Method for predicting residual life of rubber hose

Abstract

The invention provides a method for predicting the residual life of a rubber hose, and belongs to the technical field of industrial data prediction based on machine learning. In the service process of the rubber hose, running state data of the rubber hose are collected, and an effective state data set for degradation analysis is obtained through preprocessing; aiming at the degradation characteristics of the rubber hose under the action of pressure fatigue and material aging, multiple degradation related characteristics are extracted from multi-source data, and health indexes comprehensively reflecting the overall degradation state of the rubber hose are constructed; introducing working condition information of a pressure level and a temperature condition, establishing a rubber hose service life degradation model, and identifying model parameters through historical operation data, so that the model can describe an evolution rule of health indexes under different working conditions; and finally, continuously acquiring latest state data and updating health indexes, inputting the data into the life degradation model, and predicting the corresponding remaining service time when the rubber hose reaches a failure threshold value.

Description

Technical Field

[0001] This invention belongs to the field of industrial data prediction technology based on machine learning, and particularly relates to a method for predicting the remaining life of rubber hoses. Background Technology

[0002] Rubber hoses are widely used in hydraulic systems, construction machinery, aerospace, petrochemicals, and rail transportation, primarily for conveying high-pressure liquids or gases. Their operational reliability directly affects the safety and stability of equipment systems. During long-term service, rubber hoses are typically subjected to alternating internal pressure, temperature changes, bending loads, and complex environmental media, making them prone to fatigue damage, material aging, and structural performance degradation, ultimately leading to leakage, bursting, or functional failure. Current engineering practices primarily employ two maintenance strategies for rubber hoses: periodic replacement and reactive maintenance. Periodic replacement is usually based on experience or fixed service cycles, failing to adequately consider the actual degradation state of the hose under real-world operating conditions, easily resulting in over-maintenance or resource waste. Reactive maintenance, on the other hand, often occurs after severe hose failure, potentially causing equipment downtime, safety accidents, or even environmental pollution, posing a high risk. Therefore, there is an urgent need for a method that can accurately predict the remaining lifespan of rubber hoses before failure, enabling condition-based maintenance and preventative maintenance.

[0003] In recent years, with the development of sensor technology and data acquisition methods, the use of operational data to predict the lifespan of structural components has gradually attracted attention. However, existing technologies for predicting the lifespan of rubber hoses still have many shortcomings: on the one hand, some methods estimate lifespan based on only a single parameter (such as service time or pressure cycle count), which is difficult to reflect the actual degradation process of rubber hoses under complex operating conditions; on the other hand, some data-driven prediction methods often directly use raw signals or simple statistical features for modeling, without fully considering the degradation mechanisms such as aging and fatigue damage of rubber hose materials, resulting in poor stability and generalization ability of the prediction results. In addition, the operating conditions of rubber hoses change frequently during actual service, and different pressure levels, temperature conditions, and load combinations have a significant impact on their degradation rate. However, existing technologies mostly use fixed model parameters for lifespan prediction, lacking effective adaptability to changes in operating conditions and making it difficult to achieve dynamic correction of remaining lifespan. Especially under long-term operating conditions, model errors tend to accumulate continuously, leading to a large deviation between the predicted results and the actual lifespan.

[0004] Therefore, developing a method for predicting the remaining life of rubber hoses that can integrate multi-source operational data, accurately characterize the degradation state of rubber hoses, and dynamically update the prediction method according to changes in operational status is of great engineering significance and application value for improving the operational safety of rubber hoses, reducing maintenance costs, and realizing intelligent operation and maintenance. Summary of the Invention

[0005] To address the above problems, this invention proposes a method for predicting the remaining life of rubber hoses, comprising the following steps: S1, During the service of the rubber hose, the internal pressure signal and the ambient temperature signal of the hose are continuously collected, and the collection period is recorded; the historical data of the operating status is obtained through preprocessing. S2. The historical data of the operating status is segmented according to a fixed time window. Degradation features are extracted from pressure and temperature signals within each time window. Degradation trend analysis is performed on the extracted degradation features. The correlation coefficient between each feature and the operating time is calculated. The degradation features are weighted based on the magnitude of the correlation coefficient. The weighted degradation features are linearly fused to construct the historical rubber hose health index. S3, based on historical rubber hose health indicators, models the evolution of health indicators as an exponential continuous degradation model, and uses a deep learning model to adaptively model the degradation rate function in the continuous degradation model, finally obtaining the life degradation model; S4 collects new operational data according to the collection cycle of S1, and processes it according to the process of S1 and S2 to obtain the observed health index value at the current moment. It is then input into the lifespan degradation model constructed in S3 to obtain the predicted health index value in the future operational cycle. When the predicted value is less than or equal to the preset threshold, the predicted value is input into the trained remaining lifespan prediction model and the remaining lifespan prediction value is output.

[0006] Preferably, in step S2, the degradation features are extracted from the pressure and temperature signals within each time window, specifically as follows: For pressure signals, features reflecting load level, load fluctuation and fatigue accumulation effect are extracted, including average pressure level, pressure fluctuation degree and pressure cycle amplitude; For temperature signals, characteristic parameters reflecting the thermal effects of the environment are extracted, including the average temperature level and the rate of temperature change. The above features are used to form a corresponding set of degradation features within each time window.

[0007] Preferably, in step S2, degradation trend analysis is performed on the extracted degradation features, the correlation coefficient between each feature and the running time is calculated, and the degradation features are weighted based on the magnitude of the correlation coefficient, specifically as follows: Based on the extracted degradation feature vector Regarding the first Individual degenerative characteristics Construct its feature sequence that changes over a time window, represented as ;in, This represents the total number of time windows, and the corresponding time series is represented as follows: Among them, the first Each degenerate feature is a degenerate feature vector. The first in Dimensional components; by extracting the values ​​of this dimension's features across all time windows, the dimensional component is formed. The feature sequence of the degradation feature changes over a time window; based on the above feature sequence and time sequence, calculate the feature sequence of the degradation feature. The correlation coefficient between a degradation feature and its operating time is used to characterize the sensitivity of the degradation feature to the lifetime degradation process. The correlation coefficient is the Pearson correlation coefficient, which is used to measure the statistical correlation between the evolution sequence and the operating time sequence of the same degradation feature in the time dimension. The weight of each degradation feature is determined according to the absolute value of the correlation coefficient. The weight of each degradation feature is adaptively assigned according to the correlation analysis results, and the degradation feature with high correlation receives a larger weight.

[0008] Preferably, in step S2, the weighted degradation features are linearly fused, specifically as follows: The degradation features within each time window are linearly fused according to their corresponding weights to obtain the health indicator value for each time window; the health indicator is defined as: ; in, For the corresponding time window The output value of the health indicator is obtained by fusing degenerative features. Indicates the first i The weights corresponding to each degenerate feature The total number of degenerate features; It is a comprehensive quantitative index obtained by weighting and fusing multiple physical degradation characteristics, including pressure level, pressure fluctuation, pressure fatigue characteristics, and temperature aging characteristics, as inputs. All values ​​are normalized to the [0,1] interval, and the weights satisfy the normalization constraint.

[0009] Preferably, the process of modeling the evolution of health indicators as an exponential, continuous degradation model is specifically as follows: Assume health indicators over continuous time The value of is represented as The evolution of health indicators is modeled as an exponential, continuous degradation model: ; in, This indicates the initial health indicators of the rubber hose. Indicates the time of the rubber hose The instantaneous degradation rate function is given below, and its magnitude reflects the degradation intensity of the rubber hose at different operating stages; is the integration variable, used to accumulate the integration of the degradation rate function over the time interval [0, t]. Using an exponential function, a continuous degradation form of health indicators is constructed, which exhibits exponential decay over time.

[0010] Preferably, the adoption of a deep learning model to adaptively model the degradation rate function in the continuous degradation model includes constructing the degradation dynamics equation of health indicators, constructing a deep neural network for degradation rate, and solving the neural ordinary differential equation and performing time-continuous modeling. Instantaneous degradation rate Modeling with deep neural networks ; For parameters A deep neural network with a multi-layered nonlinear mapping structure is used to characterize the high-order nonlinear mapping relationship between health indicator status and time variables on the evolution of degradation rate; a degradation rate prediction network is constructed. Its input is , It includes data collected from the rubber hose during its service life, and uses the currently collected data to fit the instantaneous degradation rate. This allows for the prediction of the remaining lifespan of the rubber hose at subsequent points in time. The degradation rate deep neural network structure consists of a multi-layer feature mapping structure, including a feature expansion layer, a nonlinear representation layer, and a degradation constraint output layer. In the feature expansion layer, the input vector The input is fed into the feature expansion layer, where the features are updimensionalized by superimposing weighted linear mappings and bias terms to obtain a high-dimensional state feature representation. This completes the feature expansion of the original input state; In the nonlinear representation layer, the feature vector output by the feature expansion layer is... The input is fed into the first nonlinear mapping layer, and features are reconstructed through a nonlinear activation function to obtain... ; then, The input is fed into the second linear transformation layer for feature combination and reweighting to obtain an intermediate feature representation. Then, it is mapped again through a nonlinear activation function to form a higher-order degenerate feature representation. ;Calculated by continuous superposition of multi-level linear transformations and nonlinear activation functions; In the degradation constraint output layer, higher-order degradation features are... The input is fed into the degradation rate output layer, and intermediate degradation variables are calculated through linear mapping. To ensure that the degradation rate satisfies the physical non-negativity constraint, a non-negativity activation function is applied to the intermediate degradation variables to obtain the final degradation rate function modeling result. ; Finally, in solving the ordinary differential equations and modeling the time continuity, the instantaneous degradation rate is obtained by using a deep neural network to calculate the degradation rate. By substituting the differential equation of health index degradation, a continuous-time degradation dynamics model that integrates physical constraints and deep neural networks is constructed.

[0011] Preferably, the life degradation model adopts a recursive update mechanism; during the actual operation of the rubber hose, the life degradation model adopts a recursive update mechanism; during the actual operation of the rubber hose, based on the system's future continuous time... The obtained operating data is used to calculate the latest health indicator values. The observed value was compared with the health index predictions obtained from the lifespan degradation model in S3. Compare and calculate the prediction error. ; Based on prediction error The degradation rate is corrected online; the prediction error is introduced into the input of the degradation rate network to construct the corrected degradation rate. Represented as: ; in, This is the degradation rate function output by the original lifetime degradation model; This is the error feedback gain coefficient, used to adjust the degree of influence of prediction error on degradation rate correction; At the same time, the prediction error As a monitoring signal, the parameters of the degradation rate deep network are recursively updated.

[0012] Preferably, the remaining lifetime prediction model includes a temporal convolutional feature extraction layer, a temporal attention weighting layer, and a remaining lifetime regression output layer; First, the health indicator prediction sequence The input is fed into the temporal convolutional feature extraction layer, which consists of three sequentially stacked temporal convolutional layers: a first temporal convolutional layer, a second temporal convolutional layer, and a third temporal convolutional layer. These layers are used to progressively extract short-term, medium-term, and long-term degradation features from the health indicator prediction sequence. The three temporal convolutional layers are connected in series, with the output of the previous layer serving as the input to the next. The input to the first temporal convolutional layer is the health indicator prediction sequence. ; in the Each time convolutional layer (l=1,2,3) feeds back to the previous layer. Output time feature sequence When performing a one-dimensional dilated convolution operation with l=1, the output of the previous layer... Defined as the input health indicator prediction sequence Subsequently, the result of the convolution operation is superimposed with the corresponding bias term, and then mapped using the non-linear activation function ReLU to obtain the first... Output of layer-time convolutional features; Secondly, in the temporal attention weighting layer, a temporal attention mechanism is introduced after the temporal convolutional layer to construct a temporal attention weighting layer, which adaptively weights the key degradation time segments that contribute more to the remaining lifetime; the temporal feature sequence of each time step is calculated through a fully connected layer. attention weights The attention weight vector is obtained by normalizing the vector using the Softmax function. Subsequently, the time feature sequences are weighted and fused to obtain a global degenerate representation. ; Finally, the weighted global degradation features The input is fed into the remaining lifetime regression layer, and the remaining lifetime prediction results are output through a two-stage fully connected network. ; This indicates the remaining operational time of the rubber hose under its current operating condition, from the current moment until its health indicators reach the preset failure threshold. The range of values ​​is ,in, This indicates the maximum service life of the rubber hose under design conditions. This indicates the cumulative service time corresponding to the current operating moment.

[0013] Compared with the prior art, the present invention has the following innovative features: 1) A method for constructing health indicators based on the degradation mechanism of rubber hoses is proposed. By screening and fusing multi-source degradation characteristics, the limitations of single characteristics or empirical indicators in life assessment results are avoided; 2) A life degradation model coupling operating condition information and health indicators was established. This model can describe the impact of different operating conditions such as pressure and temperature on the degradation rate, improving the applicability of the life prediction model under complex operating conditions. 3) Introduce an online update mechanism for the life degradation model. This allows the remaining life prediction results to be dynamically corrected as the actual operating conditions of the rubber hose change, overcoming the problem of gradual accumulation of prediction errors in static models.

[0014] The beneficial effects of this invention include: 1) Improve the accuracy and stability of lifespan prediction. By constructing health indicators oriented towards degradation mechanisms and introducing condition-coupled modeling, the prediction error caused by single parameters or empirical models is effectively reduced, thereby improving the accuracy and stability of remaining lifespan prediction results; 2) Enhanced adaptability to complex operating conditions. This invention can dynamically adjust the degradation model parameters according to changes in operating conditions such as pressure and temperature, making it suitable for predicting the lifespan of rubber hoses under complex operating conditions such as high pressure, pulsating loads, and frequent start-stop cycles; 3) Achieve dynamic correction and real-time assessment of remaining service life. Through an online update mechanism, the remaining service life prediction results can be continuously corrected as the actual operating conditions of the rubber hose change, avoiding the problem of inaccurate predictions in traditional static models over long-term operation; 4) Reduce maintenance costs and operational risks. Based on the life prediction results of this invention, a reasonable maintenance and replacement plan for rubber hoses can be formulated, reducing unnecessary premature replacements and minimizing safety risks and economic losses caused by sudden failures; 5) Provide decision-making basis for condition-based maintenance of rubber hoses. By quantitatively assessing the degradation status and remaining life of hoses, it provides scientific and intuitive decision support for equipment operation and maintenance personnel, promoting the transformation from experience-based maintenance to condition-based and predictive maintenance. Attached Figure Description

[0015] Figure 1 This is a flowchart of the overall method of the present invention.

[0016] Figure 2 This is a diagram of the deep neural network structure for the degradation rate of this invention.

[0017] Figure 3 This is a network structure diagram of the remaining lifetime prediction model of this invention.

[0018] Figure 4 This is a diagram illustrating the dynamic update effect of the prediction-correction mechanism for health indicators in an embodiment of the present invention.

[0019] Figure 5 This is a diagram showing the dynamic correction results of the remaining lifespan of the health indicator prediction sequence in an embodiment of the present invention.

[0020] Figure 6 This is a graph showing the prediction results of health indicators related to state-related degradation rates in an embodiment of the present invention. Detailed Implementation

[0021] This invention proposes a method for predicting the remaining life of rubber hoses, the overall process of which is as follows: Figure 1 As shown: S1. Rubber Hose Operation Data Acquisition and Processing. During the service life of the rubber hose, pressure and temperature sensors are used to continuously acquire internal pressure signals and ambient temperature signals, while recording the acquisition period. The acquired raw operational data undergoes preprocessing, specifically including noise reduction; outlier removal; linear normalization; and synchronization and alignment of multi-source data. After processing, standardized historical operational status data for degradation analysis is obtained.

[0022] S2, Degradation Feature Extraction and Health Indicator Construction. The obtained standardized historical operating status data is segmented according to fixed time windows. Degradation features are extracted from pressure and temperature signals within each time window. Degradation trend analysis is performed on the extracted features, and the correlation coefficient between each feature and operating time is calculated. The degradation features are then weighted based on the magnitude of the correlation coefficient. Subsequently, the weighted degradation features are linearly fused to construct historical rubber hose health indicators. These historical health indicators are continuous variables that monotonically change with operating time, used to characterize the overall degradation state of the rubber hose.

[0023] S3. Establishment of a rubber hose lifespan degradation model. The system uses constructed health indicators as state variables to establish a rubber hose lifespan degradation model. This model employs an exponential degradation model, identifying the degradation rate parameters based on historical operating data to accurately describe the degradation evolution of the rubber hose under actual operating conditions. By setting failure thresholds for the health indicators, a correspondence is established between the lifespan degradation model and the failure state of the rubber hose. After acquiring new operating data, the degradation rate parameters in the lifespan degradation model are recursively updated based on the latest health indicators, ensuring that the model parameters reflect the current true degradation state of the rubber hose. This dynamic update mechanism enables real-time correction and continuous prediction of the remaining lifespan of the rubber hose.

[0024] S4, Remaining Life Prediction and Dynamic Update. During the operation of the rubber hose, its health indicators are updated in real time, and the updated health indicators are input into the life degradation model. Based on the difference between the current health indicator value and the failure threshold, combined with the degradation rate of the life degradation model, the remaining operating time of the rubber hose is calculated, and the remaining life prediction result is obtained.

[0025] S1. Collection and Processing of Historical Operating Data for Rubber Hose In the process of predicting the remaining life of rubber hoses, accurate acquisition and standardized processing of operational data are fundamental to subsequent degradation feature extraction, life degradation modeling, and remaining life prediction. This invention first systematically collects operational status data of rubber hoses and then constructs a standardized operational status dataset through steps such as noise reduction, outlier removal, normalization, and multi-source data synchronization and alignment. This provides a reliable data foundation for subsequent degradation feature extraction and life analysis.

[0026] S1-1 Rubber Hose Operation Data Acquisition: During the service life of rubber hoses, pressure sensors and temperature sensors are installed at key locations on the hose to collect internal pressure signals and ambient temperature signals. The pressure sensors are used to collect real-time data on the change in internal pressure over time; this pressure signal is denoted as... The temperature sensor is used to collect the temperature signal of the operating environment of the rubber hose, and the temperature signal is denoted as... .

[0027] Simultaneously, the cumulative operating cycle of the rubber hose is recorded through a data acquisition module. The operating cycle is denoted as... The above pressure signal Temperature signal Operating cycle All according to a uniform sampling frequency The data is collected and timestamped to obtain the original running data sequence.

[0028] S1-2 Raw Operational Data Preprocessing: First, regarding the collected pressure signals and temperature signal Due to sensor measurement errors and environmental interference, the signal may contain high-frequency noise components. Therefore, the raw data is first denoised. Specifically, a sliding time window is used to smooth the pressure and temperature signals, resulting in the denoised pressure and temperature signals, respectively. and .

[0029] Secondly, after denoising, outlier data points in the pressure and temperature signals are identified and removed. Specifically, the denoised signals are calculated separately. and The mean and standard deviation over a given time period are used. If the absolute value of the difference between the denoised signal value and the mean of a pressure or temperature data point at a given moment is greater than three times the corresponding standard deviation, it is identified as an outlier and removed. This outlier removal helps prevent sensor malfunctions or sudden interference from affecting subsequent health indicator construction and lifespan prediction results.

[0030] Finally, the denoised pressure and temperature signals are normalized to eliminate the influence of differences in the dimensions of different physical quantities on the degradation analysis. Specifically, a linear normalization method is used to process the pressure and temperature signals, and the normalized pressure and temperature signals are expressed as follows: and .

[0031] Subsequently, the normalized pressure signal was analyzed based on the timestamp. Temperature signal Operating cycle Perform synchronization alignment to form a unified multi-source state data vector. Through the above processing, a standardized historical dataset of operating states is obtained for subsequent degradation feature extraction and lifetime analysis.

[0032] S2. Degradation Feature Extraction and Health Indicator Construction After completing the acquisition and preprocessing of the rubber hose operating data, the multi-source state data vector output from step S1 is obtained. As input, since the performance degradation of rubber hoses during service is not instantaneous but a gradual evolution process with the accumulation of operating time and load, it is necessary to further process the operating status data to extract features that can reflect the changing patterns of hose degradation, and on this basis, construct a unified health index to characterize the overall degree of degradation of rubber hoses.

[0033] This stage employs a process of time window segmentation, degradation feature extraction, degradation trend analysis, and health indicator construction to quantitatively describe the degradation state of rubber hoses, providing crucial input for the subsequent establishment of a lifespan degradation model. Specifically, it includes the following steps: S2-1 Multi-source state data segmentation processing based on time windows: Based on the multi-source state data vector obtained in step S1 To avoid interference from instantaneous operating condition fluctuations on degradation analysis and to ensure the continuity of degradation information over time, the multi-source state data is segmented according to a fixed time window.

[0034] Specifically, the time window length is preset. During the operation of the rubber hose, the operational data is divided into time windows, ensuring that the data within each time window reflects the overall operational status of the hose at that stage. This segmented time window processing method not only reduces the impact of data noise on degradation analysis but also effectively captures the changing trends of the rubber hose's degradation status at different service stages.

[0035] S2-2 Degeneracy Feature Extraction: Within each time window, the normalized pressure signal is used... and temperature signal The characteristic parameters that reflect the degradation state of the rubber hose are extracted and used to characterize the changes in the operating state of the rubber hose during the time period.

[0036] Specifically, in the first Within a time window, the following degradation features are extracted: 1) Regarding pressure signals This study extracts features reflecting load levels, load fluctuations, and cumulative fatigue effects, including average pressure levels, pressure fluctuation degrees, and pressure cycle amplitudes. These features characterize the stress state changes of rubber hoses under alternating pressure and have direct indicative significance for fatigue damage and structural degradation.

[0037] Among them, the average pressure level Pressure fluctuation level Pressure cycle amplitude .in, For the first The start time of each time window The time window length, Index for time windows; Indicates the first The average pressure level characteristics within a time window It is the first Characteristics of pressure fluctuations within a time window It represents the deviation of instantaneous pressure from average pressure. For the first Pressure cycle amplitude characteristics within a time window This represents the maximum value of the pressure signal within the time window. This represents the minimum value of the pressure signal within the time window.

[0038] 2) Regarding temperature signals We extracted characteristic parameters reflecting the effects of environmental heat, including average temperature level and temperature change rate. Temperature is a crucial factor affecting the aging rate of rubber materials, and its long-term trend has a significant impact on the performance degradation of rubber hoses.

[0039] Among them, the average temperature signal Rate of temperature change .

[0040] By using the above method, a corresponding set of degradation characteristics is formed within each time window, so that the degradation characteristics can comprehensively reflect the combined effects of pressure fatigue and temperature aging on the degradation process of rubber hoses.

[0041] Finally, the integrated degenerate feature vector is as follows: . This represents the total number of degenerative features.

[0042] S2-3 Degradation Trend Analysis and Feature Weighting: Since different degradation features have varying indicative abilities to the lifespan degradation process of rubber hoses, directly performing equal-weighted fusion of all features can easily introduce redundant information and even weaken the expressive power of the lifespan degradation trend. To address this issue, this invention designs a feature adaptive weighting mechanism based on degradation trend analysis.

[0043] Specifically, based on the degradation feature vector extracted in S2-2 Correlation analysis was performed between the time series formed within different time windows and the running time of the rubber hose.

[0044] Regarding the first Individual degenerative characteristics Construct its feature sequence that changes over a time window, represented as .in, This represents the total number of time windows, and the corresponding time series is represented as follows: Among them, the first Each degenerate feature is a degenerate feature vector. The first in The dimensional component. The values ​​of this dimensional feature are extracted across all time windows to form the first dimensional component. The feature sequence of the degradation feature changes over a time window. Based on the above feature sequence and time series, calculate the feature sequence of the degradation feature as a function of the time window. The correlation coefficient between a degradation feature and its runtime is used to characterize the sensitivity of that degradation feature to the lifetime degradation process. The correlation coefficient is the Pearson correlation coefficient, which measures the statistical correlation between the evolution sequence and runtime sequence of the same degradation feature over time. Its calculation formula is as follows: ; in, Indicates the first The mean of each degradation feature across all time windows. This represents the mean of the corresponding time series. For the first Pearson correlation coefficient between each degradation characteristic and runtime For degradation feature index. Indicates the first The first degradation feature in the first Values ​​taken within a time window For the first The running time corresponding to each sampling point in a time window The total number of time windows for calculating correlation coefficients is used to quantitatively assess the consistency of trends in each feature over time. The weight of each degradation feature is determined based on the absolute value of the correlation coefficient; a larger absolute value indicates a stronger indicative ability of the degradation feature to the degradation process of the rubber hose, and a more significant characterization ability for lifespan degradation.

[0045] Furthermore, based on the correlation analysis results, weights are adaptively assigned to each degradation feature, with features exhibiting high correlation receiving larger weights and those exhibiting low correlation receiving smaller weights. The weight of each feature is obtained by normalizing its correlation index. The weights of each degenerate feature are defined as follows: ; in, Indicates the first The correlation coefficient between each degradation characteristic and runtime Indicates the first The feature weights of each degradation feature in the construction of health indicators This represents the total number of degenerative features.

[0046] This adaptive weighting mechanism enables the objective determination of the weights of degenerate features, avoids relying on manual experience to set weights, and improves the robustness and repeatability of the health indicator construction process.

[0047] S2-4 Rubber Hose Health Indicator Construction: After weighting the degradation features, the degradation features are linearly fused to output a health index that characterizes the overall degradation state of the rubber hose. The health index is the calculation result of this step and serves as the input variable for the subsequent life degradation model.

[0048] Specifically, the degradation features within each time window are linearly fused according to their corresponding weights to obtain the health indicator value for each time window. This health indicator compresses multi-source, multi-dimensional degradation information into a single continuous variable, realizing the mapping from the degradation feature space to the health state space, thereby significantly reducing the complexity of subsequent lifespan degradation modeling. The health indicator is defined as: ; in, For the corresponding time window The output value of the health indicator is obtained by fusing degenerative features. This represents the weight corresponding to the i-th degenerate feature. It is a comprehensive quantitative index obtained by weighting and fusing multiple physical degradation characteristics, including pressure level, pressure fluctuation, pressure fatigue characteristics, and temperature aging characteristics, as inputs. Due to the various degradation characteristics... All values ​​are normalized to the [0,1] interval, and the weights satisfy the normalization constraint; therefore, the health indicators... The value range of is in the interval [0,1]. This indicates that the rubber hose is in near-initial healthy condition and its structural performance is intact; This indicates that the rubber hose is in a state of severe degradation or near failure.

[0049] To ensure that health indicators accurately reflect the degradation and evolution process of rubber hoses, the health indicator sequence is smoothed and subjected to monotonicity constraints to obtain a health state sequence that changes continuously with operating time. The processed health indicators serve as the state input variables for the rubber hose life degradation model, used for subsequent model establishment and remaining life prediction.

[0050] S3. Establishment of a lifespan degradation model for rubber hoses This step introduces a physically constrained neural frequent differential equation model, enabling high-precision modeling of the continuous degradation process of rubber hose health indicators while maintaining the physical interpretability of exponential degradation. First, the health indicators constructed using S2... As the basic state input for lifespan degradation modeling; secondly, a continuous degradation physical framework for health indicators decaying exponentially over time is established; by introducing physical constraint neural ordinary differential equations, data-driven adaptive modeling of degradation rate is performed to realize the prediction of degradation trajectory of health indicators in continuous time domain. Subsequently, the model parameters are trained and identified offline using a combination of data consistency loss and kinetic consistency loss. Furthermore, a prediction error feedback mechanism is introduced to recursively correct the degradation rate and model parameters online during the actual operation of the rubber hose. Through these steps, a closed-loop lifespan degradation modeling process—modeling, training, prediction, and feedback update—is constructed, ultimately outputting a future health indicator prediction sequence that can be used to calculate remaining lifespan. This process specifically includes: The health index of rubber hoses constructed based on S2 The health indicators serve as state input variables in the lifespan degradation model. This is a scalar quantity that changes continuously with operating time, used to characterize the overall degradation state of the rubber hose during service. As the rubber hose continues to operate, its health indicators gradually deteriorate. The predicted values ​​of the health indicators for future moments are obtained through the state evolution process of the life degradation model, and the preset failure criteria are met when the failure state is reached.

[0051] S3-1 Exponential Lifetime Degradation Modeling: Using the historical rubber hose health index constructed by S2 as the state input variable for life degradation modeling, assuming that the health index varies over continuous time... The value of is represented as Based on the material property degradation mechanism of rubber hoses under alternating pressure and temperature aging, their degradation process exhibits a nonlinear deterioration characteristic that gradually accelerates over time. The evolution of health indicators is modeled as an exponential continuous degradation model: ; in, This indicates the initial health indicators of the rubber hose. Indicates the time of the rubber hose The instantaneous degradation rate function is given below, and its magnitude reflects the degradation intensity of the rubber hose at different operating stages. is the integration variable, used to accumulate the integration of the degradation rate function over the time interval [0, t]. Using an exponential function, a continuous degradation form of health indicators is constructed, which exhibits exponential decay over time.

[0052] S3-2 Design of a Deep Learning Model for Lifetime Degradation Based on Physically Constrained God's Regular Differential Equations: To model the degradation rate function using data fitting This invention introduces physically constrained neural network constant differential equations to construct a deep learning model for lifespan degradation. This model combines a physical structure of degradation dynamics with a data-driven neural network to achieve adaptive modeling of complex degradation behaviors while ensuring the physical consistency of the degradation process. The process consists of three parts: constructing the degradation dynamics equations for health indicators, designing the deep neural network structure for degradation rates, solving the neural network constant differential equations, and performing time-continuous modeling.

[0053] First, the dynamic equations for the degradation of health indicators are constructed. This invention, from the perspective of degradation mechanisms, describes the evolution of health indicators as a continuous-time degradation dynamic system. This is to determine the instantaneous degradation rate. By employing adaptive modeling, this invention rewrites the degradation process of health indicators in S3-1 into a neuronormal differential equation form: ; The continuous degradation kinetic equation for health indicators is obtained. Wherein, the instantaneous degradation rate... Modeling with deep neural networks . For parameters A deep neural network with a multi-layer nonlinear mapping structure is used to characterize the high-order nonlinear mapping relationship between health indicator status and time variables on the evolution of degradation rate.

[0054] Secondly, construct a degradation rate prediction network. Its input is .in, The current usage time of the rubber hose, i.e. It includes data collected from the rubber hose during its service life, and uses the currently collected data to fit the instantaneous degradation rate. This allows for the prediction of the remaining lifespan of the rubber hose at subsequent points in time. Subsequently, the degradation rate deep neural network structure consists of a multi-layer feature mapping structure, including a feature expansion layer, a nonlinear representation layer, and a degradation constraint output layer.

[0055] Specifically, in the feature expansion layer, the input vector The input is fed into the feature expansion layer, where the features are updimensionalized by superimposing weighted linear mappings and bias terms to obtain a high-dimensional state feature representation. Complete the feature expansion of the original input state.

[0056] In the nonlinear representation layer, the feature vector output by the feature expansion layer is... The input is fed into the first nonlinear mapping layer, and features are reconstructed through a nonlinear activation function to obtain... Subsequently, The input is fed into the second linear transformation layer for feature combination and reweighting to obtain an intermediate feature representation. Then, it is mapped again through a nonlinear activation function to form a higher-order degenerate feature representation. By using a series of linear transformations and nonlinear activation functions, the network can extract the complex nonlinear degradation relationship between health indicators and time layer by layer.

[0057] In the degradation constraint output layer, higher-order degradation features are... The input is fed into the degradation rate output layer, and intermediate degradation variables are calculated through linear mapping. To ensure that the degradation rate satisfies the physical non-negativity constraint, a non-negativity activation function is applied to the intermediate degradation variables to obtain the final instantaneous degradation rate modeling result:

[0058] Finally, in solving the constant differential equations and modeling the time continuity, the instantaneous degradation rate is obtained by calculating the degradation rate using the aforementioned deep network. By substituting the differential equation of health index degradation, a continuous-time degradation dynamics model that integrates physical constraints and deep neural networks is finally constructed.

[0059] Based on the instantaneous degradation rate Furthermore, continuous-time prediction of the degradation trajectory of health indicators is performed to obtain the evolution of the health status of the rubber hose during future operation. Specifically: First, the degenerate differential equation is discretized. Let the discrete time series be... The corresponding time interval is... According to the aforementioned health indicator degradation kinetic equation After discretization, the recursive prediction formula for health indicators at time step 1 is expressed as: ; in, Indicates time Predicted values ​​of health indicators below This represents the degradation rate corresponding to the k-th time step calculated by a deep neural network, ultimately achieving continuous prediction of health indicators on discrete time scales.

[0060] S3-3 Model Parameter Identification and Training: During model training, network parameters are updated by jointly optimizing data consistency loss and degradation dynamics consistency loss. Data consistency loss is defined as: ; The degradation kinetic consistency loss is defined as: ; The overall loss function is expressed as: .in, These are the dynamic constraint weighting coefficients.

[0061] The network parameters in S3-2 are iteratively updated using the backpropagation algorithm, gradually bringing the predicted trajectory of health indicators closer to the actual degradation trajectory until the loss function converges. This continuous-time solution method avoids the error accumulation problem that occurs in traditional discrete models during long-term prediction.

[0062] S3-4 Recursive update mechanism for lifetime degradation model based on prediction error feedback: During the actual operation of the rubber hose, the life degradation model adopts a recursive update mechanism; during the actual operation of the rubber hose, based on the system's future continuous time... The obtained operating data is used to calculate the latest health indicator values. Then, the observed value is compared with the predicted health index value obtained from the lifespan degradation model in S3. Compare the results and calculate the prediction error: ; The prediction error This is used to characterize the bias of current life degradation models in depicting the actual degradation state of rubber hoses. Prediction error There are two functions in this invention: (1) Correction effect on degradation rate. Based on prediction error. The degradation rate is corrected online. Specifically, the prediction error is introduced into the input of the degradation rate network to construct the corrected degradation rate. Represented as: ; in, The instantaneous degradation rate is the output of the original lifetime degradation model in S3. This is the error feedback gain coefficient, used to adjust the degree of influence of prediction error on the degradation rate correction. This error feedback mechanism allows the degradation rate to be dynamically adjusted according to changes in the actual degradation state.

[0063] (2) Recursive update of model parameters based on error feedback. Based on the degradation rate correction, the prediction error... As a supervisory signal, the parameters of the degradation rate deep network are recursively updated. Specifically, this is achieved by minimizing the following online update loss function: ; Then, according to the comprehensive loss function of S3-3... The overall loss was further updated to By performing multiple gradient updates on the loss function, the network parameters are adjusted to reduce prediction errors, thereby enabling the lifespan degradation model to gradually conform to the actual degradation trajectory of the rubber hose.

[0064] After updating the model parameters, a new sequence of health indicators for future operating cycles is generated based on the updated lifespan degradation model. Figure 2 This is a diagram of a deep neural network structure for degradation rate.

[0065] S4. Remaining useful life prediction and dynamic updates This step, based on the construction of S2 rubber hose health indicators and the establishment of the S3 life degradation model, proposes a method for predicting and dynamically updating the remaining life of rubber hoses under online operating conditions. By using the predicted future health indicators output by the life degradation model as input, a remaining life prediction model is constructed to assess the remaining operating time of the rubber hose from its current operating state to its failure state. Simultaneously, after acquiring new operating data, a recursive update mechanism is introduced to correct the key parameters of the life degradation model in real time, enabling the model to continuously reflect changes in the actual degradation state of the rubber hose. This process specifically includes: S4-1 Failure Threshold Setting and Failure Criterion Construction: During the operation of the rubber hose, the system periodically collects new operational data and processes and fuses the data according to steps S1 and S2 to obtain the current health indicator observation values ​​in real time. Simultaneously, based on the lifespan degradation model constructed in S3, the health status of the rubber hose during its future operating cycle is predicted, resulting in a predicted sequence of health indicators for the future operating cycle. The aforementioned health indicator prediction sequence serves as input to the remaining life prediction module, used to characterize the future degradation trend of the rubber hose in its current state.

[0066] Based on the material performance indicators and engineering application requirements of the rubber hose, a health indicator failure threshold is pre-set. When the predicted values ​​of health indicators meet When the rubber hose reaches the predicted failure state at the current moment, the remaining life prediction model is used to make a prediction. If no health index is lower than the failure threshold within the prediction time window, the prediction window continues to roll forward until the prediction time point that meets the failure conditions is obtained.

[0067] S4-2 Remaining lifetime prediction model based on attention-enhanced temporal convolutional networks: This invention designs an attention-enhanced temporal convolutional network (AE-TCN) to predict and model the remaining lifespan of rubber hoses. The remaining lifespan prediction model takes as input the evolution sequence of health indicators predicted by the lifespan degradation model. This model uses the predicted health indicator sequence... As input, the health indicator prediction sequence is recursively obtained by the aforementioned lifespan degradation model based on current and historical health indicator states, and is used to characterize the degradation evolution trend of the rubber hose over a future period. A deep network automatically learns the nonlinear mapping relationship between the evolutionary form of health indicators and remaining lifespan. The output is the predicted remaining lifespan of the rubber hose at the current moment. The remaining lifetime prediction model is as follows: Figure 3 As shown, it includes a temporal convolutional feature extraction layer, a temporal attention weighting layer, and a remaining lifetime regression output layer.

[0068] First, the health indicator prediction sequence The input is fed into the temporal convolutional feature extraction layer, which consists of three sequentially stacked temporal convolutional layers: a first temporal convolutional layer, a second temporal convolutional layer, and a third temporal convolutional layer. These layers are used to extract short-term, medium-term, and long-term degradation features from the health indicator prediction sequence step by step.

[0069] Specifically, the three temporal convolutional layers are connected in series, with the output of the previous layer serving as the input of the next layer. The input of the first temporal convolutional layer is the health indicator prediction sequence. In the first Each time convolutional layer (l=1,2,3) feeds back to the previous layer. Output time feature sequence When performing a one-dimensional dilated convolution operation with l=1, the output of the previous layer... Defined as the input health indicator prediction sequence Subsequently, the result of the convolution operation is superimposed with the corresponding bias term, and then mapped using the non-linear activation function ReLU to obtain the first... The output of the temporal convolution feature layer.

[0070] Secondly, in the temporal attention weighted layer, a temporal attention mechanism is introduced after the temporal convolutional layer for key degradation time segments that contribute more to the remaining lifetime. The temporal feature sequence at each time step is calculated through a fully connected layer. attention weights The attention weight vector is then obtained by normalizing it using the Softmax function. Subsequently, the time feature sequences are weighted and fused to obtain a global degenerate representation. .

[0071] Finally, the weighted global degradation features The input is fed into the remaining lifetime regression layer, which outputs the remaining lifetime prediction result through a two-stage fully connected network. . Specifically, This indicates the remaining operational time of the rubber hose under its current operating condition, from the current moment until its health indicators reach the preset failure threshold.

[0072] The remaining service life prediction is quantified on a time scale, and its physical meaning is the expected remaining service time of the rubber hose under the condition of maintaining safety and functional requirements. The range of values ​​is ,in, This indicates the maximum service life of the rubber hose under design conditions. This indicates the cumulative service time corresponding to the current operating moment. When... When the rubber hose reaches a certain health threshold, it indicates that the hose has failed; when... A larger value indicates that the rubber hose still has a relatively long operating time before it fails.

[0073] By designing the remaining life regression output layer as described above, the model can comprehensively consider the degradation information contained in the health indicator prediction sequence and output prediction results that are highly consistent with the actual remaining life of the rubber hose.

[0074] Training and loss function design of the S4-3 remaining lifetime prediction model: Based on historical lifespan data, a supervised learning approach is used to train the remaining lifespan prediction network. The loss function is defined as: ; The model parameters are optimized using a backpropagation algorithm to gradually approximate the actual lifespan label. Based on the updated model parameters obtained in steps S3-4, a new sequence of health indicators for future operating cycles is generated using the updated lifespan degradation model. Then, the failure criterion judgment and remaining lifespan calculation in step S4-1 are executed again. By continuously repeating the closed-loop process of health indicator prediction, error calculation, model correction, and remaining lifespan recalculation, real-time correction and continuous updating of the rubber hose's remaining lifespan prediction results are achieved.

[0075] Simulation experiment: Figure 4The diagram shows the dynamic update effect of the prediction-correction mechanism on health indicators. Simulation results show that the model without the error feedback update mechanism exhibits significant prediction bias in the later stages of operation; however, by introducing a prediction error-driven recursive update mechanism, the predicted health indicators can continuously match the actual degradation trajectory. This result verifies that the proposed online update method has significant advantages in coping with changes in operating conditions and model uncertainties.

[0076] Figure 5 Simulation results show that as the runtime progresses, the remaining lifetime prediction value calculated based on the health indicator prediction sequence gradually converges to the actual failure time. This result verifies that the proposed remaining lifetime prediction method can achieve continuous correction and real-time updates, avoiding the instability problem of traditional one-time lifetime prediction methods in the early stages.

[0077] Figure 6 It can be intuitively observed that the health indicators show a continuous and monotonous downward trend over time, with the rate of decline accelerating significantly in the later stages of operation, reflecting the typical accelerated degradation characteristics of rubber hoses during long-term service. Simultaneously, the degradation rate dynamically adjusts with operating time and health status, remaining at a low level when the health status is good, and gradually increasing as the health indicators decline, demonstrating the dependence of the degradation process on the current state.

[0078] The above results demonstrate that the state- and time-based lifetime degradation model can effectively characterize the nonlinear degradation evolution of rubber hoses under actual working conditions. Compared with the traditional fixed degradation rate model, the degradation model established in this invention can dynamically adjust the degradation rate according to the current health status of the rubber hose, thereby improving the accuracy and physical consistency of health indicator predictions and providing a reliable basis for subsequent remaining lifetime prediction.

[0079] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0080] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of a rubber hose, characterized in that, Includes the following processes: S1, during the service of the rubber hose, continuously collects the internal pressure signal and the ambient temperature signal of the hose, and records the collection period. Preprocessing yields historical operational status data; S2, the historical data of the operating status is segmented according to a fixed time window, and the degradation features are extracted from the pressure signal and temperature signal within each time window; The extracted degradation features are analyzed for degradation trends. The correlation coefficient between each feature and the running time is calculated. The degradation features are weighted based on the correlation coefficient. The weighted degradation features are then linearly fused to construct historical rubber hose health indicators. S3, based on historical rubber hose health indicators, models the evolution of health indicators as an exponential continuous degradation model, and uses a deep learning model to adaptively model the degradation rate function in the continuous degradation model, finally obtaining the life degradation model; S4 collects new operational data according to the collection cycle of S1, and processes it according to the process of S1 and S2 to obtain the observed health index value at the current moment. It is then input into the lifespan degradation model constructed in S3 to obtain the predicted health index value in the future operational cycle. When the predicted value is less than or equal to the preset threshold, the predicted value is input into the trained remaining lifespan prediction model and the remaining lifespan prediction value is output.

2. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: In S2, degradation features are extracted from the pressure and temperature signals within each time window, specifically as follows: For pressure signals, features reflecting load level, load fluctuation and fatigue accumulation effect are extracted, including average pressure level, pressure fluctuation degree and pressure cycle amplitude; For temperature signals, characteristic parameters reflecting the thermal effects of the environment are extracted, including the average temperature level and the rate of temperature change. The above features are used to form a corresponding set of degradation features within each time window.

3. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: In step S2, degradation trend analysis is performed on the extracted degradation features, the correlation coefficient between each feature and the running time is calculated, and the degradation features are weighted based on the magnitude of the correlation coefficient, specifically as follows: Based on the extracted degradation feature vector Regarding the first Individual degenerative characteristics Construct its feature sequence that changes over a time window, represented as ;in, This represents the total number of time windows, and the corresponding time series is represented as follows: Among them, the first Each degenerate feature is a degenerate feature vector. The first in Dimensional components; by extracting the values ​​of this dimension's features across all time windows, the dimensional component is formed. The feature sequence of the degradation feature changes over a time window; based on the above feature sequence and time sequence, calculate the feature sequence of the degradation feature. The correlation coefficient between a degradation feature and its operating time is used to characterize the sensitivity of the degradation feature to the lifetime degradation process. The correlation coefficient is the Pearson correlation coefficient, which is used to measure the statistical correlation between the evolution sequence and the operating time sequence of the same degradation feature in the time dimension. The weight of each degradation feature is determined according to the absolute value of the correlation coefficient. The weight of each degradation feature is adaptively assigned according to the correlation analysis results, and the degradation feature with high correlation receives a larger weight.

4. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: In S2, the weighted degenerate features are linearly fused, specifically as follows: The degradation features within each time window are linearly fused according to their corresponding weights to obtain the health indicator value for each time window; the health indicator is defined as: ； in, For the corresponding time window The output value of the health indicator is obtained by fusing degenerative features. Indicates the first i The weights corresponding to each degenerate feature The total number of degenerate features; It is a comprehensive quantitative index obtained by weighting and fusing multiple physical degradation characteristics, including pressure level, pressure fluctuation, pressure fatigue characteristics, and temperature aging characteristics, as inputs. All values ​​are normalized to the [0,1] interval, and the weights satisfy the normalization constraint.

5. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: The process of modeling the evolution of health indicators as an exponential, continuous degradation model is specifically as follows: Assume health indicators over continuous time The value of is represented as The evolution of health indicators is modeled as an exponential, continuous degradation model: ； in, This indicates the initial health indicators of the rubber hose. Indicates the time of the rubber hose The instantaneous degradation rate function is given below, and its magnitude reflects the degradation intensity of the rubber hose at different operating stages; is the integration variable, used to accumulate the integration of the degradation rate function over the time interval [0, t]. Using an exponential function, a continuous degradation form of health indicators is constructed, which exhibits exponential decay over time.

6. The method for predicting the remaining life of a rubber hose as described in claim 5, characterized in that: The method employs a deep learning model to adaptively model the degradation rate function in the continuous degradation model, including constructing the degradation dynamics equation of health indicators, constructing a deep neural network for degradation rate, and solving the neural ordinary differential equation and performing time-continuous modeling. Instantaneous degradation rate Modeling with deep neural networks ; For parameters A deep neural network with a multi-layered nonlinear mapping structure is used to characterize the high-order nonlinear mapping relationship between health indicator status and time variables on the evolution of degradation rate; a degradation rate prediction network is constructed. Its input is , It includes data collected from the rubber hose during its service life, and uses the currently collected data to fit the instantaneous degradation rate. This allows for the prediction of the remaining lifespan of the rubber hose at subsequent points in time. The degradation rate deep neural network structure consists of a multi-layer feature mapping structure, including a feature expansion layer, a nonlinear representation layer, and a degradation constraint output layer. In the feature expansion layer, the input vector The input is fed into the feature expansion layer, where the features are updimensionalized by superimposing weighted linear mappings and bias terms to obtain a high-dimensional state feature representation. This completes the feature expansion of the original input state; In the nonlinear representation layer, the feature vector output by the feature expansion layer is... The input is fed into the first nonlinear mapping layer, and features are reconstructed through a nonlinear activation function to obtain... ; then, The input is fed into the second linear transformation layer for feature combination and reweighting to obtain an intermediate feature representation. Then, it is mapped again through a nonlinear activation function to form a higher-order degenerate feature representation. ;Calculated by continuous superposition of multi-level linear transformations and nonlinear activation functions; In the degradation constraint output layer, higher-order degradation features are... The input is fed into the degradation rate output layer, and intermediate degradation variables are calculated through linear mapping. ; To ensure that the degradation rate satisfies the physical nonnegativity constraint, a nonnegative activation function is applied to the intermediate degradation variables to obtain the final degradation rate modeling result. ; Finally, in solving the ordinary differential equations and modeling the time continuity, the instantaneous degradation rate is obtained by using a deep neural network to calculate the degradation rate. By substituting the differential equation of health index degradation, a continuous-time degradation dynamics model that integrates physical constraints and deep neural networks is constructed.

7. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: The lifespan degradation model employs a recursive update mechanism; during the actual operation of the rubber hose, the model is updated based on the system's future continuous time... The obtained operating data is used to calculate the latest health indicator values. The observed value was compared with the health index predictions obtained from the lifespan degradation model in S3. Compare and calculate the prediction error. ; Based on prediction error The degradation rate is corrected online; the prediction error is introduced into the input of the degradation rate network to construct the corrected degradation rate. Represented as: ； in, This represents the instantaneous degradation rate output by the original lifetime degradation model; This is the error feedback gain coefficient, used to adjust the degree of influence of prediction error on degradation rate correction; At the same time, the prediction error As a monitoring signal, the parameters of the degradation rate deep network are recursively updated.

8. The method for predicting the remaining life of a rubber hose as described in claim 1, characterized in that: The remaining lifetime prediction model includes a temporal convolutional feature extraction layer, a temporal attention weighting layer, and a remaining lifetime regression output layer. First, the health indicator prediction sequence The input is fed into the temporal convolutional feature extraction layer, which consists of three sequentially stacked temporal convolutional layers: the first temporal convolutional layer, the second temporal convolutional layer, and the third temporal convolutional layer. These layers are used to extract short-term, medium-term, and long-term degradation features from the health indicator prediction sequence step by step. The three temporal convolutional layers are connected in series, with the output of the previous layer serving as the input to the next. The input to the first temporal convolutional layer is the health indicator prediction sequence. ; in the Each time convolutional layer (l=1,2,3) feeds back to the previous layer. Output time feature sequence When performing a one-dimensional dilated convolution operation with l=1, the output of the previous layer... Defined as the input health indicator prediction sequence ; Subsequently, the result of the convolution operation is superimposed with the corresponding bias term, and then mapped using the non-linear activation function ReLU to obtain the first... Output of layer-time convolutional features; Secondly, a temporal attention mechanism is introduced after the temporal convolutional layer to construct a temporal attention weighting layer, which adaptively weights key degradation time segments that contribute more to the remaining lifetime; the temporal feature sequence of each time step is calculated through a fully connected layer. attention weights The attention weight vector is obtained by normalizing the vector using the Softmax function. Subsequently, the time feature sequences are weighted and fused to obtain a global degenerate representation. ; Finally, the weighted global degradation features The input is fed into the remaining lifetime regression layer, and the remaining lifetime prediction results are output through a two-stage fully connected network. ; This indicates the remaining operational time of the rubber hose under its current operating condition, from the current moment until its health indicators reach the preset failure threshold. The range of values ​​is ,in, This indicates the maximum service life of the rubber hose under design conditions. This indicates the cumulative service time corresponding to the current operating moment.

Images