A fault early warning method and system for a fully automatic flatbed die-cutting machine

Abstract

This invention relates to the field of die-cutting machine monitoring technology, and discloses a fault early warning method and system for a fully automatic flatbed die-cutting machine. The method includes: synchronously acquiring stator current signals and rotor speed signals to obtain a continuous operating data stream; inputting the continuous operating data stream into a dynamic state observer, using observation errors for parameter adaptive correction, and outputting a viscous friction coefficient estimation sequence; extracting an effective friction coefficient sequence from the viscous friction coefficient estimation sequence, and extracting a speed signal segment from the continuous operating data stream; performing frequency domain transformation on the speed signal segment to obtain speed ripple characteristics, and simultaneously performing linear fitting on the effective friction coefficient sequence to obtain friction degradation rate characteristics; constructing a health feature vector based on the speed ripple characteristics and the friction degradation rate characteristics, calculating the Mahalanobis distance, and generating a fault early warning command. This method can achieve accurate identification and early warning of early, minor faults in equipment.

Description

Technical Field

[0001] This invention relates to the field of die-cutting machine monitoring technology, and in particular to a fault early warning method and system for a fully automatic flatbed die-cutting machine. Background Technology

[0002] Currently, fault prediction and health diagnosis of fully automatic flatbed die-cutting machines are directly related to the operating efficiency of the production line and the consistency of finished product quality. As packaging technology develops towards higher speed and precision, die-cutting machines are constantly operating under extreme conditions of high-frequency start-stop and heavy-load cutting, and their motor drive systems and precision transmission mechanisms face severe wear and fatigue challenges.

[0003] In existing technologies, many die-cutting machine condition monitoring solutions rely on external physical sensors or simple electrical threshold alarms. These solutions typically install vibration accelerometers or temperature sensors at locations such as motor bearing housings, crankshafts, and connecting rods, comparing the collected vibration amplitude or surface temperature data with preset fixed safety thresholds; or they directly monitor the effective value of the motor driver current to prevent overload. Once the monitored value exceeds a set upper limit, the system triggers an alarm or shutdown command. However, because the evolution of mechanical faults usually involves a long early degradation process, and die-cutting conditions have highly nonlinear characteristics, the above methods have significant limitations.

[0004] Existing technologies suffer from problems such as large delays and low accuracy in early warning of die-cutting machine malfunctions. Summary of the Invention

[0005] This invention provides a fault early warning method and system for a fully automatic flatbed die-cutting machine, in order to solve the technical problems of large lag and low accuracy in fault early warning of existing die-cutting machines.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a fault early warning method for a fully automatic flatbed die-cutting machine, comprising: The stator current signal and rotor speed signal of the die-cutting machine motor drive system are continuously and synchronously acquired through multiple channels to obtain a continuous running data stream. A dynamic state observer based on the motor torque balance equation is constructed. The continuous running data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream is calculated. The observation error is used to perform parameter adaptive feedback correction and output a continuous viscous friction coefficient estimation sequence. The working cycle of the die-cutting machine is identified based on the peak or trough feature points of the rotor speed signal. The non-die-cutting work interval within each working cycle is locked, and numerical segments corresponding to the non-die-cutting work interval are extracted from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. The rotor speed signal segment corresponding to the non-die-cutting work interval is extracted from the continuous running data stream. The rotor speed signal segment is subjected to frequency domain transformation processing to obtain the spectral amplitude data within the preset monitoring frequency band and obtain the speed ripple characteristics. The effective friction coefficient sequence is subjected to sliding window linear fitting, and the slope of the fitted line is extracted to obtain the friction degradation rate characteristics. A device health feature vector is constructed using the rotational speed ripple feature and the friction degradation rate feature as feature elements. The Mahalanobis distance of the device health feature vector relative to the preset device health reference vector is calculated. If the Mahalanobis distance exceeds the preset safety threshold, a fault warning command is generated.

[0007] Secondly, the present invention provides a fault early warning system for a fully automatic flatbed die-cutting machine, comprising: The data acquisition module is used to continuously and synchronously acquire the stator current signal and rotor speed signal of the die-cutting machine motor drive system through multiple channels to obtain a continuous running data stream; The state observation module is used to construct a dynamic state observer based on the motor torque balance equation. The continuous running data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream is calculated. The observation error is used to perform parameter adaptive feedback correction and output a continuous viscous friction coefficient estimation sequence. The interval locking module is used to identify the working cycle of the die-cutting machine based on the peak or trough feature points of the rotor speed signal, lock the non-die-cutting work interval within each working cycle, and extract the numerical segment corresponding to the non-die-cutting work interval from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. The ripple extraction module is used to extract the rotor speed signal segment corresponding to the non-die-cutting work interval from the continuous running data stream, perform frequency domain transformation processing on the rotor speed signal segment, obtain the spectral amplitude data within the preset monitoring frequency band, and obtain the speed ripple characteristics. The degradation analysis module is used to perform sliding window linear fitting on the effective friction coefficient sequence, extract the slope of the fitted line, and obtain the friction degradation rate characteristics. The early warning decision module is used to construct a device health feature vector using the rotational speed ripple feature and the friction degradation rate feature as feature elements, calculate the Mahalanobis distance of the device health feature vector relative to the preset device health benchmark vector, and generate a fault early warning command if the Mahalanobis distance exceeds the preset safety threshold.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs a dynamic state observer based on the motor torque balance equation, uses observation errors for adaptive feedback correction of parameters, and outputs a continuous sequence of estimated viscous friction coefficients. Compared with traditional passive monitoring methods that rely on vibration or temperature thresholds, this invention utilizes the intrinsic coupling relationship between the electromagnetic torque of the motor and the mechanical load to deeply decouple and quantify the core parameter characterizing the health state of the machine—the viscous friction coefficient—from the electrical signal. This enables the system to keenly detect subtle changes in friction characteristics caused by grease drying, slight wear of guide rails, etc., before substantial physical damage or severe vibration occurs in mechanical components, thereby overcoming the lag of traditional monitoring methods and achieving "non-destructive perception" and accurate early warning of early hidden faults in die-cutting machines.

[0009] (2) This invention identifies the working cycle based on the characteristics of the rotation speed signal and eliminates the time period corresponding to the sudden change in die-cutting pressure, accurately locking the non-die-cutting work interval to extract effective data. Addressing the pain point of flatbed die-cutting machines generating huge periodic impact loads at the moment of paper cutting, which easily overwhelm weak fault characteristics, this invention innovatively introduces a "cycle segmentation and working condition avoidance" mechanism. This avoids the high-noise die-cutting work segment and focuses on extracting features during the relatively stable feeding or idle stages. This refined data screening strategy based on the dynamic characteristics of the working conditions significantly improves the signal-to-noise ratio of fault features, effectively avoids false alarms caused by normal process impacts, and ensures the extremely high robustness of the early warning model under high-speed, heavy-load, and complex working conditions.

[0010] (3) This invention constructs a multi-dimensional equipment health feature vector and calculates the Mahalanobis distance by fusing the rotational speed ripple characteristics in the frequency domain and the frictional degradation rate characteristics in the time domain. Among them, the rotational speed ripple characteristics can sensitively reflect short-term sudden mechanical failures such as rotor imbalance and coupling loosening, while the frictional degradation rate characteristics can accurately characterize long-term gradual performance degradation such as lubrication failure and component cumulative wear. By orthogonally fusing the features of these two dimensions through the Mahalanobis distance algorithm, not only are the differences in magnitude between different physical dimensions (rotational speed fluctuation amplitude and frictional change rate) eliminated, but the statistical correlation between feature variables is also fully utilized, thereby realizing a comprehensive and objective evaluation of the equipment health status and significantly reducing the risk of missed reports caused by monitoring a single indicator. Attached Figure Description

[0011] Figure 1This is a schematic diagram of a fault early warning method for a fully automatic flatbed die-cutting machine provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a fault early warning system for a fully automatic flatbed die-cutting machine provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a fault early warning method for a fully automatic flatbed die-cutting machine, comprising the following steps: S11, perform multi-channel continuous synchronous acquisition of stator current signal and rotor speed signal of the die-cutting machine motor drive system to obtain continuous running data stream; S12, construct a dynamic state observer based on the motor torque balance equation, continuously input the continuous running data stream as input to the dynamic state observer, calculate the observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream, and use the observation error to perform parameter adaptive feedback correction, and output a continuous viscous friction coefficient estimation sequence. S13, based on the peak or trough feature points of the rotor speed signal, identify the working cycle of the die-cutting machine, lock the non-die-cutting work interval within each working cycle, and extract the numerical segment corresponding to the non-die-cutting work interval from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. S14, extract the rotor speed signal segment corresponding to the non-die-cutting work interval from the continuous running data stream, perform frequency domain transformation processing on the rotor speed signal segment, obtain the spectral amplitude data in the preset monitoring frequency band, and obtain the speed ripple characteristics. S15, Perform sliding window linear fitting on the effective friction coefficient sequence, extract the slope of the fitted line, and obtain the friction degradation rate characteristics. S16, construct a device health feature vector using the rotational speed ripple feature and the friction degradation rate feature as feature elements, calculate the Mahalanobis distance of the device health feature vector relative to the preset device health reference vector, and generate a fault warning command if the Mahalanobis distance exceeds the preset safety threshold.

[0014] In step S11, the stator current signal and rotor speed signal of the die-cutting machine motor drive system are continuously and synchronously acquired through multiple channels to obtain a continuous operating data stream, including: Acquire the analog current signal and instantaneous speed pulse signal of the stator winding at the motor drive port of the die-cutting machine; A nanosecond-level sampling clock is used to perform synchronous analog-to-digital conversion on the analog current signal and the instantaneous speed pulse signal, and output a discrete digital sequence with timestamps. The phase lag angle is calculated based on the discrete digital sequence, and the analog current signal and the instantaneous speed pulse signal are time-domain aligned and corrected according to the phase lag angle to generate the continuous running data stream.

[0015] It should be noted that the acquisition of the simulated current signal and instantaneous speed pulse signal of the stator winding at the motor drive port of the die-cutting machine is achieved using high-precision industrial-grade sensors. Specifically, a Hall effect current sensor is connected to the three-phase windings of the motor stator to non-invasively sense the simulated current signal of the stator windings that changes over time. Simultaneously, an incremental optical encoder mounted on the motor shaft captures the A / B phase pulses generated by the motor rotation in real time, i.e., the instantaneous speed pulse signal.

[0016] It should be noted that the synchronous analog-to-digital conversion using a nanosecond-level sampling clock is implemented based on a parallel control architecture of a Field-Programmable Gate Array (FPGA). The FPGA is internally configured with a high-frequency master clock (e.g., 100MHz), which serves as a global time base and simultaneously triggers multiple high-speed analog-to-digital converter (ADC) chips. When the trigger signal arrives, the ADC quantizes and samples the analog current signal; simultaneously, a high-speed counter within the FPGA latches the current encoder pulse count value. The system packages the quantized digital current value with the latched rotational speed count value and adds a timestamp accurate to the nanosecond level of the current global clock count value, forming a strictly corresponding discrete digital sequence. This hardware-level parallel triggering mechanism eliminates the inter-channel timing jitter caused by traditional software polling methods.

[0017] It should be noted that calculating the phase lag angle based on the discrete digital sequence and performing time-domain alignment correction is to eliminate physical delay errors caused by differences in response bandwidth between different types of sensors. Typically, the response speed of a current sensor is faster than that of the low-pass filtered rotational speed signal, resulting in a slight misalignment between the two on the time axis. This will severely affect the estimation accuracy of the subsequent state observer. This step first calculates the current sequence based on the cross-correlation algorithm. With rotational speed sequence Cross-correlation function between : ; in, To calculate the window length, This represents the number of lag points. The cross-correlation calculation is mainly used to find the... Number of time delay points to reach the maximum value Then the phase lag angle (converted to time lag) is calculated. ,in The sampling period is defined as [value]. Subsequently, a fractional delay filter is constructed using Lagrange interpolation to shift the faster signal sequence backward. This ensures strict time-domain alignment with the slower signal sequence. The specific construction process of the fractional delay filter involves adjusting the hysteresis... Decompose into integer parts and decimal part (Right now For the integer part, it is directly implemented by shifting the data index. For the fractional part, it uses... Calculating interpolation coefficients using the first-order Lagrange polynomial : ; Use this coefficient to correct the signal Perform convolution operation This enables high-precision fractional-level time-domain translation, which is then combined to generate the continuously running data stream.

[0018] It is worth noting that the frequency setting of the nanosecond-level sampling clock was determined based on the Nyquist sampling theorem and statistical analysis of the highest electrical frequency of the die-cutting machine motor. Based on historical operating data, the upper limit of the fundamental frequency of the current of the die-cutting machine servo motor is... (For example, 200Hz), considering that the rotational ripple to be analyzed subsequently contains harmonics up to the 50th order, the effective signal bandwidth is 10kHz. To ensure a phase alignment accuracy better than 0.1 degrees, the sampling frequency... Must meet The system selects a standard industrial clock frequency that meets this condition (e.g., ...). The preset value is 5000ns (corresponding to a period of 5000ns). This setting ensures that every tiny transient change in the signal can be completely captured during high-speed die cutting, and that data redundancy will not occur due to oversampling.

[0019] It should be noted that the sampling frequency setting must simultaneously meet two requirements. First, it must satisfy the Nyquist sampling theorem to fully capture the highest frequency component of the signal (e.g., 10kHz), therefore the sampling frequency must be greater than 20kHz. Second, it must meet the time resolution accuracy requirements of the subsequent phase alignment algorithm. The phase alignment accuracy better than 0.1 degrees refers to the fundamental period of the signal (e.g., 1 / 50Hz = 20ms), corresponding to a time accuracy requirement of (0.1 / 360) × 20ms ≈ 5.56μs. To ensure that this timing accuracy is still met after digitization, the sampling period should be significantly smaller than this value. Therefore, selecting a sampling frequency of 200kHz (sampling period of 5μs) is reasonable and provides sufficient margin. The above formula... This is a simplified expression of the principle, the core of which lies in using an extremely high sampling rate ( To ensure the highest frequency signal ( It can also achieve fine phase analysis.

[0020] For example, assume the die-cutting machine operates at 3000 rpm, and the fundamental frequency of the motor current is 100 Hz. The FPGA uses a 100 MHz clock (10 ns level) to generate a trigger signal, driving the ADC to acquire data at a frequency of 200 kHz. The acquired current signal sequence is timestamped. The value at that location is 5.2A, but due to the low-pass filtering effect of the encoder signal processing circuit, the corresponding speed response is... The peak value was reached only at that moment. Through cross-correlation calculation, the system identified 10 sampling points between the two ( The system automatically shifts the current signal sequence backward by 10 points, ensuring that the peak current and peak rotational speed coincide perfectly on the time axis. The corrected data stream is then used as the accurate physical fact and input into the subsequent dynamics observer.

[0021] In step S12, a dynamic state observer based on the motor torque balance equation is constructed. The continuous operating data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous operating data stream is calculated. The observation error is then used for parameter adaptive feedback correction, and a continuous viscous friction coefficient estimation sequence is output, including: Construct a discretized state-space model that includes the viscous friction coefficient as an implicit state variable; The continuous running data stream is input into the discretized state-space model to deduce the prior state estimation vector; Calculate the deviation between the predicted rotational speed value in the prior state estimation vector and the measured rotational speed value in the continuous operation data stream, and calculate the Kalman gain matrix based on the deviation; The prior state estimation vector is weighted and corrected using the Kalman gain matrix to obtain the posterior state estimation vector. The steady-state converged values ​​are extracted from the posterior state estimation vector to obtain the continuous viscous friction coefficient estimation sequence.

[0022] It should be noted that the discretized state-space model, which includes the viscous friction coefficient as an implicit state variable, is constructed based on the Extended Kalman Filter (EKF) theoretical framework. This model first establishes the dynamic differential equations of the main drive motor of the die-cutting machine: ; in, The moment of inertia of the motor and load referred to the motor shaft. Let be the torque constant of the motor. The rotor angular velocity, This refers to the stator current (i.e., the current component in the input continuous operating data stream). The coefficient of viscous friction to be estimated is... This refers to the inherent Coulomb friction torque and steady-state operating load torque of the system (excluding the impact load at the moment of die cutting).

[0023] It is worth noting that the system inherently possesses Coulomb friction torque and steady-state operating load torque. The settings are based on tests of the die-cutting machine's no-load low-speed characteristics and statistical analysis of steady-state operating conditions. Specifically, during the equipment's factory calibration phase, the control motor is kept running at a constant speed (e.g., 5-10 rpm) at extremely low speeds. At this point, the viscous friction term... Approaching zero, the average electromagnetic torque measured by the system is mainly used to overcome mechanical static friction and Coulomb friction, and is calibrated as the basic resistance torque. Simultaneously, the average load level of the equipment during normal feeding and idle phases (avoiding the die-cutting work point) is statistically analyzed and superimposed on the basic resistance torque as the model's input torque. The preset constant value. This setting is intended to eliminate the interference of constant load on state observation, allowing the observer to focus on capturing the viscous friction coefficient, which changes dynamically with speed and reflects the lubrication / wear state. Tiny fluctuations.

[0024] In order to achieve parameter The system will provide real-time observation. Expand to state variables and construct an augmented state vector. Assuming a small sampling interval Intrinsic parameters Slow change (i.e.) Discretization using the Euler method yields the state transition equation: ; Simultaneously establish the observation equation ,in The measured rotational speed (i.e., the rotational speed component in the continuously running data stream). To address the noise issue, this model combines physical mechanisms with data-driven approaches, enabling the inversion of the internally unmeasurable friction coefficient by observing the dynamic relationship between rotational speed and current.

[0025] It should be noted that using observation errors for adaptive parameter feedback correction is the core mechanism for the observer's closed-loop convergence. The specific computation process is as follows: first, a time update (pre-existing prior) is performed, using the posterior estimate from the previous time step... Substitute the values ​​into the state transition equation to calculate the prior state estimate vector at the current time step. That is, predicting the current rotational speed. and coefficient of friction Next, we calculate the Kalman gain and the prior estimate covariance matrix. Combined with the observation matrix and measurement noise covariance Calculate the Kalman gain matrix Finally, a measurement update (feedback correction) is performed to calculate the deviation between the predicted and measured speed values. (i.e., observation error). This error is used to correct the prior estimate. During this process, if the predicted rotational speed is lower than the measured rotational speed, it means that the resistance parameter (friction coefficient) in the model is too large. The Kalman gain will automatically adjust the weights to reduce the resistance. The estimated value is obtained by analogy with the true value, and vice versa, thus achieving adaptive approximation of the true value.

[0026] It is worth noting that the process noise covariance matrix in the discretized state-space model... With measurement noise covariance matrix The determination was based on statistical analysis of historical operating data of the die-cutting machine under no-load conditions at the factory. Specifically, the encoder signal noise variance under static conditions was selected as... The preset value; select the state fluctuation variance of the equipment when it is running stably at rated speed, extract the white noise component through autocorrelation analysis, and diagonalize it as the preset value; The preset value. This setting ensures the observer's ability to suppress random noise while maintaining its sensitivity to changes in the friction coefficient.

[0027] For example, suppose at a certain moment The model predicts the rotational speed. The friction coefficient estimated a priori At this time, the measured rotational speed was collected from the continuously running data stream. Observation error Assume that the components corresponding to the friction coefficient in the currently calculated Kalman gain matrix are... Then the posterior-corrected coefficient of friction The system iterates this process continuously at a frequency of 200kHz, outputting continuously fluctuating values. The sequence serves as the foundational data for subsequent health analysis.

[0028] In step S13, the working cycle of the die-cutting machine is identified based on the peak or trough feature points of the rotor speed signal, the non-die-cutting work interval within each working cycle is locked, and numerical segments corresponding to the non-die-cutting work interval are extracted from the continuous viscous friction coefficient estimation sequence to obtain an effective friction coefficient sequence, including: An extreme value search is performed on the rotor speed signal to identify periodically occurring speed peaks and troughs; The time interval between two adjacent speed peaks is defined as a working cycle. Within each work cycle, the time period corresponding to the sudden change in die-cutting pressure is removed, and the remaining stable operating time period is locked as the non-die-cutting work interval. Based on the start and end timestamps of the non-die-cutting work interval, corresponding data segments are extracted from the continuous viscous friction coefficient estimation sequence and combined to generate the effective friction coefficient sequence.

[0029] It should be noted that the extreme value search and peak point identification of the rotor speed signal are implemented using a differential extreme value detection algorithm based on a sliding window. This algorithm sets a window of length... A sliding window (e.g., 5 sampling points) is used to traverse the rotor speed signal sequence obtained in step S11. For any given time If satisfied and (in If so, then determine These are candidate peak points. To prevent misjudgments caused by minute signal fluctuations, the algorithm introduces a minimum peak distance constraint. (Minimum Peak Distance) Only retains adjacent intervals greater than [a certain value]. The candidate points are used as the final speed peak points. The timestamps of two adjacent peak points are recorded as follows: and Define this time interval This is a complete work cycle.

[0030] It should be noted that the process of eliminating the time period corresponding to the sudden change in die-cutting pressure within each work cycle is based on mechanical phase mapping technology. Because the main drive mechanism of a flatbed die-cutting machine (usually a crank-connecting rod or double-lever mechanism) has strict kinematic constraints, the die-cutting action (i.e., the moment when the moving platform and the stationary platform close to generate enormous pressure) always occurs within a specific phase interval of the motor's rotation cycle. This step first calculates any time within the cycle. Normalized phase : ; Subsequently, a pre-defined die-cutting phase blacklist range is loaded. If the phase at a certain moment If a moment falls within this blacklist range, it is determined that the time corresponds to a sudden change in die-cutting pressure (at which point a huge impact load will cause distortion in the friction coefficient estimation); conversely, if the moment does not fall within this range, it is determined that the time falls within a stable operating period (such as feeding, waste discharge, or idle stroke stages). The system locks the time index set of all stable operating periods into the non-die-cutting work range.

[0031] It is worth noting that the length of the sliding window The minimum peak distance constraint and the die-cut phase blacklist range The determination was based on kinematic calibration and statistical analysis of historical operating data under standard working conditions of the die-cutting machine. Specifically, regarding the window length... The system analyzes the autocorrelation attenuation characteristics of high-frequency noise in historical speed signals and selects the signal sampling frequency. time constant of noise decorrelation The product of the two products is rounded up to the nearest integer as the preset value (e.g., ...). This ensures that the window width just covers the noise spikes but is smaller than the effective width of the actual speed peak. For The system collects the rotational speed waveform of the equipment at its rated speed and calculates the average duration of the rotational speed cycle. Select As a preset value, it effectively filters out local non-periodic fluctuation interference. For the die-cutting phase blacklist interval, the system combines the die-cutting pressure waveform collected by the pressure sensor, statistically analyzes the phase distribution range where the pressure value exceeds 10% of the peak pressure, selects the union of this range and extends it outward by 5% safety margin, and determines it as the die-cutting phase blacklist interval (e.g., ).

[0032] For example, suppose the starting peak time of a work cycle is identified as... The end of the peak time is The cycle duration is The preset die-cutting phase blacklist range is... For each moment within this period The data points, whose phase is .because If a point falls into the blacklist range, it is removed. And for... Data points, phase is It is within a safe range. The system eventually locked. and The two time periods are non-die-cutting work intervals. Based on the start and end timestamps of these time periods, the system accurately extracts the corresponding data segments from the continuous viscous friction coefficient estimation sequence output in step S12 and splices them into a clean effective friction coefficient sequence for subsequent weak fault feature extraction.

[0033] In step S14, a rotor speed signal segment corresponding to the non-die-cutting work interval is extracted from the continuous operation data stream. The rotor speed signal segment undergoes frequency domain transformation to obtain spectral amplitude data within a preset monitoring frequency band, thus obtaining the speed ripple characteristics, including: The rotor speed signal segment is windowed and truncated by applying a Hanning window function. Perform a Fast Fourier Transform on the windowed signal segment to generate a full-band amplitude spectrum; In the full-band amplitude spectrum, the low-frequency segment located between the DC component frequency and the third harmonic frequency is extracted as the preset monitoring frequency band; The root mean square deviation of the amplitude of each frequency component within the preset monitoring frequency band is calculated and used as the rotational speed ripple characteristic.

[0034] It should be noted that applying a Hanning window function to the rotor speed signal segment for windowing truncation is to suppress spectral leakage caused by truncating the non-periodic signal. Since the duration of the non-modular work interval locked in step S13 is limited, directly performing a Fourier transform would lead to spectral sidelobe interference. This step uses the following formula to modify the original speed segment... Weighted processing is performed to obtain the windowed signal. : ; in, This represents the total number of data points in the signal segment. After windowing, the amplitude at both ends of the signal smoothly decays to zero, significantly reducing the impact of high-frequency interference on the low-frequency ripple characteristics.

[0035] Subsequently, the windowing signal was adjusted. Perform a Fast Fourier Transform (FFT) to calculate the complex modulus at each frequency point and divide by . Normalization is performed to obtain the full-band amplitude spectrum that reflects the true physical quantity. .

[0036] It should be noted that extracting the low-frequency segment between the DC component frequency and the third harmonic frequency as the preset monitoring frequency band is based on a physical understanding of the frequency distribution characteristics of rotating machinery faults. First, the system calculates the time average value of this rotational speed signal segment. Convert it to the rotating fundamental frequency (Unit: Hz). Then, the frequency range of the monitoring band is defined as follows: This frequency band covers the characteristic frequencies of early mechanical faults such as rotor imbalance (1st harmonic), misalignment (2nd harmonic), and mechanical loosening (3rd harmonic), while avoiding the high-frequency electromagnetic noise region.

[0037] It should be noted that calculating the root mean square deviation (RMSD) of the amplitude of each frequency component within the preset monitoring frequency band aims to quantify the dispersion of the spectrum (i.e., the significance of fault characteristic peaks). The calculation formula is as follows: ; in, To monitor the number of frequency points within the frequency band, For the first The amplitude at each frequency point This is the arithmetic mean of all amplitudes within this frequency band. When the equipment is healthy, the spectrum of this frequency band is flat (approximately white noise), and the RMSD value is small; when a mechanical failure occurs, significant harmonic spikes will appear in the spectrum, leading to increased data distribution dispersion and a significant increase in the RMSD value.

[0038] It is worth noting that the determination of the three-fold fundamental frequency as the cutoff frequency is based on statistical analysis of a database of failure cases throughout the entire life cycle of die-cutting machines. The system performed frequency domain analysis on over 500 historical failure data points, including bearing wear, coupling loosening, and guide rail deviation, and found that 98.5% of early mechanical failures had energy concentrated within three times the fundamental frequency of the rotational speed. Therefore, selecting three times the fundamental frequency as the preset cutoff line not only covers the vast majority of potential failure characteristics but also minimizes computational redundancy from irrelevant high-frequency signals.

[0039] For example, suppose the duration of a non-die-cut speed signal segment extracted from S13 is 0.5 seconds, and the sampling rate is 2kHz (i.e. (Point). The average rotational speed of this signal segment is 3000 rpm, then the fundamental frequency is... The monitoring frequency band determined by the system is... After processing with a Hanning window FFT, amplitude data at 75 frequency points within this frequency band were obtained. If there is a slight misalignment in the device, a spike with an amplitude of 0.5 appears at 100Hz (2nd harmonic), while the amplitudes at other frequencies fluctuate around 0.01. Calculate the mean of this set of data. Root mean square deviation Compared to calculations under a perfectly healthy state (no spikes) This characteristic value sensitively reflects abnormal fluctuations in rotational speed ripple.

[0040] In step S15, the effective friction coefficient sequence is subjected to sliding window linear fitting processing, and the slope of the fitted line is extracted to obtain the friction degradation rate characteristics, including: A sliding window of a preset length is set and slides along the effective friction coefficient sequence at time steps; For the friction coefficient data within each sliding window, a linear regression model is constructed using the least squares method; Extract the slope parameter of the linear regression model to obtain the local gradient vector that changes over time; Calculate the statistical mean of the local gradient vector, and determine the statistical mean as the characteristic of the frictional degradation rate.

[0041] It should be noted that the least squares method is used to construct a linear regression model and extract the slope parameter, aiming to quantify the microscopic trend of the friction coefficient evolution over time. Let the effective friction coefficient sequence output in step S13 be... The corresponding discrete time series is Define the sliding window length as... Step size is (generally =1 to obtain continuous gradient estimates). For the th A sliding window extracts a subset of data. and the corresponding time subset Establish a linear regression model within this window. ,in The slope Let be the intercept. According to the least squares principle, the slope that minimizes the sum of squared residuals is... The calculation formula is: ; System by step By sliding the window and repeating the above calculation, a series of slope values ​​are obtained after traversing the entire sequence. This refers to the local gradient vector. Finally, the vector is calculated. Arithmetic mean: ; The mean This refers to the friction degradation rate characteristic, which represents the average rate at which the lubrication condition of the equipment deteriorates or mechanical wear intensifies during the monitoring period. A larger positive value indicates faster degradation.

[0042] It is worth noting that the preset length of the sliding window The determination of the value is based on statistical analysis of the thermal equilibrium time constant and friction coefficient fluctuation characteristics of the die-cutting machine. The system collects friction coefficient data throughout the entire process from cold start to thermal stability, calculates its autocorrelation function, and selects the number of hysteresis points corresponding to the autocorrelation coefficient decaying to 0.5 as the baseline length. Based on this, regression analysis is performed on the historical validation set, and the window length that minimizes the variance of the fitted residuals is selected as the final preset value (e.g., setting...). (Corresponding to 50 working cycles). This setting ensures that the window contains enough data points to eliminate random measurement noise, while avoiding a window that is too long and causes a delay in response to rapid sudden failures (such as fuel cut-off).

[0043] For example, assume the effective friction coefficient sequence contains 1000 data points, representing 10 minutes of device operation. The preset sliding window length is 100 points (1 minute), with a step size of 10 points. For the first window (points 1-100), the slope of the fitted line is calculated using the least squares method. For the second window (points 11-110), calculate the slope. And so on, a total of 91 local gradient values ​​were calculated. The system calculates the average of these 91 values ​​to obtain... Although this value is small in absolute terms, it is significant compared to the baseline health condition. (Caused solely by temperature drift) This clearly indicates that the equipment is in a phase of accelerated wear.

[0044] In step S16, a device health feature vector is constructed using the rotational speed ripple feature and the friction degradation rate feature as feature elements. The Mahalanobis distance of the device health feature vector relative to a preset device health reference vector is calculated. If the Mahalanobis distance exceeds a preset safety threshold, a fault warning command is generated, including: Calculate the difference vector between the device health feature vector and the preset device health benchmark vector; The Mahalanobis distance is obtained by calculating the weighted Euclidean distance of the difference vector using a preset inverse covariance matrix. If the Mahalanobis distance is greater than the preset safety threshold, a fault warning instruction containing the fault level and recommended maintenance time is generated.

[0045] It should be noted that constructing the equipment health feature vector and calculating the Mahalanobis distance are processes that utilize multidimensional statistical principles to eliminate correlations between features and quantify health deviations. First, the system uses the rotational speed ripple features obtained in step S14... The frictional degradation rate characteristics obtained in step S15 Combined into column vectors Subsequently, the preset device health baseline vector is loaded. With the preset inverse covariance matrix Calculate the difference vector. The Mahalanobis distance is calculated using the following formula. : ; This distance not only considers the Euclidean deviation between the current eigenvalue and the baseline mean, but also takes into account the distribution shape of eigenvalue changes and the coupling relationship between eigenvalues ​​through the covariance matrix. For example, if the natural fluctuation variance of speed ripple has been large historically, the covariance matrix will assign it a smaller weight, thereby avoiding false alarms caused by normal fluctuations.

[0046] It should be noted that the generation of fault warning instructions containing fault levels and recommended maintenance times is achieved using a distance-segmented lookup table mapping mechanism. The system pre-defines a hierarchical judgment table, which defines the correspondence between different Mahalanobis distance intervals and fault levels. The construction of this hierarchical judgment table is based on retrospective analysis of historical maintenance data throughout the die-cutting machine's entire lifecycle. Specifically, the system retrieves characteristic data from historical equipment fault times and calculates their corresponding Mahalanobis distances. This distance value is then mapped to the actual damage level confirmed during subsequent disassembly and repair (e.g., minor scratches, pitting, fracture failure). By statistically analyzing the distance distribution boundaries corresponding to different damage levels, grading coefficients (e.g., 2x and 4x) are determined, thereby constructing the following judgment rule: if... If so, it is judged as a Level 1 warning (minor hidden danger), and the recommended maintenance time is "at the next downtime maintenance"; if If so, it is judged as a Level 2 warning (significant fault), and the recommended maintenance time is "within 24 hours"; if If the alarm is triggered, it is classified as a Level 3 alarm (dangerous state), and the recommended maintenance time is "immediate shutdown". The system calculates this in real time. Upon entering the designated area, the system automatically assembles a command message containing the corresponding level label and time suggestion, and sends it to the Human-Machine Interface (HMI) via the industrial bus. The system then calculates... Once the area is entered, the system automatically assembles an instruction message containing the corresponding level label and time suggestion, and sends it to the human-machine interface (HMI) via the industrial bus.

[0047] It is worth noting that the device health baseline vector The inverse covariance matrix and the preset security threshold The determination was based on statistical modeling of the healthy operation data of the die-cutting machine after its break-in period. Specifically, the system collects all characteristic data within the first 100 hours after a new machine is delivered or after a major overhaul (confirmed as being in a healthy state) to construct a sample set. .calculate The arithmetic mean of each feature is used as Calculate its covariance matrix And by inverse, we get Regarding security thresholds Based on the assumption that Mahalanobis distance follows a chi-square distribution, the chi-square critical value with 2 degrees of freedom (feature dimension) and a confidence level of 99.9% is selected. The square root of (approximately 3.7) is used as the preset value.

[0048] For example, suppose the health baseline vector inverse covariance matrix (Assuming features are independent for simplicity), the preset safety threshold is... The device health feature vector calculated at the current moment. The difference vector is The squared Mahalanobis distance is Calculations yielded .because The system determines that the current device is in a healthy state and does not generate a warning command. If the characteristics change abruptly at a certain moment, the system calculates... ,because and The system generates a level-two warning instruction, indicating "significant wear exists, and it is recommended to check within 24 hours".

[0049] In summary, this invention overcomes the bottleneck of traditional reliance on vibration or temperature monitoring by constructing a dynamic state observer based on physical mechanisms, deeply decoupling from nanosecond-level synchronously acquired electrical signals and tracking the viscous friction coefficient in real time. Through an original period segmentation and working condition avoidance mechanism, it accurately locks the non-die-cutting work interval to extract effective features, effectively overcoming the masking of weak fault signals by the strong impact load noise unique to the die-cutting process. Furthermore, by combining the multi-dimensional feature fusion analysis of frequency domain speed ripple and time domain friction degradation, it achieves accurate capture and quantitative early warning of early weak fault features of fully automatic flatbed die-cutting machines, significantly improving the operational reliability of the equipment under high-speed and heavy-load conditions and the scientific nature of maintenance decisions.

[0050] Reference Figure 2 The second embodiment of the present invention provides a fault early warning system for a fully automatic flatbed die-cutting machine, comprising: The data acquisition module is used to continuously and synchronously acquire the stator current signal and rotor speed signal of the die-cutting machine motor drive system through multiple channels to obtain a continuous running data stream; The state observation module is used to construct a dynamic state observer based on the motor torque balance equation. The continuous running data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream is calculated. The observation error is used to perform parameter adaptive feedback correction and output a continuous viscous friction coefficient estimation sequence. The interval locking module is used to identify the working cycle of the die-cutting machine based on the peak or trough feature points of the rotor speed signal, lock the non-die-cutting work interval within each working cycle, and extract the numerical segment corresponding to the non-die-cutting work interval from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. The ripple extraction module is used to extract the rotor speed signal segment corresponding to the non-die-cutting work interval from the continuous running data stream, perform frequency domain transformation processing on the rotor speed signal segment, obtain the spectral amplitude data within the preset monitoring frequency band, and obtain the speed ripple characteristics. The degradation analysis module is used to perform sliding window linear fitting on the effective friction coefficient sequence, extract the slope of the fitted line, and obtain the friction degradation rate characteristics. The early warning decision module is used to construct a device health feature vector using the rotational speed ripple feature and the friction degradation rate feature as feature elements, calculate the Mahalanobis distance of the device health feature vector relative to the preset device health benchmark vector, and generate a fault early warning command if the Mahalanobis distance exceeds the preset safety threshold.

[0051] It should be noted that the fully automatic flatbed die-cutting machine fault early warning system provided in this embodiment of the invention is used to execute all the process steps of the fully automatic flatbed die-cutting machine fault early warning method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0052] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a fault warning program for a fully automatic flatbed die-cutting machine. When the processor executes the computer program, it implements the steps described in the embodiments of the fault warning method for a fully automatic flatbed die-cutting machine, for example... Figure 1The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data acquisition module.

[0053] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0054] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0055] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0056] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0057] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0058] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0059] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A full-automatic flat press flat die cutting machine fault early warning method, characterized in that, include: The stator current signal and rotor speed signal of the die-cutting machine motor drive system are continuously and synchronously acquired through multiple channels to obtain a continuous running data stream. A dynamic state observer based on the motor torque balance equation is constructed. The continuous running data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream is calculated. The observation error is used to perform parameter adaptive feedback correction and output a continuous viscous friction coefficient estimation sequence. The working cycle of the die-cutting machine is identified based on the peak or trough feature points of the rotor speed signal. The non-die-cutting work interval within each working cycle is locked, and numerical segments corresponding to the non-die-cutting work interval are extracted from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. The rotor speed signal segment corresponding to the non-die-cutting work interval is extracted from the continuous running data stream. The rotor speed signal segment is subjected to frequency domain transformation processing to obtain the spectral amplitude data within the preset monitoring frequency band and obtain the speed ripple characteristics. The effective friction coefficient sequence is subjected to sliding window linear fitting, and the slope of the fitted line is extracted to obtain the friction degradation rate characteristics. A device health feature vector is constructed using the rotational speed ripple feature and the friction degradation rate feature as feature elements. The Mahalanobis distance of the device health feature vector relative to the preset device health reference vector is calculated. If the Mahalanobis distance exceeds the preset safety threshold, a fault warning command is generated.

2. The full-automatic perfect-bond perfect-cut machine fault early warning method according to claim 1, characterized in that, The process of continuously and synchronously acquiring the stator current signal and rotor speed signal of the die-cutting machine motor drive system through multiple channels to obtain a continuous operating data stream includes: Acquire the analog current signal and instantaneous speed pulse signal of the stator winding at the motor drive port of the die-cutting machine; A sampling clock is used to perform synchronous analog-to-digital conversion on the analog current signal and the instantaneous speed pulse signal, and output a discrete digital sequence with timestamps. The phase lag angle is calculated based on the discrete digital sequence, and the analog current signal and the instantaneous speed pulse signal are time-domain aligned and corrected according to the phase lag angle to generate the continuous running data stream.

3. The full-automatic perfect-bond perfect-cut machine fault early warning method according to claim 1, characterized in that, The process involves constructing a dynamic state observer based on the motor torque balance equation. The continuous operating data stream is continuously input into the dynamic state observer. The observation error between the estimated state variables output by the dynamic state observer and the measured state variables in the continuous operating data stream is calculated. This observation error is then used for adaptive parameter feedback correction, outputting a continuous sequence of estimated viscous friction coefficients, including: Construct a discretized state-space model that includes the viscous friction coefficient as an implicit state variable; The continuous running data stream is input into the discretized state-space model to deduce the prior state estimation vector; Calculate the deviation between the predicted rotational speed value in the prior state estimation vector and the measured rotational speed value in the continuous operation data stream, and calculate the Kalman gain matrix based on the deviation; The prior state estimation vector is weighted and corrected using the Kalman gain matrix to obtain the posterior state estimation vector. The steady-state converged values ​​are extracted from the posterior state estimation vector to obtain the continuous viscous friction coefficient estimation sequence.

4. The full-automatic perfect-bond perfect-cut machine fault early warning method according to claim 1, characterized in that, The method of identifying the working cycle of the die-cutting machine based on the peak or trough feature points of the rotor speed signal, locking the non-die-cutting work interval within each working cycle, and extracting numerical segments corresponding to the non-die-cutting work interval from the continuous viscous friction coefficient estimation sequence to obtain an effective friction coefficient sequence, including: An extreme value search is performed on the rotor speed signal to identify periodically occurring speed peaks and troughs; The time interval between two adjacent speed peaks is defined as a working cycle. Within each work cycle, the time period corresponding to the sudden change in die-cutting pressure is removed, and the remaining stable operating time period is locked as the non-die-cutting work interval. Based on the start and end timestamps of the non-die-cutting work interval, corresponding data segments are extracted from the continuous viscous friction coefficient estimation sequence and combined to generate the effective friction coefficient sequence.

5. The full-automatic perfect-bond perfect-cut machine fault early warning method according to claim 1, characterized in that, The step of extracting a rotor speed signal segment corresponding to the non-die-cutting work interval from the continuous operating data stream, performing frequency domain transformation on the rotor speed signal segment, obtaining spectral amplitude data within a preset monitoring frequency band, and obtaining speed ripple characteristics includes: The rotor speed signal segment is windowed and truncated by applying a Hanning window function. Perform a Fast Fourier Transform on the windowed signal segment to generate a full-band amplitude spectrum; In the full-band amplitude spectrum, the low-frequency segment located between the DC component frequency and the third harmonic frequency is extracted as the preset monitoring frequency band; The root mean square deviation of the amplitude of each frequency component within the preset monitoring frequency band is calculated and used as the rotational speed ripple characteristic.

6. The fault early warning method for a fully automatic flatbed die-cutting machine according to claim 1, characterized in that, The step of performing a sliding window linear fitting process on the effective friction coefficient sequence, extracting the slope of the fitted line, and obtaining the friction degradation rate characteristics includes: A sliding window of a preset length is set and slides along the effective friction coefficient sequence at time steps; For the friction coefficient data within each sliding window, a linear regression model is constructed using the least squares method; Extract the slope parameter of the linear regression model to obtain the local gradient vector that changes over time; Calculate the statistical mean of the local gradient vector, and determine the statistical mean as the characteristic of the frictional degradation rate.

7. The fault early warning method for a fully automatic flatbed die-cutting machine according to claim 1, characterized in that, The process involves constructing a device health feature vector using the rotational speed ripple characteristic and the friction degradation rate characteristic as feature elements, calculating the Mahalanobis distance of the device health feature vector relative to a preset device health benchmark vector, and generating a fault warning command if the Mahalanobis distance exceeds a preset safety threshold. This command includes: Calculate the difference vector between the device health feature vector and the preset device health benchmark vector; The Mahalanobis distance is obtained by calculating the weighted Euclidean distance of the difference vector using a preset inverse covariance matrix. If the Mahalanobis distance is greater than the preset safety threshold, a fault warning instruction containing the fault level and recommended maintenance time is generated.

8. A fault early warning system for a fully automatic flatbed die-cutting machine, characterized in that, include: The data acquisition module is used to continuously and synchronously acquire the stator current signal and rotor speed signal of the die-cutting machine motor drive system through multiple channels to obtain a continuous running data stream; The state observation module is used to construct a dynamic state observer based on the motor torque balance equation. The continuous running data stream is continuously input into the dynamic state observer. The observation error between the estimated state quantity output by the dynamic state observer and the measured state quantity in the continuous running data stream is calculated. The observation error is used to perform parameter adaptive feedback correction and output a continuous viscous friction coefficient estimation sequence. The interval locking module is used to identify the working cycle of the die-cutting machine based on the peak or trough feature points of the rotor speed signal, lock the non-die-cutting work interval within each working cycle, and extract the numerical segment corresponding to the non-die-cutting work interval from the continuous viscous friction coefficient estimation sequence to obtain the effective friction coefficient sequence. The ripple extraction module is used to extract the rotor speed signal segment corresponding to the non-die-cutting work interval from the continuous running data stream, perform frequency domain transformation processing on the rotor speed signal segment, obtain the spectral amplitude data within the preset monitoring frequency band, and obtain the speed ripple characteristics. The degradation analysis module is used to perform sliding window linear fitting on the effective friction coefficient sequence, extract the slope of the fitted line, and obtain the friction degradation rate characteristics. The early warning decision module is used to construct a device health feature vector using the rotational speed ripple feature and the friction degradation rate feature as feature elements, calculate the Mahalanobis distance of the device health feature vector relative to the preset device health benchmark vector, and generate a fault early warning command if the Mahalanobis distance exceeds the preset safety threshold.

Images