Relevance Vector Machine Regression Research Articles

Machine learning-based acoustic emission technique for corrosion-induced damage monitoring in reinforced concrete structures

Early-stage detection of corrosion in reinforcement embedded inside concrete, generally considered as the major problem in structures under extreme marine environments, can help to schedule timely maintenance/repair and to avoid any catastrophic failure. However, quantitative evaluation of the degree of deterioration, especially at a very early stage, i.e., before its appearance on the surface, is extremely difficult due to the complex process in heterogeneous material. This study aims to characterize the critical damage stages in reinforced concrete specimens under accelerated conditions using the Acoustic emission (AE) damage quantification methods supported by machine learning (ML) techniques. The stages of corrosion-induced damage are predicted by a two-step predictive model that involve both regression and classification tasks. Three supervised regression models — support vector regression (SVR), relevance vector machine (RVM) and kernel ridge regression (KRR) —are first adopted to evaluate the amount of current. Then the output from regression model is given as an input to the classification problem to predict the discrete class label output. Characterized damage stages are validated based on mass loss and crack width measurements. The results show that the corrosion initiation (at day 3) and crack formation (at day 7) can be determined from the sudden change in acoustic parameters. Further, a double layer (supervised-unsupervised) model is proposed which is found be more effective in correctly detecting the progression of damage. The findings indicate that the proposed approach that combines corrosion detection by analysing AE signals with ML technique can accurately and robustly predict corrosion severity.

Prediction of UCS of fine-grained soil based on machine learning part 2: comparison between hybrid relevance vector machine and Gaussian process regression

Prediction of UCS of fine-grained soil based on machine learning part 2: comparison between hybrid relevance vector machine and Gaussian process regression

Machine learning based swift online capacity prediction of lithium-ion battery through whole cycle life

The machine learning (ML) based methods show great promise for the online battery capacity prediction. However, the feasibility and simplicity of indispensable feature extraction are both challenging to existing ML based methods, which restrict the implementation in electric vehicle usages. To reduce the computational burden and simultaneously enhance the accuracy of prediction results, this paper proposes a ML based approach for swift capacity prediction leveraging fractional charging voltage segments. A total number of 21 voltage feature segments (VFSs) are intercepted with different voltage ranges to analyze the influence of voltage intervals on the capacity prediction. Meanwhile, three advanced ML models including random forest regression (RFR), relevance vector machine (RVM) and Gaussian process regression (GPR) are meticulously designed to build a mapping function between the intercepted VFSs and battery capacity. The battery capacity prediction can be attained subsequently based on the established matched relationship using the VFS captured in real-time. The experimental and contrastive results show that the best model can accurately predict battery capacity through whole cycle life with the maximum average relative error of only 1.95%. This work emphasizes the application potential of combining straightforward and reliable feature with ML algorithms for online battery capacity prediction.

Incremental Capacity Curve Health-Indicator Extraction Based on Gaussian Filter and Improved Relevance Vector Machine for Lithium–Ion Battery Remaining Useful Life Estimation

Accurate prediction of the remaining useful life (RUL) of lithium–ion batteries is the focus of lithium–ion battery health management. To achieve high–precision RUL estimation of lithium–ion batteries, a novel RUL prediction model is proposed by combining the extraction of health indicators based on incremental capacity curve (IC) and the method of improved adaptive relevance vector machine (RVM). First, the IC curve is extracted based on the charging current and voltage data. To attenuate the noise effects on the IC curve, Gaussian filtering is used and the optimal filtering window is determined to remove the noise interference. Based on this, the peak characteristics of the IC curve are analyzed and four groups of health indicators are extracted, and the strong correlation between health indicators and capacity degradation is determined using Pearson correlation analysis. Then, to optimize the traditional fixed kernel parameter RVM model, an RVM regression model whose kernel parameters are optimized by the Bayesian algorithm is established. Finally, four sets of datasets under CS2 battery in the public dataset of the University of Maryland are carried out for experimental validation. The validation results show that the improved RVM model has better short–term prediction performance and long–term prediction stability, the RUL prediction error is less than 20 cycles, and the mean absolute error is less than 0.02. The performance of the improved RVM model is better than that of the traditional RVM model.

Rheological properties of cemented paste backfill and the construction of a prediction model

The Cemented Paste Backfill (CPB) yield stress is a key rheological parameter for paste filling technology, which has significant practical value for pipeline optimization and equipment selection of pipeline conveying systems. However, the slurry yield stress is affected by many factors. In order to accurately analyze and predict the CPB yield stress, this study uses the sparrow search algorithm to optimize the relevance vector machine (SSA-RVM) and proposes the prediction model of SSA-RVM CPB yield stress regression. Based on 136 sets of rheological tests for copper mine, different waste rock/tailing sand ratios, mass concentrations, and water-cement ratios select to predict the yield stress at different training set ratios (78%, 85%, 92%). Compared with the traditional relevance vector machine (RVM) regression model, the SSA-RVM regression prediction model has higher prediction accuracy. In addition, the coefficient of determination R2 of the predicted and true values obtained from the SSA-RVM model increased by 0.0407, 0.0438, and 0.0500 for the training set ratios of 78%, 85%, and 92%, respectively. The results suggested that SSA-RVM can efficiently predict the CPB yield stress, which can be a reference for the design of paste-filled pipe conveying systems.

An optimal relevance vector machine with a modified degradation model for remaining useful lifetime prediction of lithium-ion batteries

Predicting the remaining useful life (RUL) of lithium-ion batteries (LIBs) is vital to ensure the safety and reliability of battery-powered systems. Owing to its ability to represent uncertainties in predictions, relevance vector machine (RVM) regression is effective for short-term predictions, but it has low accuracy for long-term predictions. To improve its performance, an integrated RUL prediction method is proposed based on optimal relevance vectors (RVs) and a modified degradation model (MDM) with the Hausdorff distance (HD), i.e., ORV-MDMHD. First, phase space reconstruction is introduced to produce the inputs to the RVM, which can enhance the long-range dependence of the capacity data for prediction. Then, for RVM regression, its critical parameter, the kernel width, is set to a numerical range rather than a single value. As a result, a series of RVM models is trained, and different sets of RVs are then obtained to represent the original large-scale training data. Moreover, an MDM is designed to fit each set of RVs, which generates one of degradation curves to simulate the possible battery aging process. Then, a curve similarity measure, the HD, is used to select the optimal fitted curve that is most similar to the actual degradation curve. Finally, the predicted RUL of the battery can be calculated by extrapolating the optimal curve to the failure threshold. The experimental results for two cases of batteries indicate that the proposed prediction method can provide more stable prediction with higher accuracy, especially for long-term prediction of the LIB RUL.

An Integrated Method for Bearing State Change Identification and Prognostics Based on Improved Relevance Vector Machine and Degradation Model

Bearings are classified as one of the most safety-critical components in industrial machinery, and their prognostics have proven to be an efficient way to reduce costly unplanned maintenance and guarantee reliability and safety. However, accurate data-driven remaining useful life (RUL) prediction remains a significant challenge because of nonlinear degradation and prediction uncertainty. In this study, an integrated prognostics method is designed for a rolling element bearing to identify its health state changes adaptively and further improve the accuracy of RUL prediction, particularly for its long-term prediction. First, a new health indicator (HI) is proposed based on fuzzy <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$c$ </tex-math></inline-formula> -means clustering and a modified confidence value (CV). The former makes the HI sensitive to state changes, and the latter improves its monotonicity and smoothness. A relevance vector machine (RVM) regression model functions with multikernel widths with a modified degradation model to improve the accuracy of RUL prediction, even when fewer samples are available for long-term prediction. The experimental results of the two case studies indicate that the proposed HI provides a better representation for bearing monotonic and smooth degradation processes while automatically identifying its first predicting time and failure threshold. Moreover, the proposed RUL prediction method exhibits stable performance for various prediction ranges and higher accuracy for bearings with different degradation rates.

Early Warning Scheme of COVID-19 related Internet Public Opinion based on RVM-L Model

Internet public opinion is affected by many factors corresponding to insufficient data in the very short period, especially for emergency events related to the outbreak of coronavirus disease 2019 (COVID-19). To effectively support real-time analysis and accurate prediction, this paper proposes an early warning scheme, which comprehensively considers the multiple factors of Internet public opinion and the dynamic characteristics of burst events. A hybrid relevance vector machine and logistic regression (RVM-L) model is proposed that incorporates multivariate analysis, which adopts Lagrange interpolation to fill in the gaps and improve the forecasting effect based on insufficient data for COVID-19-related events. In addition, a novel metric critical interval is introduced to improve the early warning performance. Detailed experiments show that compared with existing schemes, the proposed RVM-L-based early warning scheme can achieve the prediction accuracy up to 96%, and the intervention within the critical interval can reduce the number of public opinions by 60%.

Crack Prediction Based on Wavelet Correlation Analysis Least Squares Support Vector Machine for Stone Cultural Relics

Preventive protection of cultural relics is to make use of all the science and technology beneficial to the research and protection of archaeological heritage to predict the disease of cultural relics. The existing preventive cultural relics protection system has made some achievements in environmental monitoring, but the analysis and utilization of large data of cultural relics are still insufficient. In this paper, under the idea of multisource information fusion, a least squares support vector machine regression method based on multivariate time series wavelet correlation analysis is proposed to achieve accurate crack prediction of stone cultural relics. Firstly, the correlation of multivariate time series of stone cultural relics are quantitatively analyzed and the validity of characteristic variables of the crack is discriminated by wavelet correlation analysis; then, a least squares support vector machine prediction model is constructed based on the correlation obtained from the analysis; finally, the good performance of the method is verified by using the environmental monitoring data of the rock mass fracture in the North Qianfo Cliff of Dafo Temple in Binzhou City of Shaanxi Province. The experimental results show that the proposed method is more effective than the traditional backpropagation neural network, support vector machine, and relevance vector machine regression methods. This method is universal and easy to implement for multisource data prediction of nonmovable cultural relics diseases. It provides a scientific theoretical reference for the preventive protection of cultural relics.

A Non-destructive Ultrasonic Testing Approach for Measurement and Modelling of Tensile Strength in Rubbers

Currently, non-destructive testing is widely used to investigate various mechanical and structural properties of materials. In the present study, non-destructive ultrasonic testing was applied to study the relationship between the tensile strength value and the velocity of longitudinal ultrasonic waves. For this purpose, fourteen specimens of composites with different formulations were prepared. The tensile strength of the composites and the velocity of longitudinal ultrasonic waves inside them was measured. The relevance vector machine regression analysis, as a new methodology in supervised machine learning, was used to define a mathematical expression for the functional relationship between the tensile strength and the velocity of longitudinal ultrasonic waves. The accuracy of the mathematical expression was tested based on standard statistical indices, which proved the expression to be an efficient model. Based on these results, the developed model has the capability of being used for the online measurement of the tensile strength of rubber with the proposed formulation in the rubber industry.

Analysis and modeling of ground motion coherency at uniform site conditions

In this work, nonstationary ground motion coherency analysis and modeling are performed using wavelet analysis and relevance vector machine regression. Earthquake ground motion data from four events recorded at dense seismograph SMART-1 array in north-south and east-west horizontal directions are used to investigate the lagged coherency behavior. Wavelet transform is used to compute the nonstationary lagged coherency that characterizes the space-time variation of seismic ground motion. Lagged coherency behavior with frequency, distance, and time is examined. It is shown that the lagged coherency is a time-varying quantity. The results implied that the lagged coherency on uniform soil is relatively insensitive to the earthquake events, and depends mainly on time, frequency, and separation distance. Relevance Vector Machines (RVM) regression is used to develop a model for nonstationary lagged coherency that characterizes the space-time variation of seismic ground motion. Earthquake ground motion in the in the north-south and the first principal directions is used to determine the lagged coherency model. The proposed model does not require a fixed parametric functional form. The developed model can estimate the lagged coherency for different separation distances at different time instants and frequencies.

Wind Turbine Pitch System Condition Monitoring and Fault Detection Based on Optimized Relevance Vector Machine Regression

Condition monitoring and early fault detection of wind turbine faults can reduce maintenance costs and prevent cascaded failures. This article proposes a new Normal Behavior Modeling (NBM) method to predict wind turbine electric pitch system failures using supervisory control and data acquisition (SCADA) information. The proposed method is particularly effective for online monitoring applications at a reasonable computational complexity. Briefly, in the data preprocessing stage of the proposed method, in order to remove interferential information and improve data quality, the operational state codes from turbine programmable logic controller are applied to filter SCADA data. In the modeling process, we designed a NBM method using optimized relevance vector machine (RVM) regression, which is relatively fast and computationally efficient. An adaptive threshold by the probabilistic output of RVM is proposed and used as the rule of anomaly detection. One normal case and three typical fault cases have been studied to demonstrate the feasibility of the proposed method. The performance of the method is assessed using 38 actual pitch system faults compared with two existing methods.

Adaptive relevant vector machine based RUL prediction under uncertain conditions

Engineering systems often suffer with many uncertainties during their performance degradation processes, such as the inherent uncertainties associated with the degradation progression over time and the inevitable uncertainties caused by change of loading, operation and usage conditions. In order to improve the accuracy of remaining useful life (RUL) prediction, this study takes these common uncertainties into consideration via an improved relevance vector machine (RVM) approach, which can describe accurately the degradation process from fault to failure. Firstly, based on historical data, a multi-step RVM regression model is established offline, in which the uncertainties are represented by the variances of Gaussian distributions of parameters and then are quantified as time-varying variables. Then, an adaptive RVM model is trained and the time-varying variables are updated by the expectation–maximization (EM) algorithm. For on-line prediction, given the real-time data, the RUL is forecasted by the first hitting time (FHT) method in probability perspective. The proposed method is demonstrated by two case studies on a high-speed train’s traction system. The results can show the effectiveness of the proposed method.

The Probabilistic Model and Forecasting of Power Load Based on Variational Bayesian Expectation Maximization and Relevance Vector Machine

As the surging demands of secure power supply and reliable power system, the power load approximation and forecasting are becoming more significant and more important. Different from the current research work, we integrate power load approximation and forecasting based on the Gaussian mixture model and relevance vector machine. In order to estimate the parameters of GMM, the variational bayesian expectation maximization algorithm are employed. Based on the estimation results, the relevance vector machine and bayesian regression model are built to forecast the power load and its labels. The simulation results show that the proposed algorithms can closely approximate the power load profiles and forecast the power load with lower errors.

A new hybrid method for the prediction of the remaining useful life of a lithium-ion battery

The lithium-ion battery has become the main power source of many electronic devices, it is necessary to know its state-of-health and remaining useful life to ensure the reliability of electronic device. In this paper, a novel hybrid method with the thought of error-correction is proposed to predict the remaining useful life of lithium-ion battery, which fuses the algorithms of unscented Kalman filter, complete ensemble empirical mode decomposition (CEEMD) and relevance vector machine. Firstly, the unscented Kalman filter algorithm is adopted to obtain a prognostic result based on an estimated model and produce a raw error series. Secondly, a new error series is constructed by analyzing the decomposition results of the raw error series obtained by CEEMD method. Finally, the new error series is utilized by relevance vector machine regression model to predict the prognostic error which is adopted to correct the prognostic result obtained by unscented Kalman filter. Remaining useful life prediction experiments for batteries with different rated capacities and discharging currents are performed to show the high reliability of the proposed hybrid method.

Prognostics-Based LED Qualification Using Similarity-Based Statistical Measure With RVM Regression Model

Light-emitting diodes (LEDs) are widely used for general lighting and display applications. As the demand for LEDs has grown, the need to quickly qualify them has emerged. To address this issue, this paper introduces a prognostics-based qualification method using an efficient relevance vector machine (RVM) regression model that reduces the qualification testing time of LEDs from 6000 h (as recommended by industry standards) to 210 h. The developed method predicts LED remaining useful life (RUL) by calculating the accumulated sum of products of similarity weights and historical LED RUL values at the 210th hour. Specifically, a similarity weight, defined as the degree of affinity between two different LED's degradation trends, is derived from the difference between a test unit's degradation trend and a training unit's degradation trend. Likewise, the RVM is used to represent a unit's degradation behavior and facilitates the reduction of unit-to-unit variations by precisely capturing transient degradation dynamics.

Prediction and synthesis of novel layered double hydroxide with desired basal spacing based on relevance vector machine

A Quantitative Structure Property Relationship (QSPR) model for the basal spacing of layered double hydroxide is developed in the present work by using generic algorithms feature selection, and relevance vector machine regression. The relative error of the developed model is 0.78% in cross-validation. Then, the QSPR model is applied to recommend a LDH with desired basal spacing for synthesis. The synthesized LDH meets the design requirement, thereby confirming the prediction power of the developed model.

Does Nonlinear Modeling Play a Role in Plasmid Bioprocess Monitoring Using Fourier Transform Infrared Spectra?

The monitoring of biopharmaceutical products using Fourier transform infrared (FT-IR) spectroscopy relies on calibration techniques involving the acquisition of spectra of bioprocess samples along the process. The most commonly used method for that purpose is partial least squares (PLS) regression, under the assumption that a linear model is valid. Despite being successful in the presence of small nonlinearities, linear methods may fail in the presence of strong nonlinearities. This paper studies the potential usefulness of nonlinear regression methods for predicting, from in situ near-infrared (NIR) and mid-infrared (MIR) spectra acquired in high-throughput mode, biomass and plasmid concentrations in Escherichia coli DH5-α cultures producing the plasmid model pVAX-LacZ. The linear methods PLS and ridge regression (RR) are compared with their kernel (nonlinear) versions, kPLS and kRR, as well as with the (also nonlinear) relevance vector machine (RVM) and Gaussian process regression (GPR). For the systems studied, RR provided better predictive performances compared to the remaining methods. Moreover, the results point to further investigation based on larger data sets whenever differences in predictive accuracy between a linear method and its kernelized version could not be found. The use of nonlinear methods, however, shall be judged regarding the additional computational cost required to tune their additional parameters, especially when the less computationally demanding linear methods herein studied are able to successfully monitor the variables under study.

Active learning using transductive sparse Bayesian regression

Active learning, one of large and important branches in machine learning and data mining, aims to build an accurate learning model with a relatively small number of labeled points which are chosen actively by the constructed learning model. Active learning algorithms can play an important role when the cost of obtaining labels of data points is expensive. However, the active learning algorithms for the relevance vector machine regression, one of the famous kernel methods for the regression due to its sparseness, Bayesian characteristics, and kernel properties, have not been established yet, whilst those for other kernel methods are extensively developed. In this paper, we propose a transductive relevance vector machine to construct the active learning algorithm for relevance vector machine based a sparse Bayesian kernel regression model. The proposed method obtains its basis vectors from the unlabeled data set as well as the labeled one. Then, we also suggest three querying strategies which uses only the relevance vectors automatically selected by the developed model for active selection for data points to be labeled. Applied to several artificial and real data sets, the proposed method show significantly more accurate results than the benchmark, the random selection, while maintaining the sparse characteristic of the traditional relevance vector machines.

Relevance vector machine and multivariate adaptive regression spline for modelling ultimate capacity of pile foundation

This study examines the capability of the Relevance Vector Machine (RVM) and Multivariate Adaptive Regression Spline (MARS) for prediction of ultimate capacity of driven piles and drilled shafts. RVM is a sparse method for training generalized linear models, while MARS technique is basically an adaptive piece-wise regression approach. In this paper, pile capacity prediction models are developed based on data obtained from the literature and comprise in-situ pile loading tests and Cone Penetration Test (CPT) results. Equations are derived from the developed RVM and MARS models, and the prediction results are compared with those obtained from available CPT-based methods. Sensitivity has been carried out to determine the effect of each input parameter. This study confirms that the developed RVM and MARS models predict ultimate capacity of driven piles and drilled shafts reasonably well, and outperform the available methods.