Reliability evaluation and residual life prediction method based on multi-source degradation data fusion

Abstract

The invention discloses a reliability evaluation and residual life prediction method based on multi-source degradation data fusion, and belongs to the technical field of life prediction, and the method comprises the steps: collecting degradation data from a plurality of data sources of equipment, and carrying out the preprocessing; data weighted fusion is realized; inputting the fused data into a Weibull model, and calculating to obtain shape parameters and scale parameters of the overall failure time distribution of the equipment, so as to carry out global prediction, including residual life and reliability evaluation of the equipment; performing significance analysis on the global prediction result, and screening out key elements influencing the overall performance of the equipment; and when the global prediction result is lower than a preset threshold value, extracting relevant feature data of the key elements and inputting the relevant feature data into a Weibull model to obtain residual life and reliability prediction of each key element. According to the method, through multi-source data fusion, a global-local prediction architecture and a dynamic adjustment strategy, the prediction precision, the diagnosis capability and the adaptability are effectively improved, and a more reliable basis is provided for preventative maintenance of equipment.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of life prediction, and particularly relates to a reliability evaluation and residual life prediction method based on multi-source degradation data fusion. BACKGROUND

[0002] In modern industrial systems, the reliability evaluation and residual life prediction of equipment are crucial for ensuring production safety, reducing maintenance costs, and improving production efficiency. Traditional reliability evaluation and life prediction methods mainly rely on a single data source. However, these single data sources often have limitations, such as incomplete data, noise interference, or incomplete information coverage, which limit the accuracy and reliability of the prediction results. With the development of industrial Internet of Things and big data technology, more and more equipment generates multi-source data during operation, including sensor data, maintenance records, environmental data, etc. How to effectively fuse these multi-source degradation data to improve the accuracy of reliability evaluation and residual life prediction has become a hot and difficult problem in current research.

[0003] Traditional models assume that the equipment degradation process follows a fixed pattern, such as simple linear degradation or constant failure rate, which does not conform to the actual complex nonlinear degradation process, and the prediction accuracy is limited. Existing technologies lack diagnostic capabilities at the component level, making it difficult to accurately locate the key components that cause overall performance degradation of the equipment, and cannot provide specific guidance for maintenance decisions. In addition, once the model parameters are determined, they remain fixed and cannot be dynamically adjusted according to the actual operating state of the equipment, resulting in lagging prediction results that cannot reflect the true state of the equipment in real time. Existing methods are mostly based on a single model, such as using only a simple regression model, lacking a multi-model fusion mechanism, and unable to fully utilize the advantages of different models to further improve prediction accuracy. Especially when faced with multi-source heterogeneous data, existing technologies lack effective data fusion strategies, making it difficult to integrate data from different sensors and different operating conditions to form a unified equipment health evaluation system.

[0004] To solve the above problems, the present application provides a reliability evaluation and residual life prediction method based on multi-source degradation data fusion. SUMMARY

[0005] Therefore, the present application provides a reliability evaluation and residual life prediction method based on multi-source degradation data fusion, which fuses multi-source data to comprehensively capture equipment degradation characteristics, introduces an adaptive global-local prediction architecture to accurately locate key components and dynamically adjust prediction strategies, and realizes more accurate, reliable, and real-time equipment life prediction, providing strong technical support for preventive maintenance of industrial equipment.

[0006] To achieve the above purpose, the present application provides the following technical solutions: The method for reliability evaluation and residual life prediction based on multi-source degradation data fusion comprises the following steps: S1, multi-source data acquisition: collect degradation data from multiple data sources of the equipment and perform preprocessing; S2, multi-source data fusion: calculate the weight of each data source, and perform weighted fusion on the data of each data source to obtain fused degradation data; S3, global dynamic prediction: input the fused degradation data into the Weibull model to calculate the Weibull shape parameter and scale parameter of the overall failure time distribution of the equipment, and calculate the global prediction result including the residual life and reliability of the overall equipment through the Weibull shape parameter and scale parameter of the overall failure time distribution of the equipment; S4, key element screening: perform significance analysis on the performance data of each element in the fused degradation data, calculate the significance index of each element, and screen out the key elements affecting the overall performance of the equipment; S5, local dynamic prediction: when the global prediction result is less than a preset threshold, extract feature data related to the key elements from the fused degradation data, input the feature data into the Weibull model, calculate the Weibull shape parameter and scale parameter of the failure time distribution of each key element, and calculate the residual life prediction value and reliability of each key element.

[0007] As a further preferred embodiment of the present application, S3 specifically comprises the following steps: S31, input the fused degradation data into the Weibull model, analyze the fused degradation data, and estimate the overall shape parameter and overall scale parameter of the Weibull model using the maximum likelihood estimation method as follows: wherein, is the failure time of the i-th degradation data, is the average failure time of the degradation data, is the number of degradation data; S32, set the failure probability threshold to and calculate the overall residual life RUL and overall reliability of the equipment according to the Weibull model as follows: wherein, is the current running time of the equipment;

[0008] As a further preferred embodiment of the present application, S4 specifically comprises the following steps:​​​​ S41, collect performance data of each element in the device, including historical degradation data and failure mode of the element; S42, use variance analysis to compare whether there is a significant difference in the performance indicators of different elements, take each element as a group, calculate the mean and variance of each group, and then compare between groups, calculate the F value and corresponding p value of each element, wherein the F value is used to measure the ratio of the difference between groups to the difference within groups, and the p value is used to judge whether the difference is significant; S43, set the threshold value of p value as , if the p value of the element is less than , it is determined that the element has a significant impact on the overall performance of the device, and is screened as a key element.

[0009] As a further preferred embodiment of the present application, S5 specifically comprises the following steps: S51, set the threshold values of overall remaining useful life RUL and overall reliability , if the overall remaining useful life RUL or the overall reliability is less than the threshold value, local dynamic prediction is performed; S52, extract relevant feature data of the key element from the fused degradation data; S53, perform Weibull model parameter estimation on the feature data of each key element to obtain the Weibull shape parameter and scale parameter of the failure time distribution of each key element; S54, calculate the remaining life prediction value and reliability of each key element according to the Weibull shape parameter and scale parameter of the failure time distribution of each key element.

[0010] As a further preferred embodiment of the present application, when the global prediction result is less than the set threshold value, and the significance index of the mth element reaches the corresponding , local dynamic prediction of the element is triggered, and the calculation formula of the significance index is as follows: wherein, represents the prediction threshold value of the mth element, is an adjustment coefficient, is the total number of elements, represents the power loss of the mth element, represents the average power loss, is a sensitivity coefficient.

[0011] ​As a further preferred embodiment of the present application, a nonlinear coefficient is introduced in the Weibull model and a time-varying coefficient , the traditional Weibull model is improved to obtain a nonlinear shape parameter and a time-varying scale parameter , and the overall device remaining life and overall reliability and the remaining life prediction value and reliability of each key element are calculated through the corresponding nonlinear shape parameter and time-varying scale parameter , wherein the calculation formula of the nonlinear shape parameter and time-varying scale parameter is as follows: wherein is the initial shape parameter, is or , is a nonlinear coefficient, and t is time; wherein, is the initial scale parameter, is or , and b is a time-varying coefficient.

[0012] As a further preferred embodiment of the present application, S2 specifically comprises the following steps: S21, the information entropy of each data source after preprocessing is calculated, and the weight of each data source is determined, and the calculation formula is: wherein, represents the information entropy of the i th data source, represents the weight of the i th data source; S22, the weight of each data source is weighted and fused with the corresponding data to obtain the fused degradation data; S23, the improved kernel principal component analysis KPCA is performed on the fused degradation data, the key features capable of representing the degradation state of the device are extracted, and the data dimension is reduced.

[0013] The present application at least has the following beneficial effects: 1. Multi-source data fusion improves prediction accuracy: the present application comprehensively captures the performance changes of the device under various working conditions by fusing multi-source degradation data. The weight of each data source is calculated by using information entropy and weighted fusion, the key features and complementary information of multi-source data are retained, compared with the traditional single data source method, the fused data can more accurately reflect the actual degradation state of the device, providing high-quality basis for subsequent prediction, thereby improving the accuracy of reliability evaluation and remaining life prediction.

[0014] 2. Enhanced Diagnostic Capabilities with a Global-Local Predictive Architecture: This innovative architecture combines global and local dynamic prediction. Global prediction assesses the remaining lifespan and reliability of the equipment as a whole, while local prediction precisely targets key components affecting overall equipment performance when the global prediction falls below a set threshold, enabling in-depth diagnosis from the equipment level to the component level. Unlike traditional approaches that focus solely on the overall equipment or uniformly predict all components, this architecture can promptly identify potential equipment failure risks and accurately pinpoint the source of failure, providing maintenance personnel with more targeted maintenance information, effectively improving equipment maintenance efficiency and reducing maintenance costs.

[0015] 3. Enhanced Generalization Ability through Nonlinear Modeling: The traditional Weibull model is improved by introducing nonlinear shape parameters and time-varying scale parameters to describe the complex nonlinear degradation process of equipment and key components. Dynamic optimization of model parameters enables the model to better fit actual degradation curves. Unlike conventional fixed-form degradation models, the improved model possesses stronger generalization ability, applicable to different types of equipment and complex, ever-changing industrial environments, significantly broadening the application scope of the Weibull model and enhancing the versatility and practicality of the prediction method.

[0016] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0017] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of the overall process of the reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion of the present invention. Figure 2 This is a schematic diagram of the S3 process of the reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion of the present invention. Figure 3 This is a schematic diagram of the S4 process of the reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion of the present invention. Figure 4 This is a schematic diagram of the S5 process of the reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion of the present invention. Figure 5 This is a schematic diagram of the S2 process of the reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] like Figures 1-5 As shown, this invention provides a reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion, comprising the following steps: S1. Multi-source data acquisition: Collect degraded data from multiple data sources of the device and perform preprocessing.

[0020] S11. Multi-source data acquisition: In industrial settings, equipment data comes from various types of sensors, maintenance recording systems, and environmental monitoring equipment. This can include collected accelerated stress data under different accelerated stress conditions and real-time degradation data during normal operation.

[0021] S111 Sensor Data Acquisition: Taking a common motor equipment on an industrial production line as an example, the vibration sensor installed on it can capture the vibration amplitude data of the motor in real time. The acquisition frequency can be set according to the operating characteristics of the equipment, such as acquiring 100 data points per second; the temperature sensor can obtain the temperature change during the operation of the motor, and acquires data once per minute; the pressure sensor is used to measure the pressure value of the environment or working load of the motor. The acquisition frequency is also determined according to the actual application scenario, such as acquiring once every 10 seconds.

[0022] S112. Maintenance record system data acquisition: Extract the motor maintenance history from the enterprise's internal equipment maintenance management system, including the specific time of each maintenance, the parts repaired, the reason for the maintenance, etc.; at the same time, obtain the replacement parts record, recording the name of the replaced parts, the replacement time, and the running time of the equipment during the replacement, etc.

[0023] S113. Environmental monitoring equipment data acquisition: Obtain humidity data of the working environment from environmental monitoring equipment installed in the equipment operating area, once per hour; dust concentration data can be collected once every half hour. These data are of great significance for assessing the degradation rate of the equipment under specific environments.

[0024] S12, Data Preprocessing S121. Data Cleaning: For data acquired by sensors, filtering algorithms are used to remove high-frequency noise interference. For example, a low-pass filter is used, with a suitable cutoff frequency set, such as 10Hz, to effectively filter out noise signals above this frequency, retaining the low-frequency signals reflecting the true operating state of the equipment. Statistical methods are used to detect and remove outlier data points. Taking temperature sensor data as an example, its mean and standard deviation are calculated. A threshold can typically be set at the mean ± 3 times the standard deviation; data points exceeding this range are considered outliers and removed.

[0025] S122. Data Normalization: Data collected from different data sources is normalized to ensure they have the same dimensions and numerical range. The min-max normalization method is used to normalize all data to the interval [0, 1]. For the original data x, the normalized value x' is calculated as: x' = (x - xmin) / (xmax - xmin), where xmin and xmax are the minimum and maximum values ​​collected from the data source, respectively.

[0026] S123. Time Alignment: Due to inconsistent timestamps from different data sources, time alignment is required. Using the vibration sensor's timestamp as a reference, interpolation is performed on data from other data sources. For example, for missing data from a temperature sensor at a certain moment, the temperature value at that moment can be calculated using linear interpolation based on data values ​​from adjacent time points, ensuring consistency with the vibration sensor's data in the time dimension.

[0027] S2. Multi-source data fusion: Calculate the weight of each data source and perform weighted fusion of the data from each data source to obtain the fused degraded data.

[0028] S2 optimization includes the following steps: S21. Calculate the information entropy of each data source after preprocessing and determine the weight of each data source. Information entropy reflects the degree of uncertainty of the information contained in the data source. The larger the information entropy, the more uncertain the information contained in the data source, and the smaller the weight; conversely, the smaller the information entropy, the more certain the information contained in the data source, and the larger the weight.

[0029] The formula for calculating the weight is: in, This represents the information entropy of the i-th data source. This represents the weight of the i-th data source, where n is the number of data sources; Assume there are three data sources after preprocessing: vibration sensor data source A, temperature sensor data source B, and humidity sensor data source C. For data source A, count the frequency of each data value and then calculate the probability of each data value. For example, if data source A has 1000 data values, and a specific data value 5.2 appears 50 times, then its probability of occurrence is 50 / 1000 = 0.05. Based on the information entropy formula: ,in, Let be the probability of the j-th data value appearing in data source A, and let be the probability of the j-th data value appearing in all distinct data values ​​in data source A. By calculating and summing the results, we can obtain the information entropy H(A) of data source A. Similarly, we can calculate the information entropy H(B) and H(C) of data sources B and C, respectively.

[0030] S22. Multiply the weight of each data source by its corresponding data, and then add the weighted data from all data sources together to obtain the fused degraded data.

[0031] The weights are calculated based on the information entropy of each data source. Assuming the information entropies of data sources A, B, and C are H(A) = 0.5, H(B) = 0.3, and H(C) = 0.6 respectively, and the number of data sources n = 3, then the weight calculation formula is: , Substitute the numerical values ​​into the calculation: Similarly, .

[0032] The weight of each data source is multiplied by its corresponding data, and then the results are summed to obtain the fused degraded data. For example, at a certain time t, the data value of vibration sensor data source A is a(t), the data value of temperature sensor data source B is b(t), and the data value of humidity sensor data source C is... Then the degraded data d(t) after fusion is: Substituting the above weight values, we get: S23. Improved Kernel Principal Component Analysis (KPCA) is applied to the fused degraded data to extract key features characterizing the equipment's degradation state and reduce data dimensionality. A suitable kernel function, such as a Gaussian kernel or a polynomial kernel, is selected to map the fused degraded data into a high-dimensional feature space. In this high-dimensional feature space, the covariance matrix of the data is calculated, and its eigenvalues ​​and eigenvectors are solved. The dominant eigenvectors are selected based on the magnitude of the eigenvalues ​​to form a projection matrix. The fused degraded data is then projected onto a low-dimensional feature space using this projection matrix to obtain the extracted key features, thus reducing data dimensionality.

[0033] For example: The Gaussian kernel function is chosen as the kernel function for KPCA, and its expression is: Where xi and xj are the fused degenerate data vectors, and σ is the kernel function width parameter, which can be determined based on data characteristics through methods such as cross-validation. The fused degenerate data is mapped to a high-dimensional feature space using a Gaussian kernel function, and the covariance matrix of the data in this space is calculated. The eigenvalues ​​and eigenvectors of the covariance matrix are solved, and the eigenvectors corresponding to the largest m eigenvalues ​​are selected to form the projection matrix.

[0034] The fused degradation data is projected onto a low-dimensional feature space using a projection matrix to obtain the extracted key features, thereby reducing the data dimensionality and retaining the key information in the data that characterizes the device degradation state.

[0035] S3. Global Dynamic Prediction: Input the fused degradation data into the Weibull model to calculate the Weibull shape parameters and scale parameters of the overall failure time distribution of the equipment. Calculate the global prediction results using the Weibull shape parameters and scale parameters of the overall failure time distribution of the equipment, including the remaining life and reliability of the overall equipment.

[0036] S3 optimization includes the following steps: S31. Input the fused degenerate data into the Weibull model, analyze the fused degenerate data, and use the maximum likelihood estimation method to estimate the overall shape parameters of the Weibull model. and overall scale parameters The principle of maximum likelihood estimation is to find the model parameter values ​​that maximize the probability of a given data point occurring. The Weibull model can also be trained using LSTM.

[0037] S311. Constructing the likelihood function: Assuming there are n samples in the fused degraded data, and the failure time of each sample is ti (i = 1, 2, ..., n), then the likelihood function of the Weibull model is... for: Taking the logarithm of the likelihood function yields the log-likelihood function: ; S312. Solving for the parameters: Estimate the parameters k and λ by taking the derivative and setting it to zero. Take the partial derivative with respect to k: Take the partial derivative with respect to λ: Solving these two equations together yields estimates of the overall shape parameter k and the overall scale parameter λ.

[0038] Received and The simplified formula is as follows: in, It is the first The expiration time of each degraded data point. It is the average time to failure of degraded data. It is the amount of degraded data; S32. Based on the equipment's operating requirements and safety standards, set the failure probability threshold as follows: For example, for some critical equipment, it can be The threshold is set to 0.1, meaning that when the probability of equipment failure reaches 10%, maintenance or replacement is considered necessary. The overall remaining life (RUL) and overall reliability of the equipment are then calculated based on the Weibull model. As shown below: in, The current running time of the device;

[0039] S4. Key Component Screening: Perform significance analysis on the performance data of each component in the fused degradation data, calculate the significance index of each component, and screen out the key components that affect the overall performance of the equipment.

[0040] S4 optimization includes the following steps: S41. Collect performance data of each component in the equipment, including historical degradation data and failure modes of the components.

[0041] Taking electric motor equipment as an example, its key components may include bearings, windings, rotors, etc.

[0042] Bearing components: Collect historical vibration data, temperature data, and failure modes (such as fatigue spalling, wear, etc.) of the bearing. For example, vibration amplitude data, temperature change data, and failure records caused by fatigue spalling during the bearing's past operation.

[0043] Winding elements: Collect historical data on insulation resistance, temperature, and failure modes (such as short circuit, open circuit, etc.) of the windings. The insulation resistance measurement data can reflect the aging degree of the windings, while the temperature data is closely related to the thermal aging process of the windings.

[0044] Rotor components: Collect dynamic balance data, speed fluctuation data, and failure modes (such as deformation, cracks, etc.) of the rotor. Dynamic balance data can reflect the operating stability of the rotor, while speed fluctuation data may be related to the mechanical state of the rotor.

[0045] S42. Use analysis of variance to compare whether there are significant differences in the performance indicators of different components. Treat each component as a group, calculate the mean and variance of each group, and then perform inter-group comparisons. Calculate the F-value and corresponding p-value for each component. The F-value is used to measure the ratio of inter-group difference to intra-group difference, and the p-value is used to determine whether the difference is significant.

[0046] Taking the performance data of three components—bearing, winding, and rotor—as an example, it is assumed that 20 performance data points for each component at different time points were collected.

[0047] For the data set of bearing elements, calculate its mean μ_bearing and variance σ²_bearing; for the data set of winding elements, calculate its mean μ_winding and variance σ²_winding; for the data set of rotor elements, calculate its mean μ_rotor and variance σ²_rotor.

[0048] Between-group comparisons: Calculate the F-value and p-value for each component. The F-value is calculated as follows: F = [S²_between / S²_within] where S²_between is the between-group mean square, calculated as: S²_between = Σn_i×(μ_i - μ_total)² / (a - 1). S²_within is the within-group mean square, calculated as: S²_within = ΣΣ(x_ij - μ_i)² / (N - a) where a is the number of component groups (here a = 3), n_i is the number of data in the i-th component group, μ_i is the mean of the i-th component group, μ_total is the total mean of all data, x_ij is the j-th data in the i-th component group, and N is the total number of data (here N = 3×20 = 60).

[0049] After calculating the F value, the corresponding p value is found according to the F distribution table and the degrees of freedom (a - 1, N - a).

[0050] S43. Set the threshold value of p as... If the p-value of the component is less than If the component has a significant impact on the overall performance of the equipment, it is determined to be a key component.

[0051] Significance assessment: Set a threshold α for the p-value, typically α = 0.05. If the p-value of a component is less than α, then the component is considered to have a significant impact on the overall performance of the equipment and is selected as a critical component. For example, if the calculated p-value for a bearing component is 0.03 < α = 0.05, then the bearing is a critical component; if the p-value for a winding component is 0.06 > α = 0.05, then the winding is not a critical component; and if the p-value for a rotor component is 0.04 < α = 0.05, then the rotor is a critical component.

[0052] As another preferred embodiment of the present invention, when the global prediction result is less than a set threshold, and the significance index of the m-th element reaches the corresponding threshold... At that time, a local dynamic prediction of the element is triggered, the The calculation formula is as follows: in, This represents the prediction threshold for the m-th element. To adjust the coefficient, The total number of components, This represents the power loss of the m-th element. Indicates average power loss. This is the sensitivity coefficient.

[0053] S5. Local dynamic prediction: When the global prediction result is less than the preset threshold, the feature data related to the key components are extracted from the fused degradation data, input into the Weibull model, and the Weibull shape parameters and scale parameters of the failure time distribution of each key component are calculated. The remaining lifetime prediction value and reliability of each key component are also calculated.

[0054] S5 specifically includes the following steps: S51, Set Overall Remaining Life (RUL) and Overall Reliability The threshold, if the overall remaining lifetime RUL or overall reliability If the value is less than the threshold, local dynamic prediction is performed. S52. Extract relevant feature data of key components from the fused degradation data; taking the selected bearings and rotors as key components as examples, extract feature data related to bearing vibration, temperature, rotor dynamic balance, and speed fluctuation from the fused degradation data.

[0055] S53. Estimate the Weibull model parameters for the characteristic data of each key component to obtain the Weibull shape parameters of the failure time distribution of each key component. and scale parameters ; S54. Weibull shape parameters based on the failure time distribution of each critical component. and scale parameters Calculate the predicted remaining lifetime value for each critical component. and reliability .

[0056] The traditional Weibull model describes the failure distribution of equipment using the shape parameter k and the scale parameter λ. However, it assumes a fixed distribution pattern for equipment degradation, making it difficult to adapt to the complex nonlinear degradation processes of real-world equipment. This invention improves the traditional Weibull function by introducing nonlinear shape parameters and time-varying scale parameters: nonlinear coefficients are introduced into the Weibull model. and time-varying coefficients An improvement to the traditional Weibull model yields nonlinear shape parameters. and time-varying scale parameters And through the corresponding nonlinear shape parameters and time-varying scale parameters Calculate the remaining life and overall reliability of the entire equipment, as well as the predicted remaining life and reliability of each critical component, where nonlinear shape parameters are considered. and time-varying scale parameters The calculation formula is as follows: Where is the initial shape parameter. for or , Here, t represents the nonlinear coefficient, and t represents time. in, These are the initial scale parameters. for or b is a time-varying coefficient.

[0057] The advantages of introducing nonlinear shape parameters and time-varying scaling parameters are: Adapting to complex degradation trends: By introducing nonlinear coefficients, the model can capture the nonlinear characteristics of degradation rate changing over time. For example, in the early stages of equipment operation, the degradation rate may be slow, but it gradually accelerates over time. Nonlinear shape parameters can enable the model to fit this trend more accurately.

[0058] Dynamically reflecting scale changes: The introduction of time-varying coefficients allows the scale parameters to adjust over time. This reflects the changing dispersion of failure time distribution at different operational stages. For example, during the break-in period, failure times may be relatively concentrated with low dispersion; while during the aging period, due to various factors, the dispersion of failure times increases. Time-varying scale parameters can better reflect this dynamic change, thereby improving the model's accuracy in fitting real-world data.

[0059] Improving the accuracy of remaining service life prediction: By calculating nonlinear shape parameters and time-varying scale parameters, the model can more accurately predict the remaining service life of equipment and components. Taking rotating machinery as an example, during its degradation process, due to the influence of complex mechanical forces and environmental factors, degradation exhibits obvious nonlinear characteristics. Using the improved Weibull model, the remaining service life of mechanical components at different stages can be estimated more accurately, providing a more reliable basis for preventive maintenance and reducing downtime and maintenance costs caused by unexpected failures.

[0060] Enhanced Reliability Assessment: The improved model more accurately reflects the reliability level of equipment and components under actual operating conditions. Traditional Weibull models may underestimate or overestimate equipment reliability when faced with complex degradation processes. The introduction of nonlinear and time-varying coefficients allows the model to dynamically adjust reliability assessment results based on the equipment's real-time status and historical degradation data, providing a more accurate reference for equipment operation decisions and risk management.

[0061] This invention first integrates information from various data sources of the equipment through multi-source data fusion technology, comprehensively capturing the performance changes of the equipment under complex operating conditions. Compared with traditional single-data-source methods, it can more accurately reflect the actual degradation state of the equipment, improve prediction accuracy, and provide a high-quality basis for reliability assessment and remaining life prediction.

[0062] Secondly, the construction of the global-local prediction architecture enables this method not only to grasp the remaining lifespan and reliability of the equipment at an overall level, but also to accurately locate key components affecting equipment performance when the global prediction results are abnormal, achieving in-depth diagnosis from the equipment level to the component level. This helps to promptly identify potential failure risks and accurately locate the source of failure, providing maintenance personnel with more targeted maintenance information, effectively improving equipment maintenance efficiency and reducing maintenance costs, changing the traditional mode of only focusing on the overall equipment or uniformly predicting all components, and optimizing maintenance strategies.

[0063] Furthermore, the traditional Weibull model is improved by introducing nonlinear shape parameters and time-varying scale parameters, enabling it to better describe the complex nonlinear degradation process of equipment and key components. This dynamic optimization of model parameters allows the model to better fit the actual degradation curve, enhances its generalization ability, broadens its application scope, and is suitable for different types of equipment and complex and ever-changing industrial environments. This significantly improves the versatility and practicality of the prediction method and solves the problem that conventional fixed-form degradation models do not match the actual complex degradation process.

[0064] Furthermore, the dynamic adjustment strategy of this invention can adjust the prediction in real time according to the actual operating status of the equipment, ensuring that the prediction results reflect the true condition of the equipment in a timely manner. The set threshold can trigger local dynamic prediction in a timely manner, re-evaluate the equipment and key components, avoid prediction lag, cope with the uncertainty of equipment operating status, improve the real-time performance and adaptability of prediction, and ensure that the prediction results are accurate and reliable.

[0065] In summary, this invention significantly improves the accuracy, diagnostic capabilities, and adaptability of equipment reliability assessment and remaining life prediction, providing reliable technical support for preventive maintenance of industrial equipment and yielding significant economic and social benefits.

[0066] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion, characterized in that, Includes the following steps: S1. Multi-source data acquisition: Collect degraded data from multiple data sources of the device and perform preprocessing; S2. Multi-source data fusion: Calculate the weight of each data source and perform weighted fusion of the data from each data source to obtain the fused degraded data; S3. Global Dynamic Prediction: Input the fused degradation data into the Weibull model to calculate the Weibull shape parameters and scale parameters of the overall failure time distribution of the equipment. Calculate the global prediction results, including the remaining life and reliability of the overall equipment, using the Weibull shape parameters and scale parameters of the overall failure time distribution of the equipment. S4. Key Component Screening: Perform significance analysis on the performance data of each component in the fused degradation data, calculate the significance index of each component, and screen out the key components that affect the overall performance of the equipment. S5. Local dynamic prediction: When the global prediction result is less than the preset threshold, the feature data related to the key components are extracted from the fused degradation data, input into the Weibull model, and the Weibull shape parameters and scale parameters of the failure time distribution of each key component are calculated. The remaining lifetime prediction value and reliability of each key component are also calculated.

2. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 1, characterized in that: S3 specifically includes the following steps: S31. Input the fused degenerate data into the Weibull model, analyze the fused degenerate data, and use the maximum likelihood estimation method to estimate the overall shape parameters of the Weibull model. and overall scale parameters As shown below: in, It is the first The expiration time of each degraded data point. It is the average time to failure of degraded data. It represents the amount of degraded data; S32. Set the failure probability threshold as follows: And calculate the overall remaining life (RUL) and overall reliability of the equipment based on the Weibull model. As shown below: in, The current running time of the device; 。 3. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 1, characterized in that: S4 specifically includes the following steps: S41. Collect performance data of each component in the equipment, including historical degradation data and failure modes of the components; S42. Use analysis of variance to compare whether there are significant differences in the performance indicators of different components. Treat each component as a group, calculate the mean and variance of each group, and then perform inter-group comparisons. Calculate the F-value and corresponding p-value for each component. The F-value is used to measure the ratio of inter-group difference to intra-group difference, and the p-value is used to determine whether the difference is significant. S43. Set the threshold value of p as... If the p-value of the component is less than If the component has a significant impact on the overall performance of the equipment, it is determined to be a key component.

4. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 1, characterized in that: S5 specifically includes the following steps: S51, Set Overall Remaining Life (RUL) and Overall Reliability The threshold, if the overall remaining lifetime RUL or overall reliability If the value is less than the threshold, local dynamic prediction is performed. S52. Extract relevant feature data of key components from the fused degradation data; S53. Estimate the Weibull model parameters for the characteristic data of each key component to obtain the Weibull shape parameters of the failure time distribution of each key component. and scale parameters ; S54. Weibull shape parameters based on the failure time distribution of each critical component. and scale parameters Calculate the predicted remaining lifetime value for each critical component. and reliability .

5. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 1, characterized in that: When the global prediction result is less than the set threshold, and the significance index of the m-th element reaches the corresponding threshold... At that time, a local dynamic prediction of the element is triggered, the The calculation formula is as follows: in, This represents the prediction threshold for the m-th element. To adjust the coefficient, The total number of components, This represents the power loss of the m-th element. Indicates average power loss. This is the sensitivity coefficient.

6. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 2 or 4, characterized in that: Introducing nonlinear coefficients into the Weibull model and time-varying coefficients An improvement to the traditional Weibull model yields nonlinear shape parameters. and time-varying scale parameters And through the corresponding nonlinear shape parameters and time-varying scale parameters Calculate the remaining life and overall reliability of the entire equipment, as well as the predicted remaining life and reliability of each critical component, where nonlinear shape parameters are considered. and time-varying scale parameters The calculation formula is as follows: Where is the initial shape parameter. for or , Here, t represents the nonlinear coefficient, and t represents time. in, These are the initial scale parameters. for or b is a time-varying coefficient.

7. The reliability assessment and remaining lifetime prediction method based on multi-source degradation data fusion according to claim 1, characterized in that: S2 specifically includes the following steps: S21. Calculate the information entropy of each data source after preprocessing and determine the weight of each data source. The calculation formula is as follows: in, This represents the information entropy of the i-th data source. This represents the weight of the i-th data source; S22. The weight of each data source is weighted and fused with its corresponding data to obtain the fused degraded data; S23. Improved kernel principal component analysis (KPCA) is performed on the fused degradation data to extract key features that characterize the degradation state of the equipment and reduce the data dimensionality.

Images