical problem that it is difficult to accurately predict faults caused by multimodal synchronization mismatch during asynchronous transfer of integrated film product processing equipment.

[0004] The first aspect disclosed in this application provides a fault prediction method for an integrated membrane product processing equipment. The method includes: acquiring the main material mechanism and auxiliary material mechanism of the integrated membrane product processing equipment; collecting the main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism, respectively, whereby the material parameters include material type, thickness, tension reference value, and bonding fabric distance; defining multimodal synchronization feature terms for the asynchronous transfer process according to the main material parameters and auxiliary material parameters, whereby the multimodal synchronization feature terms include bonding position offset features, tension fluctuation features, step distance consistency features, and control delay features; collecting a multimodal synchronization feature dataset for the asynchronous transfer process using the multimodal synchronization feature terms; calculating the coefficient of variation of the multimodal synchronization feature dataset based on a short-term window moving average algorithm, outputting the multimodal coefficient of variation; converting the multimodal coefficient of variation into a fault probability to obtain a fault prediction probability; and issuing a fault signal alert to the integrated membrane product processing equipment when the fault prediction probability is greater than a preset fault probability.

[0005] Another aspect of this application discloses a fault prediction system for an integrated membrane product processing equipment. The system includes: a processing mechanism acquisition unit for acquiring the main material mechanism and auxiliary material mechanism of the integrated membrane product processing equipment; a material parameter acquisition unit for acquiring the main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism, respectively, whereby the material parameters include material type, thickness, tension reference value, and bonding fabric distance; a feature item definition unit for defining multimodal synchronization feature items for the asynchronous transfer process according to the main material parameters and auxiliary material parameters, whereby the multimodal synchronization feature items include bonding position offset features, tension fluctuation features, step distance consistency features, and control delay features; a feature item acquisition unit for acquiring a multimodal synchronization feature dataset for the asynchronous transfer process using the multimodal synchronization feature items; a fault probability conversion unit for calculating the coefficient of variation of the multimodal synchronization feature dataset based on a short-term window sliding mean algorithm, outputting a multimodal coefficient of variation, and converting the multimodal coefficient of variation into a fault probability to obtain a fault prediction probability; and a fault signal alert unit for alerting the integrated membrane product processing equipment when the fault prediction probability is greater than a preset fault probability.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The aforementioned fault prediction method for integrated film product processing equipment involves acquiring key parameters of the main and auxiliary material handling mechanisms within the equipment (including material type, thickness, tension reference value, and bonding fabric distance) of the materials used. Based on these parameters, multiple characteristic items are defined to evaluate the synchronicity of the asynchronous transfer process, such as bonding position offset, tension fluctuation, step distance consistency, and control delay. Subsequently, multimodal synchronicity data during the asynchronous transfer process is collected based on these characteristics. For this data, a short-term window moving average algorithm is used to calculate the coefficient of variation, which is then converted into a fault prediction probability for equipment operation. When this prediction probability exceeds a set risk threshold, a fault warning signal is proactively issued to the equipment, thereby achieving early fault identification and prevention.

[0008] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0010] Figure 1 This is a flowchart illustrating a fault prediction method for an integrated membrane product processing equipment in one embodiment.

[0011] Figure 2 This is a fault prediction system architecture diagram of an integrated membrane product processing equipment in one embodiment.

[0012] Explanation of reference numerals in the attached drawings: 11. Processing mechanism acquisition unit; 12. Material parameter acquisition unit; 13. Feature item definition unit; 14. Feature item acquisition unit; 15. Fault probability conversion unit; 16. Fault signal reminder unit. Detailed Implementation

[0013] This application provides a fault prediction method and system for integrated film product processing equipment, solving the technical problem that it is difficult to accurately predict faults caused by multimodal synchronization mismatch during asynchronous transfer of integrated film product processing equipment.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0016] Example 1, as Figure 1 As shown, this application provides a fault prediction method for integrated membrane product processing equipment, the method comprising:

[0017] Obtain the main material mechanism and auxiliary material mechanism of the integrated membrane product processing equipment.

[0018] In this embodiment of the application, in the integrated membrane product processing equipment, the main material mechanism and the auxiliary material mechanism are used to process different types of membrane materials. The main material mechanism is responsible for conveying and processing the main base membrane, i.e., the main membrane, which typically serves as the structural foundation of the entire membrane product. The auxiliary material mechanism, on the other hand, is used to synchronously convey and bond membrane layers with specific functions, i.e., functional membranes, such as those with scratch-resistant, waterproof, conductive, or optical properties. Before the equipment operates, the existence and operating status of these two types of mechanisms need to be identified to facilitate subsequent material parameter acquisition, synchronization characteristic analysis, and fault prediction and handling. This step ensures that subsequent processing steps can be performed based on the correct equipment structural information, which helps to achieve high-precision bonding and collaborative processing between the main membrane and the functional membrane.

[0019] Furthermore, this application provides a main material handling mechanism and a secondary material handling mechanism for obtaining integrated membrane product processing equipment, and the method further includes:

[0020] A first type of processing equipment and a second type of processing equipment are acquired during the asynchronous transfer process. The first type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the main material mechanism, and the second type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the auxiliary material mechanism. The operating parameters of the first type of processing equipment are synchronized according to the operating parameters of the main material mechanism; the operating parameters of the first type of processing equipment are synchronized according to the operating parameters of the auxiliary material mechanism.

[0021] Preferably, in the integrated film product processing, the asynchronous transfer stage involves the coordinated operation of multiple devices. First, based on the structural connection relationship between the main material mechanism (i.e., the main film transport section) and the auxiliary material mechanism (i.e., the functional film transport section), devices with direct or indirect transmission connections to the main material mechanism are identified as Class I processing equipment, such as main transport rollers, leveling rollers, and main drive units. Devices that form a transmission cooperation with the auxiliary material mechanism are identified as Class II processing equipment, such as auxiliary film unwinding devices, functional film bonding rollers, and auxiliary tension regulators. Subsequently, the relevant control parameters of the Class I processing equipment are synchronously adjusted according to the operating parameters of the main material mechanism (e.g., the main film conveying speed, tension setting, and synchronization signal) to ensure that all equipment related to the main film operates at a uniform process rhythm, avoiding film deformation or misalignment caused by speed differences or inconsistent tension. Similarly, the operating parameters of the Class II processing equipment are synchronously processed according to the operating parameters of the auxiliary material mechanism (e.g., the tension requirements of the functional film, bonding cycle time, etc.) to ensure that the functional film can achieve high-precision bonding with the main film with precise step distance and tension. The entire process coordinates the collaborative actions of the main membrane and functional membrane related equipment through a two-way synchronization mechanism, laying a data foundation for the stable acquisition of multimodal synchronization characteristics and fault prediction in the future.

[0022] The main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism are collected respectively. The material parameters include material type, thickness, tension reference value and bonding fabric distance.

[0023] In one embodiment, during the initial operation of the equipment, parameters of the membrane materials used in the main material mechanism and the auxiliary material mechanism are collected to obtain the corresponding main material parameters and auxiliary material parameters for the main material mechanism and the auxiliary material mechanism. This process mainly involves obtaining the basic physical and technological characteristics of each membrane material from the membrane material process database, including material type, thickness, tension reference value, and bonding fabric distance. The material type refers to the type of membrane material used, such as PET film, PI film, release film, or functional coated film, used to determine its processing characteristics; the thickness refers to the thickness of the membrane material to determine its flexibility and stress response characteristics in bonding, die-cutting, and other processes; the tension reference value is the set ideal tension range of the membrane material during transmission to ensure that it does not undergo excessive stretching or relaxation during transport and maintains stable transport; the bonding fabric distance refers to the relative positional distance that the main film and the functional film should maintain during bonding, and is a key parameter for achieving high-precision bonding control. Accurate collection and recording of these parameters provides necessary raw data support for subsequent synchronization analysis, feature extraction, and fault prediction, ensuring that the processing process maintains stable and efficient operation under different membrane material combinations.

[0024] Based on the main material parameters and auxiliary material parameters, a multimodal synchronization characteristic term for the asynchronous transfer process is defined. The multimodal synchronization characteristic term includes bonding position offset characteristic, tension fluctuation characteristic, step distance consistency characteristic, and control delay characteristic.

[0025] In one embodiment, after obtaining the material parameters of the main film and the functional film, multimodal synchronization characteristics in the asynchronous transfer process are further established based on these parameters to comprehensively reflect the collaborative state of the main material and the auxiliary material in the bonding process. These characteristics include, but are not limited to, bonding position offset characteristics, tension fluctuation characteristics, step pitch consistency characteristics, and control delay characteristics. Among them, the bonding position offset feature is used to measure whether there is a lateral or longitudinal alignment deviation between the main film and the functional film during bonding, such as misalignment or inaccurate overlapping of the two films in space, which is usually detected by a position sensor; the tension fluctuation feature reflects the stability of the tension of the main film and the functional film during operation. If the tension fluctuation of the film material is too large, it may cause wrinkles, folds or uneven bonding of the film surface. Tension data usually comes from real-time feedback from a tension sensor; the step distance consistency feature is used to evaluate whether the film feeding cycle of the main material and the auxiliary material is consistent during the bonding cycle. If there is a difference in step distance, it is easy to cause bonding error or cycle disorder, affecting bonding accuracy. It is usually detected by a speed sensor; the control delay feature is used to describe the time delay of the main and auxiliary material mechanisms in responding to control commands. If there is a significant time difference in response, it may lead to inconsistent action steps, thereby destroying synchronization. The time delay data usually comes from real-time feedback from the control command acquisition module. The establishment of these feature terms is based on the properties of the material itself and the dynamic performance during the actual bonding process. By jointly monitoring these four types of features, potential synchronization problems in the bonding of main and auxiliary materials can be fully captured, providing core variable support for subsequent fault risk modeling.

[0026] The multimodal synchronization feature dataset of the asynchronous reposting process is collected using the aforementioned multimodal synchronization feature terms.

[0027] In one embodiment, after defining multimodal synchronization characteristics such as bonding position offset, tension fluctuation, step distance consistency, and control delay, real-time data acquisition is performed around these characteristics during asynchronous bonding. Specifically, a multi-source sensor network (such as position sensors, tension sensors, speed encoders, and control signal acquisition modules) deployed on the main material mechanism and auxiliary material mechanism continuously acquires data corresponding to each characteristic during equipment operation. For example, tension sensors record real-time tension value changes, and the control signal acquisition module collects the response delay between command issuance and actual execution. By summarizing these acquired data, a multimodal synchronization characteristic dataset containing multiple dimensions and multiple time points is constructed. This dataset can dynamically reflect the changes in the collaborative state of the main and auxiliary materials during asynchronous bonding and serves as the core data foundation for subsequent synchronization assessment and fault prediction.

[0028] Furthermore, this application provides a multimodal synchronization feature dataset for asynchronous transfer processes acquired through a multi-source sensor network using the multimodal synchronization feature terms; wherein the multi-source sensor network includes a position sensor, a tension sensor, a speed sensor, and a control command acquisition module; after receiving the multimodal synchronization feature dataset, timestamp alignment processing is performed, and the processed multimodal synchronization feature dataset is output.

[0029] Preferably, to comprehensively and accurately acquire multimodal synchronous characteristic data involving the collaborative processing of main and auxiliary materials during asynchronous transfer, a multi-source sensor network covering both the main and auxiliary material mechanisms is constructed. This multi-source sensor network includes position sensors, tension sensors, speed sensors, and a control command acquisition module. Specifically, the position sensors are used to detect the bonding position of the main film and functional film in real time, capturing minute displacement deviations and acquiring bonding offset characteristics; the tension sensors are deployed in the conveying path to continuously acquire tension fluctuations of the main and auxiliary materials during operation, reflecting the material's stress stability; the speed sensors are installed on the conveying rollers or drive shaft to record the step distance and conveying speed of the main and auxiliary materials per unit time, quantifying step distance consistency; and the control command acquisition module captures control signals issued by the system and feedback responses from each actuator, thereby calculating control response delays and acquiring control delay characteristics. The aforementioned sensors will transmit the collected data back to the system, forming a multimodal synchronous feature dataset, including but not limited to fit position offset features, tension fluctuation features, step distance consistency features, and control delay features. These features are all accompanied by high-precision timestamps after acquisition. To ensure that the multimodal data is aligned in time, the timestamps of these asynchronously acquired data streams will be aligned based on a unified reference clock (such as the PLC system time). This will output an aligned multimodal synchronous feature dataset arranged in a unified time sequence, ensuring that feature data from different sensors can be analyzed and compared within the same time frame, providing accurate and consistent input data for subsequent variability assessment and fault prediction.

[0030] The coefficient of variation of the multimodal synchronization feature dataset is calculated based on the short-term window moving average algorithm, and the multimodal coefficient of variation is output. The multimodal coefficient of variation is then converted into a fault probability to obtain the fault prediction probability.

[0031] In one embodiment, after collecting and aligning the multimodal synchronization feature dataset, these feature data are analyzed using a short-term window moving average algorithm to quantify their temporal fluctuations, thereby assessing the stability of device operation. Specifically, the length of the moving window (e.g., sampling points from the most recent 5 seconds or several periods), the moving step size (e.g., moving once per second), and a protection constant for numerical stability are first set to avoid abnormal situations such as division by zero. Subsequently, for each multimodal feature (e.g., tension, fit offset, etc.), within the set moving window, the moving mean and moving standard deviation (std) of the data segment are calculated. These two parameters reflect the central trend and fluctuation degree of the feature over a short period of time. Then, for each feature item, its coefficient of variation is calculated in each window, and the coefficients of variation of all modalities are integrated into a multimodal coefficient of variation vector, reflecting the overall fluctuation characteristics of each synchronization feature within the current time period. Then, the multimodal coefficient of variation vector is input into a pre-trained fault probability transformation model (such as a sigmoid model based on logistic regression). The model outputs a fault prediction probability based on the distribution characteristics of historical normal and fault data. This probability, ranging from 0 to 1, represents the likelihood of a potential fault occurring in the equipment under the current condition. This process not only allows for real-time sensing of fluctuations in various synchronicity features but also quantifies complex feature fluctuation data into a probabilistic indicator for judgment, providing a reliable basis for fault early warning.

[0032] Furthermore, this application provides a method for calculating the coefficient of variation of the multimodal synchronization feature dataset based on a short-term window moving average algorithm, and outputting the multimodal coefficient of variation. The method further includes:

[0033] Initialize the sliding window parameters, which include window length, sliding step size, and stability protection constant; collect multimodal synchronization feature data sequences with the sliding step size within the window length; calculate the sliding mean and sliding standard deviation for each modal synchronization feature data sequence; calculate the coefficient of variation based on the sliding mean and sliding standard deviation, and output the multimodal coefficient of variation.

[0034] Preferably, when calculating the coefficient of variation, the sliding window parameters are first initialized, including the window length, sliding step size, and stability protection constant. The window length represents the data time period or number of samples covered in each statistical analysis; for example, it can be set to the most recent 100 sampling points. The sliding step size represents the interval at which the window moves forward each time; for example, it slides once every 10 sampling points. The stability protection constant is used to prevent division by zero in the coefficient of variation calculation when the mean is close to zero, and is usually set to a positive number less than 0.01. Subsequently, in the multimodal synchronous feature dataset, the time series data of each modality (such as tension, offset, etc.) is segmented according to the preset sliding step size, with each segment having a length equal to the window length, forming multiple data segments. For each modal data segment, the moving average of all segments is calculated to represent the central trend of the feature within each time period. Then, the moving standard deviation of all segments is calculated to reflect the dispersion of the feature value within each time period. Subsequently, the sliding standard deviation of each mode is divided by the sum of the sliding mean and the stability protection constant to calculate the multimodal coefficient of variation (CV), thereby enabling sensitive monitoring and quantitative expression of the fluctuations in the bonding process, and serving as an important input basis for subsequent failure probability assessment.

[0035] Furthermore, this application provides a method for converting the multimodal variation coefficients into fault probability to obtain fault prediction probability, comprising:

[0036] The multimodal coefficients of variation are combined into a vector to obtain a multimodal coefficient of variation vector; a fault probability conversion model is constructed, the fault probability conversion model includes a conversion interval [0, 1]; the multimodal coefficient of variation vector is input into the fault probability conversion model for prediction, and the fault prediction probability mapped in the conversion interval is obtained.

[0037] Optionally, after calculating the coefficient of variation for each modal synchronization feature, the coefficients of variation calculated for each synchronization feature (such as fit position offset, tension fluctuation, step size consistency, and control delay) within the current sliding window are arranged in a fixed order to form a multimodal coefficient of variation vector. For example, [CV 贴合位置偏移 CV 张力波动 CV 步距一致性 CV 控制延迟This multimodal coefficient of variation (CV) vector comprehensively reflects the stability of various key synchronicity characteristics of the equipment within the current time period. Subsequently, a fault probability conversion model is constructed. The goal of this model is to map the CV vector to a fault probability value between 0 and 1. The model typically employs logistic regression, whose core utilizes the sigmoid function for nonlinear mapping. The output value always lies within the conversion interval [0,1], representing the likelihood of a fault occurring in the current system state. Then, the CV vector is input into the fault probability conversion model. Based on the trained parameters (weights and biases), the model performs a weighted summation of the input vector and maps it to a fault prediction probability. The closer the fault prediction probability is to 1, the higher the risk of a potential equipment fault. Through this process, the conversion from raw fluctuation data to fault probability is achieved, providing a quantitative basis for subsequent judgment and early warning.

[0038] Furthermore, this application provides a method for constructing a fault probability conversion model, including:

[0039] A historical sample library is constructed, which includes multimodal coefficient of variation vector samples under normal operating conditions, multimodal coefficient of variation vector samples under known fault conditions, and fault labels. A logistic regression function is defined using the sigmoid function. Based on the multimodal coefficient of variation samples, the multimodal coefficient of variation samples under known fault conditions, and the fault labels, the logistic regression function is trained until the accuracy of fault prediction reaches a preset accuracy threshold.

[0040] Optionally, to achieve accurate mapping between the multimodal coefficient of variation (CV) vector and the failure probability, a large number of asynchronous reposting task data samples are collected during actual operation or testing to construct a historical sample library. Each sample in the historical sample library contains a CV vector within a time window and is labeled according to the equipment's operating state at that time. The samples are divided into two categories: normal operating condition samples and corresponding failure labels, and failure operating condition samples and corresponding failure labels. Subsequently, a logistic regression function is defined using the sigmoid function, and a failure probability conversion model is constructed using this logistic regression function. The samples from the historical sample library are then input into the failure probability conversion model. In the initial state, a set of parameters (including the weights and biases of each feature term) are randomly set, and the failure probability of each sample is predicted based on the current parameters. Afterward, the prediction results are compared with the true labels, the model's prediction error (such as mean squared error) is calculated, and the parameters are adjusted using optimization algorithms (such as gradient descent) based on these errors. The goal is to make the model's prediction results closer to the actual labels. This process is repeated continuously, and each round is called a training iteration or training round. In each iteration, the model updates its parameters and repeats the prediction-comparison-adjustment cycle. As iterations progress, the model's predictive ability gradually improves, and the prediction error continuously decreases. The entire training process continues until the model's accuracy on the validation dataset (data not used in training) reaches a preset accuracy threshold. The trained logistic regression model can be used to convert the real-time calculated multimodal coefficient of variation vector into the probability of equipment failure prediction, enabling online early warning and fault prevention.

[0041] Furthermore, this application provides a method for defining a logistic regression function using the sigmoid function, the expression of which includes: Among them, P fault Here, ω represents the fault prediction probability, ω0 is the bias term, and ω represents the basic offset of the model. i Let CV be the weight coefficient of the i-th modal feature. i Let be the coefficient of variation of the i-th modal feature, reflecting the fluctuation intensity of a certain synchronicity feature, n be the total number of multimodal synchronicity feature terms, and e be the base of the natural logarithm.

[0042] Optionally, to convert the multimodal coefficient of variation vector into fault prediction probability, the sigmoid function is used to define the logistic regression function, whose expression is: Among them, P fault ω represents the final output fault prediction probability; ω0 is the bias term, representing the model's basic offset or constant term; ω i CV represents the weighting coefficient of the i-th modal synchronization feature, used to measure the degree of influence of this feature on the failure probability. idenoted as the coefficient of variation of the i-th modal synchronicity feature, reflecting the intensity of its fluctuation within a short-term window; n is the total number of multimodal synchronicity feature terms, i.e., the number of coefficients of variation input into the model; e is the base of the natural logarithm, approximately 2.718, used to calculate the exponential function value. This formula, by weighting and summing the variability of each feature term and applying a sigmoid function for nonlinear mapping, strictly limits the output value to the range of 0 to 1, intuitively representing the probability of equipment failure at the current moment.

[0043] Furthermore, this application provides that the value range of the stability protection constant is

[10] . -6 10 -2 ].

[0044] Optionally, to ensure the numerical stability and physical interpretability of the coefficient of variation calculation, a non-zero stability protection constant is set, and the value range of this stability protection constant is

[10] . -6 10 -2 This range avoids division by zero or near-zero anomalies without significantly affecting the coefficient of variation's reflection of real fluctuations. It is suitable for data normalization processing of different synchronicity feature dimensions, ensuring the accuracy of early fault warnings.

[0045] Furthermore, the method provided in this application includes:

[0046] Obtain the duration of the multimodal variation coefficients; use the duration as an incremental learning parameter to perform feedback optimization learning on the logistic regression function to obtain a retrained fault probability conversion model; update the fault prediction probability based on the retrained fault probability conversion model.

[0047] Optionally, to enhance the adaptive capability of the fault probability conversion model and enable it to continuously adjust and optimize prediction accuracy during long-term operation, the coefficient of variation of each modality's synchronicity characteristics will be continuously monitored during operation, and it will be detected whether the coefficient of variation is consistently higher than a preset fluctuation threshold for a certain period of time (i.e., abnormal fluctuations continue to occur). When abnormally persistent coefficient of variation is identified, the duration of this state is recorded, for example, the number of consecutive windows with high variation or the corresponding time length (e.g., 30 seconds, 5 minutes, etc.). If the duration exceeds a certain learning trigger threshold (e.g., the continuous abnormality exceeds 1 minute), this information will be used as a feedback signal to trigger the incremental learning mechanism. In this mechanism, the multimodal coefficient of variation vector during the period of continuous abnormality and its corresponding actual equipment state (whether a fault has occurred) are labeled and appended to the original historical sample library. This new data serves as an incremental sample set for retraining the original logistic regression model. In this process, the parameters of the original model are used as initial values, and newly collected abnormal continuous samples are introduced to retrain the logistic regression function. During training, standard gradient descent or batch update algorithms are used to readjust the weights and biases to adapt to the new data distribution and improve the model's predictive ability under current operating conditions. After feedback learning is completed, an updated logistic regression model is generated—a retrained fault probability transformation model—which can more accurately capture abnormal pattern changes that may occur during long-term operation. Finally, the currently collected multimodal variation coefficient vector is input into the updated model, outputting the latest fault prediction probability. This prediction value serves as the basis for the final judgment on whether to trigger a fault warning, thereby achieving adaptive evolution and dynamic optimization of the model. This learning mechanism based on feedback of abnormal duration effectively enhances the model's robustness and accuracy in diverse operating environments and is an important component for achieving intelligent prediction.

[0048] When the predicted fault probability is greater than the preset fault probability, a fault signal is sent to the integrated membrane product processing equipment.

[0049] In one embodiment, after completing the multimodal coefficient of variation analysis of the current asynchronous transfer process and outputting the corresponding fault prediction probability through the fault probability conversion model, this probability value is compared with a pre-set fault risk threshold. This preset fault probability is usually set based on historical data statistics, experience judgment, or risk control requirements. For example, 0.6 or 0.8 indicates that when the model judges that the current operating condition has a probability of more than 60% or 80% leading to a fault, an intervention mechanism should be triggered immediately. When the current fault prediction probability is detected to exceed the preset threshold, that is, the equipment operating status is in a potentially high-risk fault range, a fault alert mechanism will be triggered immediately. For example, a fault warning signal will be sent to the equipment control system, prompting the equipment to enter standby, self-test, or deceleration mode; real-time alarm information will be displayed on the human-machine interface (HMI), indicating possible abnormal parts or fluctuation characteristics; and a fault warning notification will be sent to the maintenance team through the console, mobile terminal, or information platform. Through the above fault signal alerts, it can be ensured that fault risks are identified in time before they occur and response measures are taken, thereby minimizing the risk of material waste, equipment damage, or downtime caused by synchronization anomalies and ensuring the continuous and stable operation of the membrane product processing equipment.

[0050] In summary, the embodiments of this application have at least the following technical effects:

[0051] This application embodiment first acquires the main material mechanism and auxiliary material mechanism of the integrated film product processing equipment; then, it collects the main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism, respectively. The material parameters include material type, thickness, tension reference value, and bonding fabric distance; then, according to the main material parameters and auxiliary material parameters, it defines multimodal synchronization feature terms for the asynchronous transfer process. The multimodal synchronization feature terms include bonding position offset features, tension fluctuation features, step distance consistency features, and control delay features; further, it collects a multimodal synchronization feature dataset for the asynchronous transfer process using the multimodal synchronization feature terms; then, it calculates the coefficient of variation of the multimodal synchronization feature dataset based on the short-term window moving average algorithm, outputs the multimodal coefficient of variation, and converts the multimodal coefficient of variation into a fault probability to obtain a fault prediction probability; finally, when the fault prediction probability is greater than the preset fault probability, it sends a fault signal reminder to the integrated film product processing equipment. These technologies collectively solve the technical problem of difficulty in accurately predicting faults caused by multimodal synchronization mismatch during asynchronous transfer of integrated film product processing equipment. They achieve the technical effect of early warning of faults through multimodal synchronization feature analysis and coefficient of variation calculation, thereby improving equipment operation stability and production efficiency.

[0052] Example 2, based on the same inventive concept as the fault prediction method for an integrated membrane product processing equipment in the foregoing examples, such as... Figure 2As shown, this application provides a fault prediction system for an integrated film product processing equipment. The system includes: a processing mechanism acquisition unit 11 for acquiring the main material mechanism and auxiliary material mechanism of the integrated film product processing equipment; a material parameter acquisition unit 12 for acquiring the main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism, respectively, wherein the material parameters include material type, thickness, tension reference value, and bonding fabric distance; and a feature item definition unit 13 for defining multimodal synchronization feature items for the asynchronous transfer process according to the main material parameters and auxiliary material parameters, wherein the multimodal synchronization feature items include bonding position. The multimodal synchronization features include offset characteristics, tension fluctuation characteristics, step-width consistency characteristics, and control delay characteristics. Feature acquisition unit 14: acquires a multimodal synchronization feature dataset for the asynchronous transfer process using the aforementioned multimodal synchronization feature items. Fault probability conversion unit 15: calculates the coefficient of variation of the multimodal synchronization feature dataset based on a short-term window moving average algorithm, outputs the multimodal coefficient of variation, and converts the multimodal coefficient of variation into a fault probability to obtain a fault prediction probability. Fault signal alert unit 16: when the fault prediction probability is greater than a preset fault probability, it sends a fault signal alert to the integrated membrane product processing equipment.

[0053] Furthermore, the processing mechanism acquisition unit 11 is also used to perform the following method:

[0054] A first type of processing equipment and a second type of processing equipment are acquired during the asynchronous transfer process. The first type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the main material mechanism, and the second type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the auxiliary material mechanism. The operating parameters of the first type of processing equipment are synchronized according to the operating parameters of the main material mechanism; the operating parameters of the first type of processing equipment are synchronized according to the operating parameters of the auxiliary material mechanism.

[0055] Furthermore, the feature acquisition unit 14 is also used to perform the following method:

[0056] A multimodal synchronization feature dataset for the asynchronous transfer process is collected using a multi-source sensor network with the multimodal synchronization feature terms. The multi-source sensor network includes a position sensor, a tension sensor, a speed sensor, and a control command acquisition module. After receiving the multimodal synchronization feature dataset, timestamp alignment processing is performed, and the processed multimodal synchronization feature dataset is output.

[0057] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0058] Initialize the sliding window parameters, which include window length, sliding step size, and stability protection constant; collect multimodal synchronization feature data sequences with the sliding step size within the window length; calculate the sliding mean and sliding standard deviation for each modal synchronization feature data sequence; calculate the coefficient of variation based on the sliding mean and sliding standard deviation, and output the multimodal coefficient of variation.

[0059] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0060] The multimodal coefficients of variation are combined into a vector to obtain a multimodal coefficient of variation vector; a fault probability conversion model is constructed, the fault probability conversion model includes a conversion interval [0, 1]; the multimodal coefficient of variation vector is input into the fault probability conversion model for prediction, and the fault prediction probability mapped in the conversion interval is obtained.

[0061] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0062] A historical sample library is constructed, which includes multimodal coefficient of variation vector samples under normal operating conditions, multimodal coefficient of variation vector samples under known fault conditions, and fault labels. A logistic regression function is defined using the sigmoid function. Based on the multimodal coefficient of variation samples, the multimodal coefficient of variation samples under known fault conditions, and the fault labels, the logistic regression function is trained until the accuracy of fault prediction reaches a preset accuracy threshold.

[0063] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0064] The logistic regression function is defined using the sigmoid function, and its expression includes: Among them, P fault Here, ω represents the fault prediction probability, ω0 is the bias term, and ω represents the basic offset of the model. i Let CV be the weight coefficient of the i-th modal feature. i Let be the coefficient of variation of the i-th modal feature, reflecting the fluctuation intensity of a certain synchronicity feature, n be the total number of multimodal synchronicity feature terms, and e be the base of the natural logarithm.

[0065] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0066] The stability protection constant takes values ​​in the range of

[10] . -6 10 -2 ].

[0067] Furthermore, the fault probability conversion unit 15 is also used to perform the following method:

[0068] Obtain the duration of the multimodal variation coefficients; use the duration as an incremental learning parameter to perform feedback optimization learning on the logistic regression function to obtain a retrained fault probability conversion model; update the fault prediction probability based on the retrained fault probability conversion model.

[0069] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0070] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0071] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A fault prediction method for integrated membrane product processing equipment, characterized in that, The method includes: Acquire the main material mechanism and auxiliary material mechanism of the integrated membrane product processing equipment; The main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism are collected respectively. The material parameters include material type, thickness, tension reference value and bonding fabric distance. Based on the main material parameters and auxiliary material parameters, a multimodal synchronization characteristic term for the asynchronous transfer process is defined. The multimodal synchronization characteristic term includes bonding position offset characteristic, tension fluctuation characteristic, step distance consistency characteristic, and control delay characteristic. The multimodal synchronization feature dataset of the asynchronous transfer process is collected using the aforementioned multimodal synchronization feature terms; The coefficient of variation of the multimodal synchronization feature dataset is calculated based on the short-term window moving average algorithm, and the multimodal coefficient of variation is output. The multimodal coefficient of variation is then converted into a fault probability to obtain the fault prediction probability. When the predicted fault probability is greater than the preset fault probability, a fault signal is sent to the integrated membrane product processing equipment. The method further includes calculating the coefficient of variation of the multimodal synchronization feature dataset based on the short-term window moving average algorithm and outputting the multimodal coefficient of variation. Initialize the sliding window parameters, which include the window length, sliding step size, and stability protection constant; The multimodal synchronization feature data sequence is acquired within the window length using the sliding step size. For each modal synchronization feature data sequence, calculate the moving mean and moving standard deviation; The coefficient of variation is calculated based on the moving mean and the moving standard deviation, and the multimodal coefficient of variation is output. The multimodal variation coefficients are converted into fault probability to obtain fault prediction probability. The method includes: The multimodal variation coefficients are combined into a vector to obtain a multimodal variation coefficient vector; Construct a fault probability conversion model, wherein the fault probability conversion model includes a conversion interval, and the conversion interval is [0, 1]; The multimodal variation coefficient vector is input into the fault probability conversion model for prediction, and the fault prediction probability mapped in the conversion interval is obtained.

2. The fault prediction method for an integrated membrane product processing equipment as described in claim 1, characterized in that, Methods for constructing fault probability conversion models include: Construct a historical sample library, which includes multimodal coefficient of variation vector samples under normal operating conditions, multimodal coefficient of variation vector samples under known fault conditions, and fault labels; Define the logistic regression function using the sigmoid function; Based on the multimodal coefficient of variation vector samples under normal operating conditions, the multimodal coefficient of variation vector samples under known fault conditions, and fault labels, the logistic regression function is trained until the accuracy of fault prediction reaches a preset accuracy threshold.

3. The fault prediction method for an integrated membrane product processing equipment as described in claim 2, characterized in that, The logistic regression function is defined using the sigmoid function, and its expression includes: ； in, For fault prediction probability, This is the bias term, representing the basic offset of the model. The weight coefficients for the i-th modal feature are... Let be the coefficient of variation of the i-th modal feature, reflecting the fluctuation intensity of a certain synchronicity feature, and n be the total number of multimodal synchronicity feature terms. is the base of the natural logarithm.

4. The fault prediction method for an integrated membrane product processing equipment as described in claim 1, characterized in that, The range of values ​​for the stability protection constant is: .

5. The fault prediction method for an integrated membrane product processing equipment as described in claim 1, characterized in that, The method for obtaining the main material mechanism and auxiliary material mechanism of integrated membrane product processing equipment also includes: Acquire a first type of processing equipment and a second type of processing equipment in the asynchronous transfer process, wherein the first type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the main material mechanism, and the second type of processing equipment is the equipment in the integrated film product processing equipment that has a transmission relationship with the auxiliary material mechanism; The operating parameters of the first type of processing equipment are synchronously processed according to the operating parameters of the main material mechanism; The operating parameters of the second type of processing equipment are synchronized according to the operating parameters of the auxiliary material mechanism.

6. The fault prediction method for an integrated membrane product processing equipment as described in claim 1, characterized in that, A multimodal synchronization feature dataset of the asynchronous transfer process is collected using a multi-source sensor network with the aforementioned multimodal synchronization feature terms; The multi-source sensor network includes a position sensor, a tension sensor, a speed sensor, and a control command acquisition module. After receiving the multimodal synchronization feature dataset, timestamp alignment processing is performed, and the processed multimodal synchronization feature dataset is output.

7. The fault prediction method for an integrated membrane product processing equipment as described in claim 2, characterized in that, The method includes: The duration of obtaining the multimodal variation coefficient; The logistic regression function is subjected to feedback optimization learning using the duration as an incremental learning parameter to obtain a retrained fault probability conversion model; The fault prediction probability is updated based on the retrained fault probability transformation model.

8. A fault prediction system for integrated membrane product processing equipment, characterized in that, The system is used to execute the fault prediction method for an integrated membrane product processing equipment according to any one of claims 1-7, including: Processing mechanism acquisition unit: acquires the main material mechanism and auxiliary material mechanism of the integrated membrane product processing equipment; Material parameter acquisition unit: Collects the main material parameters and auxiliary material parameters corresponding to the main material mechanism and auxiliary material mechanism respectively. The material parameters include material type, thickness, tension reference value and bonding fabric distance; Feature definition unit: According to the main material parameters and auxiliary material parameters, define the multimodal synchronization feature of the asynchronous transfer process. The multimodal synchronization feature includes bonding position offset feature, tension fluctuation feature, step distance consistency feature and control delay feature. Feature acquisition unit: Acquires multimodal synchronization feature dataset of asynchronous transfer process using the multimodal synchronization feature items; Fault probability conversion unit: Calculates the coefficient of variation of the multimodal synchronization feature dataset based on the short-term window moving average algorithm, outputs the multimodal coefficient of variation, and converts the multimodal coefficient of variation into a fault probability to obtain the fault prediction probability; Fault signal alert unit: When the predicted fault probability is greater than the preset fault probability, a fault signal alert is sent to the integrated membrane product processing equipment.

Images