A method and system for predicting the real-time remaining life of hot rolling work rolls

Abstract

The present invention relates to a method and system for predicting the real-time remaining life of a hot rolling working roll. The method for predicting the real-time remaining life of a hot rolling working roll first collects the original eddy current signal sequence of the roll to be predicted, uses a time-frequency analysis method to extract characteristic parameters for characterizing the performance degradation of the roll, and then establishes a characteristic parameter data set based on the characteristic parameters. Then, a probability density function of the solved remaining life of the roll is constructed in combination with the characteristic parameter data set, and the running time of the roll is used as input to predict the remaining life of the roll in real time based on the constructed roll performance degradation evaluation model. Based on the lossless eddy current signal, the prediction method adopts a time-frequency domain feature extraction method to effectively suppress the influence of noise on the characteristic value, retain the effective information in the signal, and based on the degradation trajectory of the characteristic value, uses the remaining life prediction model of the Wiener process to predict the remaining life of the roll in real time, while having little impact on the working state of the roll.

Description

Technical Field

[0001] The present invention relates to the technical field of roll maintenance, and in particular to a method and system for predicting the real-time remaining life of a hot rolling work roll. Background Art

[0002] Rollers are the primary working components and tools on a rolling mill that produce continuous plastic deformation of metal. During operation, they are not only subjected to the rolling force of the rolled product but, in some cases, hot-rolled work rolls, are also subject to various factors, including thermal fatigue, thermal stress, and chemical corrosion. These factors can often lead to surface cracking, wear, and oxidation, and in severe cases, roll failure. Therefore, to ensure mill efficiency and reduce frequent downtime for inspection and maintenance, roll health monitoring is essential. This not only determines the health of the rolls during operation but also predicts their remaining service life. This provides a reliable basis for proactive roll maintenance, minimizes failures and downtime losses, maximizes the equipment's potential, and facilitates decision-makers in determining maintenance timing.

[0003] Traditional remaining life prediction methods require a large amount of data to support the analysis of equipment degradation processes. For rolling mills, this is restricted by many factors such as test costs, technical conditions, and time costs. Collecting a large amount of failure data on rolling mills takes too long and requires a large amount of capital, making it difficult to obtain sufficient failure data. Therefore, traditional remaining life prediction methods are not suitable for rolling mills.

[0004] Currently, the remaining life of rolls is often predicted through certain inspection methods during roll downtime. This makes it difficult to predict the remaining life of rolls in real time without affecting their operating status, making it difficult to provide a basis for timely roll grinding or replacement. Technicians have developed a prediction system that requires the addition of a coupling agent between the probe and the roll. This technology significantly affects the roll's real-time operating status and only detects internal defects, making it less effective in detecting surface and near-surface performance degradation. Summary of the Invention

[0005] Based on this, it is necessary to address the technical problem in the existing technology that it is difficult to accurately predict the real-time remaining life of the hot rolling working roll without affecting the working state of the hot rolling working roll. The present invention provides a method and system for predicting the real-time remaining life of the hot rolling working roll.

[0006] The present invention discloses a method for predicting the real-time remaining life of a hot rolling work roll, which comprises the following steps:

[0007] 1. Establishing a feature parameter dataset

[0008] A raw eddy current signal sequence is collected from the surface of a roll to be predicted, and characteristic parameters representing the roll performance degradation are extracted using a time-frequency analysis method. The characteristic parameter dataset is then established based on the characteristic parameters. The raw eddy current signal sequence includes an initial, non-operating pulsed eddy current signal and a real-time, operating pulsed eddy current signal. The characteristic parameters include crack, oxidation, and wear characteristics of the roll surface.

[0009] 2. Constructing a Roller Performance Degradation Assessment Model

[0010] (1) According to the current working state of the roll, the characteristic value threshold ω of the roll failure is determined. Based on the Wiener process and introducing the characteristic value threshold ω, the probability density function of the remaining life of the roll is established:

[0011]

[0012] Where, t represents the time when the equipment life is reached. k represents the number of revolutions that the roller has run at the current moment. T represents the life. f Tk (t) is the probability density function of the remaining life of the roller, and the probability density function is k The unknown parameters under a represents the drift parameter, and the drift parameter a is set to obey the mean value μ a And the variance is The normal distribution of σ is represented by the diffusion parameter.

[0013] (2) According to the characteristic parameter data set, the time t k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k .

[0014] (3) The unknown parameter Θ is obtained k Substituting it into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model.

[0015] 3. Predicting the remaining life of the roll

[0016] The operating time point of the roll is used as input, and the roll performance degradation assessment model is used to calculate the remaining life of the roll at each moment, thereby achieving real-time prediction of the remaining life of the hot rolling work roll.

[0017] As a further improvement of the above solution, the method for collecting the original eddy current signal sequence includes the following steps:

[0018] S11: Divide the working surface of the roller into a plurality of detection areas.

[0019] S12: Setting the sampling frequency of pulsed eddy current detection on the roller.

[0020] S13: Perform preliminary pulsed eddy current detection on each detection area on the roller surface in the initial non-operating stage according to the set sampling frequency, and perform real-time pulsed eddy current detection in the operating stage, so as to obtain the non-operating pulsed eddy current signal and the operating pulsed eddy current signal, that is, the original eddy current signal sequence.

[0021] As a further improvement of the above solution, in step S11, a detection probe senses the change in the magnetic field on the roller surface and generates a detection signal.

[0022] When dividing the working surface of the roller, first, based on the size of the detection probe and the roller, the roller surface is divided into n1 circumferential regions along the circumferential direction, and then each circumferential region is divided into n2 axial regions along the axial direction of the roller, thereby dividing the working surface of the roller into n detection regions, where n = n1 × n2.

[0023] As a further improvement of the above solution, after collecting each signal in the original eddy current signal sequence on the roller surface, each signal is further subjected to denoising processing.

[0024] As a further improvement to the above solution, the method for establishing the characteristic parameter data set includes the following steps:

[0025] S21: taking the non-operating pulsed eddy current signal as a reference signal, and then taking the operating pulsed eddy current signals of the roller in the operating stage as an initial sample group.

[0026] The periods of the sample signals of each group in the initial sample group and the reference signal are consistent.

[0027] S22: performing differential processing on each group of sample signals and the reference signal to obtain multiple groups of differential signals, and decomposing each group of differential signals using a variational mode decomposition method to obtain multiple intrinsic mode functions associated with each group of differential signals.

[0028] S23: Performing Hilbert transform on the plurality of intrinsic mode functions to obtain a Hilbert spectrum of the pulsed eddy current detection signal, and superimposing corresponding frequency components in the Hilbert spectrum in the entire time domain to obtain a marginal spectrum of the pulsed eddy current signal.

[0029] S24: extracting the marginal spectrum peak value in the marginal spectrum graph and using it as an intermediate feature value, obtaining the marginal spectrum peak value data of each detection area at each time point, and then drawing a data graph of the running time and the corresponding marginal spectrum peak value.

[0030] S25: performing linear fitting processing on the multiple marginal spectrum peaks in the data graph, so that the multiple target feature quantities MS obtained after the processing are approximately linearly related to the running time.

[0031] S26: Construct the feature parameter data set according to multiple target feature quantities MS.

[0032] As a further improvement of the above solution, the characteristic parameter data set is expressed as:

[0033]

[0034] Among them, MS(t ik ) represents t ik The characteristic value at the moment. 1≤i≤n, n is the number of samples tested. 1≤k≤m i , m i is the number of all detection time points of the i-th sample. i The feature dataset corresponding to the i-th sample is and MS=(MS1,...,MS n ). The increment of the eigenvalue is expressed as:

[0035]

[0036] Among them, ΔMS ik =MS ik -MS i，k-1 Corresponding to the i-th sample from time point t k-1 At time t k The degradation increment.

[0037] As a further improvement of the above scheme, the degradation process of the roller conforms to the Wiener process with linear drift, that is:

[0038] MS(t)=at+σB(t)

[0039] Where MS(t) is the characteristic quantity of the roller that characterizes the performance degradation. B(t) is a standard Brownian motion used to characterize the randomness of the degradation process.

[0040] As a further improvement of the above scheme, the Bayesian update and EM algorithm are used to estimate the time t k The unknown parameter Θ under k .

[0041] The present invention also discloses a hot rolling work roll real-time remaining life prediction system, which adopts any of the above hot rolling work roll real-time remaining life prediction methods. The prediction system includes: a signal acquisition module, a signal processing module, a model generation module and a calculation module.

[0042] The signal acquisition module is used to collect an original eddy current signal sequence of the roller surface to be predicted.

[0043] The signal processing module uses a time-frequency analysis method to extract characteristic parameters used to characterize the roll performance degradation, and then establishes the characteristic parameter dataset based on these characteristic parameters. The raw eddy current signal sequence includes an initial, non-operating pulsed eddy current signal and a real-time, operating pulsed eddy current signal. The characteristic parameters include crack characteristics, oxidation characteristics, and wear characteristics of the roll surface.

[0044] The model generation module is used to determine the characteristic value threshold of the roller failure according to the current working state of the roller, and introduce the characteristic value threshold based on the Wiener process to establish the probability density function of the remaining life of the roller. k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k The unknown parameter Θ is obtained k Substituting it into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model.

[0045] The calculation module is used to take the operating time point of the roll as input, and use the roll performance degradation assessment model to calculate the remaining life of the roll at each moment, thereby realizing real-time prediction of the remaining life of the hot rolling work roll.

[0046] As a further improvement of the above solution, the prediction system further includes: a signal generating module and a detection probe.

[0047] The signal generating module is used to generate a square wave signal to the roller.

[0048] The detection probe is used to convert the detected magnetic field changes on the roller surface into detection signals and send them to the signal acquisition module. The signal acquisition module converts the received detection signals into digital signals and sends them to the signal processing module for subsequent processing.

[0049] Compared with the prior art, the technical solution disclosed in the present invention has the following beneficial effects:

[0050] 1. This real-time remaining life prediction method for hot-roll work rolls utilizes a time-frequency domain feature extraction method based on lossless eddy current signals, effectively suppressing the impact of noise on eigenvalues ​​and preserving the effective information in the signals. The differential signals are processed using variational mode decomposition, and the multiple intrinsic mode functions associated with each differential signal are then Hilbert transformed to obtain eigenvalues ​​that characterize the performance of the hot-roll work rolls. Based on the degradation trajectory of the eigenvalues, a roll performance degradation assessment model based on the Wiener process is employed to accurately predict the remaining life of the hot-roll work rolls in real time, while minimizing the impact on the rolls' operating conditions. Compared to traditional model-based and experience-based remaining life assessment methods, this prediction method offers improved adaptability, prediction accuracy, and robustness, avoiding the drawbacks of traditional approaches, such as the difficulty and inaccuracy of mechanism modeling and the limited number of experimental data samples.

[0051] In addition, this prediction method can not only be used for real-time remaining life prediction of hot rolling work rolls, but also for real-time remaining life prediction of support rolls and cold rolling work rolls. It can also realize online real-time acquisition of roll surface degradation conditions, and has high practical value.

[0052] 2. The beneficial effects of the real-time remaining life prediction system for hot rolling work rolls are the same as those of the above-mentioned prediction method, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for predicting the real-time remaining life of a hot rolling work roll in Example 1 of the present invention;

[0054] Figure 2 for Figure 1 Schematic diagram of the process of establishing the characteristic parameter data set and constructing the roll performance degradation evaluation model;

[0055] Figure 3 Schematic diagram of the surface area division of the hot rolling work roll in Example 1 of the present invention;

[0056] Figure 4 The hot rolling working rolls in Example 1 of the present invention are k =Probability density function of remaining life at time 2.5;

[0057] Figure 5 The hot rolling working rolls in Example 1 of the present invention are k =Probability density function of remaining life at time 4;

[0058] Figure 6 The hot rolling working rolls in Example 1 of the present invention are k =Probability density function of remaining life at time 5;

[0059] Figure 7 α-λ performance index diagram of the hot rolling work roll in Example 1 of the present invention (α=0.2);

[0060] Figure 8 is the parameter μ in the roll performance degradation evaluation model in Example 1 of the present invention ak Estimation results at different measurement times. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0063] See also Figure 1 and Figure 2 This embodiment provides a method for predicting the real-time remaining life of a hot rolling work roll, which includes the following steps:

[0064] 1. Establishing a feature parameter dataset

[0065] A raw eddy current signal sequence is collected from the roll surface to be predicted. Time-frequency analysis is used to extract characteristic parameters that characterize roll degradation. A characteristic parameter dataset is then created based on these characteristic parameters. The raw eddy current signal sequence includes both the initial, non-operating pulsed eddy current signal and the real-time, operating pulsed eddy current signal. The characteristic parameters include crack, oxidation, and wear characteristics on the roll surface.

[0066] During operation, the rollers are subjected to cyclic thermal stress (for hot rolling work rolls) and contact stress, which can cause cracks, wear, oxidation and other losses on the surface, as well as changes in material structure. These factors directly affect the electrical conductivity, magnetic permeability and material continuity of the roller surface, and indirectly affect the intensity and conduction path of the eddy current. Therefore, by analyzing the characteristics of the eddy current response signal, the degradation information of the hot rolling work rolls can be obtained.

[0067] In this embodiment, a pulsed eddy current testing test bench can be built near the roller to analyze the roller's eddy current response signal. The test bench can include a signal generation module, a detection probe, a signal acquisition module, and a signal processing module. The signal generation module uses a signal generator of some existing models to generate a pulse square wave. The detection probe is a homemade probe of mutual inductance type, mainly composed of a magnetic core, a detection coil, an excitation coil, and a shielding copper foil, which is used to excite and receive the magnetic field. The signal acquisition module can use an acquisition card of some existing models with a high sampling rate. The signal processing module can use a computer to facilitate subsequent signal processing and analysis.

[0068] Specifically, the method for collecting the original eddy current signal sequence includes the following steps:

[0069] S11: Divide the working surface of the roller into multiple detection areas.

[0070] In step S11, the above-mentioned detection probe can generate a detection signal based on the magnetic field changes on the roller surface. When dividing the working surface of the roller, the roller surface is first divided into n1 circumferential regions along the circumference of the roller based on the size of the detection probe and the roller. Each circumferential region is then divided into n2 axial regions along the axial direction of the roller, thereby dividing the working surface of the roller into n detection regions. Here, n = n1 × n2.

[0071] See also Figure 3 In this embodiment, based on the actual relative size of the detection probe and the roller, a zone (circumferential zone) can be defined for every 40° of roller rotation. This means that nine zones can be defined on a circumference. Furthermore, the roller is further divided into zones in the axial direction, taking into account the detection probe and the axial length of the roller. Thus, the entire working surface of the roller is defined as one zone in the axial direction. Therefore, the entire working surface of the roller is divided into n = 9 × 1 = 9 detection zones.

[0072] S12: Set the sampling frequency for pulsed eddy current testing of the roller.

[0073] S13: Perform preliminary pulsed eddy current detection on each detection area on the roller surface in the initial non-operating stage according to the set sampling frequency, and perform real-time pulsed eddy current detection in the operating stage, thereby obtaining non-operating pulsed eddy current signals and operating pulsed eddy current signals, that is, the original eddy current signal sequence.

[0074] After collecting each signal in the original eddy current signal sequence on the roller surface, each signal can be subjected to denoising processing.

[0075] In this embodiment, before the roller is first operated (i.e., in its brand new state after leaving the factory), surface eddy current detection is performed on nine detection areas of the roller surface, and the measured pulsed eddy current signals are used as reference signals. Only detection is required sequentially. The pulsed eddy current signals detected every 10 seconds of roller operation are then used as sample signals of the initial sample group, thereby establishing a characteristic parameter data set. Specifically, the method for establishing the characteristic parameter data set may include the following steps:

[0076] S21: The non-operating pulsed eddy current signal is used as a reference signal, and then the operating pulsed eddy current signals of the roller in the operating stage are used as an initial sample group, wherein the periods of the sample signals in the initial sample group are consistent with those of the reference signal.

[0077] S22: Perform differential processing on each set of sample signals and the reference signal to obtain multiple sets of differential signals. A preset number of cycles is set for the differential signals before variational mode decomposition. The number of cycles of the differential signals obtained after differential processing of the collected pulsed eddy current detection signals should be no less than the preset number of cycles. If the number of cycles of the differential signals is less than the preset number of cycles, the differential signals are periodically extended to meet the preset number of cycles. The consistency of the number of cycles of the detection signals facilitates subsequent data processing.

[0078] After differentiating the signals of the initial sample group from the reference signal, multiple groups of differential signals can be obtained. Then, the variational mode decomposition method is used to decompose each group of differential signals, thereby obtaining multiple intrinsic mode functions associated with each group of differential signals.

[0079] In this embodiment, the number of intrinsic mode functions is set to q, and the constrained variational model is:

[0080]

[0081] Where δ is the Dirac distribution; {μ q}={μ1,μ2,…,μ q} is the intrinsic mode function obtained after decomposition; {ω q}={ω1,ω2,…,ω q} is the center frequency of each natural mode function; x is the original signal.

[0082] Introducing the penalty term and Lagrange multiplier λ, the specific steps of the variational mode decomposition algorithm are as follows:

[0083] Step 1: Initialization Let n = 0;

[0084] Step 2: Let n = n + 1;

[0085] Step 3: Set q = 0, q = q + 1, and update μ q、ω q ,λ;

[0086]

[0087]

[0088]

[0089] Step 4: Repeat Step 2 and Step 3 until the following equation is satisfied for the given discrimination accuracy e, and then stop the iteration and output μ p .

[0090]

[0091] S23: Perform Hilbert transform on multiple intrinsic mode functions to obtain the Hilbert spectrum of the pulsed eddy current detection signal, and superimpose the corresponding frequency components in the Hilbert spectrum in the entire time domain to obtain the marginal spectrum of the pulsed eddy current signal. The specific steps are as follows:

[0092] For signal x k (t) Perform variational mode decomposition to obtain the intrinsic mode function μ ik (t), whose Hilbert transform is as follows:

[0093]

[0094] Where: μ ik Hilbert transform of (t), n k Represents the number of intrinsic mode functions obtained after variational mode decomposition of the differential signal.

[0095] According to the intrinsic mode function μ ik (t) and Hilbert transform function Get the corresponding analytical signal Analyzing the signal The expression is as follows:

[0096]

[0097] in,

[0098] Where: A ik (t) is the instantaneous amplitude; is the instantaneous phase.

[0099] Make the Hilbert spectrum of the analytical signal, which is represented as follows:

[0100]

[0101]

[0102] Where: ω ik (t) is the instantaneous angular frequency; f ik (t) is the instantaneous frequency.

[0103] By superimposing the corresponding frequency components of the above Hilbert spectrum in the entire time domain, we can obtain the marginal spectrum, which is expressed as:

[0104]

[0105] Where: H k (ω, t) represents the Hilbert spectrum of the signal, h k (ω) represents the marginal spectrum of the signal.

[0106] S24: extracting the marginal spectrum peak value in the marginal spectrum graph and using it as the intermediate feature value, obtaining the marginal spectrum peak value data at each time point in each detection area, and then drawing a data graph of the running time and the corresponding marginal spectrum peak value.

[0107] S25: Perform linear fitting on the multiple marginal spectrum peaks in the data graph, so that the multiple target feature quantities MS obtained after the processing are approximately linearly related to the running time. In this embodiment, step S5 is to perform mathematical processing on the marginal spectrum peaks on the vertical axis. The expression of the mathematical transformation is:

[0108]

[0109] Where: MS ik is the latest feature quantity, Pm ik is the marginal spectrum peak of the differential signal. After processing, it can be obtained that the target feature quantity MS has an approximately linear relationship with the running time.

[0110] Comprehensively organize and obtain the changing trend of characteristic quantities along with the life of hot rolling work rolls, in preparation for subsequent real-time remaining life prediction.

[0111] S26: Construct a feature parameter data set based on multiple target feature quantities MS.

[0112] 2. Constructing a Roller Performance Degradation Assessment Model

[0113] (1) According to the current working state of the roll, the characteristic value threshold ω of the roll failure is determined. Based on the Wiener process and the characteristic value threshold ω, the probability density function of the remaining life of the roll is established:

[0114]

[0115] Where, t represents the time when the equipment life is reached. k represents the number of revolutions that the roller has run at the current moment. T represents the life. fTk (t) is the probability density function of the remaining life of the roller, and the probability density function at time t k The unknown parameters under a represents the drift parameter and σ represents the diffusion parameter.

[0116] The experimental data for this example was obtained through fatigue and wear equivalent experiments on hot-roll work rolls. A series of pulsed eddy current (PEC) data was acquired through condition monitoring of hot-roll work roll specimens. Real-time marginal spectrum peak values ​​were obtained using a signal processing method using variational mode decomposition and Hilbert transform. Mathematical processing of the marginal spectrum peak values ​​on the ordinate yielded a target characteristic value, MS, which exhibits an approximately linear relationship with run time.

[0117] In this embodiment, due to the limitations of the experimental detection equipment conditions and the size of the hot-rolled working roll specimen, and the fact that the hot-rolled working roll specimen did not show obvious damage, no obvious signal difference was measured before 20,000 revolutions. After 20,000 revolutions, due to the increase in operating time and the accumulation of damage, the pulsed eddy current signal began to change, and the remaining life of the roll could be predicted.

[0118] The target characteristic quantity MS changes with the operation of the hot rolling work roll, which can be specifically expressed as:

[0119]

[0120] Among them, MS(t ik ) represents t ik The characteristic value at the moment. 1≤i≤9, 1≤k≤m i , m i is the number of all detection time points of the i-th sample. i The characteristic data set corresponding to the i-th sample is And MS = (MS1, ..., MS9). The increment of the eigenvalue is expressed as:

[0121]

[0122] Among them, ΔMS ik =MS ik -MS i，k-1 Corresponding to the i-th sample from time point t k-1 At time t k The degradation increment.

[0123] The degradation process of the hot rolling work roll conforms to the Wiener process with linear drift, that is, MS(t) = at + σB(t), where MS(t) represents the characteristic quantity of the hot rolling work roll that characterizes the performance degradation, a represents the drift parameter, B(t) is a standard Brownian motion used to characterize the randomness of the degradation process, and σ represents the diffusion parameter. In order to better describe the uncertainty in the degradation data, the drift parameter a is set to obey the mean μ a And the variance is A normal distribution is used to characterize the influence of random factors. A vector is used to describe the unknown parameters of the model.

[0124] When k = 25000, that is, t k =2.5,m k =25000,

[0125] In this embodiment, the characteristic value threshold value ω=1 when the hot rolling work roll fails is determined based on the current working state of the roll. According to the concept of first arrival time, let R k is the hot rolling working roll at time t k = 2.5, the corresponding remaining life is defined as R k =inf{r k :MS(t k +r k )≥1|MS 0：k Therefore, the above-mentioned probability density function of the remaining life of the roller is established.

[0126] (2) According to the characteristic parameter data set, the time t k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k .

[0127] In this embodiment, the time t can be estimated by using the Bayesian update and EM algorithm in a collaborative manner. k = 2.5 when the probability density function has unknown parameters The specific process includes the following:

[0128] Assume a ik Denotes the drift parameter corresponding to a single eigenvalue sample i, and let Ω k =(a 1k ,...,a 9k ), Δt h =t h -t h-1 , h=0,1,...,25000. When MS 0：k and Ω kWhen all can be obtained through observation, the corresponding complete log-likelihood function is:

[0129]

[0130] set up Indicates that the j-th step is based on the degradation data MS 0：k For unknown parameters Θ k The estimated value, given the known MS 0：k；i and Under the conditions of a ik The posterior estimated distribution of is also in line with the normal distribution, so its mean is The variance is a ik The posterior distribution of can be updated by the Bayesian formula:

[0131]

[0132] in:

[0133]

[0134]

[0135] Using the EM algorithm, we first obtain the latent variable Ω k The expectation of the complete log-likelihood function is then maximized in the M-step to the expectation obtained in the E-step. The E-step calculates the expected value of Ω. k of Expectations In M-step, The latest parameter estimates are available in:

[0136]

[0137]

[0138]

[0139] After 100 iterations of Bayesian updating and EM algorithm, and Convergence, we can get

[0140] (3) The unknown parameter Θ is obtained k Substituting it into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model.

[0141] In the previous article, we find the unknown parameter Θ k, from this we can obtain the probability density function of the remaining life of the hot rolling work roll when k = 25000:

[0142]

[0143] That is, the remaining life of the hot rolling work roll is 46,700 revolutions at the time k = 25,000. k =2.5 when the probability density function of the roller life is as follows Figure 4 shown.

[0144] Similarly, when k = 40000, that is, t k =4,m k =40000, Set the initial iteration parameters:

[0145] After 100 iterations of Bayesian updating and EM algorithm, and Convergence, we can get From this, we can obtain the probability density function of the remaining life of the hot rolling work roll when k = 40000:

[0146]

[0147] That is, the remaining life of the hot rolling work roll is 17,900 revolutions at the time k = 40,000. k The probability density function of the remaining life at the time of =4 is as follows Figure 5 shown.

[0148] When k = 50000, that is, t k =5,m k =5, ω=1, set the initial iteration parameters:

[0149] After 100 iterations of Bayesian updating and EM algorithm, and Convergence, we can get From this, we can get the probability density function of the remaining life of the hot rolling work roll when k = 50000: Figure 6 shown.

[0150] Through full cycle testing, the actual life of the hot rolling work roll specimen was found to be 57,500 revolutions.

[0151] 3. Predicting the remaining life of the roll

[0152] The operating time point of the roll is taken as input, and the roll performance degradation assessment model is used to calculate the remaining life of the roll at each moment, thereby realizing real-time prediction of the remaining life of the roll.

[0153] In order to reflect the remaining life prediction performance by calculating the α-λ index, we can make real-time life prediction for the hot rolling work roll every 10r. Figure 7 The α-λ performance results corresponding to α = 0.2 are given for the predicted remaining life. To simplify the calculation, we take a time point every 2,500 revolutions for calculation to show the quality of the prediction performance of this method. Figure 8 Given the parameter μ in the roll performance degradation evaluation model ak Estimation results at different measurement times.

[0154] As shown in the α-λ performance index plot, the proposed RLS prediction method struggles to obtain good parameter estimates in the early stages of the forecast due to the limited available degradation data. This results in a significant deviation from the actual RLS. However, as degradation data accumulates, the resulting RLS estimates become increasingly accurate. While the accuracy of the early predictions is not as good as that of the later ones, it remains within an acceptable range and meets the requirements of the α-λ performance index, demonstrating that the proposed method can continuously improve the RLS prediction accuracy as more degradation data becomes available.

[0155] In summary, the real-time remaining life prediction method for rolling mill rolls provided in this embodiment utilizes a time-frequency domain feature extraction method based on lossless eddy current signals, effectively suppressing the impact of noise on eigenvalues ​​and preserving the signal's valid information. The differential signals are processed using variational mode decomposition, and multiple intrinsic mode functions associated with each differential signal are then Hilbert transformed to obtain eigenvalues ​​representing the performance of the hot rolling work rolls. Based on the degradation trajectories of the eigenvalues, a Wiener process remaining life prediction model is used to predict the remaining life of the hot rolling work rolls in real time. Compared to traditional model-based and empirically-based remaining life assessment methods, the prediction method described in this embodiment offers improved adaptability, prediction accuracy, and robustness, avoiding the drawbacks of traditional approaches, such as the difficulty and inaccuracy of mechanism modeling and the limited number of experimental data samples. Furthermore, this prediction method can be used not only for real-time remaining life prediction of hot rolling work rolls, but also for backup rolls and cold rolling work rolls. It also enables online and real-time acquisition of roll surface degradation, thus possessing high practical value.

[0156] Example 2

[0157] This embodiment provides a system for predicting the real-time remaining life of hot rolling work rolls, which utilizes the method for predicting the real-time remaining life of hot rolling work rolls described in Example 1. The prediction system includes a signal acquisition module, a signal processing module, a model generation module, and a calculation module. It may also include a signal generation module and a detection probe.

[0158] The signal acquisition module is used to collect an original eddy current signal sequence of the roller surface to be predicted.

[0159] The signal processing module uses time-frequency analysis to extract characteristic parameters that characterize roll performance degradation. This data set is then constructed based on these characteristic parameters. The raw eddy current signal sequence includes both the initial, non-operating pulsed eddy current signal and the real-time, operating pulsed eddy current signal. The characteristic parameters include crack, oxidation, and wear characteristics on the roll surface.

[0160] The model generation module is used to determine the characteristic value threshold of the roll failure according to the current working state of the roll, and introduce the characteristic value threshold based on the Wiener process to establish the probability density function of the remaining life of the roll. k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k The unknown parameter Θ will be found k Substituting it into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model.

[0161] The calculation module is used to take the operating time point of the roll as input, and use the roll performance degradation assessment model to calculate the remaining life of the roll at each moment, thereby realizing real-time prediction of the remaining life of the hot rolling work roll.

[0162] The signal generating module is used to generate square wave signals to the roller.

[0163] The detection probe is used to convert the magnetic field changes sensed on the roller surface into detection signals and send them to the signal acquisition module. The signal acquisition module converts the received detection signals into digital signals and sends them to the signal processing module for subsequent processing.

[0164] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for predicting the real-time remaining life of a hot rolling work roll, characterized in that: It includes the steps of:

1. Establishing a feature parameter dataset A raw eddy current signal sequence of a roll surface to be predicted is collected, and characteristic parameters for characterizing the roll performance degradation are extracted using a time-frequency analysis method, and then the characteristic parameter data set is established based on the characteristic parameters; wherein the raw eddy current signal sequence includes an initial non-operating pulsed eddy current signal and a real-time operating pulsed eddy current signal; and the characteristic parameters include crack characteristics, oxidation characteristics, and wear characteristics of the roll surface; 2. Constructing a Roller Performance Degradation Assessment Model (1) According to the current working state of the roll, the characteristic value threshold ω of the roll failure is determined. Based on the Wiener process and introducing the characteristic value threshold ω, the probability density function of the remaining life of the roll is established: Where, t represents the time when the equipment life is reached; k represents the number of revolutions that the roller has run at the current moment; T represents the life; f Tk (t) is the probability density function of the remaining life of the roller, and the probability density function is k The unknown parameters under a represents the drift parameter, and the drift parameter a is set to obey the mean value μ a And the variance is Normal distribution; σ represents the diffusion parameter; (2) According to the characteristic parameter data set, the time t k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k ; (3) The unknown parameter Θ is obtained k Substituting the probability density function of the remaining life of the roll into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model; 3. Predicting the remaining life of the roll The operating time point of the roll is used as input, and the roll performance degradation assessment model is used to calculate the remaining life of the roll at each moment, thereby achieving real-time prediction of the remaining life of the hot rolling work roll.

2. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 1, characterized in that: The method for collecting the original eddy current signal sequence comprises the following steps: S11: Dividing the working surface of the roller into a plurality of detection areas; S12: Setting the sampling frequency of pulsed eddy current testing of the roller; S13: Perform preliminary pulsed eddy current detection on each detection area on the roller surface in the initial non-operating stage according to the set sampling frequency, and perform real-time pulsed eddy current detection in the operating stage, so as to obtain the non-operating pulsed eddy current signal and the operating pulsed eddy current signal, that is, the original eddy current signal sequence.

3. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 2, characterized in that: In step S11, a detection probe senses the change in the magnetic field on the roller surface and generates a detection signal; Among them, when dividing the working surface of the roller, the roller surface is first divided into n1 circumferential areas along the circumferential direction according to the size of the detection probe and the roller, and then each circumferential area is divided into n2 axial areas along the axial direction of the roller, and then the working surface of the roller is divided into n detection areas; among them, n = n1×n2.

4. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 1, characterized in that: After collecting each signal in the original eddy current signal sequence on the roller surface, each signal is subjected to denoising processing.

5. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 1, characterized in that: The method for establishing the characteristic parameter data set comprises the following steps: S21: using the non-operating pulsed eddy current signal as a reference signal, and then using the operating pulsed eddy current signals of the roller in the operating stage as an initial sample group; Wherein, the periods of the sample signals of each group in the initial sample group and the reference signal are consistent; S22: performing differential processing on each group of sample signals and the reference signal to obtain multiple groups of differential signals, and decomposing each group of differential signals using a variational mode decomposition method to obtain multiple intrinsic mode functions associated with each group of differential signals; S23: performing Hilbert transform on the plurality of intrinsic mode functions to obtain a Hilbert spectrum of the pulsed eddy current detection signal, and superimposing corresponding frequency components in the Hilbert spectrum in the entire time domain to obtain a marginal spectrum of the pulsed eddy current signal; S24: extracting the marginal spectrum peak value from the marginal spectrum graph and using it as an intermediate feature value, obtaining the marginal spectrum peak value data at each time point in each detection area, and then drawing a data graph of the running time and the corresponding marginal spectrum peak value; S25: performing linear fitting processing on the multiple marginal spectrum peaks in the data graph, so that the multiple target feature quantities MS obtained after the processing are approximately linearly related to the running time; S26: Construct the feature parameter data set according to multiple target feature quantities MS.

6. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 5, characterized in that: The characteristic parameter data set is expressed as: Among them, MS(t ik ) represents t ik The characteristic value at the moment; 1≤i≤n, n is the number of samples detected; 1≤k≤m i , m i is the number of all detection time points of the i-th sample; MS i The feature dataset corresponding to the i-th sample is and MS=(MS1,...,MS n ); where the increment of the eigenvalue is expressed as: Among them, ΔMS ik =MS ik -MS i，k-1 Corresponding to the i-th sample from time point t k-1 At time t k The degradation increment.

7. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 1, characterized in that: The degradation process of the roller conforms to the Wiener process with linear drift, that is: MS(t)=at+σB(t) Where MS(t) is the characteristic quantity of the roller that characterizes the performance degradation; B(t) is a standard Brownian motion used to characterize the randomness of the degradation process.

8. The method for predicting the real-time remaining life of a hot rolling work roll according to claim 1, characterized in that: Using Bayesian update and EM algorithm to estimate the time t k The unknown parameter Θ under k .

9. A system for predicting the real-time remaining life of hot rolling work rolls, characterized in that: The method for predicting the real-time remaining life of a hot rolling work roll according to any one of claims 1 to 8 is adopted; the prediction system comprises: A signal acquisition module, which is used to acquire an original eddy current signal sequence of a roller surface to be predicted; a signal processing module, which uses a time-frequency analysis method to extract characteristic parameters for characterizing the performance degradation of the roll, and then establishes the characteristic parameter data set based on the characteristic parameters; wherein the original eddy current signal sequence includes an initial non-operating pulsed eddy current signal and a real-time operating pulsed eddy current signal; and the characteristic parameters include crack characteristics, oxidation characteristics, and wear characteristics of the roll surface; The model generation module is used to determine the characteristic value threshold when the roll fails according to the current working state of the roll, and introduce the characteristic value threshold based on the Wiener process to establish the probability density function of the remaining life of the roll; according to the characteristic parameter data set, the time t k and its characteristic parameters at historical moments as input, and find the k The unknown parameter Θ under k ; The unknown parameter Θ is obtained k Substituting the probability density function of the remaining life of the roll into the probability density function of the remaining life of the roll, the solved probability density function of the remaining life of the roll is obtained, that is, the roll performance degradation assessment model; and The calculation module is used to take the operating time point of the rolling mill roll as input, calculate the remaining life of the rolling mill roll at each moment using the rolling mill roll performance degradation assessment model, and thus realize real-time prediction of the remaining life of the hot rolling working roll.

10. The hot rolling work roll real-time remaining life prediction system according to claim 9, characterized in that: The prediction system further includes: a signal generating module, configured to generate a square wave signal to the roller; and The detection probe is used to convert the sensed changes in the magnetic field on the roller surface into a detection signal and send it to the signal acquisition module; the signal acquisition module converts the received detection signal into a digital signal and sends it to the signal processing module for subsequent processing.

Images