Rolling mill lining plate inclination detection method based on multi-axis inclination angle sensing

Abstract

The invention relates to a rolling mill lining plate inclination detection method based on multi-axis inclination angle sensing, and belongs to the technical field of rolling mill equipment state monitoring. According to the method, initial reference inclination angle data of a lining plate detection area are obtained, dynamic reference data are generated through abrasion-reference linkage self-calibration, a lining plate multi-dimensional coupling mapping model is constructed, and a dynamic threshold system is set; collecting real-time inclination angle data, fusing the real-time inclination angle data with physical quantity compensation data, and preprocessing to generate effective data; on the basis of comparison of effective data and dynamic reference data, suppressing vibration interference, generating a regional inclination offset through regional-global bidirectional iterative calculation, and training a trend prediction model to obtain an inclination development trend; and generating an abnormal judgment result through multi-region data cross verification and redundancy verification, tracing the induction factors, constructing a multi-stage early warning mechanism, triggering early warning, outputting complete abnormal information, and generating an operation report. Inclination detection precision and early warning timeliness are improved, and reliable support is provided for safe operation of the rolling mill and lining plate maintenance.

Description

Technical Field

[0001] This invention belongs to the field of rolling mill equipment condition monitoring technology, specifically relating to a rolling mill liner tilt detection method based on multi-axis tilt angle sensing. Background Technology

[0002] As a core production equipment in the metallurgical industry, the operational stability of rolling mills directly affects product quality and production safety. Liners, as key load-bearing components of the rolling mill, play a crucial role in protecting the rolls, transmitting rolling force, and ensuring rolling accuracy. Under long-term heavy loads, high-speed operation, and material impact conditions, liners are prone to tilting due to uneven wear, changes in installation clearance, and load fluctuations. Failure to detect and intervene in a timely manner will lead to roll stress imbalance, increased product dimensional deviations, and in severe cases, liner detachment, mill shutdown, and other safety accidents, resulting in significant economic losses.

[0003] Currently, the detection of mill liner tilt mainly relies on manual inspection, single threshold alarms, or simple sensor detection methods, which have several technical shortcomings: First, traditional detection methods mostly use fixed benchmark data for comparison, failing to consider the benchmark drift caused by long-term wear of the liner. As operating time accumulates, the detection accuracy continuously decreases, easily leading to false positives or false negatives. Second, existing early warning mechanisms mostly set thresholds based on a single tilt offset, lacking a comprehensive consideration of the tilt development trend and the scope of the abnormality's impact. They can only provide simple alarms and cannot provide graded response schemes based on the severity of the abnormality, making it difficult for operators to accurately detect tilts. First, the priority of handling issues must be grasped. Second, the early warning information often only includes basic anomaly alerts and does not integrate key information such as the location of the abnormal area and the tracing of the inducing factors. Operators need to spend a lot of extra time to investigate the problem, which delays the opportunity to deal with it. Third, the early warning mechanism lacks closed-loop optimization capabilities. The detection and early warning processes are independent of each other, and the processing effect cannot be fed back to the threshold adjustment stage, resulting in poor early warning adaptability and difficulty in adapting to the complex and ever-changing operating conditions of the rolling mill. Fourth, some detection methods do not effectively suppress interference factors such as vibration and temperature changes generated by the operation of the rolling mill, resulting in insufficient data reliability and further affecting the accuracy of the early warning.

[0004] Therefore, in order to address the problems of fixed benchmarks, single warnings, incomplete information, lack of closed-loop optimization, and weak anti-interference capabilities in existing mill liner tilt detection technologies, it is necessary to develop a detection method that can dynamically adapt to the liner status, provide accurate and graded warnings, integrate complete abnormal information, and achieve closed-loop optimization, so as to improve the accuracy of mill liner tilt detection and the timeliness and effectiveness of warnings, and ensure the safe and stable operation of the mill. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for detecting the tilt of rolling mill liners based on multi-axis tilt sensing. The objective of this invention can be achieved through the following technical solutions: A method for detecting the tilt of rolling mill liners based on multi-axis tilt sensing includes: S1: Obtain the initial reference tilt angle data of the mill liner detection area, introduce a wear-reference linkage self-calibration mechanism, correct the initial reference tilt angle data, and generate dynamic reference data; construct a multi-dimensional coupled mapping model of the liner, and set a dynamic threshold system based on historical fault and normal working condition data; S2: Acquire real-time tilt angle data of each detection area during the operation of the rolling mill, synchronously integrate physical quantity compensation data and perform preprocessing; identify abnormal data points in the time-series tilt data based on the dynamic threshold system, correct the abnormal data points, and generate preprocessed valid data; S3: Based on the comparison results between the effective data and the dynamic benchmark data, the vibration interference signal is suppressed, and the regional tilt offset is generated through comparison calculation; combined with the force distribution characteristics of the liner in the multi-dimensional coupling model of the liner, the regional weight is optimized by using the regional-global bidirectional iterative calculation logic; based on the time-series tilt data, the trend prediction model is trained to generate the liner tilt development trend prediction result; S4: Generate tilt anomaly judgment results through cross-validation and redundancy verification of multi-regional data; trace the inducing factors of tilt anomalies; based on the dynamic threshold system, set up a multi-level early warning mechanism according to the severity and development trend of the anomaly; generate multi-level early warning signals after triggering the early warning; output the anomaly area, tilt parameters and the conclusion of the traceability of the inducing factors of the anomaly; and generate an operation detection report.

[0006] Specifically, the calibration process of the wear-reference linkage self-calibration mechanism includes: continuously acquiring the cumulative running time of the liner and the actual wear data of each maintenance operation record, and establishing a database in combination with the wear resistance characteristics of the liner material; mining the correspondence between the cumulative running time and the actual wear amount through correlation analysis algorithm, coupling analysis with the tilt angle change data, and generating tilt angle correction coefficients for different wear stages; calling the corresponding correction coefficients in stages based on the wear process of the liner operation to successively correct the initial reference tilt angle data; and comparing the current tilt angle data with the corrected dynamic reference data to verify the correction effect.

[0007] Specifically, the construction process of the liner multi-dimensional coupling mapping model includes: classifying and acquiring liner material characteristic parameters, actual gap data during installation, load data, speed fluctuation data, and continuous monitoring data of ambient temperature and humidity at different operating stages of the rolling mill; denoising and normalizing various types of data, eliminating invalid interference data, calculating the correlation coefficient with the liner inclination angle change, and generating the influence weight of the inclination angle change; dividing the analysis unit based on different working condition combinations, establishing a quantitative mapping relationship between each influencing factor and the inclination angle change within each analysis unit, integrating the quantitative mapping relationship, and generating the liner multi-dimensional coupling mapping model.

[0008] Specifically, the update process of the dynamic threshold system is as follows: periodically acquire data on the current wear stage of the liner and real-time changes in ambient temperature and humidity; call the operating condition data similar to the current wear stage and environmental conditions in the preset historical database and extract the corresponding threshold parameters; combine the current operating data with historical similar operating condition data to adjust the upper and lower limits of the threshold, and substitute the new threshold parameters into the actual detection process to monitor the accuracy and false alarm rate of the early warning.

[0009] Specifically, the steps for fusing physical quantity compensation data include: acquiring temperature data, vibration intensity data, and load fluctuation data in the stress area of ​​the liner; assigning corresponding fusion weights to the temperature data, vibration intensity data, and load fluctuation data based on the influence weights of physical quantities in the multi-dimensional coupling mapping model of the liner; and weighting and fusing the physical quantity compensation data with the real-time tilt angle data through calculation.

[0010] Specifically, the preprocessing procedure includes: denoising the fused tilt data; identifying and filtering interference signals by analyzing the frequency characteristics of the data; dynamically adjusting the filter cutoff frequency; selecting a fixed-length sliding window based on a time-series data smoothing algorithm; calculating the mean or median of the data within the window; and replacing the original data at the center of the window with the mean or median of the data within the window. After processing, verifying the temporal continuity of the data; if data breakpoints exist, using linear interpolation to fill the breakpoints.

[0011] Specifically, the process of suppressing vibration interference signals includes: acquiring vibration intensity data during the operation of the rolling mill in real time, identifying the type, frequency range, and amplitude variation law of the vibration signal through spectrum analysis, adjusting the parameters of the working condition adaptive anti-interference and noise reduction technology, and monitoring the amplitude variation of the effective tilt angle signal in real time.

[0012] Specifically, the implementation process of the region-global bidirectional iterative calculation logic includes: based on the force distribution characteristics of each detection region in the multi-dimensional coupling mapping model of the liner, assigning initial weights to each local detection region; weighted summing of the tilt offset of each local region with the corresponding initial weight to obtain the initial global composite tilt state; calculating the deviation value between the tilt offset of each local region and the initial global composite tilt state, and adjusting the weight of the corresponding region; substituting the adjusted region weights into the weighted summation operation again to obtain a new global composite tilt state; repeating the deviation calculation, weight adjustment and global synthesis steps until the difference between two adjacent global composite tilt states is less than the preset judgment standard, stopping the iteration, and outputting the final global composite tilt state.

[0013] Specifically, the training process of the trend prediction model includes: based on the historical tilt data accumulated from the long-term operation of the liner, organizing it into a time series dataset in chronological order, and simultaneously acquiring the working condition-related data and environmental impact data corresponding to each period of historical tilt data; performing preprocessing to remove outliers, fill in missing values, and perform data normalization to map the data to a unified range; dividing the training set and validation set, constructing the model framework using a time series analysis algorithm, inputting the training set data, iteratively adjusting the core parameters of the model, optimizing the model with the goal of minimizing the prediction error of the validation set, generating a development trend prediction model, and predicting the magnitude and direction of the liner tilt change under different working conditions.

[0014] Specifically, the implementation process of the multi-region data cross-validation and redundancy verification includes: extracting the tilt offset, time-series change curve and local anomaly characteristic parameters of each detection region, and dividing the verification groups according to the classification of adjacent regions, symmetrical regions, and force-related regions; performing bidirectional comparison of the tilt data of each region within the same group, and calculating the synchronization deviation of tilt changes between regions; simultaneously retrieving the redundant backup data of each detection region, comparing the real-time acquired data with the redundant backup data point by point, eliminating abnormal region data, and including the data that has passed both the cross-validation within the group and the redundant backup comparison into the anomaly judgment analysis.

[0015] Specifically, the process of tracing the inducing factors of tilt anomalies includes: extracting working condition data, wear correlation data, and environmental impact data for the period when tilt anomalies occur from the multi-dimensional coupling mapping model of the liner, comparing them item by item with the corresponding data under normal working conditions, and calculating the deviation value; based on the influence weight priority of each factor in the multi-dimensional coupling mapping model of the liner, checking them one by one, analyzing the causal relationship between the identified abnormal factors and tilt anomalies, obtaining the core inducing factors and related inducing factors, and generating a source tracing chain of abnormal inducing factors.

[0016] Specifically, the implementation process of the multi-level early warning mechanism is as follows: combining the tilt offset standards of different intervals in the dynamic threshold system, the tilt development trend prediction results, and the impact range of tilt anomalies on the operation of the rolling mill, the early warning level is divided; when any level of early warning is triggered, the location of the abnormal area, the specific tilt parameters, the conclusion of the source tracing of the abnormal inducing factors, and the tilt development trend prediction results are integrated to generate an early warning information package.

[0017] The beneficial effects of this invention are as follows: By introducing a wear-reference linkage self-calibration mechanism, the initial reference tilt angle data is dynamically corrected by combining the cumulative running time of the liner, the actual wear amount and material characteristics, and dynamic reference data adapted to the real-time status of the liner is generated. This completely avoids the problem of decreased detection accuracy caused by the accumulation of traditional fixed reference over running time, significantly reduces the probability of misjudgment and missed judgment, and ensures the accuracy and reliability of tilt detection results.

[0018] This invention, based on a dynamic threshold system, integrates tilt offset standards, development trends, and the scope of impact on rolling mill operation to classify early warning levels, forming differentiated response schemes. Operators can quickly determine the priority of handling according to the early warning level, avoiding the delays caused by the ambiguity of priorities in the traditional single alarm mode. Minor anomalies only require routine attention, while severe anomalies can initiate emergency handling procedures, significantly improving the pertinence and efficiency of anomaly response.

[0019] Once an early warning is triggered, this invention automatically integrates the location of the abnormal area, specific tilt parameters, conclusions on the causes of the abnormality, and predictions of the tilt trend to form a complete early warning information package. Compared to traditional early warning systems that only provide basic prompts, operators do not need to conduct extensive additional investigations. They can directly locate the core issues and related influencing factors based on the integrated information, significantly shortening the preparation time for problem handling and providing a window of opportunity for timely intervention.

[0020] This invention feeds back operational data and tilt state changes after anomaly handling to the dynamic benchmark calibration and dynamic threshold system optimization stages. By continuously adjusting the benchmark data and threshold parameters, the detection and early warning mechanism can dynamically adapt to complex operating conditions such as mill load fluctuations, environmental changes, and liner wear processes. Compared to the traditional model where detection and early warning are independent, this significantly improves the long-term adaptability and stability of the technical solution, avoiding early warning failures caused by changes in operating conditions.

[0021] This invention simultaneously fuses compensation data from multiple physical quantities, such as temperature, vibration intensity, and load fluctuations, and combines this with adaptive anti-interference and noise reduction technology to specifically suppress interference signals such as vibration and electromagnetic interference. Simultaneously, a preprocessing workflow removes abnormal data points and fills data gaps, forming a comprehensive anti-interference design encompassing "multi-physical quantity fusion - adaptive noise reduction - data correction." This effectively addresses the shortcomings of traditional detection methods where data is easily affected by environmental and operational conditions, ensuring the authenticity and continuity of real-time tilt angle data and subsequent analysis results.

[0022] By detecting the tilting state of the liner plates, providing timely graded early warnings, and quickly locating the root cause of the problem, this invention can proactively avoid safety hazards such as roll stress imbalance, product dimensional deviations, and liner plate detachment caused by worsening liner plate tilting, thereby reducing the number of unplanned mill downtimes. Simultaneously, targeted handling plans developed based on early warning information and source tracing conclusions can optimize liner plate maintenance strategies, extend liner plate service life, reduce equipment maintenance costs and production losses, and provide reliable technical support for the safe, stable, and efficient operation of rolling mill equipment in the metallurgical industry. Attached Figure Description

[0023] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0024] Figure 1 This is a schematic flowchart of a rolling mill liner tilt detection method based on multi-axis tilt angle sensing according to the present invention. Figure 2 This is a flowchart of the wear-reference linkage self-calibration mechanism in this invention. Detailed Implementation

[0025] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0026] Please see Figures 1-2 A method for detecting the tilt of rolling mill liners based on multi-axis tilt sensing, comprising: S1: Obtain the initial reference tilt angle data of the mill liner detection area, introduce a wear-reference linkage self-calibration mechanism, correct the initial reference tilt angle data, and generate dynamic reference data; construct a multi-dimensional coupled mapping model of the liner, and set a dynamic threshold system based on historical fault and normal working condition data; S2: Acquire real-time tilt angle data of each detection area during the operation of the rolling mill, synchronously integrate physical quantity compensation data and perform preprocessing; identify abnormal data points in the time-series tilt data based on the dynamic threshold system, correct the abnormal data points, and generate preprocessed valid data; S3: Based on the comparison results between the effective data and the dynamic benchmark data, the vibration interference signal is suppressed, and the regional tilt offset is generated through comparison calculation; combined with the force distribution characteristics of the liner in the multi-dimensional coupling model of the liner, the regional weight is optimized by using the regional-global bidirectional iterative calculation logic; based on the time-series tilt data, the trend prediction model is trained to generate the liner tilt development trend prediction result; S4: Generate tilt anomaly judgment results through cross-validation and redundancy verification of multi-regional data; trace the inducing factors of tilt anomalies; based on the dynamic threshold system, set up a multi-level early warning mechanism according to the severity and development trend of the anomaly; generate multi-level early warning signals after triggering the early warning; output the anomaly area, tilt parameters and the conclusion of the traceability of the inducing factors of the anomaly; and generate an operation detection report.

[0027] Specifically, the calibration process of the wear-benchmark linkage self-calibration mechanism includes: firstly, continuously collecting the cumulative running time of the liner and the actual wear data of each maintenance operation record, and establishing a database in combination with the wear resistance characteristics of the liner material; secondly, mining the correspondence between the cumulative running time and the wear amount through correlation analysis algorithm, and then coupling the correspondence with the tilt angle change data to form tilt angle correction coefficients for different wear stages; thirdly, calling the corresponding correction coefficients in stages according to the wear process of the liner operation to correct the initial benchmark tilt angle data step by step; fourthly, verifying the correction effect by comparing the current tilt angle data with the corrected dynamic benchmark data. If the deviation exceeds the reasonable range, the correction coefficients are readjusted until the dynamic benchmark data can accurately reflect the benchmark level of the liner under the current actual wear state.

[0028] Specifically, the construction process of the liner multi-dimensional coupling mapping model includes: firstly, classifying and collecting liner material characteristic parameters, actual gap data during installation, load data at different operating stages of the rolling mill, speed fluctuation data, and continuous monitoring data of ambient temperature and humidity; secondly, denoising and normalizing the collected data to remove invalid interference data; thirdly, using a multi-factor correlation analysis algorithm to calculate the correlation coefficient between various data and the liner inclination angle change, and determining the influence weight of each factor on the inclination angle change; fourthly, dividing the analysis units according to different working condition combinations (different combinations of load and speed, different ranges of ambient temperature and humidity), and establishing a quantitative mapping relationship between each influencing factor and the inclination angle change within each analysis unit; and fifthly, integrating the mapping relationships of all analysis units to form a liner multi-dimensional coupling mapping model covering all working conditions, providing a basis for subsequent threshold setting and anomaly tracing.

[0029] Specifically, the update process of the dynamic threshold system is as follows: periodically collect data on the current wear stage of the liner (determined by a combination of wear detection and runtime) and real-time changes in ambient temperature and humidity; call up operating condition data similar to the current wear stage and environmental conditions in the historical database and extract the corresponding threshold parameters; use a parameter optimization algorithm, combined with the current operating data and historical similar operating condition data, to adjust the upper and lower limits of the threshold so that the threshold can adapt to the current wear state of the liner and environmental changes; after the update is completed, the new threshold parameters are substituted into the actual detection process to monitor the accuracy of the early warning and the false alarm rate. If the false alarm rate exceeds the allowable range, more historical data is retrieved again for optimization until the threshold parameters can accurately distinguish between normal tilt fluctuations and abnormal tilt states.

[0030] This embodiment uses the tilt detection of liners in hot rolling mills in the metallurgical industry as an application scenario to illustrate the specific implementation process of the present invention. Implementation of benchmark self-calibration and dynamic threshold preset Wear-reference linkage self-calibration implementation First, continuously collect the cumulative running time T_acc of the liner and the actual wear amount W_act of each maintenance operation record. Combine this with the wear resistance characteristics parameter M_mat of the liner material (such as hardness and wear resistance coefficient) to establish a database DB={(T_acc1,W_act1,M_mat),(T_acc2,W_act2,M_mat),...}.

[0031] Association analysis algorithms (such as the Pearson correlation coefficient method) are used to mine data associations: the correlation coefficient between cumulative running time and actual wear is calculated. r_TW=Cov(T_acc,W_act) / (σ_Tacc×σ_Wact) Where Cov(·) is the covariance and σ is the standard deviation; then the correlation coefficient is coupled with the tilt angle change data Δθ_hist to generate the tilt angle correction coefficient K_cal=f(r_TW,Δθ_hist,M_mat) for different wear stages, and the function f(·) is a nonlinear mapping relationship trained based on historical data.

[0032] K_cal is called in stages according to the liner wear process to successively correct the initial reference tilt angle data B0: B_dyn=B0×K_cal, where B_dyn is the dynamic reference data.

[0033] Verify the correction effect: Collect the current tilt angle data B_curr, calculate the deviation E_B=|B_curr-B_dyn|, if E_B exceeds the preset reasonable range [E_min,E_max], then readjust the mapping parameters of K_cal and repeat the correction until E_B∈[E_min,E_max].

[0034] Construction of multi-dimensional coupling mapping model for liner Data collected by category: wear resistance parameter of liner material M_mat, actual installation gap data G_ins, mill load data L_run, speed fluctuation data S_fluc, and ambient temperature and humidity data H_env (temperature T_env, humidity RH_env).

[0035] Data preprocessing: The mean filtering method is used to denoise X_denoise=(ΣX_i) / N (X is the original data, N is the amount of data in the filtering window); invalid interference data is removed by normalization X_norm=(X-X_min) / (X_max-X_min) (X_min and X_max are the extreme values ​​of the data).

[0036] Calculate the influence weight: Use the grey relational analysis method to calculate the correlation degree between each data and the change in tilt angle Δθ, r_i=(minmin|Δθ-X_i|+ζmaxmax|Δθ-X_i|) / (|Δθ-X_i|+ζmaxmax|Δθ-X_i|) (ζ is the resolution coefficient), and generate the influence weight ω_i=r_i / Σr_i (Σr_i is the sum of all correlation degrees).

[0037] Establish quantitative mapping relationships: Divide the analysis units according to the combination of working conditions (different combinations of L_run and S_fluc, different intervals of H_env), and construct a mapping model Δθ=ω_mat×M_mat+ω_ins×G_ins+ω_run×L_run+ω_fluc×S_fluc+ω_env×H_env within each unit. Integrate the mapping relationships of all units to obtain the multi-dimensional coupling mapping model M_couple of the liner.

[0038] Dynamic threshold system settings Based on historical fault data DB_fault and normal operating condition data DB_norm, and combined with the influence weights ω_i of each factor in M_couple, a dynamic threshold Th_dyn=[Th_low,Th_high] is trained and generated, where Th_low=g(DB_norm,ω_i) and Th_high=h(DB_fault,ω_i), and g(·) and h(·) are threshold calculation functions.

[0039] Specifically, the steps for fusing physical quantity compensation data include: deploying data acquisition units in the key stress areas of the liner and at positions near the liner on the mill stand to simultaneously collect temperature data, vibration intensity data, and load fluctuation data, ensuring that the acquisition time is completely synchronized with the sampling time of the real-time tilt angle data; assigning corresponding fusion weights to the temperature data, vibration intensity data, and load fluctuation data according to the influence weights of each physical quantity in the multi-dimensional coupling mapping model of the liner; and weighting and fusing the three types of physical quantity compensation data with the real-time tilt angle data through a data fusion algorithm to correct sensor drift deviation caused by temperature changes, tilt angle data fluctuations caused by vibration interference, and instantaneous tilt angle distortion caused by load fluctuations, thereby improving the accuracy of the real-time tilt angle data.

[0040] Specifically, the preprocessing procedure includes: first, using an adaptive filtering algorithm to denoise the fused tilt angle data; by analyzing the frequency characteristics of the data, identifying and filtering high-frequency vibration interference signals and electromagnetic interference signals; and dynamically adjusting the filter cutoff frequency to adapt to the signal characteristics under different operating conditions. Next, using a time-series data smoothing algorithm to process the denoised data, selecting a fixed-length sliding window, calculating the mean or median of the data within the window, and replacing the original data at the center of the window with this statistical value to eliminate random fluctuations in the data. After processing, verifying the temporal continuity of the data; if data breaks exist, using linear interpolation to fill the breaks, ensuring that the preprocessed effective data is continuous, stable, and free from significant interference.

[0041] Physical quantity compensation data fusion Data acquisition units are deployed in the stress area of ​​the liner to simultaneously acquire temperature data T_comp, vibration intensity data V_comp, and load fluctuation data L_comp, ensuring that the acquisition time is completely synchronized with the sampling time of the real-time tilt angle data θ_raw.

[0042] Extract the influence weights ω_T, ω_V, and ω_L of each physical quantity from M_couple, assign a weight ω_θ to θ_raw (ω_θ+ω_T+ω_V+ω_L=1), and calculate θ_fuse=ω_θ×θ_raw+ω_T×T_comp+ω_V×V_comp+ω_L×L_comp using a weighted fusion algorithm.

[0043] Preprocessing implementation Noise reduction: An adaptive filtering algorithm is used to analyze the frequency characteristics F_θ of θ_fuse, identify the frequency of the interference signal F_dist, and dynamically adjust the filter cutoff frequency F_cut to filter high-frequency interference signals: θ_filter=Filter(θ_fuse,F_cut,F_dist).

[0044] Data smoothing: Select a fixed-length sliding window N_win and use the mean method to calculate smoothed data: θ_smooth=(Σθ_filter(i)) / N_win (i is the data index within the window).

[0045] Temporal continuity verification: If there are data breakpoints (x_a, θ_a) and (x_b, θ_b), use linear interpolation to fill them: θ_interp=θ_a+(x-x_a)×(θ_b-θ_a) / (x_b-x_a) (x is the x-coordinate of the breakpoint).

[0046] Abnormal data correction: Based on Th_dyn, abrupt and jump anomalous data points θ_abn are identified. Continuous valid data segments θ_prev and θ_next before and after the anomalous data point are selected. The neighborhood similarity feature Sim=Corr(θ_abn neighborhood, θ_prev neighborhood) is calculated to generate fill data θ_fill=Sim×θ_prev+(1-Sim)×θ_next. Combined with time-series trend fitting correction: θ_valid=θ_fill+k×(θ_next-θ_prev) (k is the trend coefficient), finally generating the preprocessed valid data θ_valid.

[0047] Specifically, the process of suppressing vibration interference signals includes: firstly, real-time acquisition of vibration intensity data during mill operation; secondly, identification of the vibration signal type (periodic or random vibration), frequency range, and amplitude variation pattern through spectrum analysis; thirdly, adjustment of the core parameters of the adaptive anti-interference and noise reduction technology based on the identification results; fourthly, setting targeted notch filter parameters for corresponding frequencies for periodic vibrations to accurately suppress vibration interference within that frequency range; fifthly, dynamically adjusting the noise reduction amplitude threshold to suppress interference signals in stages according to the magnitude of vibration intensity; and sixthly, real-time monitoring of the amplitude change of the effective tilt angle signal during noise reduction to avoid loss of effective tilt angle information due to excessive noise reduction, ensuring that the signal after suppressing interference can truly reflect the actual tilt state of the liner.

[0048] Specifically, the implementation process of the region-global bidirectional iterative calculation logic includes: first, referring to the force distribution characteristics of each detection region in the multi-dimensional coupling mapping model of the liner, assigning initial weights to each local detection region, with the initial weights of regions with concentrated force and easy wear being higher than those of other regions; then, weighted summing of the tilt offset of each local region with the corresponding initial weight to obtain the initial global composite tilt state; calculating the deviation between the tilt offset of each local region and the initial global composite tilt state, and adjusting the weight of the corresponding region according to the magnitude of the deviation value—the larger the deviation value, the larger the weight adjustment range, so as to enhance the influence of the region on the global composite result; substituting the adjusted region weights into the weighted summation operation again to obtain a new global composite tilt state; repeating the above deviation calculation, weight adjustment and global synthesis steps until the difference between two adjacent global composite tilt states is less than the preset judgment standard, stopping the iteration, and outputting the final global composite tilt state.

[0049] Specifically, the training process of the trend prediction model includes: collecting historical tilt data accumulated from long-term operation of the liner, and organizing it into a time-series dataset in chronological order; simultaneously collecting operating condition-related data (load, speed) and environmental impact data (temperature and humidity) for each period corresponding to the historical tilt data, and mapping them one-to-one with the time-series tilt data; preprocessing all training data to remove outliers, fill in missing values, and perform data normalization to map the data to a unified range; dividing the data into training and validation sets, with the training set used for model training and the validation set used for model performance evaluation; constructing the model framework using a time-series analysis algorithm, inputting the training set data, and iteratively adjusting the core parameters of the model (such as time window size and feature extraction dimension) to optimize the model with the goal of minimizing the prediction error of the validation set; after training, testing the model with new historical data, and if the prediction accuracy does not meet the preset requirements, supplementing more training data and repeating the training process until the model can accurately predict the magnitude and direction of liner tilt changes under different operating conditions.

[0050] Vibration interference signal suppression Real-time acquisition of mill vibration intensity data V_run, and obtaining vibration signal type (periodic / random), frequency range F_V, and amplitude variation pattern A_V through spectrum analysis.

[0051] Adjust the adaptive anti-interference and noise reduction parameters for the operating conditions: For periodic vibration, set the notch filter parameter Para_notch=(F_V,Q) (Q is the quality factor) to suppress the interference at that frequency; for random vibration, dynamically adjust the noise reduction amplitude threshold A_th=f(A_V) to suppress interference in stages: θ_denoise=θ_valid-α×A_V×F_V (α is the noise reduction coefficient, which is dynamically adjusted with A_V), and at the same time monitor the change in the effective signal amplitude ΔA_θ to ensure ΔA_θ_θ_max (A_θ_max is the maximum allowable amplitude loss of the effective signal).

[0052] Calculation of regional tilt offset Based on the comparison results of θ_denoise and B_dyn, the tilt offset of each detection region is calculated: Δθ_reg = θ_denoise - B_dyn.

[0053] Region-Global Bidirectional Iterative Computation Based on the force distribution characteristics of the liner in M_couple, an initial weight ω_reg0 is assigned to each local detection area (ω_reg0 is higher in areas with concentrated force).

[0054] Initial global composite tilt state calculation: θ_global0=Σ(Δθ_reg×ω_reg0).

[0055] Deviation calculation and weight adjustment: E_reg=|Δθ_reg-θ_global0|, adjustment weight ω_reg1=ω_reg0×(1+E_reg / ΣE_reg); Normalization is performed using ω_reg1_norm = ω_reg1 / Σω_reg1.

[0056] Iteratively update the global synthesis state: θ_global1=Σ(Δθ_reg×ω_reg1_norm), repeat the bias calculation, weight adjustment, and global synthesis steps until |θ_global1-θ_global0| is the convergence criterion), stop the iteration, and output the final global synthesis tilt state θ_global_final.

[0057] Trend prediction model training and prediction Construct a time-series dataset: Collect historical tilt data of the liner during long-term operation, and organize it in chronological order as X_seq=[θ_valid1,θ_valid2,...,θ_validn]. Simultaneously collect the corresponding time period working condition correlation data Y_seq=[L_run,S_fluc,H_env], and form a training dataset DB_train={(X_seq1,Y_seq1),...,(X_seqm,Y_seqm)}.

[0058] Data preprocessing: Remove outliers (based on the 3σ criterion), fill in missing values ​​(using interpolation), and normalize X_seq_norm=(X_seq-X_seq_mean) / X_seq_std (X_seq_mean is the mean, and X_seq_std is the standard deviation).

[0059] Split the training and validation sets: DB_train_split=(DB_train_tr,DB_train_val) (split proportionally), use time series analysis algorithms (such as ARIMA model) to build the model framework, and set the model parameters (p,d,q) (p is the autoregressive order, d is the difference order, and q is the moving average order).

[0060] Model training and optimization: Input DB_train_tr, calculate prediction error Loss=Σ(θ_pred-θ_true) 2(θ_pred is the predicted value, θ_true is the true value), iteratively adjust (p,d,q) to minimize the loss, verify the model performance through DB_train_val, if the verification error Loss_val > Loss_th (Loss_th is the error threshold), supplement the training data and repeat the training, finally generate the trend prediction model M_pred, and output the prediction result of the liner tilt development trend θ_trend=M_pred(X_seq_new,Y_seq_new).

[0061] Specifically, the implementation process of the multi-region data cross-validation and redundancy verification includes: extracting the tilt offset, time-series change curve and local anomaly characteristic parameters of each detection region, and dividing the verification groups according to the classification of adjacent regions, symmetrical regions, and force-related regions; performing bidirectional comparison of the tilt data of each region within the same group, and calculating the synchronization deviation of tilt changes between regions; simultaneously retrieving the redundant backup data of each detection region, comparing the real-time acquired data with the redundant backup data point by point, eliminating abnormal region data, and including the data that has passed both the cross-validation within the group and the redundant backup comparison into the anomaly judgment analysis.

[0062] Specifically, the process of tracing the inducing factors of tilt anomalies includes: extracting working condition data, wear correlation data, and environmental impact data for the period when tilt anomalies occur from the multi-dimensional coupling mapping model of the liner, comparing them item by item with the corresponding data under normal working conditions, and calculating the deviation value; based on the influence weight priority of each factor in the multi-dimensional coupling mapping model of the liner, checking them one by one, analyzing the causal relationship between the identified abnormal factors and tilt anomalies, obtaining the core inducing factors and related inducing factors, and generating a source tracing chain of abnormal inducing factors.

[0063] Specifically, the implementation process of the multi-level early warning mechanism is as follows: First, combining the tilt offset standards of different intervals in the dynamic threshold system, the tilt development trend prediction results, and the impact range of tilt anomalies on mill operation, three levels of early warning are divided. Each early warning level corresponds to a clear judgment condition—the first level of early warning corresponds to the situation where the tilt offset does not exceed the critical impact range and the development trend is gentle; the second level of early warning corresponds to the situation where the tilt offset is close to the critical standard or the development trend is accelerating; the third level of early warning corresponds to the situation where the tilt offset exceeds the critical standard and the development trend is drastic, which may affect the safe operation of the mill. When any level of early warning is triggered, the location of the abnormal area, specific tilt parameters, the conclusion of the source tracing of the abnormal inducing factors, and the tilt development trend prediction results are integrated to form a complete early warning information package; then, according to the early warning level... Matching the corresponding information push methods, the first-level warning is notified to operators through pop-up prompts on the rolling mill control interface and on-site audio-visual prompts. The second-level warning, in addition to the above prompts, sends warning information and preliminary handling suggestions to the technical control terminal. The third-level warning, based on the second level, simultaneously pushes emergency handling notices and key control requirements to the operation and maintenance management terminal. At the same time, the data recording module is linked to record the warning trigger time, warning level, abnormal data, handling measures, and post-handling tilt state changes in real time, forming a warning handling archive. Subsequently, by collecting the post-handling liner tilt data, the accuracy of the warning level determination and the effectiveness of the handling measures are verified. The verification results are fed back to the dynamic threshold system optimization stage to adjust the judgment conditions of each level of warning, thereby improving the adaptability and accuracy of the warning mechanism.

[0064] In this embodiment, multi-regional data cross-validation and redundancy verification are performed. Extract Δθ_reg, time-series change curve Curve_reg, and local anomaly feature parameter Para_reg from each detection area, and divide them into verification groups Group_reg={G1,G2,...} according to adjacent areas, symmetrical areas, and force-related areas.

[0065] Intra-group cross-validation: Calculate the synchronicity deviation of regional tilt changes within the same group, E_sync = |Δθ_reg_i - Δθ_reg_j| (i,j∈Gk).

[0066] Redundancy verification: Retrieve redundant backup data θ_backup from each region and calculate the point-by-point deviation between real-time data and backup data E_redun=|θ_denoise-θ_backup|.

[0067] Anomaly detection: If E_syncE_redun_redun_th (E_sync_th and E_redun_th are verification thresholds), then the data is valid and included in the anomaly detection analysis, generating a tilt anomaly detection result Res_abn` (normal / abnormal).

[0068] Tracing the inducing factors of tilting abnormality Extract abnormal period operating condition data L_abn, wear correlation data W_abn, and environmental impact data H_abn from M_couple, and compare them item by item with normal operating condition data L_norm, W_norm, and H_norm to calculate the deviations ΔL=L_abn-L_norm, ΔW=W_abn-W_norm, and ΔH=H_abn-H_norm.

[0069] According to the influence weight priority ω_rank (from high to low) of each factor in M_couple, check them one by one: if |ΔL|>ΔL_th (ΔL_th is the load deviation threshold), then analyze the cause of ΔL (such as raw material distribution imbalance); if ΔL is not abnormal, check ΔW and ΔH in turn, analyze the causal relationship between abnormal factors and Res_abn, determine the core inducing factor F_core and the related inducing factor F_related, and generate the abnormal inducing factor tracing chain Chain_abn={F_core,F_related,...}.

[0070] Implementation of multi-level early warning mechanism Based on the tilt offset standard of Th_dyn, θ_trend, and the range of impact of the anomaly on the mill operation Range_abn, the warning level is divided into Level_alert={L1,L2,L3}.

[0071] When an alert is triggered, the abnormal area location Pos_abn, the specific tilt parameters Δθ_reg, θ_global_final, Chain_abn, and θ_trend are integrated to generate an alert information package.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for detecting the tilt of rolling mill liners based on multi-axis tilt angle sensing, characterized in that, include: S1: Obtain the initial reference tilt angle data of the mill liner detection area, introduce a wear-reference linkage self-calibration mechanism, correct the initial reference tilt angle data, and generate dynamic reference data; Construct a multi-dimensional coupled mapping model for the liner, and set a dynamic threshold system based on historical fault and normal operating condition data; S2: Acquire real-time tilt angle data of each detection area during the operation of the rolling mill, synchronously integrate physical quantity compensation data and perform preprocessing; Based on the dynamic threshold system, abnormal data points in time-series tilted data are identified, the abnormal data points are corrected, and preprocessed valid data is generated. S3: Based on the comparison results between the effective data and the dynamic benchmark data, the vibration interference signal is suppressed, and the regional tilt offset is generated through comparison calculation; combined with the force distribution characteristics of the liner in the multi-dimensional coupling model of the liner, the regional weight is optimized by using the regional-global bidirectional iterative calculation logic. A trend prediction model is trained based on the aforementioned time-series tilt data to generate prediction results of the liner tilt development trend. S4: Generate tilt anomaly determination results through cross-validation and redundancy verification of multi-region data; The system traces the inducing factors of tilt anomalies, sets up a multi-level early warning mechanism based on the dynamic threshold system according to the severity and development trend of the anomalies, generates multi-level early warning signals after triggering an early warning, outputs the abnormal area, tilt parameters and the conclusion of the tracing of the inducing factors of the anomalies, and generates an operation detection report.

2. The method according to claim 1, characterized in that, In S1, the calibration process of the wear-reference linkage self-calibration mechanism specifically includes: continuously acquiring the cumulative running time of the liner and the actual wear data of each maintenance operation record, and establishing a database in combination with the wear resistance characteristics of the liner material; mining the correspondence between the cumulative running time and the actual wear amount through correlation analysis algorithm, coupling analysis with the tilt angle change data, and generating tilt angle correction coefficients for different wear stages; calling the corresponding correction coefficients in stages based on the wear process of the liner operation to successively correct the initial reference tilt angle data; comparing the current tilt angle data with the corrected dynamic reference data to verify the correction effect.

3. The method according to claim 1, characterized in that, In S1, the construction process of the liner multi-dimensional coupling mapping model specifically includes: classifying and acquiring liner material characteristic parameters, actual gap data during installation, load data, speed fluctuation data, and continuous monitoring data of ambient temperature and humidity at different operating stages of the rolling mill; performing noise reduction and normalization on various types of data, eliminating invalid interference data, calculating the correlation coefficient with the liner inclination angle change, and generating the influence weight of the inclination angle change; dividing the analysis unit based on different working condition combinations, establishing a quantitative mapping relationship between each influencing factor and the inclination angle change within each analysis unit, integrating the quantitative mapping relationship, and generating the liner multi-dimensional coupling mapping model.

4. The method according to claim 1, characterized in that, In S1, the update process of the dynamic threshold system is as follows: periodically acquire data on the current wear stage of the liner and real-time changes in ambient temperature and humidity; call the operating condition data similar to the current wear stage and environmental conditions in the preset historical database and extract the corresponding threshold parameters; combine the current operating data and historical similar operating condition data to adjust the upper and lower limits of the threshold, and substitute the new threshold parameters into the actual detection process to monitor the accuracy and false alarm rate of the early warning.

5. The method according to claim 1, characterized in that, In S2, the specific steps of fusing physical quantity compensation data include: acquiring temperature data, vibration intensity data, and load fluctuation data in the stress area of ​​the liner; assigning corresponding fusion weights to the temperature data, vibration intensity data, and load fluctuation data based on the influence weights of physical quantities in the multi-dimensional coupling mapping model of the liner; and weighting and fusing the physical quantity compensation data with the real-time tilt angle data through calculation.

6. The method according to claim 1, characterized in that, In S2, the specific implementation process of the preprocessing flow includes: denoising the fused tilt angle data; identifying and filtering interference signals by analyzing the frequency characteristics of the data; dynamically adjusting the filter cutoff frequency; selecting a sliding window of fixed length based on a time-series data smoothing algorithm; calculating the mean or median of the data within the window; replacing the original data at the center of the window; after processing, verifying the temporal continuity of the data; if there are data breakpoints, using linear interpolation to fill the breakpoints.

7. The method according to claim 1, characterized in that, In S3, the specific process of suppressing vibration interference signals includes: acquiring vibration intensity data during the operation of the rolling mill in real time, identifying the type, frequency range and amplitude variation law of the vibration signal through spectrum analysis, adjusting the parameters of the working condition adaptive anti-interference and noise reduction technology, and monitoring the amplitude variation of the effective tilt angle signal in real time.

8. The method according to claim 1, characterized in that, In S3, the specific implementation process of the region-global bidirectional iterative calculation logic includes: based on the force distribution characteristics of each detection region in the multi-dimensional coupling mapping model of the liner, assigning initial weights to each local detection region; weighted summing of the tilt offset of each local region with the corresponding initial weight to obtain the initial global composite tilt state; calculating the deviation value between the tilt offset of each local region and the initial global composite tilt state, and adjusting the weight of the corresponding region; substituting the adjusted region weights into the weighted summation operation again to obtain a new global composite tilt state; repeating the deviation calculation, weight adjustment and global synthesis steps until the difference between two adjacent global composite tilt states is less than the preset judgment standard, stopping the iteration, and outputting the final global composite tilt state.

9. The method according to claim 1, characterized in that, In S3, the training process of the trend prediction model specifically includes: based on the historical tilt data accumulated from the long-term operation of the liner, organizing it into a time series dataset in chronological order, and simultaneously acquiring the working condition-related data and environmental impact data corresponding to each period of historical tilt data; performing preprocessing to remove outliers, fill in missing values, and perform data normalization to map the data to a unified range; dividing the training set and validation set, constructing the model framework using a time series analysis algorithm, inputting the training set data, iteratively adjusting the core parameters of the model, optimizing the model with the goal of minimizing the prediction error of the validation set, generating a development trend prediction model, and predicting the magnitude and direction of the liner tilt change under different working conditions.

10. The method according to claim 1, characterized in that, In S4, the specific implementation process of the multi-region data cross-verification and redundancy verification includes: extracting the tilt offset, time-series change curve and local anomaly characteristic parameters of each detection region, and dividing the verification groups according to the classification of adjacent regions, symmetrical regions, and force-related regions; performing bidirectional comparison of the tilt data of each region within the same group, and calculating the synchronization deviation of tilt changes between regions; simultaneously retrieving the redundant backup data of each detection region, comparing the real-time acquired data with the redundant backup data point by point, eliminating abnormal region data, and including the data that has passed both the cross-verification within the group and the redundant backup comparison into the anomaly judgment analysis.

11. The method according to claim 1, characterized in that, In S4, the specific process of tracing the inducing factors of tilt anomalies includes: extracting the operating condition data, wear correlation data, and environmental impact data of the tilt anomaly occurrence period from the liner multi-dimensional coupling mapping model, comparing them item by item with the corresponding data under normal operating conditions, and calculating the deviation value; based on the influence weight priority of each factor in the liner multi-dimensional coupling mapping model, checking them one by one, analyzing the causal relationship between the identified abnormal factors and the tilt anomaly, obtaining the core inducing factors and related inducing factors, and generating a traceability chain of abnormal inducing factors.

12. The method according to claim 1, characterized in that, In S4, the specific implementation process of the multi-level early warning mechanism is as follows: combining the tilt offset standards of different intervals in the dynamic threshold system, the tilt development trend prediction results, and the impact range of tilt anomalies on the operation of the rolling mill, the early warning level is divided; when any level of early warning is triggered, the location of the abnormal area, the specific tilt parameters, the conclusion of the source of the abnormal inducing factors, and the tilt development trend prediction results are integrated to generate an early warning information package.

Images