A method for predicting the wear state of steel-based ceramic wear-resistant components

Abstract

This application provides a method for predicting the wear state of steel-based ceramic wear-resistant components, relating to the field of material life assessment technology. The method includes the following steps: emitting excitation eddy currents containing a first frequency and a second frequency into a steel substrate, and receiving the response signal modulated by the state of the steel substrate and the ceramic-steel interface; processing the response signals corresponding to the first and second frequencies respectively, extracting the first phase change of the first reference signal corresponding to the initial health state of the component relative to the first frequency, and the second phase change of the second reference signal corresponding to the second frequency; calculating the difference between the first and second phase changes to characterize the change in thermal conductivity state of the ceramic-steel interface; obtaining the real-time operating temperature of the wear-resistant component, and based on the real-time operating temperature and the difference, assessing the current state level of the ceramic-steel interface, and predicting the remaining safe service time of the ceramic wear-resistant layer by combining the trend of the difference over time.

Description

Technical Field

[0001] This application relates to the field of material life assessment technology, specifically to a method for predicting the wear state of steel-based ceramic wear-resistant components. Background Technology

[0002] Steel-based ceramic composite wear-resistant components (such as ceramic-lined pipes and composite wear-resistant plates) are widely used in harsh wear environments in industries such as metallurgy, mining, and power due to the extremely high hardness and wear resistance of their surface ceramic layer. However, the typical failure mode of these components is often not the gradual wear of the ceramic working layer, but rather the early damage (such as microcracks and debonding) at the interface between the ceramic layer and the steel substrate under thermal stress and impact loads, ultimately leading to a sudden failure with large-area peeling of the ceramic layer. This interface failure is highly concealed, develops rapidly, and once it occurs, it will cause complete loss of component function and production interruption.

[0003] Currently, condition monitoring and life assessment of such components mainly rely on macroscopic inspections after periodic shutdowns (such as measuring the remaining thickness of the ceramic layer) or holistic monitoring based on vibration and acoustic emission. These methods are extremely insensitive to early degradation of the interface bonding state: thickness measurements cannot detect micro-damage within the interface; and holistic signals (such as vibration frequency) change weakly in the early stages of interface damage, are easily drowned out by noise, and cannot distinguish the specific location and mode of damage.

[0004] Therefore, existing technologies are unable to provide effective early warnings in the early stages of interface damage initiation and expansion, and cannot provide accurate remaining life predictions for predictive maintenance, often leaving users in the dilemma of over-maintenance or sudden failure. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide a method for predicting the wear state of a steel-based ceramic wear-resistant component, wherein the wear-resistant component includes a steel substrate and a ceramic wear-resistant layer bonded to its working surface, comprising the following steps:

[0006] An excitation eddy current containing a first frequency and a second frequency is emitted into the steel substrate, and a response signal modulated by the state of the interface between the steel substrate and the ceramic-steel bonding is received; wherein the first frequency is higher than the second frequency;

[0007] Process the response signals corresponding to the first frequency and the second frequency respectively, and extract the first phase change amount of the first frequency relative to the first reference signal corresponding to the initial health state of the component, and the second phase change amount of the second frequency relative to the second reference signal corresponding to the initial health state of the component;

[0008] Calculate the difference between the first phase change and the second phase change, wherein the difference characterizes the change in the thermal conductivity state of the ceramic-steel interface;

[0009] The real-time operating temperature of the wear-resistant component is obtained, and the current state level of the ceramic-steel interface is evaluated based on the difference between the real-time operating temperature and the difference.

[0010] Based on the current state level and the trend of the difference over time, the remaining safe service time of the ceramic wear-resistant layer is predicted.

[0011] According to the technical solution provided in this application, an excitation eddy current containing a first frequency and a second frequency is emitted to the steel substrate through an eddy current sensor array; the eddy current sensor array is arranged on the non-working surface of the steel substrate corresponding to the area covered by the ceramic wear-resistant layer.

[0012] According to the technical solution provided in this application, predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the changing trend of the difference over time includes the following steps:

[0013] Based on the differences obtained at multiple monitoring times, their variation over time is analyzed to establish a degradation trend.

[0014] Based on the degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time.

[0015] According to the technical solution provided in this application, the step of analyzing the variation law of the difference obtained at multiple monitoring times to establish a degradation trend includes the following steps:

[0016] Data from each sensor unit in the eddy current sensor array at multiple monitoring times are collected to obtain the time series of differences for each sensor unit;

[0017] The time series of differences of all sensor units at the same monitoring time are mapped according to their spatial location to generate multiple sets of two-dimensional damage distribution snapshots describing the degree of interface damage in the entire monitoring area.

[0018] Comparative analysis was performed on all the two-dimensional damage distribution snapshots arranged in chronological order to extract temporal and spatial features;

[0019] Based on the time dimension features and the spatial dimension features, a degradation trend is established.

[0020] According to the technical solution provided in this application, the extraction of time dimension features and spatial dimension features includes the following steps:

[0021] Calculate the curve of the average damage index over time for the entire monitoring area or a specific area of ​​interest, and extract the instantaneous slope and acceleration of the curve as time dimension features.

[0022] Identify and track the centroid location, area, and morphological evolution of high-damage regions appearing in the two-dimensional damage distribution snapshot, and use them as spatial dimension features.

[0023] According to the technical solution provided in this application, establishing a degradation trend based on the time dimension features and the spatial dimension features includes the following steps:

[0024] Based on the instantaneous slope and acceleration in the time dimension features and the morphological evolution of the high-damage region in the spatial dimension features, the first candidate degradation path is deduced based on the fracture mechanics principle of interface damage.

[0025] The change curves of the time dimension features and the evolution sequences of the spatial dimension features are input into the temporal prediction network to obtain the second candidate degradation path.

[0026] The degree of agreement between the first candidate degradation path and historical long-term data, and the degree of agreement between the second candidate degradation path and recent data are evaluated, and the two are weighted and fused to generate a degradation trend.

[0027] According to the technical solution provided in this application, after generating the degradation trend, the following steps are also included:

[0028] Retrieve the historical failure case database. Each historical failure case in the database includes a sequence of historical monitoring data for the entire process of similar wear-resistant components from initial damage to final failure, as well as the corresponding actual failure time and failure mode.

[0029] The change curve of the time dimension characteristics and the evolution sequence of the spatial dimension characteristics of the current wear-resistant component up to the latest monitoring time are matched with the data of the same period of each historical failure case in the historical failure case database to select at least one reference historical case.

[0030] Extract the actual degradation path of the reference historical case in the subsequent development process as a benchmark reference degradation path;

[0031] The baseline degradation path is compared with the degradation trend, and the deviation between the two at key nodes is calculated.

[0032] The step of estimating the time required for the ceramic-steel interface to reach a preset critical condition based on the degradation trend and the current state level, and using this time as the remaining safe service time, includes the following steps:

[0033] If the deviation is less than or equal to the first preset threshold, then based on the degradation trend and the current state level, the time required for the ceramic-steel interface state to reach the preset critical condition is estimated and used as the remaining safe service time.

[0034] According to the technical solution provided in this application, after comparing the benchmark degradation path with the degradation trend and calculating the deviation between the two at key nodes, the method further includes the following steps:

[0035] If the deviation is greater than the first preset threshold, the evolution rate and inflection point of the degradation trend are dynamically corrected according to the deviation to obtain the corrected degradation trend.

[0036] Based on the modified degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time.

[0037] According to the technical solution provided in this application, after evaluating the current state level of the ceramic-steel interface, the method further includes the following steps:

[0038] The real-time temperature distribution field of the non-working surface of the steel substrate is obtained; the real-time temperature distribution field is obtained by a temperature sensor array arranged at the corresponding position of the eddy current sensor array;

[0039] At each monitoring moment, the difference quantity obtained by the eddy current sensor array, which reflects the change in the interface heat conduction state, is mapped into an eddy current anomaly distribution field according to its spatial location.

[0040] The real-time temperature distribution field and the eddy current anomaly distribution field at the same moment are spatially superimposed and compared to obtain the confidence level of the current state level.

[0041] The step of predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the trend of the difference over time includes the following steps:

[0042] If the confidence level is high confidence, then the remaining safe service time of the ceramic wear-resistant layer is predicted based on the current state level and the trend of the difference over time.

[0043] According to the technical solution provided in this application, obtaining the confidence level of the current state level includes the following steps:

[0044] When the comparative analysis results show that the spatial overlap area between one or more local high-temperature regions identified in the real-time temperature distribution field and one or more high-anomaly regions identified in the eddy current anomaly distribution field exceeds a preset area threshold, the confidence level of the current state level is determined to be high confidence.

[0045] Compared with the prior art, the beneficial effects of this application are as follows:

[0046] I. Achieved highly sensitive and specific detection of early degradation of interfacial bonding. This invention, by calculating the phase difference between high-frequency and low-frequency eddy current response signals, can specifically capture subtle changes in the electromagnetic properties of the steel substrate surface caused by interfacial damage. Since phase parameters are extremely sensitive to the electromagnetic properties of materials (such as changes in conductivity and permeability caused by localized temperature rise due to changes in interfacial thermal resistance), and the dual-frequency differential technique effectively suppresses common-mode interference such as overall performance degradation of the steel substrate, this method can detect interfacial microcracks, early debonding, and other damage that are undetectable by traditional methods, achieving a breakthrough in monitoring dimensions from the overall to the interfacial level, and from macroscopic to micro / nanoscale.

[0047] Second, a direct, quantitative prediction pathway from physical signals to remaining service life has been established. This invention establishes a complete technical chain from "eddy current phase difference → change in interface thermal conduction state → interface damage level → remaining safe service time". By combining the difference in interface state with real-time operating conditions (temperature) for condition assessment, and further analyzing the temporal evolution trend of this difference, the development process of interface damage can be dynamically predicted, thereby achieving early warning of the risk of ceramic wear-resistant layer spalling and quantitative prediction of remaining safe service time. This provides a reliable technical tool for shifting from on-time maintenance or reactive repair to condition-based predictive maintenance.

[0048] Third, it improves the reliability and engineering applicability of condition assessment results. By introducing real-time operating temperature to compensate and correct the detection signals, the impact of environmental and operating condition fluctuations on monitoring results is effectively eliminated, improving the accuracy of condition level assessment. Simultaneously, the final output of remaining safe service time is an intuitive and clear engineering parameter, greatly facilitating decision-making by equipment maintenance personnel and helping to formulate optimal maintenance and replacement plans. This maximizes component lifespan while ensuring safety, reducing operation and maintenance costs. Attached Figure Description

[0049] Figure 1 A flowchart illustrating the steps of the method for predicting the wear state of steel-based ceramic wear-resistant components provided in this application. Detailed Implementation

[0050] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] Example 1

[0053] As mentioned in the background section, to address the problems in the prior art, this application proposes a method for predicting the wear state of steel-based ceramic wear-resistant components. The wear-resistant component includes a steel substrate and a ceramic wear-resistant layer bonded to its working surface, such as... Figure 1 As shown, it includes the following steps:

[0054] S1. Emit excitation eddy currents containing a first frequency and a second frequency to the steel substrate, and receive the response signal modulated by the state of the interface between the steel substrate and the ceramic-steel bonding; wherein the first frequency is higher than the second frequency;

[0055] S2. Process the response signals corresponding to the first frequency and the second frequency respectively, and extract the first phase change amount of the first frequency relative to the first reference signal corresponding to the initial health state of the component, and the second phase change amount of the second frequency relative to the second reference signal corresponding to the initial health state of the component.

[0056] S3. Calculate the difference between the first phase change and the second phase change, wherein the difference characterizes the change in the thermal conductivity state of the ceramic-steel interface.

[0057] S4. Obtain the real-time operating temperature of the wear-resistant component, and evaluate the current state level of the ceramic-steel interface based on the difference between the real-time operating temperature and the difference.

[0058] S5. Based on the current state level and the trend of the difference over time, predict the remaining safe service time of the ceramic wear-resistant layer.

[0059] Specifically, steel-based ceramic wear-resistant components refer to composite components made by bonding a layer of ceramic material (ceramic wear-resistant layer) to the wear-bearing working surface of steel material as the base (steel substrate) through processes such as welding, sintering, spraying, or inlaying. Common types include ceramic composite wear-resistant steel plates, pipe liners, fan blades, pump casing liners, and vulnerable parts of mineral processing equipment. Excited eddy currents refer to the annular current induced in the adjacent conductive material (here, the steel substrate) by alternating current passing through an excitation coil. In this method, the excited eddy currents are actively generated by external detection equipment (such as an eddy current flaw detector or a dedicated sensor). First frequency and second frequency refer to the frequencies corresponding to the two different frequencies of alternating current applied to the excitation coil. The first frequency is designed to be a relatively high frequency (e.g., in the range of several hundred kHz to several MHz), and the resulting eddy current exhibits a significant "skin effect," with a shallow penetration depth, and is mainly sensitive to changes in the electromagnetic properties near the ceramic-steel interface and the surface of the steel substrate. The second frequency is designed to be a lower frequency (e.g., in the range of tens to hundreds of kHz), with a deeper eddy current penetration depth, reflecting more information about the steel matrix and deeper regions. Response signal: refers to the electromagnetic induction signal picked up by the receiving coil or the same coil (in self-comparison mode) after the excited eddy current in the steel matrix is ​​modulated by its own electromagnetic properties (such as conductivity and permeability) and interface conditions (such as bonding integrity and thermal conductivity). This signal contains information such as amplitude and phase. Phase change: refers to the offset of the response signal in phase angle relative to its initial reference signal. In this method, the phase change is calculated for the response signals at the first and second frequencies respectively. Phase is extremely sensitive to small changes in the electromagnetic properties of the material, making it particularly suitable for monitoring early, slow damage. Difference: refers to the difference between the first and second phase change calculated through specific mathematical operations (such as subtraction, ratio, normalized difference, etc.). The core objective is to amplify and extract signal characteristics primarily caused by changes in the thermal conductivity state of the ceramic-steel interface through differential or combined processing of dual-frequency signals, while suppressing or subtracting the influence of common-mode interference factors such as uniform aging of the steel substrate and uniform temperature changes. The thermal conductivity state of the ceramic-steel interface refers to the efficiency and uniformity of heat transfer from the ceramic layer to the steel substrate (or vice versa) at the interface. When the interface has cracks, delamination, increased porosity, or foreign matter (such as an oxide layer), its thermal resistance increases, and the thermal conductivity state deteriorates. This method utilizes the principle that the phase response of electromagnetic eddy currents is indirectly affected by the thermophysical properties of the material (related to thermal conductivity) to characterize the change in this state through a difference quantity. The current state level refers to the qualitative or semi-quantitative classification of the interface health level assessed based on real-time operating temperature and the difference quantity, using predetermined rules, empirical formulas, or machine learning models (such as classifiers). For example, it can be divided into several levels such as "healthy," "early degradation," "moderate damage," "severe damage," and "near failure."

[0060] Description of the implementation steps: Transmitting and receiving modulated eddy current signals: An eddy current detector equipped with dual-frequency excitation function is used. Under the initial health condition of the component (e.g., new from the factory or after overhaul), the probe is placed at a selected monitoring point on the non-working surface of the steel substrate. The instrument simultaneously or rapidly alternately transmits preset excitation signals of a first frequency (high frequency f_H) and a second frequency (low frequency f_L), and receives the corresponding response signals at the same location, storing them as a first reference signal S_H0 and a second reference signal S_L0, respectively. In subsequent service monitoring, this operation is repeated at the same location to obtain the response signals S_Ht and S_Lt at the current moment. Signal processing and phase change extraction: The detector's built-in signal processing unit performs phase comparisons between S_Ht and S_H0, and between S_Lt and S_L0, respectively. The phase shift ΔΦ_Ht (first phase change) of the first frequency response signal relative to its reference signal at the current moment, and the phase shift ΔΦ_Lt (second phase change) of the second frequency response signal relative to its reference signal, are calculated. Calculating the difference in the interface thermal state: The data processing system (which can be a computer built into the detector or an external computer) calculates the difference D_t according to a preset algorithm. A typical implementation is to calculate the difference: D_t = ΔΦ_Ht - k * ΔΦ_Lt, where k is a calibration coefficient used to balance the responses of the two frequency signals to common background factors, so that D_t mainly reflects the change in the interface state. An increase in the absolute value of D_t or a trend of increase usually indicates that the interface thermal conductivity is deteriorating (increased thermal resistance). Assessing the current state level in conjunction with temperature: Near the monitoring point, the real-time operating temperature T_t of the wear-resistant component is simultaneously acquired through methods such as embedded thermocouples or infrared thermometry. A state assessment model is established, with input (T_t, D_t) and output as the current state level L_t. The model can be an empirical formula based on a physical failure model (e.g., setting thresholds for D_t in different temperature ranges) or a classification model trained on historical data (such as support vector machines or decision trees). The model considers the inherent effects of temperature on the electromagnetic and thermophysical properties of materials, thus more accurately separating the D_t changes caused by damage. Predicting remaining safe service time: The above monitoring is performed continuously and periodically (e.g., daily, weekly) to obtain a series of time points t_i and their corresponding state levels L_i and differences D_i. The trend of D_i over time is analyzed (e.g., linear growth, exponential growth). Combining the current state level L_current (L_t from the latest monitoring) and the trend of D_i, extrapolation, a degradation rate-based lifetime model, or a machine learning time series prediction algorithm is used to deduce the time required for D_t to reach a preset "failure threshold" or for the state level to reach "near failure." This time is the predicted remaining safe service time.

[0061] The core technical principle of this implementation lies in dual-frequency eddy current differential detection and thermo-electric coupling indirect sensing. High-frequency eddy currents are sensitive to microscopic defects near the interface (such as microcracks and debonding) because these defects alter the local electromagnetic field distribution and eddy current paths, thus affecting the phase. Low-frequency eddy currents reflect the overall condition of the matrix more comprehensively. The difference in phase change between the two (D_t) is designed to be sensitive to the key failure precursor parameter of "interfacial thermal conductivity state." This is because mechanical damage to the interface (cracking, delamination) directly leads to changes in its thermal contact resistance, and the thermophysical properties (thermal conductivity, heat capacity) and electromagnetic properties (electrical conductivity, magnetic permeability) of materials are correlated under certain conditions (such as the Wiedemann-Franz law for metals), making it possible to indirectly assess the thermal state through precise electromagnetic measurements. Introducing real-time temperature compensation eliminates the direct interference of operating temperature fluctuations on electromagnetic measurements, allowing D_t to more purely characterize the thermal state changes caused by damage.

[0062] In a preferred embodiment, an excitation eddy current containing a first frequency and a second frequency is emitted into the steel substrate via an eddy current sensor array; the eddy current sensor array is arranged on the non-working surface of the steel substrate corresponding to the area covered by the ceramic wear-resistant layer.

[0063] Specifically, an eddy current sensor array refers to a detection device that integrates multiple eddy current sensor units according to certain rules (such as a matrix grid, a ring, or a linear arrangement along a specific path) onto a probe housing or flexible substrate. Each sensor unit typically includes an excitation coil and a receiving coil (or uses a self-comparison coil), and can work independently or collaboratively to transmit and receive eddy current signals. Its types can be rigid PCB arrays, flexible printed coil arrays, or arrays assembled from multiple independent miniature probes. A non-working surface of a steel substrate refers to the steel substrate surface on a wear-resistant part that is not covered with a ceramic layer and does not directly bear wear impacts; it is usually located on the back or side of the part. This is an ideal location for performing non-destructive testing without affecting the component's performance.

[0064] Description of Implementation Steps: Array Design and Arrangement: Based on the size and shape of the wear-resistant component to be monitored (such as a large wear-resistant liner), and the contour of the area covered by the ceramic wear-resistant layer, design an eddy current sensor array covering the area. The array shape can be rectangular, circular, or a matching irregular shape. The spacing between the sensor elements in the array is determined according to the required spatial resolution (e.g., a 20mm × 20mm grid). Installation and Fixing: During component installation or overhaul, permanently or semi-permanently install the eddy current sensor array (e.g., by bonding with high-temperature adhesive, fixing with mechanical clamps, or integrating it into a mounting base) on the non-working surface of the steel substrate, ensuring that its position corresponds to the working area of ​​the ceramic wear-resistant layer in spatial projection. The installation must ensure good contact between the sensor coil plane and the surface of the steel substrate to optimize coupling efficiency. Excitation and Scanning: During monitoring, the individual sensor elements in the array are excited sequentially or simultaneously (depending on hardware capabilities) using a multiplexer or dedicated array drive circuit, causing each element to emit excitation eddy currents containing a first frequency and a second frequency downwards (towards the ceramic layer) towards the steel substrate. Signal reception and correlation: Each sensor unit synchronously or sequentially receives modulated response signals from a local area directly below it. The response signals from all units are aggregated into a data acquisition system and stored in association with the unit's ID (corresponding to its spatial coordinates). In this way, a single monitoring operation can obtain two-dimensional spatial distribution data of the steel substrate-interface state beneath the entire ceramic wear-resistant layer coverage area.

[0065] In a preferred embodiment, predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the trend of the difference over time includes the following steps:

[0066] Based on the differences obtained at multiple monitoring times, their variation over time is analyzed to establish a degradation trend.

[0067] Based on the degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time.

[0068] Specifically, monitoring time: refers to the specific time point at which eddy current detection and data collection are performed according to a predetermined plan (such as periodic inspection) or triggered by an event (such as equipment maintenance window), denoted as t_1, t_2, …, t_n. Degradation trend: refers to the mathematical model or qualitative judgment extracted from historical variance data using mathematical modeling methods (such as curve fitting, time series analysis, machine learning regression) that describes the change in interface state (using variance as a proxy indicator) over time. For example, identifying linear growth, exponential acceleration, or fluctuation around a certain threshold in variance. Preset critical condition: refers to the technical indicator threshold set based on engineering experience, laboratory testing, or theoretical calculations, indicating that the ceramic-steel interface is about to fail or can no longer guarantee safe service. It can be a direct absolute or relative growth value of variance D_t, or a higher-level state level associated with variance (such as the lower limit of the "severe damage" level). Remaining safe service time: refers to the predicted time from the current monitoring time until the interface state reaches the "preset critical condition". This is a dynamically updated prediction value.

[0069] Description of Implementation Steps: Constructing a Time Series Dataset: During the service life of wear-resistant components, detection is performed at multiple discrete monitoring times t_i (i=1, 2, ..., n), recording the difference data D_i corresponding to each time t_i. For array detection, D_i can be an average value, a value of a specific key area, or a comprehensive indicator of the entire distribution. Analyzing Change Patterns and Establishing Degradation Trends: The collected data pairs (t_i, D_i) are plotted on a time-difference coordinate system. Establishing a Trend Model Using Data Analysis Methods. Simple Model: If the data shows a clear trend, the least squares method can be used for linear (D = a*t + b), exponential (D = c*exp(d*t)), or polynomial fitting. The fitted curve equation represents the degradation trend. Advanced Model: Time series prediction algorithms, such as autoregressive integral moving average models and long short-term memory networks, can be used to train and model the {D_i} sequence. This model can capture more complex nonlinear change patterns and serve as a mathematical expression of the degradation trend. Extrapolating remaining time based on the current state: Determining the critical value D_crit: Based on preset critical conditions, determine the corresponding critical value D_crit for the difference (or the boundary of the D value range corresponding to the state level). Extrapolation calculation: Substitute the current time t_now and the current difference D_now into the established degradation trend model. Using this model, calculate or predict the time t_crit required when the difference D reaches D_crit. Calculating the remaining time: Remaining safe service time RUL = t_crit - t_now. Considering the state level: If the current state level L_current is already high (e.g., "moderate damage"), a safety factor may be introduced during model extrapolation, or a more conservative failure threshold may be used for calculation to provide a safer early warning.

[0070] The core principle of this implementation method is the combination of lifespan prediction based on data extrapolation and a physical model of failure. It considers that the degradation process of components (manifested in this case as the deterioration of interfacial thermal conductivity) has inherent regularities that can be revealed through historical data. By capturing this regularity through mathematical modeling (establishing a degradation trend) and combining it with a clear failure criterion (preset critical conditions), the current health status indicators can be extrapolated forward on the timeline until they intersect with the failure criterion. The time corresponding to the intersection point is the predicted failure time. The current state level, as prior knowledge, is used to calibrate or constrain this extrapolation process, improving the rationality of the prediction.

[0071] In a preferred embodiment, the step of analyzing the variation of the difference obtained at multiple monitoring times over time to establish a degradation trend includes the following steps:

[0072] Data from each sensor unit in the eddy current sensor array at multiple monitoring times are collected to obtain the time series of differences for each sensor unit;

[0073] The time series of differences of all sensor units at the same monitoring time are mapped according to their spatial location to generate multiple sets of two-dimensional damage distribution snapshots describing the degree of interface damage in the entire monitoring area.

[0074] Comparative analysis was performed on all the two-dimensional damage distribution snapshots arranged in chronological order to extract temporal and spatial features;

[0075] Based on the time dimension features and the spatial dimension features, a degradation trend is established.

[0076] Specifically, the sensor unit difference time series refers to the sequence of difference values ​​D_{ij}(t_k) obtained at each monitoring time t_k (k=1, 2, ..., m) starting from a fixed spatial location (i, j). It describes the evolution of the interface state at that specific spatial point over time. The two-dimensional damage distribution snapshot refers to a static image reflecting the spatial distribution of interface damage (indicating difference) in the entire monitoring area at the same monitoring time. This image is formed by arranging and visualizing the difference values ​​D_{ij}(t) of all sensor units (assuming M rows and N columns) in the array according to their actual physical location (two-dimensional coordinates i, j) in the array (e.g., drawing a heat map, contour map, or grid table). The time dimension features refer to the statistical or dynamic features extracted from the difference data of the same spatial region (which can be global or local) over time. The spatial dimension features refer to the features extracted from the two-dimensional damage distribution snapshot at the same time point that describe the spatial morphology and distribution of damage.

[0077] The implementation steps describe how to generate a unit time series: For each sensor unit (coordinate (i, j)) in the eddy current sensor array, extract the difference data D_{ij}(t_1), D_{ij}(t_2), ..., D_{ij}(t_m) measured at all historical monitoring times t_1, t_2, ..., t_m, forming an independent difference time series for that unit. Generate damage distribution snapshots at each time point: For each monitoring time t_k, extract the difference values ​​{D_{ij}(t_k)} of all M×N sensor units at that moment, and fill them into an M-row N-column matrix Snapshot(t_k) according to their corresponding row index i and column index j. This matrix (or its visualization) is the two-dimensional damage distribution snapshot at time t_k. Repeat this operation for all monitoring times to obtain a set of snapshots {Snapshot(t_1), Snapshot(t_2), ..., Snapshot(t_m)} arranged in time sequence.

[0078] Comparative analysis of snapshots extracts spatiotemporal features: Extracting temporal features: The average value of all differences in each snapshot of the entire monitoring area at each moment can be calculated as Mean_D(t_k), resulting in a sequence {Mean_D(t_1), ..., Mean_D(t_m)}. This average sequence is analyzed to extract features, such as calculating its first-order difference (approximate instantaneous slope) and second-order difference (approximate instantaneous acceleration), and observing its variation patterns. Extracting spatial features: Image processing or spatial data analysis is performed on each snapshot (t_k). For example, a difference threshold is set to identify all connected pixel regions in the snapshot that exceed this threshold (i.e., "high-damage patches"). The area, centroid coordinates, equivalent diameter, and shape parameters of each patch are calculated. The evolution of these patches in different snapshots at different times (e.g., appearance, merging, expansion, movement) is tracked. Degradation trends are established based on spatiotemporal features: Extracted temporal features (such as the accelerated growth of average damage) and spatial features (such as the exponential expansion of the area of ​​the largest damage patch and the movement of patches in a certain direction) are combined to construct a more comprehensive degradation model. For example, the model can be described as: "Interfacial damage initially undergoes uniform micro-degradation, gradually forming macroscopic damage patches in stress concentration areas (manifested as specific locations in space). The area of ​​these patches expands at a nonlinear rate, accompanied by an accelerated increase in the global average damage level." This comprehensive description, combining the temporal evolution rate and spatial evolution pattern, provides a richer and more robust "degradation trend" than simply looking at a numerical sequence.

[0079] The principle behind this implementation is damage evolution analysis that integrates spatiotemporal information. It recognizes that the failure of the ceramic-steel interface is not a uniform, synchronous process; it often begins at local weak points (stress concentrations, manufacturing defects), and then the damaged area expands in both time and space. This method elevates one-dimensional time series analysis to the level of "sequences of time series" (i.e., time series analysis of the spatial field). By generating and analyzing continuous two-dimensional snapshots, it can simultaneously capture the temporal dynamics (how fast) and spatial patterns (where it begins and how it expands), which better reflects the actual physical failure process.

[0080] In a preferred embodiment, the extraction of temporal and spatial features includes the following steps:

[0081] Calculate the curve of the average damage index over time for the entire monitoring area or a specific area of ​​interest, and extract the instantaneous slope and acceleration of the curve as time dimension features.

[0082] Identify and track the centroid location, area, and morphological evolution of high-damage regions appearing in the two-dimensional damage distribution snapshot, and use them as spatial dimension features.

[0083] Specifically, the mean damage index (MDI) is a scalar value obtained by mathematically averaging or weighted averaging the differences in the values ​​of all sensor units within the entire monitoring area (or a designated area of ​​interest) at a given moment. This index comprehensively reflects the overall average level of interface damage in the area and is a key macroscopic indicator in the time dimension. The weighted average can be adjusted according to the importance of the location (e.g., stress concentration areas have higher weights). The instantaneous slope is the rate of change of the MDI time series Mean_D(t) near a given monitoring moment t_k. It can be approximated by calculating the first difference of the series (i.e., (Mean_D(t_k) - Mean_D(t_{k-1})) / (t_k - t_{k-1})), with units of difference per unit of time. It indicates how quickly the damage deteriorates at the "current moment." The instantaneous acceleration is the rate of change of the instantaneous slope over time, reflecting whether the degradation process is uniform, accelerating, or decelerating. It can be approximated by calculating the second difference of the MDI time series, with units of difference per unit of time. 2Positive values ​​indicate accelerating degradation, while negative values ​​indicate decelerating degradation. High-damage regions: These refer to connected pixel regions in a two-dimensional damage distribution snapshot where the difference exceeds a pre-defined empirical or statistical threshold (e.g., more than twice the standard deviation of the global average). These regions are considered hotspots where interface damage is relatively significant or developing rapidly. Centroid location, area, and morphological evolution: Centroid location: Refers to the geometric center coordinates of the high-damage region (e.g., (x_c, y_c)), used to track whether the damage core has moved in space. Area: Refers to the number of pixels contained in the high-damage region or its converted actual physical area, used to quantify the scale of the damage. Morphological evolution: Refers to the dynamic changes in the geometric features of the high-damage region, such as its shape (e.g., circle, elongated shape, irregular shape), boundary complexity (e.g., fractal dimension), and orientation, in continuous time snapshots.

[0084] Implementation steps: Calculate and extract time-dimensional features: Generate the average damage curve: For a two-dimensional damage distribution snapshot {Snapshot(t_1), ..., Snapshot(t_m)} arranged in chronological order, calculate the average difference Mean_D(t_k) of all pixels (corresponding to all sensor units) in each snapshot Snapshot(t_k), obtaining the time series {Mean_D(t_1), ..., Mean_D(t_m)}. Plot this as an "average damage exponent-time" curve. Calculate the instantaneous slope: Perform numerical differentiation on the above average damage time series (such as first-order central difference, three-point difference, or differentiation after Savitzky-Golay filtering smoothing), and calculate the instantaneous slope S(t_k) at each monitoring time t_k. This slope sequence {S(t_1), ..., S(t_m)} describes the dynamic change of the overall degradation rate. Calculate instantaneous acceleration: The instantaneous slope sequence {S(t_k)} is numerically differentiated again to obtain the instantaneous acceleration A(t_k) at each monitoring time t_k. The acceleration sequence {A(t_1), ..., A(t_m)} reveals whether the degradation process is accelerating or slowing down. S(t_k) and A(t_k) are output as the core temporal dimension features. Identify and extract spatial dimension features: Identify high-damage regions: For each snapshot (Snapshot(t_k), image segmentation techniques are applied. First, a difference threshold Thresh is determined (this can be a fixed value or an adaptive value based on the snapshot's own statistical characteristics, such as the mean + 1.5 × standard deviation). All pixels in the snapshot where D_{ij} > Thresh are marked as "high-damage" pixels. Region connectivity and attribute calculation: Using a connected component analysis algorithm (such as an 8-neighborhood-based labeling algorithm), adjacent "high-damage" pixels are aggregated into independent "high-damage regions". For each identified region, calculate its: Centroid position: (x_c, y_c), representing the approximate center position of the region on the component surface. Area: the total number of pixels within the region. Morphological parameters: such as aspect ratio, roundness, eccentricity, etc., describing its shape. Evolution tracking: Match the high-damage regions identified at the current time t_k with the regions at the previous time t_{k-1} (based on the similarity of centroid position and area). Track the evolution trajectory of each region: the birth of new regions, the disappearance of old regions, the merging or splitting of regions, the increase or decrease of area, the movement of the centroid, and the change of shape (e.g., stretching from a circle to an ellipse, which may represent crack propagation along a certain direction). Output the centroid trajectory {(x_c(t), y_c(t))}, the area change curve {Area(t)}, and the morphological evolution description of each region as the core spatial dimension features.

[0085] The principle behind this implementation lies in combining macroscopic statistical trends with microscopic spatial evolution through feature engineering. Temporal features (slope, acceleration) provide global dynamic information about the degradation process, answering the crucial question of "how fast is the overall deterioration, and is it accelerating?" Spatial features (centroid, area, and morphological evolution of high-damage regions) provide localized and patterned information about the degradation process, answering the question of "where does the damage begin, and how does it grow and spread?" This separation and quantification allows subsequent modeling to utilize both overall trends and local details, more closely resembling the real-world picture of damage development (typically starting locally and then causing failure by affecting overall performance).

[0086] In a preferred embodiment, establishing the degradation trend based on the time dimension features and the spatial dimension features includes the following steps:

[0087] Based on the instantaneous slope and acceleration in the time dimension features and the morphological evolution of the high-damage region in the spatial dimension features, the first candidate degradation path is deduced based on the fracture mechanics principle of interface damage.

[0088] The change curves of the time dimension features and the evolution sequences of the spatial dimension features are input into the temporal prediction network to obtain the second candidate degradation path.

[0089] The degree of agreement between the first candidate degradation path and historical long-term data, and the degree of agreement between the second candidate degradation path and recent data are evaluated, and the two are weighted and fused to generate a degradation trend.

[0090] Specifically, fracture mechanics refers to the branch of solid mechanics that studies the crack propagation laws, fracture criteria, and life prediction of cracked bodies under external loads. In this context, it refers to analogizing damage (such as delamination and cracking) at the ceramic-steel interface to cracks within the material, and using classical or modified fracture mechanics models (such as the Paris equation to describe fatigue crack propagation: da / dN = C*(ΔK)) to describe the crack propagation. mThe evolution law is deduced using the following method: 'a' is the crack size (corresponding to the size of the damaged area), 'N' is the number of cycles (corresponding to time), and 'ΔK' is the stress intensity factor amplitude. First candidate degradation path: refers to the possible trajectory of future interface damage development, predicted through theoretical derivation or physical model calculation, based on physical principles (fracture mechanics in this case) and extracted spatiotemporal features. It represents prediction based on "first principles." Temporal prediction network: refers to a class of deep learning models specifically designed for processing time series data, such as recurrent neural networks, long short-term memory networks, and Transformer time series models. These networks can automatically learn complex temporal dependencies and dynamic patterns from historical data. Second candidate degradation path: refers to the future damage development trajectory directly predicted by the network through the data patterns learned internally, after inputting historical spatiotemporal feature data into the temporal prediction network. It represents prediction based on "data-driven" methods. Weighted fusion: refers to combining two or more prediction paths (here, the first and second candidate paths) according to certain weights to form a final, more reliable prediction path (i.e., degradation trend). The weights can be dynamically allocated based on factors such as the historical prediction accuracy of each path and its consistency with recent data.

[0091] Description of Implementation Steps: Deducing the First Candidate Degradation Path: Model Selection and Parameterization: Based on the service load characteristics of the wear-resistant component (e.g., cyclic impact, thermal cycling), a suitable fracture mechanics model is selected. For example, for components primarily subjected to cyclic mechanical stress, a fatigue crack propagation model can be chosen. Model parameters (e.g., material constants C, m) can be obtained through laboratory standard specimen testing or used as initial empirical values. Feature Input and Deduction: Extracted spatial dimension features are used as key inputs. For example, the equivalent radius 'a' of the current main high-damage region is considered as the crack size, and A(t_k) (instantaneous acceleration) or load spectrum information is correlated with ΔK. Time dimension features (e.g., S(t_k)) are used as a verification or calibration reference for the model output (da / dt). Based on the current state and the model, numerical integration is used to deduce the future change of 'a' (damage size) over time, and the corresponding global average damage index is calculated (this can be achieved by establishing an empirical relationship between 'a' and Mean_D). This derived "time-average damage" curve (possibly with spatial morphology description) is the first candidate degradation path. It is based on physical laws, and long-term trends may be more consistent with reality. Generating a second candidate degradation path: Data preparation: Standardize the spatiotemporal feature sequence arranged in time series and divide it into training and validation sets. Input features can include Mean_D, S, A at multiple historical moments, as well as Area, centroid coordinates, etc., of major high-damage regions. Output labels can be the predicted Mean_D values ​​for several future time steps. Network training and prediction: Construct an LSTM or Transformer time-series prediction network. Train the network with historical data to learn to predict the future average damage index from past feature sequences (spatial features can also be predicted simultaneously). After training, input the latest feature sequence into the network to obtain the predicted Mean_D sequence for a future period. This prediction curve obtained from pure data learning is the second candidate degradation path. It may be more sensitive to capturing patterns of recent data changes. Evaluation and weighted fusion: Consistency evaluation: Evaluate the consistency between the first candidate path and historical long-term data (e.g., the Mean_D curve for the entire past monitoring period). Calculate its root mean square error or correlation coefficient. A high degree of agreement indicates that the physical model has a good grasp of historical patterns. Evaluate the agreement between the second candidate path and recent data (e.g., Mean_D values ​​of the most recent 3-5 monitoring points). Calculate its prediction error within the recent time window. A high degree of agreement indicates that the data-driven model accurately captures the latest trends. Dynamic weighting: Based on the above two agreement evaluation results, dynamically allocate weights w_phy and w_data (w_phy + w_data = 1). For example, if the physical model has a very high long-term agreement, then w_phy is higher; if the data model's recent predictions are extremely accurate, then w_data is higher. Expert experience or more complex optimization algorithms can also be introduced to determine the weights.Generating the final degradation trend: The values ​​of the two candidate paths at each future prediction time point are linearly weighted and averaged according to the assigned weights: Final_Trend(t) = w_phy * Path_Phy(t) + w_data * Path_Data(t). This fused curve is the final degradation trend that integrates physical laws and data intelligence.

[0092] The principle behind this implementation is the heterogeneous fusion of a physical model and a data-driven model. The first path (physical model) is based on a deep understanding of failure mechanisms, exhibiting better interpretability and physical consistency during extrapolation, and may be more stable, especially in long-term predictions with sparse data or data exceeding historical ranges. The second path (data-driven model) learns patterns entirely from data, capturing complex nonlinear interactions and unknown factors that the physical model might overlook, and may be more sensitive to short-term fluctuations and abrupt changes. The weighted fusion strategy aims to leverage the strengths of both approaches, dynamically balancing their contributions through an evaluation mechanism to generate a prediction trajectory that conforms to physical laws and closely reflects actual observations.

[0093] This implementation significantly improves the accuracy and reliability of predictions by overcoming the biases that may exist in a single model. When the physical model parameters are inaccurate, the data model can correct them; when abnormal noise or abrupt pattern changes occur in the data, the physical model can provide stability constraints. The fused predictions are generally more robust and accurate than any single model. Moreover, it adapts to different degradation stages: in the early stages of degradation, when data patterns may not be obvious, the weight of the physical model can be higher; in the accelerated degradation phase, the data-driven model has a stronger ability to capture rapid changes, and its weight can be adaptively increased. This dynamic adjustment allows the prediction system to flexibly respond to the characteristics of different stages.

[0094] In a preferred embodiment, after generating the degradation trend, the following step is further included:

[0095] Retrieve the historical failure case database. Each historical failure case in the database includes a sequence of historical monitoring data for the entire process of similar wear-resistant components from initial damage to final failure, as well as the corresponding actual failure time and failure mode.

[0096] The change curve of the time dimension characteristics and the evolution sequence of the spatial dimension characteristics of the current wear-resistant component up to the latest monitoring time are matched with the data of the same period of each historical failure case in the historical failure case database to select at least one reference historical case.

[0097] Extract the actual degradation path of the reference historical case in the subsequent development process as a benchmark reference degradation path;

[0098] The baseline degradation path is compared with the degradation trend, and the deviation between the two at key nodes is calculated.

[0099] The step of estimating the time required for the ceramic-steel interface to reach a preset critical condition based on the degradation trend and the current state level, and using this time as the remaining safe service time, includes the following steps:

[0100] If the deviation is less than or equal to the first preset threshold, then based on the degradation trend and the current state level, the time required for the ceramic-steel interface state to reach the preset critical condition is estimated and used as the remaining safe service time.

[0101] Furthermore, after comparing the baseline degradation path with the degradation trend and calculating the deviation between them at key nodes, the process further includes the following steps:

[0102] If the deviation is greater than the first preset threshold, the evolution rate and inflection point of the degradation trend are dynamically corrected according to the deviation to obtain the corrected degradation trend.

[0103] Based on the modified degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time.

[0104] Specifically, the historical failure case database refers to a structured repository containing complete records of failures of similar wear-resistant components (with identical or similar designs, materials, processes, and operating conditions) during past service. Each record is a case, containing monitoring data for the entire process from commissioning to final failure (i.e., the spatiotemporal characteristic sequence collected according to this method), the actual failure time, and the failure mode determined through post-failure inspection and analysis (such as "large-area peeling of the ceramic layer," "interfacial thermal fatigue cracking," "edge erosion failure," etc.). Contemporaneous data refers to historical data segments from historical cases that correspond to the lifespan percentage (or absolute time calculated from the start of monitoring) of a certain component and the lifespan stage of the currently monitored component "up to the latest monitoring time." This is used for horizontal comparison within the "same stage." Reference historical cases refer to one or more historical failure cases selected from the historical database through similarity matching, whose historical monitoring data is most similar to the current component's monitoring data in terms of development trends and characteristic patterns. Baseline reference degradation path: This refers to the actual damage development trajectory extracted from historical reference cases, starting from a state point contemporaneous with the current component and continuing until the final actual failure (usually expressed as the average damage index changing over time). This is a real degradation path verified by actual outcomes. Critical nodes: These are moments or state points of significant importance on the degradation path, such as the moment when damage first accelerates, the moment when damage reaches a certain severity threshold, or the moment when the spatial damage pattern undergoes a sudden change. These nodes serve as anchor points for assessing prediction bias. Dynamic correction: This refers to adjusting the parameters, weights, or evolution function of the prediction model (such as a fusion model) based on the deviation between the current prediction (degradation trend) and the historical actual path (baseline reference path) at critical nodes, so that the predicted future path can better align with similar historical experiences.

[0105] The implementation steps describe the process of retrieving historical failure cases from a pre-built cloud or local historical failure case database. This retrieves all historical case records that are the same as or highly similar to the currently monitored wear-resistant component. Similarity matching is used to filter reference cases: Data alignment: The time-dimensional feature curves (Mean_D(t), S(t), A(t)) and spatial-dimensional feature evolution sequences (changes in high-damage areas) of the current component up to the latest time t_now are time-aligned with the historical data of each historical case in the database within the same runtime or lifespan percentage range. Similarity calculation: Using dynamic time warping, Euclidean distance, cosine similarity, or more complex sequence similarity measurement algorithms, the similarity score between the current component data and the data of each historical case from the same period is calculated. The similarity of the time curve shape and the similarity of the spatial evolution pattern can be combined (e.g., whether the damage all expands from the corners). Filtering: Based on the similarity scores, one or more historical cases (e.g., the top 3) with the highest scores are selected as reference historical cases. Extracting the Benchmark Degradation Path: From each selected historical reference case, extract the complete sequence of real changes in the average damage index (or main spatial characteristics) from its "time point contemporaneous with the current component t_now" to its actual failure time. This real sequence is a benchmark degradation path. Comparing and Calculating Deviation: Identifying Key Nodes: On the benchmark degradation path, manually or automatically identify several key nodes, such as T1 (damage acceleration point) and T2 (reaching the "severe damage" level). Extrapolating the Current Trend: Extrapolate the degradation trend from t_now to a sufficiently long time range. Calculating Node Deviation: For each key node, compare the actual time T_ref required for the benchmark path to reach that node's state with the time T_pred predicted by the current degradation trend to reach the same state. Calculate the absolute time deviation |T_ref - T_pred| or the relative deviation |T_ref - T_pred| / T_ref. Bias-based conditional judgment and correction: Case 1: Small deviation (≤ first preset threshold): This indicates that the current degradation trend of the component is highly consistent with the actual development path of the most similar historical cases, and the prediction reliability is high. In this case, the originally generated degradation trend and the remaining safe service time calculated based on it can be directly used without significant correction. The historical case library plays a role in enhancing confidence.

[0106] Scenario 2: Large deviation (> first preset threshold): This indicates that although historical data from the same period are similar, the current predicted trend deviates significantly from the historical actual path. Dynamic correction is required in this case.

[0107] Adjust the evolution rate: If the current predicted T_pred is generally earlier than T_ref (the prediction is too pessimistic), then the slope of the degradation trend (evolution rate) is reduced proportionally; otherwise, it is increased.

[0108] Correcting inflection points: If the order or form of key nodes is inconsistent, adjust the function or threshold related to inflection points in the prediction model.

[0109] Furthermore, the evolution rate is corrected by analyzing deviation patterns. For example, if the baseline path shows that damage accelerates faster than currently predicted after critical node T1, the slope (evolution rate) of the current degradation trend after T1 is increased proportionally (e.g., by linear or nonlinear interpolation based on the magnitude of the deviation). Conversely, the rate is decreased. The inflection point is corrected by analyzing the temporal sequence of critical nodes. If the T2 node in the baseline path occurs significantly later than predicted, the threshold conditions or logic triggering the "T2 state" in the prediction model are appropriately adjusted to shift the predicted inflection point later, closer to historical experience. Corrections may involve adjusting the parameters of the physical model (e.g., the C and m values ​​in the crack propagation formula), adjusting the internal weights of the data-driven model, or adjusting the weights w_phy and w_data of the fusion of both. Through these corrections, a new post-correction degradation trend is obtained that is more consistent with the historical reference case path.

[0110] Re-fusion: The deviation can also be used as a feedback signal to readjust the fusion weights w_phy and w_data of the physical model and the data-driven model.

[0111] After the above corrections, the final degradation trend calibrated by the case library is obtained.

[0112] Output: Regardless of whether it is corrected, the final output is a clear prediction of the remaining safe service time, and you can add a note on whether it has been calibrated with historical cases and the confidence level.

[0113] In a preferred embodiment, after assessing the current state level of the ceramic-steel interface, the method further includes the following steps:

[0114] The real-time temperature distribution field of the non-working surface of the steel substrate is obtained; the real-time temperature distribution field is obtained by a temperature sensor array arranged at the corresponding position of the eddy current sensor array;

[0115] At each monitoring moment, the difference quantity obtained by the eddy current sensor array, which reflects the change in the interface heat conduction state, is mapped into an eddy current anomaly distribution field according to its spatial location.

[0116] The real-time temperature distribution field and the eddy current anomaly distribution field at the same moment are spatially superimposed and compared to obtain the confidence level of the current state level.

[0117] The step of predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the trend of the difference over time includes the following steps:

[0118] If the confidence level is high confidence, then the remaining safe service time of the ceramic wear-resistant layer is predicted based on the current state level and the trend of the difference over time.

[0119] In a preferred embodiment, obtaining the confidence level of the current state level includes the following steps:

[0120] When the comparative analysis results show that the spatial overlap area between one or more local high-temperature regions identified in the real-time temperature distribution field and one or more high-anomaly regions identified in the eddy current anomaly distribution field exceeds a preset area threshold, the confidence level of the current state level is determined to be high confidence.

[0121] Specifically, dual-field data is acquired synchronously: While installing the eddy current sensor array, a temperature sensor array is simultaneously placed at its corresponding location (e.g., next to each eddy current unit, or integrated within the same probe). This ensures that the spatial coverage and resolution of the two arrays are essentially consistent. At each monitoring time t_k, the data acquisition system is synchronously triggered. The eddy current detection unit transmits dual-frequency excitation and receives signals, which are then processed to obtain the difference D_{ij}(t_k) of each sensor unit, thereby generating the eddy current anomaly distribution field EC_Field(t_k). The temperature acquisition unit synchronously reads the readings T_{ij}(t_k) of all temperature sensors, generating the real-time temperature distribution field Temp_Field(t_k). Spatial overlay and comparison analysis: Data registration: Since the installation positions of the two arrays are fixed and known, the data grids of EC_Field(t_k) and Temp_Field(t_k) are first spatially aligned (registered) to ensure that the coordinates (i, j) point to the same physical location of the component in both fields. Anomaly Region Identification: In EC_Field(t_k), one or more "high outlier regions" (i.e., connected regions with significantly higher differences than the background value) are identified based on a preset threshold or adaptive segmentation algorithm. In Temp_Field(t_k), one or more "local high-temperature regions" or "local low-temperature regions" are identified based on a threshold (e.g., higher than a certain value above the average temperature of the region) or relative hot-cold comparison (Note: Interface damage leads to increased thermal resistance, which may cause heat to accumulate on the ceramic layer side, thus relatively lowering the temperature of the corresponding area on the back of the steel substrate; however, under certain working conditions, local high temperatures may also occur due to frictional heating, etc. Specific characteristics need to be determined based on the actual physical model. For the sake of generality, they can be collectively referred to as "temperature anomaly regions"). Spatial Correlation Analysis: The core is to analyze the degree of overlap between the two types of anomaly regions in spatial location. For example, calculate the spatial intersection area between each "high eddy current anomaly region" and all "temperature anomaly regions". Or, calculate the spatial correlation coefficient between two distribution field matrices. Obtain the confidence level of the current state level: Rule Determination: Establish determination rules based on spatial overlap. For example: High confidence: If the identified major "high eddy current anomaly region" and "temperature anomaly region" highly overlap in spatial location and outline (e.g., centroid distance less than a certain threshold, and overlap area greater than 70%), then the current "state level" assessed based on eddy current data is determined to have high confidence. This strongly indicates that the electromagnetic anomaly detected by eddy current is indeed caused by actual physical changes in the interface thermal conduction state (leading to temperature field anomalies), rather than electromagnetic interference or other false signals. Medium / low confidence: If the two only partially overlap or their locations are significantly different, the confidence is reduced. Possible reasons include: the temperature field being interfered with by other heat sources, noise in eddy current detection, or damage not yet significantly affecting macroscopic thermal conduction.At this point, the system will label the status level as "Medium" or "Low" confidence and indicate that further checks or continued observation are needed. Confidence-based predictive decision-making: The subsequent "predicting the remaining safe service time of the ceramic wear-resistant layer" process will only be automatically triggered and executed when the system determines the confidence level to be high. If the confidence level is medium or low, the system can issue a "review required" alarm, suspend automatic life prediction, and suggest manual review, increasing monitoring frequency, or checking sensor status, thereby avoiding erroneous predictions and decisions based on unreliable data.

[0122] Furthermore, anomaly regions are identified from the dual fields: For the real-time temperature distribution field Temp_Field: First, necessary preprocessing (such as filtering and denoising) is performed. Then, the statistical properties of the entire field (such as mean μ_T, standard deviation σ_T) are calculated. A temperature anomaly threshold is set, for example, Th_T = μ_T ± n * σ_T (where n is a coefficient, such as 2 or 3; "+" indicates high temperature, "-" indicates low temperature, or both are considered). Connectivity analysis is performed on the binarized image to mark all local high-temperature (or low-temperature) regions {R_T1, R_T2, ...} that meet the conditions, and the pixel set of each region is recorded. For the eddy current anomaly distribution field EC_Field: Similarly, its statistical properties (mean μ_E, standard deviation σ_E) are calculated. An eddy current anomaly threshold is set, for example, Th_E = μ_E + m * σ_E (usually only positive anomalies are considered). Connectivity analysis is performed to identify all high outlier regions {R_E1, R_E2, ...}, and the pixel set for each region is recorded. Spatial overlap area is calculated: for each possible pair of regions (R_Ti, R_Ej), the area of ​​their spatial intersection (overlapping portion), Overlap_Area(i, j), is calculated. This can be achieved by calculating the number of intersection elements of the two region pixel sets and multiplying it by the actual area represented by a single pixel. Simultaneously, a relative overlap ratio can be calculated, for example, Overlap_Ratio = Overlap_Area / min(Area(R_Ti), Area(R_Ej)), for a fairer evaluation. Confidence level is determined based on overlap area thresholds: all identified temperature and eddy current anomaly regions are traversed. A preset area threshold is applied for determination. For example, a rule is set: a region (R_Ti, R_Ej) has an absolute overlap area Overlap_Area(i, j) > S_th (a preset absolute area threshold, such as 50 mm). 2And / or the relative overlap ratio Overlap_Ratio > R_th (a preset ratio threshold, such as 40%). Judgment logic: If the above conditions are met, the system determines that at the current monitoring time, the temperature field evidence and the eddy current field evidence are spatially validly corroborated, therefore the current state level assessed by the eddy current data has high confidence. Otherwise: If no pair of overlapping regions meets the threshold condition, the confidence level is determined to be insufficient (medium or low). The system may further analyze whether the temperature anomaly is not obvious or whether the eddy current anomaly location is deviated.

[0123] The principle behind this implementation is multi-physics information fusion and cross-validation. It is based on the same fundamental physical fact: damage to the ceramic-steel interface (such as cracking or delamination) leads to a significant increase in thermal resistance, thereby simultaneously altering two observable physical fields: the electromagnetic properties near the interface (affecting eddy currents); the heat flow path across the interface; and the temperature distribution on the back of the steel substrate. Therefore, a real interface damage should theoretically generate corresponding spatial anomaly signals in both the "eddy current anomaly distribution field" and the "temperature distribution field." By comparing whether these two are spatially consistent, the "real damage signal" can be effectively distinguished from "interference or noise in a single field measurement." This is a data fusion strategy based on physical mechanisms, significantly improving the specificity and reliability of state perception.

[0124] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for predicting the wear state of a steel-based ceramic wear-resistant component, the wear-resistant component comprising a steel substrate and a ceramic wear-resistant layer bonded to its working surface, characterized in that... Includes the following steps: An excitation eddy current containing a first frequency and a second frequency is emitted into the steel substrate, and a response signal modulated by the state of the interface between the steel substrate and the ceramic-steel bonding is received; wherein the first frequency is higher than the second frequency; Process the response signals corresponding to the first frequency and the second frequency respectively, and extract the first phase change amount of the first frequency relative to the first reference signal corresponding to the initial health state of the component, and the second phase change amount of the second frequency relative to the second reference signal corresponding to the initial health state of the component; Calculate the difference between the first phase change and the second phase change, wherein the difference characterizes the change in the thermal conductivity state of the ceramic-steel interface; The real-time operating temperature of the wear-resistant component is obtained, and the current state level of the ceramic-steel interface is evaluated based on the difference between the real-time operating temperature and the difference. Based on the current state level and the trend of the difference over time, predict the remaining safe service time of the ceramic wear-resistant layer; Specifically, an excitation eddy current containing a first frequency and a second frequency is emitted into the steel substrate via an eddy current sensor array; the eddy current sensor array is arranged on the non-working surface of the steel substrate corresponding to the area covered by the ceramic wear-resistant layer. The step of predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the trend of the difference over time includes the following steps: Based on the differences obtained at multiple monitoring times, their variation over time is analyzed to establish a degradation trend. Based on the degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time. The process of analyzing the variation patterns of the differences obtained at multiple monitoring times to establish a degradation trend includes the following steps: Data from each sensor unit in the eddy current sensor array at multiple monitoring times are collected to obtain the time series of differences for each sensor unit; The time series of differences of all sensor units at the same monitoring time are mapped according to their spatial location to generate multiple sets of two-dimensional damage distribution snapshots describing the degree of interface damage in the entire monitoring area. Comparative analysis was performed on all the two-dimensional damage distribution snapshots arranged in chronological order to extract temporal and spatial features; Based on the aforementioned time-dimensional and spatial-dimensional features, a degradation trend is established; The extraction of temporal and spatial features includes the following steps: Calculate the curve of the average damage index over time for the entire monitoring area or a specific area of ​​interest, and extract the instantaneous slope and acceleration of the curve as time dimension features. Identify and track the centroid location, area, and morphological evolution of high-damage regions appearing in the two-dimensional damage distribution snapshot, and use them as spatial dimension features; The process of establishing a degradation trend based on the time dimension features and the spatial dimension features includes the following steps: Based on the instantaneous slope and acceleration in the time dimension features and the morphological evolution of the high-damage region in the spatial dimension features, the first candidate degradation path is deduced based on the fracture mechanics principle of interface damage. The change curves of the time dimension features and the evolution sequences of the spatial dimension features are input into the temporal prediction network to obtain the second candidate degradation path. The degree of agreement between the first candidate degradation path and historical long-term data, and the degree of agreement between the second candidate degradation path and recent data are evaluated, and the two are weighted and fused to generate a degradation trend.

2. The method for predicting the wear state of steel-based ceramic wear-resistant components according to claim 1, characterized in that: After generating the degradation trend, the following steps are also included: Retrieve the historical failure case database. Each historical failure case in the database includes a sequence of historical monitoring data for the entire process of similar wear-resistant components from initial damage to final failure, as well as the corresponding actual failure time and failure mode. The change curve of the time dimension characteristics and the evolution sequence of the spatial dimension characteristics of the current wear-resistant component up to the latest monitoring time are matched with the data of the same period of each historical failure case in the historical failure case database to select at least one reference historical case. Extract the actual degradation path of the reference historical case in the subsequent development process as a benchmark reference degradation path; The baseline degradation path is compared with the degradation trend, and the deviation between the two at key nodes is calculated. The step of estimating the time required for the ceramic-steel interface to reach a preset critical condition based on the degradation trend and the current state level, and using this time as the remaining safe service time, includes the following steps: If the deviation is less than or equal to the first preset threshold, then based on the degradation trend and the current state level, the time required for the ceramic-steel interface state to reach the preset critical condition is estimated and used as the remaining safe service time.

3. The method for predicting the wear state of steel-based ceramic wear-resistant components according to claim 2, characterized in that: After comparing the baseline degradation path with the degradation trend and calculating the deviation between them at key nodes, the method further includes the following steps: If the deviation is greater than the first preset threshold, the evolution rate and inflection point of the degradation trend are dynamically corrected according to the deviation to obtain the corrected degradation trend. Based on the modified degradation trend and the current state level, the time required for the ceramic-steel interface to reach the preset critical condition is estimated, which is taken as the remaining safe service time.

4. The method for predicting the wear state of steel-based ceramic wear-resistant components according to claim 1, characterized in that: After assessing the current state level of the ceramic-steel interface, the method further includes the following steps: The real-time temperature distribution field of the non-working surface of the steel substrate is obtained; the real-time temperature distribution field is obtained by a temperature sensor array arranged at the corresponding position of the eddy current sensor array; At each monitoring moment, the difference quantity obtained by the eddy current sensor array, which reflects the change in the interface heat conduction state, is mapped into an eddy current anomaly distribution field according to its spatial location. The real-time temperature distribution field and the eddy current anomaly distribution field at the same moment are spatially superimposed and compared to obtain the confidence level of the current state level. The step of predicting the remaining safe service time of the ceramic wear-resistant layer based on the current state level and the trend of the difference over time includes the following steps: If the confidence level is high confidence, then the remaining safe service time of the ceramic wear-resistant layer is predicted based on the current state level and the trend of the difference over time.

5. The method for predicting the wear state of steel-based ceramic wear-resistant components according to claim 4, characterized in that: Obtaining the confidence level of the current state level includes the following steps: When the comparative analysis results show that the spatial overlap area between one or more local high-temperature regions identified in the real-time temperature distribution field and one or more high-anomaly regions identified in the eddy current anomaly distribution field exceeds a preset area threshold, the confidence level of the current state level is determined to be high confidence.

Images