Printing slip fault diagnosis method and system based on image recognition
By fusing machine vision and encoder signals, and utilizing least-squares fitting and spectral analysis during the speed change test phase, the problem of traditional methods being unable to detect viscoelastic slippage of the rubber roller was solved, enabling early and accurate identification and closed-loop control of printing slippage faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-11
- Publication Date
- 2026-04-07
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional monitoring methods cannot accurately detect the viscoelastic slippage of the rubber roller during dynamic operation, resulting in quality defects such as printing ghosting and misregistration, and lacking early fault diagnosis and warning mechanisms.
By fusing machine vision and encoder signals, the stress relaxation time constant of the rubber roller is determined using least squares fitting during the speed change test. Combined with dense optical flow field and spectrum analysis, a damping attenuation modulus is constructed for fault diagnosis and closed-loop control.
It enables early and accurate identification and closed-loop control of printing slip faults, improves the accuracy and robustness of fault diagnosis, and avoids the problem of low signal-to-noise ratio under non-steady-state conditions in traditional methods.
Smart Images
Figure CN121811208A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image recognition technology, and in particular to a method and system for diagnosing printing slippage faults based on image recognition. Background Technology
[0002] The operational stability of modern high-speed printing machinery directly determines the quality of printed products. Among them, the synchronization performance of the printing unit is the core indicator. In related technologies, printing presses usually use signals fed back by spindle encoders or servo motors to monitor the operating speed, and work with tension control systems to maintain stable transmission of the printing area. This monitoring method relies on rigid transmission, which assumes that there is a fixed linear proportional relationship between the linear velocity of the rubber roller surface and the angular velocity of the spindle drive.
[0003] However, in actual high-speed printing processes, in order to ensure sufficient ink transfer, the surface of the printing roller is coated with a rubber material with significant viscoelastic properties. This viscoelastic property causes the roller to exhibit nonlinear creep and stress relaxation phenomena during dynamic operation, especially when accelerating, decelerating, or subjected to external disturbances.
[0004] There is a non-negligible dynamic slip error between the actual linear velocity on the surface of the rubber roller and the theoretical linear velocity calculated by the encoder. This slip is a complex nonlinear hysteresis process that is highly related to the material properties of the rubber roller, real-time acceleration, and running speed. Traditional encoder-based rigid monitoring methods cannot detect this microscopic viscoelastic slip, which leads to the control system misjudging the current state and causing quality defects such as printing ghosting, periodic ink streaks, or even misregistration. Moreover, such faults are often only discovered manually after a large number of waste products have been generated, and there is a lack of effective early fault diagnosis and warning mechanisms.
[0005] Therefore, a diagnostic scheme is needed that can accurately quantify the viscoelastic slip characteristics of the rubber roller and separate the true fault characteristics from complex dynamic signals in order to achieve early and accurate identification and closed-loop control of printing slip faults. Summary of the Invention
[0006] To achieve early and accurate identification and closed-loop control of printing slip faults, this application provides a printing slip fault diagnosis method and system based on image recognition.
[0007] According to a first aspect of the embodiments of this application, a printing slip fault diagnosis method based on image recognition is provided, comprising: The printing press is controlled to perform a speed change test under no-load and with the printing unit in contact. The theoretical linear velocity is acquired through the spindle encoder, and the actual velocity of the rubber roller surface is acquired through machine vision. Based on the theoretical linear velocity and the actual velocity of the rubber roller surface, a pre-constructed least-squares objective function is iteratively solved to determine the stress relaxation time constant of the rubber roller. The printing area image sequence and the spindle encoder signal are acquired, and the dense optical flow field of each frame image is calculated. Spatial statistical pooling is performed on the dense optical flow field to obtain a one-dimensional visual velocity signal. Combining the stress relaxation time constant and the real-time acceleration obtained by differentiating the spindle encoder signal, the cumulative phase including rubber roller hysteresis compensation is calculated. The cumulative phase is used to resample the one-dimensional visual velocity signal in the angular domain to obtain the phase domain velocity signal. The phase domain velocity signal is subjected to spectral analysis to extract the order domain features. Based on the order domain features and the real-time running linear velocity, a damping attenuation modulus is constructed. The damping attenuation modulus is compared with a preset fault threshold, and hierarchical diagnosis and closed-loop feedback control are performed according to the comparison results.
[0008] This enables early and accurate identification and closed-loop control of printing slippage faults.
[0009] Optionally, the printing press is controlled to perform a speed change test under no-load and with the printing units in contact, including: controlling the printing press to sequentially perform a rapid acceleration action, a constant speed action, and a rapid deceleration action to form a standardized speed step process, so as to test the printing press under the speed step process.
[0010] Optionally, the stress relaxation time constant of the rubber roller is determined using the following least-squares objective function: Where G represents the least squares objective function, Let be the stress relaxation time constant to be solved. This represents the summation of all sampling points within the test period, where t is the sampling time, u(t) is the actual surface velocity of the rubber roller measured by machine vision optical flow at time t, v(t) is the theoretical linear velocity measured by the spindle encoder at time t, and R is the physical radius of the printing roller. Let be the angular acceleration at time t.
[0011] Optionally, performing spatial statistical pooling on the dense optical flow field includes: selecting a region of interest in the center of each frame of the printed image sequence, the size of which is set based on the printing width; extracting the longitudinal velocity component of each pixel in the dense optical flow field along the paper feed direction; performing median statistical calculation on all extracted longitudinal velocity components, and determining the median value obtained from the calculation as the one-dimensional visual velocity signal at the current sampling time.
[0012] Optionally, by combining the stress relaxation time constant with the real-time acceleration obtained by differentiating the spindle encoder signal, the cumulative phase including rubber roller hysteresis compensation is calculated as follows: using a Savitzky-Golay filter to smooth and differentiate the velocity data corresponding to the spindle encoder signal to obtain the real-time acceleration; using the stress relaxation time constant to construct a dynamic compensation term to offset the phase hysteresis of rubber deformation; and converting the linear velocity in the time domain into the cumulative phase by integrating the dynamic compensation term.
[0013] Optionally, the accumulated phase is determined by the following formula: ,in, For a moment The cumulative phase, The starting time of integration, For integration variables, For a moment Smooth linear velocity, This is the standard circumference of the printing roller. Let be the stress relaxation time constant. The radius of the printing roller. For a moment Real-time acceleration, This is the sign function used to extract the velocity direction.
[0014] In this way, by using an integral algorithm that includes viscoelastic compensation, the rubber deformation error caused by acceleration can be dynamically corrected when calculating the phase, ensuring that the resampled signals are equidistant in the angular domain, thus avoiding the spectral blurring caused by ignoring material deformation in traditional angular domain resampling.
[0015] Optionally, the one-dimensional visual velocity signal is obtained by angular domain resampling of the accumulated phase, including: constructing an equally spaced phase grid sequence with a fixed phase resolution, and using the mapping relationship of the accumulated phase increasing monotonically with time and the interpolation algorithm to map the time-varying one-dimensional visual velocity signal to the phase grid sequence to obtain the phase-varying phase domain velocity signal.
[0016] In this way, the non-stationary frequency conversion signal that changes with time is transformed into a stationary signal that changes with angle, so that the periodic fault characteristics caused by mechanical defects such as eccentricity or surface damage are represented by a fixed period in the signal, thus solving the problem of frequency ambiguity that may occur in direct spectrum analysis during speed change.
[0017] Optionally, spectral analysis is performed on the phase-domain velocity signal to extract order-domain features, including: taking the arithmetic mean of the phase-domain velocity signal as the DC component, subtracting the DC component from the phase-domain velocity signal to obtain the phase-domain vibration signal; performing a fast Fourier transform on the vibration phase-domain vibration signal to obtain an order-domain spectral sequence containing the order and spectral line amplitude, so as to extract order-domain features from the order-domain spectral sequence; the horizontal axis of the order-domain spectral sequence represents the order of the number of vibration cycles that occur per rotation.
[0018] In this way, by using order domain analysis, the fault characteristics and operating speed are decoupled, so that specific mechanical faults such as gear meshing or rubber roller sticking and slipping can correspond to fixed order coordinates, which makes it easy to quickly locate the fault source by combining mechanical parameters.
[0019] Optionally, the damping modulus is determined by the following formula: Where M is the damping modulus, S(k) represents the set of fault characteristic orders, which includes integer multiples of the rubber roller's rotational frequency and discrete order points in their neighborhood; N is the number of orders, and S(k) is the spectral amplitude corresponding to order k. Total energy across the entire frequency band. For background noise energy, To prevent positive numbers with a denominator of zero, exp is an exponential function with the natural constant as its base. For real-time running linear speed, This is the rated maximum design speed of the printing press.
[0020] In this way, by constructing a dimensionless damping modulus, not only is the proportion of fault frequency energy considered, but also a velocity exponent term is introduced to compensate for the hardening effect of viscoelastic materials at high frequencies.
[0021] According to a second aspect of the embodiments of this application, a printing slip fault diagnosis system based on image recognition is provided, comprising: a processor and a memory, wherein the memory stores computer program instructions, and the computer program instructions, when executed by the processor, implement the steps of the printing slip fault diagnosis method based on image recognition provided in the first aspect of this application.
[0022] The technical solutions provided by the embodiments of this application may include the following beneficial effects: by introducing the fusion of machine vision and encoder signals, the inherent rheological parameters of the rubber roller, namely the stress relaxation time constant, are accurately identified by using the least squares fitting in the speed test stage. In the subsequent phase calculation, a dynamic compensation model that can offset the viscoelastic hysteresis is constructed, and the damping attenuation modulus is combined to realize the quantitative assessment of the severity of the fault. This effectively solves the problems of low signal-to-noise ratio and inability to distinguish between normal speed fluctuations and abnormal slip faults under non-steady-state conditions in traditional methods, and improves the accuracy and robustness of fault diagnosis of slip faults in printing presses.
[0023] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description
[0024] Figure 1 This is a flowchart illustrating a printing slip fault diagnosis method based on image recognition according to an exemplary embodiment; Figure 2 This is a schematic diagram of the speed curve during the speed change test. Figure 3 This is a schematic diagram of the results of spectrum analysis using existing techniques; Figure 4 This is a schematic diagram of the phase domain order spectrum according to an embodiment of this application; Figure 5 This is a schematic diagram illustrating the structure of a printing slip fault diagnosis system based on image recognition, according to an exemplary embodiment. Detailed Implementation
[0025] To achieve early and accurate identification and closed-loop control of printing slip faults, embodiments of this application provide a printing slip fault diagnosis method and system based on image recognition. Figure 1 This is a flowchart illustrating an image recognition-based printing slip fault diagnosis method according to an exemplary embodiment, such as... Figure 1 As shown, the method includes the following steps.
[0026] In step S101, the printing press is controlled to perform a speed change test under no-load and with the printing unit in contact. The theoretical linear velocity is collected by the spindle encoder and the actual velocity of the rubber roller surface is collected by machine vision. Based on the theoretical linear velocity and the actual velocity of the rubber roller surface, the pre-constructed least squares objective function is iteratively solved to determine the stress relaxation time constant of the rubber roller.
[0027] In one embodiment, controlling the printing press to perform a speed change test in an unloaded state with the printing units in contact includes: controlling the printing press to sequentially perform a rapid acceleration action, a constant speed action, and a rapid deceleration action to form a standardized speed step process, so as to test the printing press under the speed step process.
[0028] Precise timing speed commands can be sent from the controller to the servo drive system of the printing press to create a test excitation source with rich frequency response characteristics.
[0029] For example, the total test duration can be set to 30 seconds. From the 0th to the 8th second interval, the printing press can be controlled to accelerate linearly from a standstill at an acceleration of 0.8 meters per square second to an intermediate speed of 200 meters per minute. From the 8th to the 15th second interval, the printing press can maintain a constant speed of 200 meters per minute, and its dynamic response tends to be steady. From the 15th to the 20th second interval, a rapid deceleration action can be performed, reducing the speed quickly to 50 meters per minute at a deceleration of 1.2 meters per square second. Finally, a controlled stop can be performed from the 20th to the 30th second interval.
[0030] Throughout the test, the printing press was in an idle state without inserting the printing material, and the cylinder or servo motor could be controlled to push the impression cylinder and the printing plate cylinder to fully fit together, with the contact pressure set to the standard working pressure, such as 0.35 MPa.
[0031] The viscoelastic characteristics of the rubber material of the rubber roller are not obvious in steady-state operation, i.e., the uniform speed stage. At this time, the elastic deformation tends to be saturated and the damping force is almost zero, making it difficult to effectively separate the viscosity parameters through direct observation data.
[0032] The changes in inertial force brought about by rapid acceleration and deceleration can maximize the excitation of the unbalanced stress response inside the rubber roller material, producing a significant transient slip phenomenon. By artificially creating intense dynamic working conditions, sample data with high signal-to-noise ratio and high dynamic range can be provided for subsequent parameter identification, ensuring that the identified model parameters can truly reflect the physical limits of the rubber roller under extreme working conditions, thereby better realizing the identification of slip defects in the printing press.
[0033] Figure 2 This is a schematic diagram of the speed curve during the gear shift test, such as... Figure 2 As shown, under continuous speed change conditions including acceleration, constant speed and deceleration, there is a certain difference between the theoretical linear velocity curve collected by the encoder and the one-dimensional visual velocity signal curve measured and calculated by the machine vision system.
[0034] Figure 2This intuitively reflects that due to the viscoelasticity of rubber materials, the one-dimensional visual velocity signal has a significant physical lag compared to the theoretical linear velocity, and there is a dynamic deviation between the two in terms of values, which can provide a data basis for subsequent calculation of the stress relaxation time constant.
[0035] In one embodiment, the stress relaxation time constant of the rubber roller is determined by the following least-squares objective function: Where G represents the least squares objective function, Let be the stress relaxation time constant to be solved. This represents the summation of all sampling points within the test period, where t is the sampling time, u(t) is the actual surface velocity of the rubber roller measured by machine vision optical flow at time t, v(t) is the theoretical linear velocity measured by the spindle encoder at time t, and R is the physical radius of the printing roller. Let be the angular acceleration at time t.
[0036] The machine vision acquisition system can use an industrial camera with a global shutter to capture the microscopic texture of the rubber roller surface at a frame rate of, for example, 200 frames per second, and calculate the actual linear velocity at each moment t using an optical flow algorithm.
[0037] The angular velocity signal is acquired by a high-resolution incremental encoder mounted on the spindle, for example, 5000 pulses per revolution, and the theoretical linear velocity is obtained by combining it with the radius of the rubber roller. In order to eliminate the time deviation between the two acquisition systems, the time synchronization can be achieved in advance by hardware trigger signal.
[0038] The angular acceleration in the least squares objective function is obtained by performing a difference operation on the theoretical linear velocity and dividing it by the radius. A low-pass filter is applied before the difference operation to suppress high-frequency quantization noise.
[0039] When the drive shaft changes with acceleration, the stress relaxation characteristics of the rubber material change from... Characterized by this, the actual linear velocity of the rubber roller surface will exhibit a hysteresis decay relative to the driving speed, which is proportional to the acceleration.
[0040] The solution process can employ the Gauss-Newton iteration method or the Levenberg-Marquardt algorithm. For example, the initial iteration value can be set to 0.01 seconds, and the time between two consecutive iterations can be less than 0.01 seconds. Stop the iteration and output the optimal solution.
[0041] In actual industrial settings, measurements at a single moment are often affected by random noise such as mechanical vibration, changes in stroboscopic illumination, and dust obstruction. The least squares method, by using data points throughout the entire test period for global optimization, can effectively smooth out the influence of random noise using statistical principles and converge to the most accurate physical true value.
[0042] The determined stress relaxation time constant reflects the degree of hysteresis in the response of the current rubber roller material to speed changes, enabling the diagnostic process to adapt to rubber rollers with different hardness or different aging levels.
[0043] In step S102, the printed image sequence and spindle encoder signal are acquired, the dense optical flow field of each frame image is calculated, and spatial statistical pooling is performed on the dense optical flow field to obtain a one-dimensional visual velocity signal.
[0044] In one embodiment, performing spatial statistical pooling on a dense optical flow field includes: selecting a region of interest in the center of each frame of a printed image sequence, the size of which is set based on the printing width; extracting the longitudinal velocity component of each pixel within the region of interest along the paper feed direction in the dense optical flow field; performing median statistical calculation on all extracted longitudinal velocity components, and determining the median value obtained from the calculation as the one-dimensional visual velocity signal at the current sampling time.
[0045] The selected region of interest can be located at the geometric center of the image. The physical size of the region of interest corresponds to a rectangular area of, for example, 100 mm by 100 mm on the actual printed surface, with a pixel resolution of 512 by 512, avoiding the lens distortion areas that may exist at the edges of the image.
[0046] The calculation of dense optical flow fields can employ the Farneback algorithm. For example, by setting the pyramid to 3 layers and the window size to 15 pixels, it is possible to simultaneously capture large displacements and minute texture deformations. After calculation, each pixel within the region of interest is assigned a two-dimensional velocity vector. Extract only the longitudinal component along the paper's travel direction. .
[0047] Construct a numerical set containing multiple velocity samples. To obtain the one-dimensional visual velocity at the current moment, a sorting operation can be performed, and the value located in the middle of the sequence can be selected as the output.
[0048] The printing environment is complex, and images may contain random ink splatters, reflective bright spots on the paper surface, or electrostatically adsorbed paper dust. These interfering points may be misidentified as outliers with abnormally high or low speeds in optical flow calculations, such as calculating a speed of 0 or more than 10 times the normal speed. The median statistical characteristics have robust anti-interference properties, and most areas of the printing press surface are usually normal printing textures, which allows the calculation results to accurately lock the true movement speed of the roller surface, thereby avoiding interference from possible local noise.
[0049] In this way, a high-precision, high-signal-to-noise ratio time-series velocity curve, i.e., a one-dimensional visual velocity signal, can be extracted from two-dimensional image data containing a large amount of redundant information and noise interference. The one-dimensional visual velocity signal can keenly capture the vibration of the rubber roller surface.
[0050] In step S103, the cumulative phase including roller hysteresis compensation is calculated by combining the stress relaxation time constant and the real-time acceleration obtained by differentiating the spindle encoder signal, and the phase domain velocity signal is obtained by angular domain resampling of the one-dimensional visual velocity signal using the cumulative phase.
[0051] In one embodiment, the calculation of the cumulative phase including rubber roller hysteresis compensation, by combining the stress relaxation time constant with the real-time acceleration obtained by differentiating the spindle encoder signal, includes: using a Savitzky-Golay filter to smooth and differentiate the velocity data corresponding to the spindle encoder signal to obtain the real-time acceleration; using the stress relaxation time constant to construct a dynamic compensation term to offset the phase hysteresis of rubber deformation; and converting the linear velocity in the time domain, combined with the dynamic compensation term, into the cumulative phase through integration.
[0052] The Savitzky-Golay filter, or SG filter for short, is a digital smoothing filtering algorithm based on polynomial least squares fitting. Its core advantage is that it can preserve the trend and higher-order derivative information of the signal to the greatest extent while smoothing noise.
[0053] To balance noise suppression and dynamic capture capabilities, for example, the sliding window length can be set to 31 sampling points and the fitting polynomial order can be 3. Without introducing significant phase delay, the smooth encoder can quantize the high-frequency jitter caused by the error to obtain a smooth velocity curve, thereby calculating accurate real-time acceleration.
[0054] The construction of the dynamic compensation term involves substituting the determined stress relaxation time constant into the correction model to calculate the instantaneous micro-slippage caused by inertia and viscous damping at each moment. The final integral calculation uses the trapezoidal rule or Simpson's integral formula to continuously accumulate the corrected effective velocity from the start of the test, thereby obtaining the true physical angle rotated by the rubber roller.
[0055] In unsteady states, such as acceleration and deceleration, the rigid assumption that the traditional encoder pulse count equals the rotation angle completely fails. Due to the hysteresis effect of the rubber layer of the rubber roller, when the encoder shows that it has rotated 360 degrees, the surface of the rubber roller may actually have only rotated 358 degrees, resulting in acceleration lag or 362 degrees, resulting in deceleration lead.
[0056] If the encoder signal is used directly for angular domain resampling without compensation, the resampled signal will undergo nonlinear stretching and deformation on the angular axis. This deformation manifests as a severe spectral smearing effect in the frequency domain, where the originally sharp single-frequency fault feature peaks spread out and widen, or even get submerged in the background noise, leading to diagnostic failure.
[0057] By introducing a compensation mechanism based on a physical model, phase errors caused by the flexibility of the mechanical transmission chain and the viscoelasticity of materials can be eliminated, ensuring the accuracy of the phase.
[0058] In one embodiment, the accumulated phase is determined by the following calculation: ,in, For a moment The cumulative phase, The starting time of integration, For integration variables, For a moment Smooth linear velocity, This is the standard circumference of the printing roller. Let be the stress relaxation time constant. The radius of the printing roller. For a moment Real-time acceleration, This is the sign function used to extract the velocity direction.
[0059] This represents the theoretical number of revolutions per unit time, multiplied by Implicit in the phase definition, the theoretical angular velocity is normalized to the number of revolutions. When the printing press is accelerating, the correction factor is less than 1, and the actual angle turned is less than the theoretical angle, compensating for hysteresis loss; when the printing press is decelerating, the correction factor is greater than 1, compensating for inertial lead.
[0060] The purpose of this is to ensure the algorithm's versatility in reverse or reciprocating motion tests and to ensure the correctness of the directional logic. In the numerical calculation implementation, the integral is approximated by summation, and an overflow prevention and reset mechanism is set up, that is, when the phase exceeds a certain threshold, the overflow is reset. Modulo is used, but in resampling maps, the absolute cumulative value is usually preserved to maintain monotonicity.
[0061] The slippage of the rubber roller is a path-dependent historical accumulation process. The current phase state depends not only on the current instantaneous acceleration, but also on the historical state accumulation at each previous moment. Through integral form, the phase evolution trajectory of the entire motion process can be completely traced.
[0062] In one embodiment, the method of obtaining a phase domain velocity signal by angular domain resampling of a one-dimensional visual velocity signal using accumulated phase includes: constructing an equally spaced phase grid sequence with a fixed phase resolution; using the mapping relationship of the accumulated phase increasing monotonically with time and an interpolation algorithm, mapping the time-varying one-dimensional visual velocity signal to the phase grid sequence to obtain a phase domain velocity signal that varies with phase.
[0063] We can first define a fixed angular increment resolution, for example Based on the total number of rotations of the rubber roller during the test, for example, 1000 rotations, an arithmetic sequence is generated starting from 0 degrees, with a step size of 0.1 degrees, and ending at 360000 degrees, to obtain the phase grid sequence.
[0064] Using the calculated cumulative phase as a lookup table, the non-integer time point corresponding to each fixed phase grid point is found through inverse function mapping. Since the original sampling is based on equal time intervals, the non-integer time points usually fall outside the sampling points. Therefore, the cubic spline interpolation algorithm can be used to calculate the corresponding one-dimensional visual velocity value based on the original velocity samples adjacent to the non-integer time points.
[0065] During speed change, the frequency of the vibration signal is a variable frequency signal that changes with the rotational speed. Direct Fourier transform will produce severe frequency ambiguity. After mapping the signal to the angular domain, the vibration components that are synchronized with the rotational speed, such as the eccentricity of the rubber roller, gear meshing, and the rotation of the bearing balls, have periodic waveforms in the angular domain.
[0066] In this way, the phase domain velocity signal after resampling changes the horizontal axis from time to angle. Regardless of whether the printing press is accelerating, decelerating or running at a constant speed, the fluctuations caused by physical defects in the rubber roller, such as pits or flat spots on the surface, can be represented as a constant periodic signal in the phase domain signal. This simplifies the complex problem of non-stationary signal processing into the problem of stationary signal processing.
[0067] In step S104, the phase domain velocity signal is subjected to spectral analysis to extract the order domain features. Based on the order domain features and the real-time running linear velocity, a damping attenuation modulus is constructed. The damping attenuation modulus is compared with a preset fault threshold. Based on the comparison result, hierarchical diagnosis and closed-loop feedback control are performed.
[0068] In one embodiment, performing spectral analysis on the phase-domain velocity signal to extract order-domain features includes: taking the arithmetic mean of the phase-domain velocity signal as the DC component, subtracting the DC component from the phase-domain velocity signal to obtain the phase-domain vibration signal; performing a fast Fourier transform on the vibration phase-domain vibration signal to obtain an order-domain spectral sequence containing the order and spectral line amplitude, so as to extract order-domain features from the order-domain spectral sequence; the horizontal axis of the order-domain spectral sequence represents the order of the number of vibration cycles that occur per rotation.
[0069] Removing the DC component can eliminate the influence of macroscopic operating speed on microscopic vibration amplitude. To prevent spectral leakage, a Hanning window or Blackman window can be added to the signal before performing the transformation. The length of the Fast Fourier Transform can be an integer power of 2, such as 4096 points or 8192 points, to obtain sufficient frequency resolution.
[0070] The X-axis unit of the transformed spectrum is order. For example, order 1 means that the rubber roller vibrates once per revolution, which usually corresponds to eccentricity or dynamic imbalance. In printing fault diagnosis, characteristic frequencies are often related to mechanical structural parameters such as the number of gear teeth, roller diameter, and number of bearing balls. These characteristics are usually fixed in the order spectrum, which makes it possible to accurately locate specific fault sources.
[0071] For example, if a significant peak appears at a non-integer order, and the non-integer order belongs to the characteristic order of any gear or bearing, and if it can be determined by looking up a table that the non-integer order often corresponds to a certain pattern of stick-slip self-excited vibration of the rubber roller, the source of the fault can be quickly located.
[0072] In one embodiment, the damping modulus is determined by the following calculation: Where M is the damping modulus, S(k) represents the set of fault characteristic orders, which includes integer multiples of the rubber roller's rotational frequency and discrete order points in their neighborhood; N is the number of orders, and S(k) is the spectral amplitude corresponding to order k. Total energy across the entire frequency band. For background noise energy, To prevent positive numbers with a denominator of zero, exp is an exponential function with the natural constant as its base. For real-time running linear speed, This is the rated maximum design speed of the printing press.
[0073] The denominator of the formula for calculating the damping modulus This represents the effective signal energy base after removing background white noise. It is a positive number, for example , used for numerical stability.
[0074] The dynamic modulus of rubber materials is a function of frequency and temperature, exhibiting a significant hardening effect. With increasing linear velocity... As the aerodynamic stiffness increases, the high-frequency dynamic stiffness of the rubber will increase, which may cause the vibration amplitude excited by the same physical defects at high speed to decrease, or be masked by the increase in aerodynamic damping.
[0075] The exponential term in the formula for calculating the damping modulus It plays a role in speed gain. The higher the real-time operating linear speed, the larger the value of the exponent, which can amplify the weak fault characteristics at high speeds and offset the signal attenuation that may be caused by physical damping.
[0076] The damping attenuation modulus can maintain relative consistency during low-speed or high-speed full-load operation, which can better reflect the physical health status of the rubber roller and solve the problem of setting alarm thresholds under variable speed conditions using traditional amplitude monitoring methods.
[0077] In one embodiment, the total energy across the entire frequency band is determined by summing the squares of the amplitudes of all spectral lines in the order domain spectrum sequence; the background noise energy is determined by selecting spectral lines belonging to the high-frequency non-characteristic region in the order domain spectrum sequence and determining the average energy of the spectral lines belonging to the high-frequency non-characteristic region.
[0078] Different models of printing presses, or even the same machine installed on different foundations, will have different foundation vibration levels. By subtracting background noise and using total energy as a reference, the diagnostic bias caused by individual machine differences is eliminated, allowing it to be quickly applied to printing presses of different widths or brands without the need for tedious parameter calibration for each machine.
[0079] Figure 3 This is a schematic diagram of the results of spectrum analysis using existing techniques. Figure 3 The image shows the spectrum obtained by performing a fixed-time-window Fast Fourier Transform on a one-dimensional visual velocity signal under variable speed conditions.
[0080] Because the signal has non-stationary characteristics, the vibration frequency drifts with the change of velocity, resulting in a severe energy smearing phenomenon in the spectrum. The fault characteristic energy is dispersed in a wide frequency band, making it impossible to identify fixed characteristic peaks with indicative significance, which makes it difficult for existing technologies to achieve fault diagnosis.
[0081] Figure 4 This is a schematic diagram of the phase domain order spectrum according to an embodiment of this application. Figure 4 The paper demonstrates the results of converting a non-stationary time-domain signal into a stationary phase-domain signal after dynamic phase mapping and angular domain resampling, followed by order spectrum analysis.
[0082] like Figure 4 The fault energy shown is precisely focused on specific orders, such as first-order fault features and second-and-fifth-order fault features, forming sharp characteristic peaks and effectively suppressing background noise.
[0083] The preset fault thresholds can include warning thresholds and blocking thresholds; the warning thresholds can be set, for example, based on the 95th percentile of the statistical distribution of historical health data; the blocking thresholds can be set based on the critical value when a visible ink line defect appears.
[0084] During the break-in period when the rubber roller is first installed, for example, operating data can be collected continuously for one week, and the probability density function of the damping attenuation modulus can be calculated, with the 95th percentile as the baseline.
[0085] The blocking threshold can be combined with the results of manual quality inspection. When the quality inspector first finds a slight ink line defect on the printed product, the value of the damping attenuation modulus at this time is recorded, and the critical value blocking threshold is set according to the damping attenuation modulus when a visible ink line defect appears.
[0086] Fault evolution is a gradual process. The rubber roller goes from slight aging leading to uneven surface hardness to eventually producing periodic stripes visible to the naked eye. There is a certain deterioration time window in between. The dual threshold strategy enables the fault detection process to cover the entire life cycle of the fault, achieving more accurate fault diagnosis for the printing press.
[0087] In one embodiment, performing graded diagnosis and closed-loop feedback control based on the comparison results includes: when the damping attenuation modulus is less than the warning threshold, determining that the printing press's roller is in a healthy state and maintaining the current operating parameters; when the damping attenuation modulus is greater than or equal to the warning threshold and less than the blocking threshold, determining that the printing press's roller has early aging characteristics and outputting a maintenance prompt signal; when the damping attenuation modulus is greater than or equal to the blocking threshold, determining that the roller has a serious stick-slip fault, sending a deceleration command to the main control unit, and reducing the printing line speed until the damping attenuation modulus falls back below the blocking threshold.
[0088] When implementing closed-loop control, a yellow alert can be displayed on the human-machine interface for early aging conditions, suggesting that operators check the roller pressure or clean the surface during the next roll change interval. For severe fault conditions, motion control can be intervened using the industrial Ethernet communication protocol, for example, automatically reducing the printing press speed at a slope of 5 meters per minute per second.
[0089] The occurrence of stick-slip vibration requires specific speed and friction coefficient matching conditions. Reducing the speed can often break the critical condition of energy accumulation. Taking automatic speed reduction measures can make the rubber rollers leave the resonance zone, thereby temporarily eliminating ink streaks without immediately stopping the machine, ensuring the continuity of production, and providing a longer available time for subsequent rubber roller replacements.
[0090] Figure 5 This is a schematic diagram illustrating the structure of an image recognition-based printing slip fault diagnosis system 1000 according to an exemplary embodiment. (Refer to...) Figure 5The image recognition-based printing slip fault diagnosis system 1000 includes a processor 1100 and a memory 1200. The memory 1200 stores computer program instructions, which, when executed by the processor 1100, implement all or part of the steps of the image recognition-based printing slip fault diagnosis method in this application.
[0091] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.
[0092] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.
Claims
1. A printing slip fault diagnosis method based on image recognition, characterized in that, include: The printing press is controlled to perform a speed change test under no-load and with the printing unit in contact. The theoretical linear velocity is collected by the main shaft encoder and the actual velocity of the rubber roller surface is collected by machine vision. Based on the theoretical linear velocity and the actual velocity of the rubber roller surface, the pre-constructed least squares objective function is iteratively solved to determine the stress relaxation time constant of the rubber roller. The system acquires the image sequence of the printed area and the spindle encoder signal, calculates the dense optical flow field of each frame, and performs spatial statistical pooling on the dense optical flow field to obtain a one-dimensional visual velocity signal. By combining the stress relaxation time constant with the real-time acceleration obtained by differentiating the spindle encoder signal, the cumulative phase including rubber roller hysteresis compensation is calculated, and the cumulative phase is used to perform angular domain resampling of the one-dimensional visual velocity signal to obtain the phase domain velocity signal. Spectral analysis is performed on the phase domain velocity signal to extract order domain features. Based on the order domain features and the real-time running linear velocity, a damping attenuation modulus is constructed. The damping attenuation modulus is compared with a preset fault threshold, and hierarchical diagnosis and closed-loop feedback control are performed based on the comparison results.
2. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, The printing press is controlled to perform a speed change test under no-load and with the printing units in contact. This includes controlling the printing press to perform rapid acceleration, constant speed and rapid deceleration actions in sequence to form a standardized speed step process, so as to test the printing press under the speed step process.
3. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, The stress relaxation time constant of the rubber roller is determined using the following least-squares objective function: Where G represents the least squares objective function, Let be the stress relaxation time constant to be solved. This indicates that the summation is performed on all sampling points within the test period. Sampling time, For a moment The actual surface speed of the rubber roller measured by machine vision optical flow method. For a moment The theoretical linear velocity measured by the spindle encoder The physical radius of the printing roller. For a moment angular acceleration.
4. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, Performing spatial statistical pooling on dense optical flow fields includes: In each frame of the printed image sequence, a region of interest is selected in the center, and the size of the region of interest is set based on the printing width. Extract the longitudinal velocity component of each pixel in the region of interest along the paper feed direction in the dense optical flow field; Perform median statistical calculations on all extracted longitudinal velocity components, and determine the median value obtained from the calculation as the one-dimensional visual velocity signal at the current sampling time.
5. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, Combining the stress relaxation time constant with the real-time acceleration obtained by differentiating the spindle encoder signal, the cumulative phase including roller hysteresis compensation is calculated as follows: Real-time acceleration is obtained by smoothing and differentiating the velocity data corresponding to the spindle encoder signal using a Savitzky-Golay filter. A dynamic compensation term to counteract the phase lag of rubber deformation is constructed using the stress relaxation time constant; the linear velocity in the time domain is combined with the dynamic compensation term and converted into the cumulative phase through integration.
6. The printing slip fault diagnosis method based on image recognition according to claim 5, characterized in that, The cumulative phase is determined by the following formula: ,in, For a moment The cumulative phase, The starting time of integration, For integration variables, For a moment Smooth linear velocity, This is the standard circumference of the printing roller. Let be the stress relaxation time constant. The radius of the printing roller. For a moment Real-time acceleration, This is the sign function used to extract the velocity direction.
7. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, Phase-domain velocity signals are obtained by angular-domain resampling of one-dimensional visual velocity signals using accumulated phase, including: A phase grid sequence with fixed phase resolution is constructed. By utilizing the mapping relationship that the cumulative phase increases monotonically with time and the interpolation algorithm, the time-varying one-dimensional visual velocity signal is mapped to the phase grid sequence to obtain the phase domain velocity signal that varies with phase.
8. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, Spectral analysis of phase-domain velocity signals to extract order-domain features includes: The arithmetic mean of the phase domain velocity signal is taken as the DC component, and the phase domain velocity signal is subtracted from the DC component to obtain the phase domain vibration signal. A fast Fourier transform is performed on the phase domain vibration signal to obtain an order domain spectrum sequence containing the order and spectral line amplitude, so as to extract order domain features from the order domain spectrum sequence; the horizontal axis of the order domain spectrum sequence represents the order of the vibration cycle number that occurs per rotation.
9. The printing slip fault diagnosis method based on image recognition according to claim 1, characterized in that, The damping modulus is determined by the following formula: ,in, For damping attenuation modulus, The fault characteristic order set contains integer multiples of the rubber roller rotation frequency and discrete order points in their neighborhood; N is the number of orders. For order The corresponding spectral line amplitude, Total energy across the entire frequency band. For background noise energy, To prevent positive numbers with a denominator of zero, It is an exponential function with the natural constant as its base. For real-time running linear speed, This is the rated maximum design speed of the printing press.
10. A printing slip fault diagnosis system based on image recognition, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the image recognition-based printing slip fault diagnosis method according to any one of claims 1-9.