A liner service evaluation method and device based on digital twinning and a medium

Abstract

The application relates to the technical field of digital twin simulation, and discloses a liner service evaluation method and device based on digital twin and a medium, the method comprising the following steps: constructing an initial digital twin model of a target mill liner; in the process of mill operation, observing the wear degree of the target mill liner according to non-invasive sensing signals outside the mill to obtain an observation vector of the target mill liner; taking the observation vector as a state variable to be estimated, predicting the virtual sensing characteristic value of the initial digital twin model at the next moment, and simultaneously obtaining the actual observation vector corresponding to the current liner; correcting the parameters of the initial digital twin model according to the residual error between the actual observation vector and the virtual sensing characteristic value; deducing the wear state of the current liner according to the corrected digital twin model, and outputting an evaluation report of the liner; and the application can improve the efficiency of abnormal diagnosis and repair of energy storage equipment.

Description

Technical Field

[0001] This invention relates to the field of digital twin simulation technology, and in particular to a method, equipment and medium for evaluating the service life of liners based on digital twins. Background Technology

[0002] In existing technologies, wear assessment of mill liners typically employs digital twin methods based on physical models or finite element simulations. These methods rely on the geometric and material parameters of the liner, along with pre-calibrated fixed wear coefficients, to predict wear conditions through simulation. This assessment process requires accurate pre-obtaining boundary conditions and offline testing for parameter verification and calibration after model establishment. The fundamental reason for this approach is that once the wear model parameters are determined, they are fixed within the model and cannot be automatically adjusted based on actual responses during mill operation, leading to a high dependence of the system on the accuracy of initial parameters and the stability of operating conditions.

[0003] During the actual continuous operation of the mill, operating parameters such as the particle size distribution of the feed material, the filling rate of the grinding media, and the cylinder rotation speed fluctuate continuously. The fixed wear model parameters are unable to reflect these dynamic changes, causing the deviation between the simulation predictions and the actual wear of the liner to gradually accumulate over time. Simultaneously, the nonlinear response caused by the fluctuations in operating conditions significantly amplifies the initial parameter deviations during long-term simulations, further exacerbating the inaccuracy of the model predictions. Consequently, the evaluation results based on this digital twin model deviate significantly from the actual service condition of the liner, making it difficult to provide timely and reliable information for maintenance decisions. Summary of the Invention

[0004] In a first aspect, this disclosure provides a method for evaluating the service life of liners based on digital twins, including: Construct an initial digital twin model of the target mill liner; During the operation of the mill, the wear degree of the target mill liner is observed based on non-intrusive sensing signals from outside the mill, and the observation vector of the target mill liner is obtained. Using the observation vector as the state variable to be estimated, predict the virtual sensing feature value of the initial digital twin model at the next time step, and at the same time obtain the actual observation vector corresponding to the current liner. The parameters of the initial digital twin model are corrected based on the residual between the actual observation vector and the virtual sensing feature value. The current wear status of the liner is inferred based on the revised digital twin model, and an evaluation report of the liner is output.

[0005] In a preferred embodiment, constructing the initial digital twin model of the target mill liner includes: Obtain the geometric parameters, material property parameters, and operating parameters of the target mill liner. The geometric parameters include the outline dimensions of the liner and the installation position data of the liner in each region. The material property parameters include the density, elastic modulus, Poisson's ratio, and hardness of the liner material. The operating parameters include the mill speed, grinding media filling rate, and particle size distribution data of the feed material. A three-dimensional geometric model of the target mill liner is established based on the geometric parameters; Based on the material property parameters and working condition parameters, material property boundary conditions and motion boundary conditions are configured in the three-dimensional geometric model to obtain the initial digital twin model.

[0006] In a preferred embodiment, establishing a three-dimensional geometric model of the target mill liner based on geometric parameters includes: Based on the installation position data of the liners in each region, the outline dimensions of the liners in each region are arranged in space in the circumferential and axial directions of the inner wall of the target mill cylinder to obtain a three-dimensional assembly model of the liner group. In the 3D assembly model of the liner group, the surface of the liner in each region is divided into multiple wear calculation mesh elements, and each wear calculation mesh element is assigned corresponding material property parameters. The three-dimensional assembly model of the liner assembly, after mesh generation and material property assignment, is combined with the geometric data of the target mill cylinder to obtain the three-dimensional geometric model of the target mill liner.

[0007] In a preferred embodiment, the step of observing the wear degree of the target mill liner based on non-invasive sensing signals from outside the mill during mill operation, and obtaining the observation vector of the target mill liner, includes: Vibration sensors are deployed at the bearing housing of the target mill, acoustic emission sensors are deployed on the outer wall of the target mill cylinder, and current sensors are deployed inside the motor control cabinet of the target mill. Vibration signals during mill operation are collected using vibration sensors, acoustic emission signals during mill operation are collected using acoustic emission sensors, and motor current signals during mill operation are collected using current sensors. The vibration signal, acoustic emission signal, and motor current signal are filtered and denoised respectively to obtain the denoised vibration signal, denoised acoustic emission signal, and denoised motor current signal.

[0008] In a preferred embodiment, obtaining the denoised vibration signal, the denoised acoustic emission signal, and the denoised motor current signal includes: The time-domain features of the denoised vibration signal are extracted to obtain vibration feature indices, which include the time-domain amplitude features and frequency band energy distribution features of the vibration signal. The time-domain features of the denoised acoustic emission signal are extracted to obtain acoustic emission feature indices, which include the energy accumulation rate feature and peak frequency distribution feature of the acoustic emission signal. The time-domain features of the denoised motor current signal are extracted to obtain current feature indices, which include the steady-state mean drift characteristics and fluctuation variance characteristics of the motor current. The wear characteristic values ​​of the target mill liner are obtained by weighted fusion of the normalized vibration characteristic index, acoustic emission characteristic index and current characteristic index. The wear characteristic values ​​are output as the observation vector of the target mill liner.

[0009] In a preferred embodiment, the step of predicting the virtual sensing feature value of the initial digital twin model at the next time step, using the observed vector as the state variable to be estimated, and simultaneously obtaining the actual observed vector corresponding to the current liner, includes: Obtain the initial digital twin model and wear model parameters at the current moment; The initial digital twin model is driven forward based on the wear model parameters at the current moment to calculate the predicted value of the liner wear state at the next moment; By mapping the sensing features in the digital twin model, the predicted wear state of the liner is mapped to the virtual vibration feature value, virtual acoustic emission feature value, and virtual current feature value at the next moment. Virtual vibration characteristic values, virtual acoustic emission characteristic values, and virtual current characteristic values ​​are used as virtual sensing characteristic values.

[0010] In a preferred embodiment, the step of driving the initial digital twin model forward based on the wear model parameters at the current moment to calculate the predicted value of the liner wear state at the next moment includes: Input the wear model parameters at the current moment into the initial digital twin model, and based on the wear model parameters at the current moment and the wear state of the liner at the current moment, extrapolate a prediction cycle forward according to the preset time step; At the end of the prediction period, the wear depth increment of each wear calculation grid cell in the initial digital twin model is extracted; The current wear depth of each wear calculation grid cell is updated based on the wear depth increment to obtain the updated wear depth of each wear calculation grid cell; The updated wear depths of each wear calculation grid cell are summarized according to the region division to obtain the predicted wear state of the lining plate in each region at the next moment.

[0011] In a preferred embodiment, the step of correcting the parameters of the initial digital twin model based on the residual between the actual observed vector and the virtual sensing feature value includes: The actual observation vector and virtual sensing feature value are obtained at the same time. The actual observation vector and virtual sensing feature value are compared item by item to determine the deviation value between each item, so as to obtain the residual vector. Based on the positive and negative directions and magnitudes of the deviation values ​​in the residual vector, the adjustment direction and adjustment range of each parameter in the wear model parameters are determined respectively; The correction amount for the wear model parameters is generated based on the adjustment direction and adjustment range; The wear model parameters at the current moment are updated by superimposing the correction amount to generate the corrected wear model parameters; The corrected wear model parameters are written into the initial digital twin model, replacing the original wear model parameters in the initial digital twin model, to obtain the corrected digital twin model.

[0012] The present invention also provides a liner service evaluation device based on digital twins, the device comprising: The initial digital twin module is used to construct the initial digital twin model of the target mill liner. The wear observation module is used to observe the wear degree of the target mill liner during mill operation based on non-intrusive sensing signals from outside the mill, and obtain the observation vector of the target mill liner. The feature inference module is used to predict the virtual sensing feature values ​​of the initial digital twin model at the next time step, using the observation vector as the state variable to be estimated, and at the same time obtain the actual observation vector corresponding to the current liner. The model calibration module is used to correct the parameters of the initial digital twin model based on the residual between the actual observation vector and the virtual sensing feature value. The status assessment module is used to deduce the current wear status of the liner based on the corrected digital twin model and output an evaluation report of the liner.

[0013] The present invention also proposes a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention obtains an observation vector by real-time observation of the liner wear degree using non-invasive sensing signals. This observation vector is then used as the state variable to drive a digital twin model to predict the virtual feature value of the next time step. The residual between the actual observation and the virtual prediction is then used to correct the model parameters online, forming a closed-loop feedback mechanism of "observation-prediction-correction". This allows the digital twin model to continuously and adaptively adjust the wear parameters based on the actual response signal during operation, fundamentally overcoming the problem of accumulated prediction bias caused by the fixed wear model parameters that cannot be dynamically adjusted with changing operating conditions, as pointed out in the background art. This significantly improves the accuracy of liner wear state assessment under long-term time-domain simulation.

[0015] 2. This invention extracts multi-dimensional time-domain feature indicators from vibration signals, acoustic emission signals, and motor current signals, and then performs weighted fusion to construct a comprehensive wear observation vector encompassing multi-physical field information such as mechanical vibration, acoustic response, and electrical parameters. Based on the closed-loop correction framework established in the independent claims, this fusion strategy further enhances the completeness and anti-interference capability of the observation vector in representing the wear state of the liner, providing a more reliable data foundation for residual calculation and accurate generation of parameter correction quantities, thereby improving the convergence stability of the entire closed-loop correction process. Attached Figure Description

[0016] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings: Figure 1 The flowchart of a digital twin-based liner service evaluation method according to Embodiment 1 of the present invention is shown. Figure 2 The diagram shows a functional block diagram of a liner service evaluation device based on digital twin according to Embodiment 2 of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this disclosure.

[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer device such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be performed in a different order than that shown here.

[0019] Example 1

[0020] Figure 1 This is a schematic flowchart illustrating a digital twin-based liner service evaluation method provided in an embodiment of this disclosure. Figure 1 As shown, a service evaluation method for liners based on digital twins includes: Construct an initial digital twin model of the target mill liner; In one feasible implementation of this invention, the geometric parameters, material property parameters, and operating parameters of the target mill liner are obtained. The geometric parameters include the liner's outline dimensions and the installation position data of the liner in each region, where the outline dimensions encompass the liner's length, width, thickness, and lifting bar height. The material property parameters include the liner material's density, elastic modulus, Poisson's ratio, and hardness. The operating parameters include mill rotational speed, grinding media filling rate, and feed material particle size distribution data. A three-dimensional geometric model of the target mill liner is established based on the geometric parameters. Then, based on the material property parameters and operating parameters, material property boundary conditions and motion boundary conditions are configured in the three-dimensional geometric model to obtain an initial digital twin model. Through the above modeling process, an initial calculable three-dimensional digital mapping body can be established for the target mill liner, providing a geometric and physical basis for subsequent wear prediction.

[0021] When establishing the three-dimensional geometric model, based on the installation position data of the liners in each region, the contour dimensions of the liners in each region are spatially arranged in the circumferential and axial directions of the inner wall of the target mill cylinder, resulting in a three-dimensional assembly model of the liner assembly. Specifically, the inner wall of the cylinder is divided into multiple installation sections along the circumferential direction, and a set of liners is assembled in each section. The liners between adjacent sections are arranged alternately in the circumferential direction, while the liners in each row are arranged sequentially at predetermined intervals along the axial direction of the cylinder. Through the above assembly arrangement, the constructed three-dimensional geometric model is made consistent with the actual liner installation form of the inner wall of the mill cylinder in terms of spatial structure. Subsequently, in the three-dimensional assembly model of the liner assembly, the surface of the liner in each region is divided into multiple wear calculation mesh elements, and corresponding material property parameters are assigned to each wear calculation mesh element. Finally, the three-dimensional assembly model of the liner assembly after mesh generation and material property assignment is combined with the geometric data of the target mill cylinder to obtain a complete three-dimensional geometric model of the target mill liner. By meshing and assigning material properties element by element, the three-dimensional geometric model can support numerical calculation methods such as finite element or discrete element methods for quantitative calculation of the wear process.

[0022] In one feasible implementation, the size of the wear calculation grid cells is determined comprehensively based on the required computational accuracy and hardware resources. Optionally, for key wear areas (such as the root of the lifting strip and the center area of ​​the liner working surface), a first grid size of 5mm to 15mm is used, which can effectively capture the spatial gradient changes in wear depth within this size range. Optionally, for non-key wear areas (such as the non-working edge area of ​​the liner), a second grid size of 20mm to 40mm is used to reduce the overall number of grids and computational resource consumption. In another feasible implementation, the grid size can also be differentiated according to the different installation positions of the liner. For example, for the liner in the middle area of ​​the cylinder, since this area usually experiences the most severe wear, a finer grid is used, while for the liner at both ends of the cylinder, a relatively sparse grid is used. Through the above differentiated grid division strategy, the accuracy of wear assessment is ensured while the computational scale is reasonably controlled to balance engineering implementation efficiency.

[0023] When configuring material property boundary conditions and motion boundary conditions, the material property parameters are assigned to the corresponding wear calculation mesh elements in the three-dimensional geometric model, and the rotational motion constraints of the mill cylinder and the load boundary conditions of the grinding media on the liner are set according to the operating parameters. In one feasible implementation, the density, elastic modulus, Poisson's ratio, and hardness values ​​can be obtained by consulting material handbooks or conducting material sample tests based on the actual material of the liner. For example, for high manganese steel liners, the density is taken as 7.8×10³ kg / m³ to 7.85×10³ kg / m³, the elastic modulus is taken as 200 GPa to 210 GPa, the Poisson's ratio is taken as 0.25 to 0.30, and the hardness value is taken as 180 HB to 250 HB. In another feasible implementation, for alloy steel liners, the density ranges from 7.7 × 10³ kg / m³ to 7.85 × 10³ kg / m³, the elastic modulus from 190 GPa to 210 GPa, the Poisson's ratio from 0.28 to 0.32, and the hardness from 300 HB to 450 HB. The specific values ​​of these parameters can be set by those skilled in the art based on the actual material grade of the liner and the material inspection report provided by the supplier, or they can be obtained through on-site sampling and hardness and tensile testing. By associating material parameters with the geometric model, the digital twin model acquires the physical calculation capabilities to reflect the actual material properties of the liner.

[0024] The motion boundary conditions include at least the rotational motion parameters of the mill cylinder. The preferred range for the mill speed is 70% to 100% of the rated speed. Within this range, the mill operates in a normal cascading or precipitating manner, and the liners are most typically subjected to the impact and grinding action of the grinding media. The preferred range for the grinding media filling rate is 30% to 45%. Within this range, the grinding media can form a good cascading trajectory during the cylinder rotation, resulting in a reasonable impact load distribution on the liner surface. The preferred range for the particle size distribution of the feed material is: 20% to 40% for fine particles smaller than 0.075 mm, 30% to 50% for intermediate particles (0.075 mm to 2 mm), and 10% to 30% for coarse particles larger than 2 mm. Through the parameterized configuration of the above boundary conditions, the initial digital twin model can simulate the mill's operating state under different working conditions. When the wear parameters in the model drift with changes in operating conditions, the model is corrected in subsequent steps using sensor observations, which effectively overcomes the problem of reduced prediction accuracy caused by fixed wear parameters in traditional digital twin models.

[0025] During the operation of the mill, the wear degree of the target mill liner is observed based on non-intrusive sensing signals from outside the mill, and the observation vector of the target mill liner is obtained. In this embodiment of the invention, a vibration sensor is deployed at the bearing housing of the target mill, an acoustic emission sensor is deployed on the outer wall of the target mill cylinder, and a current sensor is deployed inside the motor control cabinet of the target mill.

[0026] Vibration signals during mill operation are collected using vibration sensors, acoustic emission signals during mill operation are collected using acoustic emission sensors, and motor current signals during mill operation are collected using current sensors.

[0027] The vibration signal, acoustic emission signal, and motor current signal are filtered and denoised respectively to obtain the denoised vibration signal, denoised acoustic emission signal, and denoised motor current signal.

[0028] Since the vibration sensor, acoustic emission sensor, and current sensor are all deployed outside the mill, no modifications are required to the mill's main structure. This allows for continuous signal acquisition during normal mill operation, avoiding production interruptions caused by downtime testing.

[0029] When deploying vibration sensors, the bearing housing position can effectively transmit the vibration energy generated by the impact between the liner and the grinding media during the operation of the mill cylinder, so that the collected vibration signal contains characteristic information related to the wear state of the liner.

[0030] In one feasible implementation, the vibration sensor employs a piezoelectric accelerometer with a measurement frequency range of 0.5 Hz to 10 kHz, a sensitivity of 50 mV / g to 100 mV / g, and a measurement range of ±50 g to ±100 g.

[0031] In another feasible implementation, the vibration sensor can also be a MEMS capacitive accelerometer with a measurement frequency range of 0Hz to 3kHz, a sensitivity of 100mV / g to 300mV / g, and a range of ±20g to ±200g. This type of sensor has good response characteristics in the low-frequency band and is small in size and low in cost, making it suitable for vibration signal acquisition of bearing housings with limited installation space.

[0032] When deploying acoustic emission sensors, the outer wall of the cylinder can effectively receive elastic wave signals generated by the liner during wear due to mechanisms such as micro-fracture, plastic deformation and friction. The frequency of these signals is much higher than that of conventional vibration signals, and they can reflect the damage evolution process at the microscale of the liner material.

[0033] In one feasible implementation, the acoustic emission sensor has a resonant frequency of 150 kHz, an operating frequency range of 100 kHz to 400 kHz, and a sensitivity of 65 dB to 75 dB.

[0034] In another feasible implementation, the resonant frequency of the acoustic emission sensor can be selected as a low-frequency model of 80kHz to 120kHz, or a high-frequency model of 200kHz to 300kHz. Sensors with different resonant frequencies have different sensitivities when detecting damage signals at different scales, and can be selected according to the actual characteristics of the mill specifications and the liner material.

[0035] When deploying current sensors, the motor control cabinet is equipped with sensors that can acquire the motor stator current signal. This current signal is directly related to the load torque borne by the mill cylinder during rotation, and the load torque is significantly affected by the friction state between the liner and the grinding media.

[0036] In one feasible implementation, the current sensor is a Hall effect type current sensor with a measurement range of 0A to 500A, an accuracy of ±0.5% to ±1.0%, and a bandwidth of 0Hz to 5kHz.

[0037] In another feasible implementation, the current sensor can also be a Rogowski coil type current sensor, with a measurement range of 0A to 1000A, an accuracy of ±0.2% to ±0.5%, and a bandwidth of 0Hz to 1MHz. This type of sensor has the characteristics of no magnetic saturation, good linearity, and fast response speed, and is suitable for the accurate acquisition of motor current signals containing a lot of harmonic components.

[0038] By filtering and denoising the vibration signal, acoustic emission signal, and motor current signal respectively, the influence of environmental noise and electromagnetic interference on the original signal can be effectively suppressed, thereby improving the signal-to-noise ratio of subsequent feature extraction.

[0039] In one feasible implementation, the vibration signal is filtered using a bandpass filter with a lower cutoff frequency of 0.5 Hz and an upper cutoff frequency of 1 kHz to filter out low-frequency DC drift and high-frequency electromagnetic interference components.

[0040] In another feasible implementation, the vibration signal can also be processed using a wavelet thresholding denoising method. The db4 wavelet basis function is selected, the decomposition level is 5, the threshold rule adopts the unbiased risk estimation criterion, and the soft thresholding function is used to shrink the detail coefficients. This method has better adaptability and denoising effect when processing non-stationary vibration signals.

[0041] In one feasible implementation, the acoustic emission signal is filtered using a bandpass filter with a lower cutoff frequency of 20kHz and an upper cutoff frequency of 400kHz to highlight the acoustic emission radio frequency band related to liner wear.

[0042] In another feasible implementation, the acoustic emission signal can be denoised using wavelet packet decomposition and reconstruction. The db8 wavelet basis function is selected, and the decomposition layer is 6 layers. After analyzing the signal components in each frequency band subspace, the frequency band components related to wear characteristics are selectively retained and the denoised signal is reconstructed. This method can better preserve the transient impact characteristics in the acoustic emission signal.

[0043] In one feasible implementation, the motor current signal is filtered using a low-pass filter with a cutoff frequency of 100Hz to effectively filter out high-frequency switching noise and harmonic interference components in the current signal, while retaining the fundamental frequency component related to load torque changes.

[0044] In another feasible implementation, the motor current signal can also be denoised using an adaptive notch filter. The notch frequency is set to an integer multiple of the power supply frequency to accurately filter out power frequency harmonic interference without affecting the amplitude and phase characteristics of the fundamental frequency component.

[0045] After denoising, the vibration signal is subjected to time-domain feature extraction to obtain vibration characteristic indices.

[0046] Vibration characteristic indicators include the time-domain amplitude characteristics and frequency band energy distribution characteristics of the vibration signal.

[0047] In one feasible implementation, the time-domain amplitude characteristics include the maximum, minimum, peak-to-peak, effective, root square amplitude, kurtosis index, and waveform factor of the vibration acceleration signal. The kurtosis index is particularly sensitive to impact vibration signals. When local wear occurs on the liner surface, the impact between the grinding media and the liner will generate non-Gaussian distributed impact vibration, resulting in a significant increase in the kurtosis index value.

[0048] In another feasible implementation, the time-domain amplitude characteristic can also be selected as the ratio of the positive peak value to the negative peak value of the vibration signal. This ratio can reflect the asymmetric characteristics of the force direction of the liner during operation. When the geometry of the liner lifting bar changes after wear, this ratio will drift regularly.

[0049] The frequency band energy distribution characteristics are obtained by dividing the frequency range into multiple sub-bands after performing a fast Fourier transform on the vibration signal, and calculating the percentage of energy in each sub-band relative to the total energy.

[0050] In one feasible implementation, the frequency band of the vibration signal is divided into three sub-bands: a low-frequency band (0.5Hz to 50Hz), a mid-frequency band (50Hz to 200Hz), and a high-frequency band (200Hz to 1kHz). The change in the energy ratio of each sub-band reflects the distribution of vibration energy of different components of the mill within different frequency ranges. The increase in the energy ratio of the mid-to-high frequency band is usually related to the aggravation of liner wear.

[0051] In another feasible implementation, the frequency band energy distribution characteristics can also be represented by the Mel frequency cepstral coefficients. After the vibration signal spectrum is nonlinearly transformed according to the Mel scale, feature parameters are extracted in the cepstral domain. These parameters have good discriminative ability in vibration signal pattern recognition.

[0052] Time-domain features are extracted from the denoised acoustic emission signal to obtain acoustic emission feature indices.

[0053] Acoustic emission characteristic indicators include the energy accumulation rate characteristics and peak frequency distribution characteristics of acoustic emission signals.

[0054] The energy accumulation rate characteristic is obtained by summing the root mean square value or absolute energy value of the acoustic emission signal within a unit time window, and using the slope of the accumulation curve within the window as the rate index. This rate directly reflects the elastic wave energy released by the liner material due to micro-fracture and plastic deformation per unit time, and is positively correlated with the instantaneous wear rate of the liner.

[0055] In one feasible implementation, the length of the unit time window is 0.5 to 2 seconds, which can balance the temporal resolution of the energy accumulation rate and computational stability.

[0056] In another feasible implementation, the energy accumulation rate can also be calculated using a recursive method based on an exponentially weighted moving average, which improves the sensitivity of the rate index to sudden changes in wear conditions by assigning a higher weighting coefficient to recent energy values.

[0057] Peak frequency distribution characteristics are obtained by extracting the frequency value corresponding to the maximum amplitude point in the power spectrum and its time trajectory after performing Fourier transform on the continuous waveform of the acoustic emission signal. When the wear of the liner plate develops from the initial stage to the severe stage, the peak frequency of the acoustic emission signal will drift from the low frequency band to the high frequency band or the opposite direction.

[0058] Time-domain features are extracted from the denoised motor current signal to obtain current feature indices.

[0059] Current characteristic indicators include the steady-state mean drift characteristics and fluctuation variance characteristics of motor current.

[0060] The steady-state mean drift characteristic is determined by averaging the motor current signal during stable operation and using the deviation of this average value relative to the no-load current reference value as an indicator. When the wear of the liner causes the rotational resistance of the mill cylinder to increase, the motor needs to output a larger electromagnetic torque to overcome the increased load, and the deviation increases accordingly.

[0061] In one feasible implementation, the sampling frequency of the motor current is 1kHz, and the length of each calculation window is 10 seconds. This window length can cover multiple power frequency cycles and obtain a stable mean estimate.

[0062] In another feasible implementation, the steady-state mean drift characteristic can also be represented by the rate of change of the mean relative to the previous moment, so as to eliminate the difficulty of comparison caused by the difference in the base value of the no-load current between different types of mills.

[0063] The fluctuation variance characteristic is calculated by measuring the variance of the motor current signal during stable operation. The magnitude of this variance reflects the severity of load fluctuations. When the wear of the liner lifting bar causes the trajectory of the grinding media to change, the amplitude and frequency of load fluctuations will change, resulting in an increase or decrease in the current variance.

[0064] The normalized vibration characteristic index, acoustic emission characteristic index, and current characteristic index are weighted and fused to obtain the wear characteristic value of the target mill liner.

[0065] In one feasible implementation, normalization adopts the minimum-maximum normalization method, which linearly maps each feature index to the interval between 0 and 1 according to its value range.

[0066] In another feasible implementation, normalization can also be achieved using Z-score standardization, which involves subtracting the mean and then dividing by the standard deviation to give each feature index a statistically significant weight in subsequent fusion.

[0067] The formula used for weighted fusion is: ; in, For wear characteristic values, The normalized vibration characteristic index, The normalized acoustic emission characteristic index. This is the normalized current characteristic index. The fusion weighting coefficients for vibration characteristic indicators The fusion weighting coefficients for acoustic emission characteristic indicators. This is the fusion weighting coefficient for current characteristic indicators.

[0068] In one feasible implementation method The value ranges from 0.35 to 0.45. The value ranges from 0.35 to 0.45. The value ranges from 0.15 to 0.25, and the sum of the three values ​​equals 1.

[0069] In another feasible implementation, for conditions where liner wear is dominated by impact fracture, the weight of acoustic emission characteristic indicators is adjusted. Set the weight to 0.45 to 0.55, and adjust the weight of the vibration characteristic index. Set to 0.25 to 0.35 to highlight the sensitivity of acoustic emission signals to microcrack propagation.

[0070] In another feasible implementation, for conditions where liner wear is dominated by abrasive grinding, the weight of vibration characteristic indicators is increased. Set the weights of the acoustic emission characteristic indicators to 0.45 to 0.55. Set to 0.25 to 0.35 to highlight the sensitivity of vibration signals to changes in macroscopic geometry.

[0071] In another feasible implementation, weighted fusion can also adopt an adaptive weight adjustment method, that is, dynamically adjust the corresponding weight coefficient according to the variance change rate of each feature index within the recent observation window. When the variance change rate of a certain type of feature index increases, its weight is increased, and vice versa, its weight is decreased, thereby realizing adaptive optimization of fusion weight as the operating conditions change.

[0072] The wear characteristic values ​​are output as the observation vector of the target mill liner.

[0073] In one feasible implementation, the observation vector contains only one dimension, namely the wear feature value itself.

[0074] In another feasible implementation, the observation vector can also contain multiple dimensions, that is, vibration characteristic index, acoustic emission characteristic index and current characteristic index are respectively used as different components of the observation vector, while retaining the independent contribution information of each characteristic index, thus forming a multi-dimensional observation vector.

[0075] Through the above-mentioned non-invasive sensor signal acquisition, denoising, feature extraction and weighted fusion process, continuous quantitative observation of the wear degree of the liner plate is realized without contacting the liner plate body or stopping the machine. The observation vector with multi-physics field comprehensive characterization capability is output, providing accurate and reliable measured data input for subsequent parameter correction of the digital twin model.

[0076] Using the observation vector as the state variable to be estimated, predict the virtual sensing feature value of the initial digital twin model at the next time step, and at the same time obtain the actual observation vector corresponding to the current liner. In this embodiment of the invention, the initial digital twin model and wear model parameters at the current moment are obtained. The wear model parameters include at least the wear coefficient, friction factor, and energy dissipation factor. These parameters collectively determine the driving force of the wear evolution function on the wear depth of the liner. By obtaining the above parameters, the model at the current moment has a physical basis for forward extrapolation. Subsequently, the initial digital twin model is driven forward extrapolation based on the wear model parameters at the current moment to calculate the predicted value of the liner wear state at the next moment.

[0077] The wear model parameters at the current moment are input into the initial digital twin model. Based on the current wear model parameters and the current wear state of the liner, a prediction cycle is extrapolated forward according to a preset time step. During the extrapolation, the finite element or discrete element solver in the digital twin model calculates the stress and displacement field distributions of each calculation node at the next moment based on the current wear state and material properties, combined with the working condition boundary conditions. In one feasible implementation, the preset time step is set to 0.5 milliseconds to 5 milliseconds. Within this time step range, both numerical stability and computational efficiency can be considered. In another feasible implementation, for scenarios with sufficient computational resources, the time step is set to 0.1 milliseconds to 0.5 milliseconds to capture more refined wear evolution dynamics. In yet another feasible implementation, for preliminary assessment or coarse screening scenarios, the time step is set to 5 milliseconds to 20 milliseconds to reduce the computation time of a single extrapolation. The specific value of the time step can be set by the operator based on the mill speed, mesh density, and available computing resources. Generally, when the ratio of the time step to the minimum size of the mesh cell and the material sound velocity satisfies the Courant stability condition, the solution process can be ensured to converge.

[0078] At the end of the prediction period, the wear depth increment of each wear calculation grid cell in the initial digital twin model is extracted. The wear depth increment is calculated as follows.

[0079] Based on the wear model parameters at the current moment, the wear depth increment of each wear calculation grid cell within the current time step is calculated according to the wear evolution function. The calculation satisfies the following mathematical relationship: ; ; in, For the first Each wear calculation grid cell is in The current wear depth at any given time, measured in units of length, preferably ranging from 0 mm to 50% of the original thickness of the liner. For the first Each wear calculation grid cell is in The increment of wear depth within a time step, with the dimension being length units. For grinding media to the first The impact velocity of the surface of each wear calculation grid cell is measured in units of length per time, with a preferred value range of 2 m / s to 15 m / s. The impact direction of the grinding media and the first The angle between the surface normals of each wear calculation grid cell is in the unit of angle, and the preferred value range is 15 degrees to 75 degrees. Within this angle range, the grinding media produces both normal impact indentation and tangential grinding on the liner surface, and the wear mechanism is most complete. For the first The hardness value of the liner material corresponding to each wear calculation grid cell is in the unit of pressure. For high manganese steel liners, the preferred value is 180HB to 250HB, and for alloy steel liners, the preferred value is 300HE to 450HB. The wear coefficient has a dimensionless value and a preferred range of values. to This coefficient reflects the proportion of unit impact energy converted into wear depth, and is jointly determined by the frictional characteristics of the liner material and the grinding media. In the calculation of the wear depth increment mentioned above, the wear coefficient... This coefficient directly determines the magnitude of the wear depth increment and is one of the core parameters for subsequent residual-based correction. Through the above calculations, a quantitative prediction of the wear depth of each grid cell is achieved at the wear mechanism level.

[0080] The current wear depth of each wear calculation grid cell is updated based on the wear depth increment, resulting in the updated wear depth for each grid cell. Subsequently, the updated wear depths of each grid cell are aggregated according to region division to obtain the predicted wear state of the liner plates in each region at the next time step. Specifically, the cylinder is divided axially into the feed end region, the middle region, and the discharge end region; circumferentially, it is divided into the lifting bar working surface region, the lifting bar back surface region, and the liner bottom plate region. The arithmetic mean of the updated wear depths of all grid cells in each region is taken as the predicted wear state for that region at the next time step. Through this region aggregation process, the fine-grid-level wear data is consolidated into a regionally readable wear index, facilitating comparison with sensor observation data on a unified spatial scale in subsequent steps.

[0081] After obtaining the predicted liner wear state for the next time step, the predicted liner wear state is mapped to virtual vibration characteristic values, virtual acoustic emission characteristic values, and virtual current characteristic values ​​for the next time step through the sensing feature mapping relationship in the digital twin model. The sensing feature mapping relationship reflects the correspondence between the liner surface wear state and external measurable sensing signals. This correspondence is pre-calibrated by applying a standard excitation signal to the digital twin model and extracting response features. The virtual vibration characteristic value is used to characterize the vibration response amplitude and frequency band energy at the mill bearing housing position under the predicted state; the virtual acoustic emission characteristic value is used to characterize the acoustic emission signal accumulation rate and peak frequency on the outer wall of the cylinder under the predicted state; and the virtual current characteristic value is used to characterize the steady-state mean and fluctuation variance of the motor current under the predicted state. These three virtual characteristic values ​​together constitute the virtual sensing characteristic value vector.

[0082] The virtual vibration characteristic value, virtual acoustic emission characteristic value, and virtual current characteristic value are output as virtual sensing characteristic values, and their forms satisfy the following mathematical relationship: ; ; in, The predicted virtual sensing feature vector has components that are virtual vibration feature values. Virtual acoustic emission characteristic values and virtual current eigenvalues . This is the vector of predicted liner wear state values ​​for the next moment, and its components include the updated wear depth of each wear calculation grid cell. This is a sensing feature mapping function that maps the wear state space to the sensing feature space. In one feasible implementation, the sensing feature mapping function... A linear mapping model is adopted, that is ,in Let b be the mapping matrix and b be the bias vector. and The result is obtained by performing multiple linear regression on simulation data of the digital twin model under various wear conditions. In another feasible implementation, the sensing feature mapping function... Nonlinear mapping models are employed, including polynomial mapping models or artificial neural network mapping models. The polynomial mapping model uses a cubic polynomial expansion to capture the nonlinear coupling between wear state and sensing features, while the artificial neural network mapping model employs a three-layer feedforward network structure and is trained using simulation data. In another feasible implementation, the sensing feature mapping function... Alternatively, a simplified model based on physical mechanisms can be used to express the relationship between wear depth and vibration characteristics as a physical expression showing that contact stiffness decreases with increasing wear depth. Regardless of the mapping method used, an effective conversion from wear state predictions to virtual sensing characteristic values ​​can be achieved, providing a data alignment basis for subsequent residual calculations.

[0083] While predicting the virtual sensing feature value at the next moment, the actual observation vector corresponding to the current liner is acquired. The actual observation vector is obtained by filtering, denoising, and feature extraction of signals collected simultaneously by vibration sensors deployed at the mill bearing housing, acoustic emission sensors on the outer wall of the cylinder, and current sensors inside the motor control cabinet. The component structure of the actual observation vector is consistent with that of the virtual sensing feature value vector, containing three components: actual vibration feature value, actual acoustic emission feature value, and actual current feature value. By performing the prediction of the virtual sensing feature value and the acquisition of the actual observation vector in parallel at the same moment, the two can be aligned in the time dimension, allowing subsequent steps to compare them item by item and calculate the residuals. This provides a reliable source of error signals for the adaptive correction of the digital twin model.

[0084] The parameters of the initial digital twin model are corrected based on the residual between the actual observation vector and the virtual sensing feature value. In this embodiment of the invention, the actual observation vector and virtual sensing feature values ​​are obtained at the same moment. The actual observation vector is obtained by collecting and extracting features from vibration sensors deployed at the mill bearing housing, acoustic emission sensors on the outer wall of the mill cylinder, and current sensors inside the motor control cabinet at the same moment during the mill's operation. The virtual sensing feature values ​​are obtained by transforming the digital twin model forward at the same moment through a sensing feature mapping relationship. The actual observation vector and the virtual sensing feature values ​​are compared item by item to determine the deviation between each item, thus obtaining the residual vector. In this step, the residual vector reflects the degree of deviation between the predicted value of the digital twin model at the current moment and the actual sensing observation value. This degree of deviation is the direct basis for subsequent parameter correction. By calculating the residual vector, quantitative error input information is provided for the closed-loop adaptive correction of the model, enabling the originally open-loop digital twin model to adjust based on the feedback of measured data.

[0085] In one feasible implementation, both the actual observation vector and the virtual sensing feature value have three components: a vibration feature component, an acoustic emission feature component, and a current feature component. The components of the residual vector are obtained by subtracting each component of the actual observation vector from the corresponding component of the virtual sensing feature value. The calculation of the residual vector satisfies the following mathematical relationship: ; ; in, for The residual vector at time step has dimensions consistent with those of each sensing feature component. for The actual observed vector at time t, its components , , These are the actual vibration characteristic value, the actual acoustic emission characteristic value, and the actual current characteristic value, respectively. for The virtual sensing feature vector at time step, its components , , These are virtual vibration characteristic values, virtual acoustic emission characteristic values, and virtual current characteristic values, respectively. , , These are the vibration residual components, acoustic emission residual components, and current residual components. Through the above item-by-item comparison and difference calculation, the comprehensive deviation of multi-physics sensing information is quantified into a three-dimensional residual vector, providing a fine-grained basis for identifying error sources for subsequent differential adjustments of various wear model parameters. When the absolute value of a certain residual component is large, it indicates that the model's prediction in that physical field dimension deviates significantly from the actual measurement, requiring focused adjustment of the sensitive parameters corresponding to that dimension.

[0086] After obtaining the residual vector, the adjustment direction and magnitude of each parameter in the wear model are determined based on the positive and negative directions and magnitudes of the deviation values ​​in the residual vector. The wear model parameters include at least the wear coefficient, friction factor, and energy dissipation factor. The wear coefficient, reflecting the proportion of unit impact energy converted into wear depth, directly determines the response amplitude of the wear evolution function and is the most sensitive core parameter in the model. The friction factor reflects the magnitude of the frictional force between the grinding media and the liner surface. This factor affects the tangential force distribution of the grinding media on the liner surface, thus influencing the spatial distribution of wear. The energy dissipation factor reflects the proportion of impact energy dissipated during collision. This factor affects the amount of effective energy transferred from the grinding media to the liner, thus affecting the overall rate of wear evolution. These three parameters jointly control the output characteristics of the wear evolution function from different physical dimensions. Deviation in any parameter will lead to a specific pattern of residual distribution between the virtual sensing feature values ​​and the actual observed vector. Therefore, it is necessary to determine the adjustment direction and adjustment range of each parameter based on the specific values ​​of each component of the residual vector in order to achieve fine-grained correction of the model parameters.

[0087] Specifically, regarding the wear coefficient When the vibration residual component Harmony emission residual components When both values ​​are positive, it indicates that the vibration response and acoustic emission response predicted by the model are lower than the measured values. In this case, the adjustment direction of the wear coefficient is determined to be positively increasing, and the adjustment range is the same as... and The weighted average values ​​are positively correlated. When both the vibration residual component and the acoustic emission residual component are negative, the adjustment direction of the wear coefficient is determined to be a negative decrease. For the friction factor... When the current residual component When the absolute value of the friction factor is large, it indicates a significant deviation between the model-predicted motor current and the measured current. Since the motor current is most sensitive to changes in the mill cylinder load torque, and the load torque is closely related to the friction force between the grinding media and the liner, the adjustment direction of the friction factor is aligned with the current residual component. The positive and negative directions are related. This applies to the energy dissipation factor. When all three residual components show the same direction of deviation, it indicates that the model's overall estimation of the impact energy transfer efficiency of the grinding media is biased. In this case, the adjustment direction of the energy dissipation factor is set to be consistent with the combined direction of the three residual components. Through the above differentiated adjustment strategy, targeted corrections are applied to wear parameters with different physical properties, avoiding the correction mismatch problem caused by uniform adjustment of a single parameter.

[0088] In one feasible implementation, the adjustment direction and magnitude are determined as follows: when the absolute value of the residual component is less than a first threshold, the adjustment magnitude of the wear parameter corresponding to that residual component is set to zero, i.e., no adjustment is performed. When the absolute value of the residual component is greater than or equal to the first threshold and less than a second threshold, the adjustment magnitude of the wear parameter corresponding to that residual component is set to the first adjustment step size. When the absolute value of the residual component is greater than or equal to the second threshold, the adjustment magnitude of the wear parameter corresponding to that residual component is set to the second adjustment step size, where the second adjustment step size is greater than the first adjustment step size. In another feasible implementation, the adjustment magnitude can also be determined using a continuous function mapping method, i.e., the adjustment magnitude and the absolute value of the residual component satisfy a linear proportional relationship or a nonlinear saturated proportional relationship. This continuous mapping method makes the parameter adjustment smoother and helps to avoid model oscillations caused by step size switching.

[0089] After determining the adjustment direction and magnitude of each wear model parameter, correction values ​​for the wear model parameters are generated. The generation of these correction values ​​satisfies the following mathematical relationship: ; ; ; in, for The correction amount for the wear coefficient at any time has the same dimensions as the wear coefficient. for The correction amount for the friction factor at any given time has the same dimensions as the friction factor. for The correction amount of the energy dissipation factor at time, whose dimensions are consistent with the energy dissipation factor. , , , These are the proportional gain coefficients for converting the corresponding residual components into correction values. The preferred range for each proportional gain coefficient is 0.01 to 0.50. This range is determined based on the fact that the single correction value should not be too large to prevent parameter oscillation, and at the same time, it can converge to the measured value within a reasonable number of corrections. , , , These are the functional relationships that map each residual component to the adjustment amount. This mapping function can be a piecewise linear function or a saturated nonlinear function. The function is a synthesis function with three residual components as independent variables, and its form can be the weighted average or maximum value of the three residual components. Through the above mathematical relationship, the residual vector is quantitatively transformed into the correction amount of each wear model parameter, realizing the accurate mapping from the residual space to the parameter space. The specific value of the proportional gain coefficient can be set according to the response sensitivity of the digital twin model. When the model parameters are more sensitive to changes in the sensing characteristics, a smaller gain coefficient is used; when the model parameters are less sensitive to changes in the sensing characteristics, a larger gain coefficient is used. This setting method ensures a reasonable match between the correction amount scale and the model response scale.

[0090] In one feasible implementation, the proportional gain coefficient , , , The adjustment can also be adaptively adjusted based on the magnitude of the absolute value of the residual components. When the absolute value of the residual components is small, a first gain coefficient is used for fine-tuning. When the absolute value of the residual components is large, a second gain coefficient is used for fast tracking; the second gain coefficient is greater than the first gain coefficient. In another feasible implementation, the correction amount can be generated using a proportional-integral-derivative control structure, where the proportional term responds to the residual at the current moment, the integral term eliminates steady-state error, and the derivative term suppresses parameter oscillations. This structure further improves the convergence speed and stability of parameter correction. In yet another feasible implementation, the correction amount can be generated using a gradient descent-based optimization method. The correction amount is generated by calculating the Jacobian matrix of the residual vector relative to each wear model parameter, so that the corrected parameters allow the residual vector to decrease along the negative gradient direction. Regardless of which of the above correction amount generation methods is used, effective correction of the wear model parameters can be achieved.

[0091] After generating the correction values ​​for the wear model parameters, the correction values ​​are used to update the wear model parameters at the current moment, generating the corrected wear model parameters. The superposition update satisfies the following mathematical relationship: ; ; ; in, for The wear coefficient after constant adjustment. for Wear coefficient before constant correction This is the correction amount for the wear coefficient. for The friction factor after constant adjustment for Friction factor before constant correction This is the correction amount for the friction factor. for The energy dissipation factor after time correction, for Energy dissipation factor before time correction This represents the correction amount for the energy dissipation factor. Through the above superposition update, the wear model parameters are adjusted from their original values ​​towards eliminating residuals, thereby achieving a gradual approximation between the model predictions and actual observations.

[0092] In one feasible implementation, after the overlay update is completed, the corrected wear model parameters can be subjected to amplitude limiting. When the corrected parameter exceeds the preset effective range, the parameter is truncated to the boundary value of the preset effective range. For example, the wear coefficient... The effective range is preset to to When the corrected wear coefficient is lower than When it is set to a value When the corrected wear coefficient is higher than When it is set to a value By limiting the amplitude, model failure can be avoided due to excessive single correction or abnormal sensor signals causing parameters to exceed limits. In another feasible implementation, after the superposition update is completed, the corrected wear model parameters can be smoothed and filtered. The corrected parameters and the parameters at the previous time step are then exponentially weighted and averaged according to preset weights to suppress high-frequency fluctuations in parameters caused by sensor signal noise, making the parameter change trajectory smoother and more stable.

[0093] The corrected wear model parameters are written into the initial digital twin model, replacing the original wear model parameters, resulting in the corrected digital twin model. After the corrected wear model parameters replace the original wear model parameters, the digital twin model will use the updated wear coefficient, friction factor, and energy dissipation factor for forward calculations in the next round of simulation. Thus, the predicted output of the digital twin model will gradually approach the actual wear state reflected by actual sensor observations, achieving adaptive tracking of operating conditions throughout the entire service life. Through the above parameter correction process, the digital twin model with fixed wear parameters acquires self-correction capabilities, fundamentally avoiding the problem of long-term prediction deviation accumulation caused by the mismatch between initial wear parameters and actual operating conditions, and significantly improving the continuous accuracy of liner wear state assessment.

[0094] The current wear status of the liner is inferred based on the revised digital twin model, and an evaluation report of the liner is output.

[0095] In this embodiment of the invention, after the modified wear model parameters are written into the digital twin model, the wear coefficient, friction factor, and energy dissipation factor in the digital twin model are all updated to their current values ​​after residual correction. Using the modified digital twin model as the calculation kernel, starting from the wear state of the liner at the current moment, time extrapolation is performed forward according to a preset extrapolation period to obtain the evolution data of the liner wear state over a continuous time series after the current moment. During this extrapolation process, the modified wear model parameters replace the original parameter values, so that the wear depth increment output by the wear evolution function at each time step matches the actual wear law under the current working condition. Through the above extrapolation, the wear depth distribution data of the current liner on each wear calculation grid cell and the summary data of the wear state of the liner in each region are obtained.

[0096] In one feasible implementation, the preset simulation period is one complete operating shift or one complete mill operation cycle after the current moment. The duration of a complete operating shift is preferably 8 to 12 hours, and the duration of a complete mill operation cycle is preferably 24 to 72 hours. In another feasible implementation, the preset simulation period can be adaptively set according to the liner wear rate. A shorter simulation period is used when the wear rate is high to enhance the timeliness of the assessment, and a longer simulation period is used when the wear rate is low to reduce the calculation frequency. In yet another feasible implementation, the preset simulation period can also be manually set by the operator according to the maintenance plan's time nodes, for example, set as the number of days remaining until the planned shutdown maintenance date. Regardless of which simulation period setting method is used, the simulation results can be matched with the time nodes of the engineering maintenance decision in the time dimension, thus providing a time benchmark with practical reference value for the generation of subsequent evaluation reports.

[0097] During the simulation process following the current moment, the digital twin model advances step by step according to a preset simulation step size. In each simulation step, the wear depth increment is calculated and the wear depth is cumulatively updated. The simulation step size can be consistent with the time step size in S103 above, or it can be adjusted according to the length of the simulation period. When the simulation period is long, the simulation step size is preferably 1 to 10 seconds to balance computational efficiency and simulation accuracy. When the simulation period is short, the simulation step size is preferably 0.1 to 1 second to obtain more refined details of the wear evolution. Through iterative calculations step by step, the digital twin model outputs the wear depth increment sequence and the cumulative wear depth sequence at each moment within the entire simulation period, forming a complete temporal evolution trajectory of the wear state.

[0098] After acquiring the wear state evolution data within the simulation period, wear state characterization parameters of the current liner are extracted to quantitatively describe the wear degree of the current liner at each spatial location. The wear state characterization parameters include at least the current wear depth, current remaining thickness, wear rate, and wear non-uniformity of each wear calculation grid cell. The current wear depth is the difference between the initial thickness and the current remaining thickness. The wear rate is the increase in wear depth per unit time, and its calculation satisfies the following mathematical relationship: ; in, For the first The wear rate of each wear calculation grid cell is measured in units of length per time, with a preferred reference range of 0.01 mm per day to 1 mm per day, which covers the wear rate range of most mill liners under normal operating conditions. and The first Each wear calculation grid cell is in Time and Wear depth over time and For two different time points within the simulation period, Greater than The difference between the two values ​​represents the time interval used to calculate the wear rate. The preferred range for this time interval is 1 to 7 days, which effectively smooths short-term fluctuations to reflect a stable wear rate trend. Wear non-uniformity characterizes the dispersion of wear depth among different wear calculation grid cells within the same region. It can be calculated as the ratio of the standard deviation to the average wear depth of each grid cell. A larger ratio indicates more uneven wear in the region, increasing the risk of stress concentration and localized spalling at the strip root. By extracting the above wear state characterization parameters, the large amount of grid-level wear data generated by the digital twin model is condensed into quantitative indicators with engineering evaluation significance, providing a data foundation for the preparation of the assessment report.

[0099] After extracting the wear state characterization parameters, the wear state of the current liner is graded and evaluated based on the values ​​of each parameter. The grading evaluation involves comparing the values ​​of each wear state characterization parameter with multiple preset threshold ranges, and classifying the liner wear state into different levels based on the comparison results. In one feasible implementation, the grading evaluation can use a single-parameter threshold method, classifying the liner wear state into three levels: slight wear, moderate wear, and severe wear, based on the current remaining thickness. Slight wear corresponds to a current remaining thickness greater than 70% of the original thickness, moderate wear corresponds to a current remaining thickness between 40% and 70% of the original thickness, and severe wear corresponds to a current remaining thickness less than 40% of the original thickness. In another feasible implementation, the grading evaluation can also use a multi-parameter comprehensive scoring method, mapping the current wear depth, wear rate, and wear unevenness to corresponding scores according to preset scoring rules, then weighting and summing the three scores to obtain a comprehensive score, and finally comparing the comprehensive score with a preset comprehensive threshold range to determine the wear level. In another feasible implementation, the graded assessment can also employ the fuzzy comprehensive evaluation method. This involves establishing a membership function of each wear state characterization parameter to the wear level, determining the membership degree of each wear level through fuzzy reasoning, and finally using the level with the highest membership degree as the assessment result. Regardless of which graded assessment method is used, the derived wear state numerical values ​​can be transformed into intuitive graded assessment conclusions, facilitating maintenance personnel to quickly grasp the overall service status of the lining.

[0100] In one feasible implementation, the specific values ​​of the threshold ranges used in the graded assessment are set by those skilled in the art based on the original thickness of the liner, material properties, and mill model. For high-manganese steel liners with a thickness of 50 mm to 80 mm, the preferred threshold for distinguishing between slight and moderate wear is 60% of the original thickness, and the preferred threshold for distinguishing between moderate and severe wear is 30% of the original thickness. For alloy steel liners with a thickness of 80 mm to 120 mm, the preferred threshold for distinguishing between slight and moderate wear is 65% of the original thickness, and the preferred threshold for distinguishing between moderate and severe wear is 35% of the original thickness. The above threshold ranges are set based on the critical points where the remaining strength and safety margin of the liner change significantly when it wears to different degrees. When the remaining thickness is less than 30% to 35% of the original thickness, the load-bearing capacity of the liner decreases significantly, and the risk of breakage and detachment increases sharply. Therefore, this range is set as the boundary for severe wear levels. Through the scientific setting of the above threshold ranges, an effective correlation is established between the graded assessment results and the structural safety of the liner.

[0101] After completing the extrapolation and classification assessment of the current liner wear state, an assessment report for the liner is generated and output. The assessment report includes at least the wear state characterization parameters of the current liner, the classification assessment results, and the spatial wear distribution information of the liner in each region. The assessment report visually displays the spatial distribution cloud map of the wear depth for each wear calculation grid cell. This cloud map uses a color-gradient mapping method to correspond different wear depth values ​​to different colors, presenting the wear distribution pattern of the entire liner on the cylinder unfolded diagram. The assessment report also displays the evolution curve of the wear rate of the liner in each region over time in the form of a trend graph, reflecting the dynamic change trend of the wear rate during mill operation. The assessment report also lists the current remaining thickness, service time, and estimated remaining life of the liner in each region in tabular form. The calculation of the estimated remaining life satisfies the following mathematical relationship: ; in, For the first The wear calculation grid cell or the first The estimated remaining lifespan of each region is measured in units of time. For the first The wear calculation grid cell or the first The original thickness of the lining in each region is measured in units of length. For the current moment The wear calculation grid cell or the first The wear depth of each region is measured in units of length. For the first The wear calculation grid cell or the first The wear rate of each region is measured in units of length per unit of time. When the calculated estimated remaining life is less than a preset safety warning threshold, a warning message is output in the assessment report. The preferred range for the safety warning threshold is that the remaining life is less than twice the length of the current simulation cycle; for example, when the simulation cycle is 8 hours, the preferred safety warning threshold is 16 hours. Through the generation and output of the above assessment report, the wear status data obtained from the digital twin model simulation is presented to maintenance personnel in an engineering-readable form, providing direct decision-making basis for determining the timing of liner replacement and formulating spare parts plans. Thus, the complete main process of the liner service assessment method based on digital twins is completed, realizing a closed-loop assessment across the entire chain, from sensor signal acquisition, model prediction, residual correction to condition simulation and assessment report output.

[0102] Example 2

[0103] like Figure 2 As shown in the figure, this embodiment also provides a functional block diagram of a liner service evaluation device based on digital twin.

[0104] The digital twin-based liner service assessment device described in this embodiment can be installed in an electronic device. Depending on the functions implemented, the digital twin-based liner service assessment device may include a twin initial modeling module, a wear observation module, a feature inference module, a model calibration module, and a condition assessment module. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0105] In this embodiment, the functions of each module / unit are as follows: The initial digital twin module is used to construct an initial digital twin model of the target mill liner. The wear observation module is used to observe the wear degree of the target mill liner based on non-intrusive sensing signals from outside the mill during mill operation, and to obtain the observation vector of the target mill liner. The feature inference module is used to predict the virtual sensing feature value of the initial digital twin model at the next moment, with the observation vector as the state variable to be estimated, and at the same time obtain the actual observation vector corresponding to the current liner. The model correction module is used to correct the parameters of the initial digital twin model based on the residual between the actual observation vector and the virtual sensing feature value. The status assessment module is used to deduce the current wear status of the liner based on the corrected digital twin model and output an evaluation report of the liner.

[0106] In detail, each module in the digital twin-based liner service evaluation device described in the embodiments of the present invention adopts the same technical means as the digital twin-based liner service evaluation method described in Embodiment 1 and Embodiment 2, and can produce the same technical effect, which will not be repeated here.

[0107] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0108] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application devices that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A service evaluation method for liner plates based on digital twins, characterized in that, The method includes: Construct an initial digital twin model of the target mill liner; During the operation of the mill, the wear degree of the target mill liner is observed based on non-intrusive sensing signals from outside the mill, and the observation vector of the target mill liner is obtained. Using the observation vector as the state variable to be estimated, predict the virtual sensing feature value of the initial digital twin model at the next time step, and at the same time obtain the actual observation vector corresponding to the current liner. The parameters of the initial digital twin model are corrected based on the residual between the actual observation vector and the virtual sensing feature value. The current wear status of the liner is inferred based on the revised digital twin model, and an evaluation report of the liner is output.

2. The service evaluation method for liners based on digital twins as described in claim 1, characterized in that, The initial digital twin model for constructing the target mill liner includes: Obtain the geometric parameters, material property parameters, and operating parameters of the target mill liner. The geometric parameters include the outline dimensions of the liner and the installation position data of the liner in each region. The material property parameters include the density, elastic modulus, Poisson's ratio, and hardness of the liner material. The operating parameters include the mill speed, grinding media filling rate, and particle size distribution data of the feed material. A three-dimensional geometric model of the target mill liner is established based on the geometric parameters; Based on the material property parameters and working condition parameters, material property boundary conditions and motion boundary conditions are configured in the three-dimensional geometric model to obtain the initial digital twin model.

3. The service evaluation method for liners based on digital twins as described in claim 2, characterized in that, The process of establishing a three-dimensional geometric model of the target mill liner based on geometric parameters includes: Based on the installation position data of the liners in each region, the outline dimensions of the liners in each region are arranged in space in the circumferential and axial directions of the inner wall of the target mill cylinder to obtain a three-dimensional assembly model of the liner group. In the 3D assembly model of the liner group, the surface of the liner in each region is divided into multiple wear calculation mesh elements, and each wear calculation mesh element is assigned corresponding material property parameters. The three-dimensional assembly model of the liner assembly, after mesh generation and material property assignment, is combined with the geometric data of the target mill cylinder to obtain the three-dimensional geometric model of the target mill liner.

4. The service evaluation method for liners based on digital twins as described in claim 1, characterized in that, During the operation of the mill, the wear degree of the target mill liner is observed based on non-intrusive sensing signals from outside the mill, resulting in an observation vector for the target mill liner, including: Vibration sensors are deployed at the bearing housing of the target mill, acoustic emission sensors are deployed on the outer wall of the target mill cylinder, and current sensors are deployed inside the motor control cabinet of the target mill. Vibration signals during mill operation are collected using vibration sensors, acoustic emission signals during mill operation are collected using acoustic emission sensors, and motor current signals during mill operation are collected using current sensors. The vibration signal, acoustic emission signal, and motor current signal are filtered and denoised respectively to obtain the denoised vibration signal, denoised acoustic emission signal, and denoised motor current signal.

5. The service evaluation method for liners based on digital twins as described in claim 4, characterized in that, The obtained denoised vibration signal, denoised acoustic emission signal, and denoised motor current signal include: The time-domain features of the denoised vibration signal are extracted to obtain vibration feature indices, which include the time-domain amplitude features and frequency band energy distribution features of the vibration signal. The time-domain features of the denoised acoustic emission signal are extracted to obtain acoustic emission feature indices, which include the energy accumulation rate feature and peak frequency distribution feature of the acoustic emission signal. The time-domain features of the denoised motor current signal are extracted to obtain current feature indices, which include the steady-state mean drift characteristics and fluctuation variance characteristics of the motor current. The wear characteristic values ​​of the target mill liner are obtained by weighted fusion of the normalized vibration characteristic index, acoustic emission characteristic index and current characteristic index. The wear characteristic values ​​are output as the observation vector of the target mill liner.

6. The service evaluation method for liners based on digital twins as described in claim 1, characterized in that, The process of using the observed vector as the state variable to be estimated, predicting the virtual sensing feature value of the initial digital twin model at the next time step, and simultaneously obtaining the actual observed vector corresponding to the current liner includes: Obtain the initial digital twin model and wear model parameters at the current moment; The initial digital twin model is driven forward based on the wear model parameters at the current moment to calculate the predicted value of the liner wear state at the next moment; By mapping the sensing features in the digital twin model, the predicted wear state of the liner is mapped to the virtual vibration feature value, virtual acoustic emission feature value, and virtual current feature value at the next moment. Virtual vibration characteristic values, virtual acoustic emission characteristic values, and virtual current characteristic values ​​are used as virtual sensing characteristic values.

7. The service evaluation method for liners based on digital twins as described in claim 6, characterized in that, The step of driving the initial digital twin model forward based on the wear model parameters at the current moment to calculate the predicted value of the liner wear state at the next moment includes: Input the wear model parameters at the current moment into the initial digital twin model, and based on the wear model parameters at the current moment and the wear state of the liner at the current moment, extrapolate a prediction cycle forward according to the preset time step; At the end of the prediction period, the wear depth increment of each wear calculation grid cell in the initial digital twin model is extracted; The current wear depth of each wear calculation grid cell is updated based on the wear depth increment to obtain the updated wear depth of each wear calculation grid cell; The updated wear depths of each wear calculation grid cell are summarized according to the region division to obtain the predicted wear state of the lining plate in each region at the next moment.

8. The service evaluation method for liners based on digital twins as described in claim 6, characterized in that, The parameter correction of the initial digital twin model based on the residual between the actual observed vector and the virtual sensing feature value includes: The actual observation vector and virtual sensing feature value are obtained at the same time. The actual observation vector and virtual sensing feature value are compared item by item to determine the deviation value between each item, so as to obtain the residual vector. Based on the positive and negative directions and magnitudes of the deviation values ​​in the residual vector, the adjustment direction and adjustment range of each parameter in the wear model parameters are determined respectively; The correction amount for the wear model parameters is generated based on the adjustment direction and adjustment range; The wear model parameters at the current moment are updated by superimposing the correction amount to generate the corrected wear model parameters; The corrected wear model parameters are written into the initial digital twin model, replacing the original wear model parameters in the initial digital twin model, to obtain the corrected digital twin model.

9. A liner service evaluation device based on digital twins, characterized in that, include: The initial digital twin module is used to construct the initial digital twin model of the target mill liner. The wear observation module is used to observe the wear degree of the target mill liner during mill operation based on non-intrusive sensing signals from outside the mill, and obtain the observation vector of the target mill liner. The feature inference module is used to predict the virtual sensing feature values ​​of the initial digital twin model at the next time step, using the observation vector as the state variable to be estimated, and at the same time obtain the actual observation vector corresponding to the current liner. The model calibration module is used to correct the parameters of the initial digital twin model based on the residual between the actual observation vector and the virtual sensing feature value. The status assessment module is used to deduce the current wear status of the liner based on the corrected digital twin model and output an evaluation report of the liner.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 8.

Images