Fault prediction system for tissue production equipment based on time series data analysis

Abstract

The application discloses a kind of based on timing data analysis's paper production equipment fault prediction system, it is related to process control technical field, to solve the technical problem that present technique cannot utilize water content mutation event as diagnostic excitation, the health state of roll gap is predicted by roll gap pressure transient response waveform, including: water content mutation event detection module is configured to detect water content mutation event by fusing liquid supply system state signal and paper physical characteristic signal online, and record event trigger time and mutation amplitude when detecting water content mutation event;Transient response waveform acquisition module is configured to take event trigger time as starting point, and continuously collect the timing signal of roll gap pressure in a predetermined time window, generate roll gap pressure transient response waveform;The application realizes the accurate prediction to wet paper towel folding machine roll gap health state by constructing water content mutation event driven roll gap pressure transient response prediction and dynamic compensation system.

Description

Technical Field

[0001] This invention relates to the field of process control technology, and more specifically, to a fault prediction system for tissue paper production equipment based on time-series data analysis. Background Technology

[0002] The wet wipe folding machine is one of the core pieces of equipment in a tissue paper production line, and its roller gap condition directly determines the folding accuracy and finished product quality. In the wet wipe production process, the raw paper is humidified by the liquid supply system before entering the folding process. Moisture content, as a key process parameter, significantly affects the paper's thickness, flexibility, and frictional characteristics. When the liquid supply pump starts or stops, the nozzle becomes clogged, the raw paper roll is replaced, or the liquid supply pressure fluctuates, the moisture content changes abruptly. This change is then transmitted to the folding machine roller gap through changes in the mechanical properties of the wet paper, causing a transient response in the roller gap pressure. The morphological characteristics of this transient response waveform—such as response speed, overshoot amplitude, oscillation characteristics, and steady-state deviation—are closely related to the mechanical stiffness, damping characteristics, and wear condition of the roller gap, containing a wealth of information about the equipment's health.

[0003] However, existing technologies lack utilization of this physical phenomenon. Currently, fault monitoring in wet wipe folding machines mainly relies on periodic manual inspections or threshold alarms for single parameters, such as monitoring whether the roller gap pressure exceeds the limit through pressure sensors or detecting abnormal impacts through vibration sensors. These methods have the following drawbacks: First, threshold alarms for single parameters cannot capture the slow degradation trend of the roller gap condition, often triggering alarms only when a fault has already occurred or seriously affected product quality, which is a passive maintenance approach. Second, existing technologies fail to transform the common disturbance event of moisture content change in wet wipe production into diagnostic excitations, and cannot extract equipment health characteristics from the transient response of roller gap pressure to moisture content impacts. Third, multi-source signals such as the liquid supply system and paper physical properties are not effectively integrated, leading to missed detections and false alarms in the detection of moisture content change events, and transient response analysis lacks accurate event triggering benchmarks. Therefore, how to utilize moisture content change events as an active excitation source and predict the roller gap health status by analyzing the transient response waveform of roller gap pressure has become a technical problem that urgently needs to be solved in this field. In view of this, we propose a fault prediction system for tissue paper production equipment based on time series data analysis. Summary of the Invention

[0004] The purpose of this invention is to provide a fault prediction system for tissue paper production equipment based on time-series data analysis, so as to solve the technical problem in the prior art that it is impossible to use sudden changes in moisture content as diagnostic excitation and predict the health status of the roll gap through the transient response waveform of the roll gap pressure.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a fault prediction system for tissue paper production equipment based on time-series data analysis, comprising: The moisture content mutation event detection module is configured to detect moisture content mutation events online by integrating the liquid supply system status signal and the paper physical property signal, and to record the event trigger time and mutation magnitude when the moisture content mutation event is detected; The transient response waveform acquisition module is configured to continuously acquire the timing signal of the roll gap pressure within a predetermined time window, starting from the event trigger time, and generate a transient response waveform of the roll gap pressure. The transient response feature extraction module is configured to extract time-domain feature parameters and model-domain dynamic parameters based on a second-order ARMA model from the transient response waveform of the roll gap pressure. The time-domain feature parameters include response delay time, peak overshoot, response time, damped oscillation period, and steady-state error. The model-domain dynamic parameters include static gain, pure time delay, damping ratio, and undamped natural frequency. The roll gap health assessment module is configured to calculate a comprehensive health index based on the extracted time-domain feature parameters and model-domain dynamic parameters, and output a graded early warning signal according to the comprehensive health index.

[0006] This invention constructs a transient response prediction and dynamic compensation system for roll gap pressure driven by moisture content mutation events, achieving accurate prediction of the roll gap health status of wet tissue folding machines. This solves the technical problem in existing technologies where moisture content mutation shocks cannot be used for equipment condition diagnosis. Specifically, this invention first integrates the liquid supply system status signal and paper physical property signal to detect moisture content mutation events online and accurately record the event trigger time and mutation amplitude, transforming moisture content fluctuations, which were originally considered disturbances during production, into active diagnostic excitations. Then, starting from the event trigger time, the transient response waveform of the roll gap pressure is continuously acquired. Response delay time, peak overshoot, response time, decay oscillation period, steady-state error, and characteristic parameters identified based on a second-order ARMA model are extracted from both the time and model domains. These parameters have a clear physical correspondence with the mechanical inertia, damping characteristics, and stiffness degradation degree of the roll gap, and can sensitively reflect the gradual evolution process of the roll gap from health to failure. Based on these parameters, a comprehensive health index is calculated and graded early warning signals are output. Warnings can be issued hours to days before the roll gap completely fails, realizing a fundamental shift from passive maintenance to predictive maintenance and significantly reducing the risk of unplanned downtime.

[0007] Preferably, the moisture content abrupt change event detection module includes: The liquid supply pump status monitoring unit is configured to collect the outlet pressure signal, drive motor current signal and pump speed signal of the liquid supply pump, and detect sudden changes in water content caused by the start-up or shutdown of the liquid supply pump or pump body abnormality by performing differential analysis on the outlet pressure signal. The nozzle status monitoring unit is configured to collect the inlet pressure signals of multiple liquid dispensing nozzles arranged along the width of the paper web, and detect sudden changes in the lateral distribution of moisture content caused by nozzle blockage or wear by analyzing the consistency and dispersion of the inlet pressure of each nozzle. The liquid supply flow monitoring unit is configured to collect the liquid supply flow signals of the main liquid supply pipe and branch pipes, and identify abnormal flow events in the liquid supply system by performing first-order differential analysis and cumulative sum control chart change point detection on the liquid supply flow signals. The paper thickness monitoring unit is configured to continuously acquire paper thickness signals through a laser displacement sensor arranged at the entrance of the folding machine, and to detect sudden changes in thickness caused by changes in moisture content by performing differential analysis on the paper thickness signals. The paper friction coefficient estimation unit is configured to estimate the current friction coefficient online by collecting the torque signal of the drive motor of the traction roller and the paper tension signal, and to detect the sudden change in friction characteristics caused by the change in moisture content by performing differential analysis on the friction coefficient. The paper tension change point detection unit is configured to continuously collect paper tension signals through tension sensors arranged between each station of the folding machine, and to perform time-series change point analysis on the tension signals using a Bayesian change point detection algorithm to detect tension anomalies caused by sudden changes in moisture content.

[0008] Preferably, the moisture content abrupt change event detection module further includes an adaptive weighted fusion unit, which is configured as follows: Calculate the anomaly probability of each detected signal source at the current moment. The anomaly probability is expressed by the formula... Calculate, where, For the first Each signal source at time The probability of an anomaly; The cumulative distribution function of the standard normal distribution; For the first Each signal source at time The measured value; , For the first The mean and standard deviation of each signal source under normal conditions; The fusion weights are dynamically assigned based on the signal-to-noise ratio of each signal source, the consistency of detection results, and the historical detection accuracy. The weight calculation formula is as follows: In the formula, For the first Each signal source at time The fusion weights; For the first The signal-to-noise ratio of each signal source within the current time window; For the first The consistency coefficient of the detection results of one signal source with other signal sources; For the first Historical detection accuracy of individual signal sources; The total number of signal sources; Calculate the weighted fusion anomaly score ,when Exceeding the preset fusion threshold The occurrence of a sudden change in moisture content was confirmed in time, and the trigger time of the event and the magnitude of the change estimated based on the change in paper thickness were recorded.

[0009] Preferably, the adaptive weighted fusion unit further includes a false alarm suppression subunit, which is configured as follows: A state where the weighted fusion anomaly score exceeds the fusion threshold must remain for a predetermined duration before being confirmed as a valid event. The confirmation condition is as follows: In the formula, Indicates the current time; Indicates the predetermined duration threshold; Indicates time The weighted fusion anomaly score; Indicates at time Confirm valid event; In the formula, For the first The timing of the confirmation event; For the first The timing of the confirmation event; This is the minimum time interval threshold.

[0010] Preferably, the transient response feature extraction module includes a time-domain feature extraction unit, which is configured as follows: Extract response delay time In the formula, The moment the event is triggered; For a moment The roller gap pressure; This represents the steady-state pressure value before the event. The predetermined deviation threshold; Extracting peak overshoot In the formula, Indicates the peak pressure of the response waveform; Extract response time In the formula, This is the verification duration used to confirm the continuation of the steady state; Extracting the decaying oscillation period In the formula, This represents the total number of positive peak values ​​detected. Indicates the first The moment corresponding to each positive peak; Extracting steady-state error In the formula, This represents the average pressure value after the response waveform reaches a steady state.

[0011] Preferably, the transient response feature extraction module includes a model domain parameter identification unit, which is configured as follows: Treating the abrupt change in moisture content as a step input, the transient response waveform of the roll gap pressure is fitted to a second-order ARMA model, with the continuous transfer function form as follows: The corresponding difference equation is In the formula, This is the transfer function from moisture content to roll gap pressure; This is the Laplace transform of the roll gap pressure; The Laplace transform of moisture content; This is the static gain; This is the pure time delay; , It is a time constant; , These are the autoregressive coefficients; , The moving average coefficient; For a sudden change in water content, step input is used. It is white noise; The parameter vector is estimated online using the recursive least squares method. The recursive formula is as follows: ; ; ; In the formula, For parameter estimation vectors; It is the gain vector; To output the measured value; For regression vectors; It is the covariance matrix; Forgetting factor; It is the identity matrix; The static gain is obtained from the estimated model parameters. The damping ratio is calculated by multiplying and summing the time constants. and undamped natural frequency ; Pure delay time Estimation of peak location using cross-correlation function: In the formula, For time delay variables; The sampling period.

[0012] Preferably, the model domain parameter identification unit is further configured as follows: Record the time evolution sequence of the static gain identified after each abrupt change in moisture content, and extract the trend term using the exponentially weighted moving average method: In the formula, Indicates the first The smoothed value of the static gain after the event; Indicates the first Static gain identified after the event; It is a smoothing factor; Degradation slope estimated by linear regression In the formula, The estimated slope representing the static gain trend; The average over time; This represents the mean of the smoothed static gain; Predicting remaining effective lifespan In the formula, This is the initial static gain; The degradation rate; The failure threshold for static gain; This refers to the current moment.

[0013] Preferably, the roll gap health assessment module is configured as follows: The extracted time-domain feature parameters and model-domain dynamic parameters are normalized by range: In the formula, For the first Normalized values ​​of each feature parameter; For the first The current values ​​of each feature parameter; This indicates the minimum value of the parameter under healthy conditions; This indicates the maximum value of the parameter in a healthy state; Calculate the comprehensive health index In the formula, This represents the total number of feature parameters. This represents the fusion weights of each parameter, and satisfies... ; Construct a multidimensional health status assessment matrix, including a response sensitivity index. Pressure stability index Stiffness degradation index and damping characteristic index In the formula, , , , , , , , , This represents the normalized value of each corresponding parameter; , , , , , , , , This represents the sub-weighting coefficients within each index.

[0014] Preferably, it further includes a dynamic compensation control module, wherein the dynamic compensation control module is configured as follows: Based on the roll gap health status output by the roll gap health assessment module and the transient response waveform predicted by the transient response feature extraction module, a feedforward compensation signal is generated and superimposed on the set value of the roll gap pressure controller. The formula for the feedforward compensation signal is as follows: The set value after superposition is In the formula, For feedforward compensation signal; This represents the estimated value of the moisture content-roll gap pressure transfer function model; A nominal model representing the object under roll gap pressure control; A time-series estimate representing the magnitude of abrupt changes in moisture content; This is the pressure setting value after superposition; Set as the reference pressure value; The dynamic compensation control module also includes a controller parameter adaptive adjustment unit, configured to dynamically adjust the controller proportional gain according to the current roll gap health status. and adjustment range limits In the formula, This represents the adaptively adjusted proportional gain. This is the nominal proportional gain; The minimum gain coefficient; The current comprehensive health index; This indicates the maximum adjustment range after adaptive adjustment; This is the nominal maximum adjustment range.

[0015] Preferably, the graded early warning signal output by the roll gap health assessment module is divided into four levels based on the comprehensive health index: ; Green indicates maintaining the regular monitoring frequency; yellow indicates increasing the monitoring frequency and recommending inspections during the planned shutdown window; orange indicates issuing an early warning and recommending maintenance within the scheduled time; and red indicates issuing an emergency warning and recommending immediate shutdown for inspection.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a roller gap pressure transient response prediction and dynamic compensation system driven by moisture content mutation events, achieving accurate prediction of the roller gap health status of a wet tissue folding machine. This solves the technical problem in existing technologies that cannot utilize moisture content mutation impacts for equipment condition diagnosis. Specifically, this invention first integrates the liquid supply system status signal and paper physical property signal to detect moisture content mutation events online and accurately record the event trigger time and mutation amplitude, transforming moisture content fluctuations, which were originally considered disturbances during production, into active diagnostic excitations. Then, starting from the event trigger time, the transient response waveform of the roller gap pressure is continuously acquired. Response delay time, peak overshoot, response time, decay oscillation period, steady-state error, and characteristic parameters identified based on a second-order ARMA model, such as static gain, pure lag time, damping ratio, and undamped natural frequency, are extracted from both the time and model domains. These parameters have a clear physical correspondence with the mechanical inertia, damping characteristics, and stiffness degradation degree of the roller gap, and can sensitively reflect the gradual evolution process of the roller gap from health to failure. Based on these parameters, a comprehensive health index is calculated and graded early warning signals are output. Warnings can be issued hours to days before the roll gap completely fails, realizing a fundamental shift from passive maintenance to predictive maintenance and significantly reducing the risk of unplanned downtime.

[0017] 2. This invention also significantly improves the accuracy and robustness of moisture content mutation event detection through multi-source signal adaptive weighted fusion and false alarm suppression mechanisms, thereby enhancing the reliability of roll gap state prediction. Specifically, this invention not only collects liquid supply system status signals such as liquid supply pump outlet pressure, nozzle inlet pressure, and liquid supply flow rate, but also integrates paper physical property signals such as paper thickness, friction coefficient, and paper tension to construct a multi-dimensional, multi-redundant detection signal source. For each signal source, an anomaly probability calculation method based on the normal cumulative distribution function is adopted, and fusion weights are dynamically allocated according to the signal-to-noise ratio, detection result consistency, and historical accuracy, so that high-reliability signal sources dominate the fusion process, while low-reliability signal sources automatically have their weights reduced. Simultaneously, through a dual mechanism of continuous verification (requiring anomaly scores to continuously exceed a predetermined threshold time) and minimum time interval filtering, transient noise from sensors, electrical interference, and repeated triggering of the same physical event are effectively eliminated. In the complex production environment of wet tissue folding machines, characterized by high humidity, strong vibration, and electromagnetic interference, this method significantly reduces the false alarm and missed detection rates of moisture content mutation events, providing a stable and reliable event triggering benchmark for subsequent transient response analysis and ensuring the accuracy and reproducibility of the prediction results.

[0018] 3. This invention also achieves adaptive optimization of the control strategy based on the prediction of the roll gap health status through the collaborative design of dynamic compensation control and health-sensing controller, further extending the service life of the equipment and improving the consistency of folding quality. Specifically, this invention generates a feedforward compensation signal based on the online identified moisture content-roll gap pressure transfer function model and superimposes it onto the setpoint of the roll gap pressure controller. This feedforward compensation signal has the same magnitude and opposite direction to the impact of sudden changes in moisture content on the roll gap pressure, and can actively offset the impact before the disturbance occurs. Compared with traditional feedback control (which can only respond after the disturbance occurs), it significantly suppresses pressure overshoot and oscillation in the transient process. At the same time, this invention feeds back the roll gap health assessment results to the controller parameter adjustment unit in real time: when the comprehensive health index is high, the controller maintains a high proportional gain and a large adjustment range to pursue the best control performance; when the health index decreases, the controller automatically reduces the gain and adjustment range to avoid applying excessive stress when the equipment is vulnerable. This "health-sensing control" strategy effectively reduces the additional wear caused by drastic pressure fluctuations to the roll gap while ensuring folding quality, extends the effective service life of the roll gap, and achieves synergistic optimization of predictive maintenance and adaptive control. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall system framework of the present invention; Figure 2 This is a schematic diagram of the moisture content mutation event detection process of the present invention; Figure 3 This is a schematic diagram of the transient response feature extraction process of the present invention; Figure 4 This is a schematic diagram of the roll gap health assessment and early warning process of the present invention; Figure 5 This is a schematic diagram of the dynamic compensation control process of the present invention. Detailed Implementation

[0020] like Figures 1 to 5 As shown, the present invention relates to a fault prediction system for tissue paper production equipment based on time-series data analysis, used for fault prediction of the roller gap state of a wet tissue folding machine, comprising: The moisture content mutation event detection module is configured to detect moisture content mutation events online by integrating the liquid supply system status signal and the paper physical property signal, and to record the event trigger time and mutation magnitude when the moisture content mutation event is detected; In embodiments of the present invention, the moisture content mutation event detection module detects moisture content mutation event types including liquid supply pump start-stop events, liquid supply pump wear events, liquid filling nozzle blockage events, liquid filling nozzle wear events, raw paper roll replacement events, and liquid supply pressure fluctuation events. Each event type is identified and classified through the abnormal patterns of each signal source.

[0021] In an embodiment of the present invention, the moisture content mutation event detection module includes: The liquid supply pump status monitoring unit is configured to collect the outlet pressure signal, drive motor current signal and pump speed signal of the liquid supply pump, and detect sudden changes in water content caused by the start-up or shutdown of the liquid supply pump or pump body abnormality by performing differential analysis on the outlet pressure signal. The nozzle status monitoring unit is configured to collect the inlet pressure signals of multiple liquid dispensing nozzles arranged along the width of the paper web, and detect sudden changes in the lateral distribution of moisture content caused by nozzle blockage or wear by analyzing the consistency and dispersion of the inlet pressure of each nozzle. The liquid supply flow monitoring unit is configured to collect the liquid supply flow signals of the main liquid supply pipe and branch pipes, and identify abnormal flow events in the liquid supply system by performing first-order differential analysis and cumulative sum control chart change point detection on the liquid supply flow signals.

[0022] In an embodiment of the present invention, the moisture content abrupt change event detection module further includes: The paper thickness monitoring unit is configured to continuously acquire paper thickness signals through a laser displacement sensor arranged at the entrance of the folding machine, and to detect sudden changes in thickness caused by changes in moisture content by performing differential analysis on the paper thickness signals. The paper friction coefficient estimation unit is configured to estimate the current friction coefficient online by collecting the torque signal of the drive motor of the traction roller and the paper tension signal, and to detect the sudden change in friction characteristics caused by the change in moisture content by performing differential analysis on the friction coefficient. The paper tension change point detection unit is configured to continuously collect paper tension signals through tension sensors arranged between each station of the folding machine, and to perform time-series change point analysis on the tension signals using a Bayesian change point detection algorithm to detect tension anomalies caused by sudden changes in moisture content.

[0023] In an embodiment of the present invention, the Bayesian variable point detection algorithm is specifically implemented as follows: Let the paper tension signal sequence be... Assume there exists a point of change at any given time, and the tension signals before and after the point of change follow the mean values ​​of... and The distributions are normal and have the same variance. A Bayesian framework is used to calculate the variance at each time step. Posterior probability of a change point ,in This represents the time step since the last change point. When the posterior probability exceeds a preset threshold (e.g., 0.7), that moment is determined to be a change point, indicating an abnormal tension caused by a sudden change in moisture content. The recursive formula for calculating the posterior probability is: .

[0024] In an embodiment of the present invention, the moisture content abrupt change event detection module further includes an adaptive weighted fusion unit, which is configured as follows: Calculate the probability of anomaly for each detected signal source at the current moment; The fusion weights of each signal source are dynamically allocated based on the signal-to-noise ratio, consistency of detection results, and historical detection accuracy of each signal source. Calculate the weighted fusion anomaly score. When the weighted fusion anomaly score exceeds the preset fusion threshold, a moisture content mutation event is confirmed to have occurred. Record the event trigger time and the mutation magnitude estimated based on the change in paper thickness.

[0025] Among them, each detection signal source at the current moment The probability of an anomaly is calculated using the following formula: In the formula: For the first Each signal source at time The probability of an anomaly; The cumulative distribution function of the standard normal distribution; For the first Each signal source at time The measured value; For the first The average value of each signal source under normal conditions; For the first The standard deviation of a signal source under normal conditions; Operational logic interpretation: Calculate the first... Each signal source at the current moment The probability of an anomaly is calculated by first determining the absolute value of the deviation between the current measurement and the normal mean, dividing it by the standard deviation to obtain the standardized deviation, and then mapping this deviation to a probability value between 0 and 1 using the cumulative distribution function of the standard normal distribution. The larger the deviation, the closer the probability of an anomaly is to 1.

[0026] The fusion weights of each signal source are dynamically allocated using the following formula: In the formula: For the first Each signal source at time The fusion weight; For the first The signal-to-noise ratio of each signal source within the current time window; For the first The consistency coefficient of the detection results of one signal source with other signal sources; For the first Historical detection accuracy of individual signal sources; The total number of signal sources; In an embodiment of the invention, the first Each signal source in the current time window Signal-to-noise ratio within The following method is used for calculation: First, calculate the mean of the signal source within the time window. and standard deviation ,but When the signal source is a physical quantity with a non-zero mean, such as pressure or flow rate, the above definition is used; when the signal source is a zero-mean signal, such as vibration, the definition of the ratio of the effective value of the signal to the effective value of the noise is used.

[0027] In an embodiment of the invention, the first Consistency coefficient of each signal source The distance between the anomalous probability of this signal source and the mean anomalous probability of all other signal sources is obtained by calculating the negative correlation distance. The specific formula is as follows: The closer this value is to 1, the more consistent the detection results of this signal source are with other signal sources.

[0028] In embodiments of the invention, the historical detection accuracy of each signal source is... Updates are based on manual feedback or cross-validation results. After each sudden change in moisture content is manually confirmed as a true or false event, the system updates the accuracy of the triggered signal source for that event: if it's a true event, the accuracy of the triggered signal source is adjusted towards 1; if it's a false event, it's adjusted towards 0. The specific update formula is: ,in For learning rate, This represents the verification result of this event (1 for a true event and 0 for a false event).

[0029] Operational logic explanation: The fusion weights of each signal source are dynamically allocated. The weight of each signal source is proportional to the product of its signal-to-noise ratio, consistency coefficient, and historical accuracy, and then normalized by dividing by the sum of the products of all signal sources to ensure that the sum of all weights is 1. Signal sources with higher signal-to-noise ratios contribute more weight; signal sources with consistent detection results with other signal sources have higher weights; and signal sources with historically high detection accuracy also have higher weights.

[0030] The weighted fusion anomaly score is calculated using the following formula: In the formula, For weighted fusion anomaly scores; when The occurrence of a sudden change in moisture content was confirmed at that time. This is the preset fusion threshold.

[0031] Explanation of the operational logic: The weighted fusion anomaly score is calculated by summing the anomaly probabilities of each signal source according to their dynamic weights. The higher the score, the greater the likelihood that multiple signal sources simultaneously point to an anomaly.

[0032] This adaptive weighted fusion method effectively suppresses false alarms caused by single sensor failure or environmental noise by dynamically adjusting the weights of each signal source. Simultaneously, it improves the detection sensitivity and reliability of moisture content mutation events by utilizing consistency verification of multi-source signals. Compared to simple voting methods with fixed weights or single threshold detection methods, this method achieves lower false alarm and false negative rates in the high-noise, high-interference environment of wet wipe production, providing accurate event trigger moments for subsequent transient response analysis.

[0033] The adaptive weighted fusion unit further includes a false alarm suppression subunit, which is configured to: require a state in which the weighted fusion abnormal score exceeds the fusion threshold for a predetermined duration to be confirmed as a valid event, and set a minimum time interval between two events to filter repeated triggering.

[0034] The valid event confirmation criteria are expressed by the following formula: In the formula: Indicates the current time; Indicates the predetermined duration threshold; Indicates time The weighted fusion anomaly score; Indicates the preset fusion threshold; Indicates at time Confirm valid event; Explanation of the operational logic: This expression is a continuous verification condition: it requires that from the current time... Backward At every point in time within the specified time period, the weighted fusion anomaly score consistently exceeds the fusion threshold. Only when this condition is met is the anomaly at point [time value missing]. A valid event has occurred. This condition avoids false alarms triggered by single-point noise spikes.

[0035] The minimum time interval filtering condition between two events is: In the formula: For the first The timing of the confirmation event; For the first The timing of the confirmation event; The minimum time interval threshold; Explanation of operational logic: This expression represents the minimum interval filtering condition: the first... The time of the second confirmation event and the first The time difference between the times of each confirmation event must be greater than or equal to the minimum time interval threshold. If the interval between two events is too short, the later event is considered a duplicate response to the earlier event and is filtered out.

[0036] This false alarm suppression mechanism effectively eliminates sensor transient noise, electrical interference, and multiple triggering of the same physical event through continuous verification and minimum interval filtering. In the complex electromagnetic environment and mechanical vibration background of continuous production in a wet wipe folding machine, this method significantly improves the signal-to-noise ratio of event detection, ensuring that each confirmed moisture content mutation event has true physical significance, and providing a stable and reliable event sequence for subsequent transient response analysis.

[0037] The transient response waveform acquisition module is configured to continuously acquire the timing signal of the roll gap pressure within a predetermined time window, starting from the event trigger time, and generate a transient response waveform of the roll gap pressure. The transient response feature extraction module is configured to extract time-domain feature parameters and model-domain dynamic parameters based on a second-order ARMA model from the transient response waveform of the roll gap pressure. The time-domain feature parameters include response delay time, peak overshoot, response time, damped oscillation period, and steady-state error. The model-domain dynamic parameters include static gain, pure time delay, damping ratio, and undamped natural frequency. In an embodiment of the present invention, the sampling frequency of the transient response waveform of the roll gap pressure is configured to be higher than the Nyquist frequency of the highest frequency component of the roll gap pressure signal, and the length of the predetermined time window is configured to be sufficient to cover the complete transient process from event triggering to the recovery of the roll gap pressure to a steady state.

[0038] In an embodiment of the present invention, the transient response feature extraction module includes a time-domain feature extraction unit, which is configured as follows: Extract the response delay time, which is the time interval from the event triggering time to the first deviation of the roll gap pressure from the pre-event steady-state value exceeding a predetermined deviation threshold; Extract the peak overshoot, which is the relative deviation between the peak value of the roll gap pressure response waveform and the steady-state value before the event; Extract the response time, which is the time required from the moment the event is triggered until the roll gap pressure enters and remains within a predetermined percentage range of the steady-state value; To extract the decaying oscillation period, when the roll gap pressure response waveform exhibits decaying oscillation characteristics, the oscillation period is extracted by filtering the response waveform and detecting the zero crossing point. Extract the steady-state error, which is the relative deviation between the average pressure value after the response waveform enters the steady state and the steady-state value before the event.

[0039] Among them, response delay time Extract using the following formula: In the formula: For response delay time; The moment the event is triggered; For a moment The roller gap pressure; This represents the steady-state pressure value before the event. The predetermined deviation threshold; Explanation of the operational logic: This formula is used to extract the response delay time: at the moment the event is triggered. Then, find the first instance where the pressure deviates from the steady-state value by more than a threshold. The moment , and the time The difference is used as the response delay time.

[0040] Peak overshoot The following formula is used for calculation: In the formula: Indicates peak overshoot (percentage); Indicates the peak pressure of the response waveform. , The predetermined time window length; Explanation of operational logic: This formula is used to extract peak overshoot: First, within the time window... Find the maximum pressure inside Then calculate the percentage of the relative deviation of the peak value from the steady-state pressure.

[0041] Response time Extract using the following formula: In the formula: Indicates response time; This is the verification duration used to confirm the continuation of the steady state; Explanation of the operational logic: This formula is used to extract the response time: finding the first moment. This makes from Starting from the beginning Over a period of time, the pressure remained within ±5% of the steady-state value. and The difference is the response time.

[0042] Decaying oscillation period The following formula is used for calculation: In the formula: The period is the decaying oscillation period; This represents the total number of positive peak values ​​detected. Indicates the first The moment corresponding to each positive peak; Explanation of the operational logic: This formula is used to calculate the damped oscillation period: extract the time of all positive peaks in the response waveform and calculate the average time interval between adjacent peak times.

[0043] steady-state error The following formula is used for calculation: In the formula: Steady-state error (percentage); This represents the average pressure value after the response waveform reaches a steady state. This is the window length used to calculate the steady-state average value.

[0044] Explanation of operational logic: This formula is used to calculate the steady-state error: calculation at the end of the time window. The average pressure over the length is taken as the final steady-state value. Then calculate the percentage of its relative deviation from the pre-event steady-state value.

[0045] This time-domain feature extraction method quantitatively characterizes the dynamic response behavior of roll gap pressure to sudden changes in moisture content from multiple dimensions. Response delay time reflects mechanical inertia and paper buffering capacity; peak overshoot reflects the roll gap's sensitivity to disturbances; response time reflects self-regulation capability; damped oscillation period reflects damping characteristics; and steady-state error reflects irreversible changes in roll gap. These characteristic parameters constitute a set of diagnostic indicators that are highly sensitive to the health status of the roll gap and have clear physical meaning, providing rich quantitative evidence for subsequent health assessments.

[0046] In an embodiment of the present invention, the transient response feature extraction module includes a model domain parameter identification unit, which is configured as follows: The sudden change in moisture content is regarded as a step input, and the transient response waveform of the roll gap pressure is fitted into a second-order ARMA model. The difference equation of the second-order ARMA model includes autoregressive coefficients and moving average coefficients. The theoretical basis for choosing the second-order ARMA model in this system is that the dynamic response of the roller gap pressure in a wet tissue folding machine to a sudden change in moisture content can be physically approximated as a second-order oscillatory system. The mechanical structure of the roller gap (pressure roller-paper-support roller) has mass and elastic elements, and its dynamic behavior can be described by second-order differential equations. Experimental comparisons show that the first-order model cannot accurately describe the overshoot and oscillatory characteristics of the response waveform, while third-order and higher-order models have the risk of overfitting and decreased parameter identification stability. Therefore, the second-order ARMA model is the optimal choice between representation accuracy and computational complexity.

[0047] The model parameters of the second-order ARMA model are estimated online using the recursive least squares method, which includes a forgetting factor to adapt to the time-varying characteristics of the system. The static gain, pure time delay, first time constant, and second time constant are derived from the estimated model parameters. The damping ratio and the undamped natural frequency are calculated based on the first time constant and the second time constant.

[0048] The continuous transfer function form of the second-order ARMA model is as follows: ; The corresponding difference equation is in the form of: ; In the formula: The transfer function from moisture content to roll gap pressure ( domain); This is the Laplace transform of the roll gap pressure; The Laplace transform of moisture content; This is the static gain; This is the pure time delay; , It is a time constant; , These are the autoregressive coefficients; , The moving average coefficient; For a sudden change in water content, step input is used. It is white noise; Explanation of the operational logic: The first equation is a continuous transfer function model, taking the sudden change in moisture content as input and the roll gap pressure as output, and using a second-order plus pure time-delay model to describe the dynamic characteristics of the system. The second equation is the corresponding discrete difference equation form, which facilitates online recursive estimation.

[0049] Online estimation of parameter vectors using recursive least squares method The recursive formula is: ; ; ; In the formula: For parameter estimation vectors; It is the gain vector; To output the measured value ( ); For the regression vector, ; It is the covariance matrix; Forgetting factor ( ); It is the identity matrix; In an embodiment of the present invention, the initial parameter vector of the recursive least squares method is set as follows: The initial covariance matrix is ​​set to ,in It is a 4×4 identity matrix. Using a larger initial covariance matrix value can ensure that the algorithm has a high parameter correction gain in the early stage of startup, so that the parameter estimates can converge to the true values ​​quickly.

[0050] Explanation of the operational logic: Equations three through five are the core recursive formulas of the recursive least squares method: For each new sample, based on the error between the current observed value and the model's predicted value, the gain vector is used to calculate the error. Correct the parameter estimates and update the covariance matrix. Forgetting factor By controlling the decay rate of historical data, the model can track the time-varying characteristics of the system.

[0051] The formula for converting estimated model parameters to transfer function parameters is: ; ; In the formula, The sampling angular frequency; Damping ratio and undamped natural frequency The following formula is used for calculation: ; Explanation of the operational logic: Equations 6 to 8 convert the ARMA parameters estimated from the difference equation into transfer function parameters, and then calculate the damping ratio and the undamped natural frequency.

[0052] Pure delay time The peak position of the input-output cross-correlation function is estimated by analysis, and its calculation formula is as follows: In the formula: For time delay variables; The sampling period.

[0053] Explanation of operational logic: The ninth equation analyzes the input signal. With output signal Find the time delay that maximizes the absolute value of the cross-correlation function. Then multiply by the sampling period to get the pure time delay.

[0054] This model-domain dynamic parameter identification method fits the transient response waveform into a second-order system model with clear physical meaning. The extracted parameters, such as damping ratio, undamped natural frequency, and static gain, directly correspond to the mechanical stiffness and damping characteristics of the roll gap. Compared to purely data-driven black-box methods, this method has stronger interpretability and generalization ability. Through online recursive identification, the system can track the slow degradation process of the roll gap health status in real time, providing reliable dynamic characteristics for remaining life prediction.

[0055] The model domain parameter identification unit is further configured to: record the time evolution sequence of the static gain, pure lag time, damping ratio and undamped natural frequency identified after each moisture content mutation event; extract the trend term and estimate the slope of the time evolution sequence of each parameter; and predict the remaining effective life of the roll gap based on the evolution rate of the parameter trend.

[0056] Among them, the The static gain identified after the event is denoted as . The absolute time of this event is denoted as (Assume the time of the first event is) The time evolution sequence is ; The trend term is extracted using the exponentially weighted moving average method: ; In the formula: Indicates the first The smoothed value of the static gain after the event; Indicates the first Static gain identified after the event; Smoothing factor ( ); Based on the smoothed static gain and its corresponding time, linear regression is used to estimate the degradation slope: In the formula: The estimated slope (degradation rate) represents the static gain trend. The average over time. ; This represents the mean of the smoothed static gain. ; The time evolution of static gain is described using an exponential degradation model: ; Log-based linear regression: , using the smooth sequence Degradation rate estimated by linear regression ; Let the static gain failure threshold be Then the remaining effective lifespan is: In the formula: This is the initial static gain; For the degradation rate (i.e., the above) (normalized value); The failure threshold for static gain; The current moment; The remaining effective lifespan; Explanation of the operational logic: The first formula uses the exponentially weighted moving average method to extract the trend term of the static gain sequence, smoothing short-term random fluctuations and highlighting long-term evolutionary trends. Smoothing factor. Controlling the decay rate of historical data The larger the value, the greater the weight of the current data. The second formula estimates the trend slope through linear regression: the event sequence number... As the independent variable, the smoothed static gain is used as the dependent variable, and the regression slope is calculated as the degradation rate. Equations three and four represent exponential degradation models: assuming the static gain degrades exponentially over time, the logarithm is taken to transform it into a linear regression form, facilitating parameter estimation. Equation five predicts remaining lifespan based on the exponential degradation model: first, the remaining lifespan is calculated from the initial value... Exponential growth to failure threshold The remaining effective lifespan is obtained by subtracting the time already elapsed from the total time required.

[0057] This remaining life prediction method utilizes static gain evolution sequences extracted from multiple abrupt moisture content events to establish a life prediction framework based on an exponential degradation model. The exponentially weighted moving average effectively filters out random errors from single identifications, while the degradation rate estimated by linear regression reflects the long-term trend of roll gap stiffness. This method can provide remaining life predictions hours to days before complete roll gap failure, offering a basis for production planning and maintenance scheduling, and enabling a shift from reactive maintenance to predictive maintenance.

[0058] The roll gap health assessment module is configured to calculate a comprehensive health index based on the extracted time-domain feature parameters and model-domain dynamic parameters, and output a graded early warning signal according to the comprehensive health index.

[0059] In an embodiment of the present invention, the roll gap health assessment module is configured as follows: The extracted time-domain feature parameters and model-domain dynamic parameters are normalized and mapped to a predetermined health score range. The comprehensive health index is calculated based on the normalized parameters. The comprehensive health index is the result of weighted fusion of the parameters, and its value range represents a continuous state from complete health to complete failure. A multidimensional health status assessment matrix is ​​constructed, which includes a response sensitivity index, a pressure stability index, a stiffness degradation index, and a damping characteristic index. Each index is calculated by combining the corresponding characteristic parameters. Based on the comprehensive health index and multidimensional health status assessment matrix, the health status of the roll gap is divided into multiple levels, and a graded early warning signal corresponding to each level is output.

[0060] The normalization of each feature parameter adopts the range normalization formula: In the formula: For the first Normalized values ​​of each feature parameter; For the first The current values ​​of each feature parameter; This indicates the minimum value of the parameter under healthy conditions; This indicates the maximum value of the parameter in a healthy state; Comprehensive Health Index The following formula is used for calculation: In the formula: A comprehensive health index; This represents the total number of feature parameters. This represents the fusion weights of each parameter, and satisfies... ; The indices in the multidimensional health status assessment matrix are calculated using the following formulas: Response sensitivity index: ; Pressure stability index: ; Stiffness degradation index: ; Damping characteristic index: ; In the formula: The response sensitivity index; It is the pressure stability index; The stiffness degradation index; The damping characteristic index; This represents the normalized value of each corresponding parameter; This represents the sub-weighting coefficients within each index; The first equation, interpreted logically, is for range normalization: mapping the current value of each feature parameter to... The interval is defined as follows: the minimum value of the parameter in a healthy state corresponds to 0, and the maximum value corresponds to 1. The larger the normalized value, the closer the parameter is to a healthy state. The second formula calculates the comprehensive health index: first, calculate the degree to which each parameter deviates from the ideal state ( The weighted sum is then subtracted from 1 to obtain the comprehensive health index. The closer the value is to 1, the healthier the overall health; the closer it is to 0, the closer it is to failure. Formulas three through six construct a multidimensional health assessment matrix, combining characteristic parameters with relevant physical meanings into four comprehensive indices. Each index is a weighted sum of the normalized values ​​of its internal sub-parameters, with the weights reflecting the contribution of each sub-parameter to that health dimension.

[0061] This multidimensional health assessment method combines a single comprehensive health index with four interpretable health dimension indices, providing both a quantitative score of overall health status and retaining diagnostic information for each physical dimension. The responsiveness index reflects the roll gap's response speed to disturbances, the pressure stability index reflects shock resistance and adjustment capability, the stiffness degradation index reflects the degree of mechanical wear, and the damping characteristic index reflects lubrication and friction conditions. This hierarchical assessment framework allows operators not only to know whether the equipment is "healthy," but also "which aspects have degraded," providing a clear direction for precise maintenance.

[0062] In an embodiment of the present invention, the graded early warning signal output by the roll gap health assessment module includes: A green health signal indicates that the overall health index is above the first threshold range, and regular monitoring frequency should be maintained. A yellow warning signal indicates that the overall health index is in the second threshold range. It is recommended to increase the monitoring frequency and conduct a check during the planned shutdown window. An orange warning signal indicates that the overall health index is in the third threshold range, issuing an alert and recommending that maintenance be scheduled within the scheduled time. A red critical signal indicates that the overall health index is below the fourth threshold range, issuing an emergency warning and recommending immediate shutdown for inspection.

[0063] Among them, the comprehensive health index The correspondence between the warning levels and the alert levels is as follows: ; In the formula: It is at the warning level; The comprehensive health index (value range 0~1); , , This is the threshold for grading; In an embodiment of the present invention, a dynamic compensation control module is further included. The dynamic compensation control module is configured to generate a feedforward compensation signal based on the roll gap health status output by the roll gap health assessment module and the transient response waveform predicted by the transient response feature extraction module, and to superimpose the feedforward compensation signal onto the set value of the roll gap pressure controller to offset the dynamic impact of sudden changes in moisture content on the roll gap pressure.

[0064] In an embodiment of the present invention, the roller gap pressure controller adopts a proportional-integral-derivative control structure, and its control law is as follows: ,in For pressure tracking error, For adaptive adjustment of proportional gain, and The pre-calibrated integral and derivative gains (remain constant during system operation). Feedforward compensation signal. The signal is superimposed on the controller output to form the overall control signal. It acts on the roller gap pressure regulating actuator.

[0065] Among them, feedforward compensation signal It is generated using the following formula: In the formula: For feedforward compensation signal; This represents the estimated value of the moisture content-roll gap pressure transfer function model; A nominal model representing the object under roll gap pressure control; A time-series estimate representing the magnitude of abrupt changes in moisture content; The superimposed settings are: In the formula: This is the pressure setting value after superposition; The reference pressure setting value; Explanation of operational logic: The first equation generates a feedforward compensation signal: the identified moisture content-roll gap pressure transfer function model... Divided by the nominal model of the controlled object Then, multiply by the time-series estimate of the moisture content fluctuation amplitude, and take the negative value as the feedforward compensation signal. The design principle of this signal is to make the output of the feedforward path equal in magnitude and opposite in direction to the impact of the disturbance on the system, thereby canceling the disturbance. The second formula calculates the superimposed pressure setpoint: the feedforward compensation signal is superimposed on the reference pressure setpoint to obtain the corrected setpoint, which is then input to the roll gap pressure controller.

[0066] This dynamic compensation control method utilizes an online-identified transfer function model to predict the impact of sudden changes in moisture content on roll gap pressure in advance, and generates a reverse compensation signal that is superimposed on the controller setpoint. Compared to traditional feedback control (which only responds after a disturbance occurs), this method achieves feedforward active compensation for moisture content shocks, effectively suppressing pressure overshoot and oscillations during transient processes. This not only improves the consistency of folding quality but also reduces the additional wear on the roll gap caused by drastic pressure fluctuations, achieving synergistic optimization of prediction and control.

[0067] In embodiments of the present invention, to ensure the engineering feasibility of the feedforward compensator, when The relative order is higher than At that time, A low-pass filter is then connected in series. ,in This is the relative order difference. This is the filter time constant (taken as 0.1 to 0.2 times the system's dominant time constant). The actual feedforward compensation signal used is... This filter makes the feedforward compensator a true rational function, satisfying the physical realizability condition.

[0068] The dynamic compensation control module further includes a controller parameter adaptive adjustment unit, which is configured to dynamically adjust the control gain and adjustment range of the roll gap pressure controller according to the current roll gap health status. When the roll gap health index decreases, the control gain and adjustment range are reduced to extend the remaining roll gap life in a gentle adjustment manner.

[0069] Among them, the controller proportional gain Adaptively adjust according to the following formula: In the formula: This represents the adaptively adjusted proportional gain. This is the nominal proportional gain; Minimum gain coefficient ( ); The current comprehensive health index; The controller's adjustment range limit is dynamically adjusted according to the following formula: In the formula: This indicates the maximum adjustment range after adaptive adjustment; This is the nominal maximum adjustment range; Explanation of operational logic: First formula: Adaptive adjustment of the controller's proportional gain: When the comprehensive health index... At higher values ​​(close to 1), the proportional gain approaches the nominal value. The controller maintains high responsiveness; when As the gain decreases, the proportional gain decreases linearly, reaching a minimum of [missing value]. . This ensures minimal controllability even when the equipment is nearing failure. The second feature is adaptive adjustment of the adjustment range limit: the maximum permissible adjustment range is proportional to the current health index. The lower the health index, the smaller the permissible pressure adjustment range, forcing the controller to operate more gently and avoiding applying excessive stress to the equipment when it is vulnerable.

[0070] This adaptive adjustment method for controller parameters directly integrates the equipment's health status into the control strategy, achieving "health-aware control." When the roll gap is healthy, the controller maintains high gain and a large adjustment range to pursue optimal control performance; when the roll gap is close to failure, the controller automatically switches to a conservative mode, sacrificing some control accuracy to extend the equipment's lifespan. This health-adaptive control strategy effectively avoids the dilemma of "making drastic adjustments when the equipment is about to fail, leading to premature scrapping," significantly extending the effective service life of the roll gap.

[0071] In embodiments of the present invention, a data storage and traceability module is also included. The data storage and traceability module is configured to store the event trigger time, mutation amplitude, corresponding roll gap pressure transient response waveform, extracted feature parameters and output health assessment results of each moisture content mutation event, forming a historical evolution database of roll gap health status for offline analysis and prediction model updates.

[0072] Explanation of operational logic: This formula is a piecewise function for tiered early warning: based on the comprehensive health index. The numerical range of the data divides the roll gap health status into four discrete levels. Each level corresponds to a different warning level and maintenance recommendation: green for healthy, yellow for attention, orange for warning, and red for critical. This hierarchical warning mechanism maps continuous equipment health indices to four discrete levels with clear maintenance implications, allowing on-site operators to quickly understand the current equipment status and take appropriate measures. Green indicates routine monitoring, yellow suggests checking during planned shutdowns, orange requires timely maintenance, and red triggers emergency shutdowns. This clear level division avoids the limitations of the traditional binary "normal / fault" classification in threshold alarms, achieving a leap from "post-event alarms" to "pre-event hierarchical predictive warnings," effectively reducing the risk of unplanned shutdowns.

[0073] In a preferred embodiment of the present invention, the parameters are set as follows: a predetermined time window seconds; sampling frequency (corresponding sampling period) (seconds); forgetting factor Smoothing factor Minimum gain coefficient Predetermined deviation threshold Verification duration Seconds; steady-state average window seconds; scheduled duration seconds; minimum time interval seconds; fusion threshold The above parameters can be calibrated offline or adaptively adjusted online according to the specific mechanical characteristics and production conditions of the wet wipe folding machine.

[0074] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A fault prediction system for tissue paper production equipment based on time-series data analysis, characterized in that, include: The moisture content mutation event detection module is configured to detect moisture content mutation events online by integrating the liquid supply system status signal and the paper physical property signal, and to record the event trigger time and mutation magnitude when the moisture content mutation event is detected; The transient response waveform acquisition module is configured to continuously acquire the timing signal of the roll gap pressure within a predetermined time window, starting from the event trigger time, and generate a transient response waveform of the roll gap pressure. The transient response feature extraction module is configured to extract time-domain feature parameters and model-domain dynamic parameters based on a second-order ARMA model from the transient response waveform of the roll gap pressure. The time-domain feature parameters include response delay time, peak overshoot, response time, damped oscillation period, and steady-state error. The model-domain dynamic parameters include static gain, pure time delay, damping ratio, and undamped natural frequency. The roll gap health assessment module is configured to calculate a comprehensive health index based on the extracted time-domain feature parameters and model-domain dynamic parameters, and output a graded early warning signal according to the comprehensive health index.

2. The fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, The moisture content abrupt event detection module includes: The liquid supply pump status monitoring unit is configured to collect the outlet pressure signal, drive motor current signal and pump speed signal of the liquid supply pump, and detect sudden changes in water content caused by the start-up or shutdown of the liquid supply pump or pump body abnormality by performing differential analysis on the outlet pressure signal. The nozzle status monitoring unit is configured to collect the inlet pressure signals of multiple liquid dispensing nozzles arranged along the width of the paper web, and detect sudden changes in the lateral distribution of moisture content caused by nozzle blockage or wear by analyzing the consistency and dispersion of the inlet pressure of each nozzle. The liquid supply flow monitoring unit is configured to collect the liquid supply flow signals of the main liquid supply pipe and branch pipes, and identify abnormal flow events in the liquid supply system by performing first-order differential analysis and cumulative sum control chart change point detection on the liquid supply flow signals. The paper thickness monitoring unit is configured to continuously acquire paper thickness signals through a laser displacement sensor arranged at the entrance of the folding machine, and to detect sudden changes in thickness caused by changes in moisture content by performing differential analysis on the paper thickness signals. The paper friction coefficient estimation unit is configured to estimate the current friction coefficient online by collecting the torque signal of the drive motor of the traction roller and the paper tension signal, and to detect the sudden change in friction characteristics caused by the change in moisture content by performing differential analysis on the friction coefficient. The paper tension change point detection unit is configured to continuously collect paper tension signals through tension sensors arranged between each station of the folding machine, and to perform time-series change point analysis on the tension signals using a Bayesian change point detection algorithm to detect tension anomalies caused by sudden changes in moisture content.

3. The fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, The moisture content abrupt change event detection module further includes an adaptive weighted fusion unit, which is configured as follows: Calculate the anomaly probability of each detected signal source at the current moment. The anomaly probability is expressed by the formula... Calculate, where, For the first Each signal source at time The probability of an anomaly; The cumulative distribution function of the standard normal distribution; For the first Each signal source at time The measured value; , For the first The mean and standard deviation of each signal source under normal conditions; The fusion weights are dynamically assigned based on the signal-to-noise ratio of each signal source, the consistency of detection results, and the historical detection accuracy. The weight calculation formula is as follows: In the formula, For the first Each signal source at time The fusion weights; For the first The signal-to-noise ratio of each signal source within the current time window; For the first The consistency coefficient of the detection results of one signal source with other signal sources; For the first Historical detection accuracy of individual signal sources; The total number of signal sources; Calculate the weighted fusion anomaly score ,when Exceeding the preset fusion threshold The occurrence of a sudden change in moisture content was confirmed in time, and the trigger time of the event and the magnitude of the change estimated based on the change in paper thickness were recorded.

4. The fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 3, characterized in that, The adaptive weighted fusion unit further includes a false alarm suppression subunit, which is configured as follows: A state where the weighted fusion anomaly score exceeds the fusion threshold must remain for a predetermined duration before being confirmed as a valid event. The confirmation condition is as follows: In the formula, Indicates the current time; Indicates the predetermined duration threshold; Indicates time The weighted fusion anomaly score; Indicates at time Confirm valid event; In the formula, For the first The timing of the confirmation event; For the first The timing of the confirmation event; This is the minimum time interval threshold.

5. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, The transient response feature extraction module includes a time-domain feature extraction unit, which is configured as follows: Extract response delay time In the formula, The moment the event is triggered; For a moment The roller gap pressure; This represents the steady-state pressure value before the event. The predetermined deviation threshold; Extracting peak overshoot In the formula, Indicates the peak pressure of the response waveform; Extract response time In the formula, This is the verification duration used to confirm the continuation of the steady state; Extracting the decaying oscillation period In the formula, This represents the total number of positive peak values ​​detected. Indicates the first The moment corresponding to each positive peak; Extracting steady-state error In the formula, This represents the average pressure value after the response waveform reaches a steady state.

6. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, The transient response feature extraction module includes a model domain parameter identification unit, which is configured as follows: Treating the abrupt change in moisture content as a step input, the transient response waveform of the roll gap pressure is fitted to a second-order ARMA model, with the continuous transfer function form as follows: The corresponding difference equation is In the formula, This is the transfer function from moisture content to roll gap pressure; This is the Laplace transform of the roll gap pressure; The Laplace transform of moisture content; This is the static gain; This is the pure time delay; , It is a time constant; , These are the autoregressive coefficients; , The moving average coefficient; For a sudden change in water content, step input is used. It is white noise; The parameter vector is estimated online using the recursive least squares method. The recursive formula is as follows: ； ； ； In the formula, For parameter estimation vectors; It is the gain vector; To output the measured value; For regression vectors; It is the covariance matrix; Forgetting factor; It is the identity matrix; The static gain is obtained from the estimated model parameters. The damping ratio is calculated by multiplying and summing the time constants. and undamped natural frequency ; Pure delay time Estimation of peak location using cross-correlation function: In the formula, For time delay variables; The sampling period.

7. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 6, characterized in that, The model domain parameter identification unit is further configured to: Record the time evolution sequence of the static gain identified after each abrupt change in moisture content, and extract the trend term using the exponentially weighted moving average method: In the formula, Indicates the first The smoothed value of the static gain after the event; Indicates the first Static gain identified after the event; It is a smoothing factor; Degradation slope estimated by linear regression In the formula, The estimated slope representing the static gain trend; The average over time; This represents the mean of the smoothed static gain; Predicting remaining effective lifespan In the formula, This is the initial static gain; The degradation rate; The failure threshold for static gain; This refers to the current moment.

8. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, The roll gap health assessment module is configured as follows: The extracted time-domain feature parameters and model-domain dynamic parameters are normalized by range: In the formula, For the first Normalized values ​​of each feature parameter; For the first The current values ​​of each feature parameter; This indicates the minimum value of the parameter under healthy conditions; This indicates the maximum value of the parameter in a healthy state; Calculate the comprehensive health index In the formula, This represents the total number of feature parameters. This represents the fusion weights of each parameter, and satisfies... ; Construct a multidimensional health status assessment matrix, including a response sensitivity index. Pressure stability index Stiffness degradation index and damping characteristic index In the formula, , , , , , , , , This represents the normalized value of each corresponding parameter; , , , , , , , , This represents the sub-weighting coefficients within each index.

9. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 1, characterized in that, It also includes a dynamic compensation control module, which is configured as follows: Based on the roll gap health status output by the roll gap health assessment module and the transient response waveform predicted by the transient response feature extraction module, a feedforward compensation signal is generated and superimposed on the set value of the roll gap pressure controller. The formula for the feedforward compensation signal is as follows: The set value after superposition is In the formula, For feedforward compensation signal; This represents the estimated value of the moisture content-roll gap pressure transfer function model; A nominal model representing the object under roll gap pressure control; A time-series estimate representing the magnitude of abrupt changes in moisture content; This is the pressure setting value after superposition; Set as the reference pressure value; The dynamic compensation control module also includes a controller parameter adaptive adjustment unit, configured to dynamically adjust the controller proportional gain according to the current roll gap health status. and adjustment range limits In the formula, This represents the adaptively adjusted proportional gain. This is the nominal proportional gain; The minimum gain coefficient; The current comprehensive health index; This indicates the maximum adjustment range after adaptive adjustment; This is the nominal maximum adjustment range.

10. A fault prediction system for tissue paper production equipment based on time-series data analysis according to claim 8, characterized in that, The graded early warning signal output by the roll gap health assessment module is divided into four levels based on the comprehensive health index: ； Green indicates maintaining the regular monitoring frequency; yellow indicates increasing the monitoring frequency and recommending inspections during the planned shutdown window; orange indicates issuing an early warning and recommending maintenance within the scheduled time; and red indicates issuing an emergency warning and recommending immediate shutdown for inspection.

Images