LSTM+MA: A Time-Series Model for Predicting Pavement IRI

Imporved model for pavement performance prediction based on recurrent neural network using LTPP database

The surface roughness in terms of International Roughness Index (IRI) captures variation in vertical surface elevations along the pavement length. Diamond grinding is a rigid pavement preservation treatment which is typically utilized to increase pavement smoothness by eliminating the surface undulations. The focus of this study is to evaluate the impact of diamond grinding on rigid pavement performance by using the mechanistic-empirical procedure. The effectiveness of a diamond grinding was assessed by evaluating the IRI before and after the treatment. Generally, smoother pavements will have lower IRI, which will result in less vehicle body and axle bounce. Consequently, the pavement will be subjected to reduced dynamic loads which may cause less pavement damage or vice versa. The longitudinal profile and axle load data from the Long-Term Pavement Performance (LTPP) in the SPS-6 experiment were utilized to evaluate the impact of diamond grinding on pavement performance. These experiments were designed to evaluate the effectiveness of rigid pavement rehabilitation treatments. The data for pavement sections from two climatic zones were evaluated before and after diamond grinding. The axle distributions for each pavement section were utilized to generate truck gross-vehicle weight (GVW) distributions. Subsequently, the truck GVW distributions were used to generate truck response on different pavement surface profiles by using TruckSim software. The impact of surface roughness on dynamic loads was evaluated by comparing axle load spectra before and after diamond grinding. The variations in the axle load spectra are used to explain the differences in the predicted pavement performance in terms of cracking, faulting and IRI.

ABSTRACTThe performance of asphalt pavement is highly affected by climate factors, such as temperature and precipitation. Different temperature and moisture conditions change the paving material properties, which consequently influence pavement performance. Considerable research has been conducted to study the effect of climate factors on pavement performance; however, there are no focused studies that have investigated the performance of asphalt pavements in cold regions. In this research, the effect of different design factors on asphalt pavement performance in cold regions is investigated utilising data extracted from the Long-Term Pavement Performance database. Only control sections with no historical maintenance or rehabilitation records were considered. The International Roughness Index (IRI) was adopted as the pavement performance measure. The IRI value is expected to increase gradually over time due to pavement deterioration. Initial screening of the data showed that most of the pavement sections experienced remarkable IRI increase followed by an unexpected decrease. An in-depth investigation revealed that frost heave was the main reason of this unexpected IRI trend as it was correlated in most of the cases to sudden monthly freezing index increases, especially for sections with no subsurface drainage and less layers thicknesses. An Artificial Neural Network with a Forward Sequential Feature Selection algorithm and regression analysis were employed to model pavement performance in cold regions and determine the most significant design factors prevailing in cold climate conditions. Moreover, a sensitivity analysis was performed to investigate the interrelation between the considered features and the IRI. As a result, several climate-related factors were found to have a significant impact on the performance of asphalt pavements in cold regions such as average temperature, freezing index, freeze/thaw, wind velocity and relative humidity.

Predicting the International Roughness Index (IRI) is crucial for maintaining road quality and ensuring the safety and comfort of road users. Accurate IRI predictions help in the timely identification of road sections that require maintenance, thus preventing further deterioration and reducing overall maintenance costs. This study aims to develop robust predictive models for the IRI using advanced machine learning techniques across different climatic conditions. Data were sourced from the Ministry of Energy and Infrastructure in the UAE for localized conditions coupled with the Long-Term Pavement Performance (LTPP) database for comparison and validation purposes. This study evaluates several machine learning models, including regression trees, support vector machines (SVMs), ensemble trees, Gaussian process regression (GPR), artificial neural networks (ANNs), and kernel-based methods. Among the models tested, GPR, particularly with rational quadratic specifications, consistently demonstrated superior performance with the lowest Root Mean Square Error (RMSE) and highest R-squared values across all datasets. Sensitivity analysis identified age, total pavement thickness, precipitation, temperature, and Annual Average Daily Truck Traffic (AADTT) as key factors influencing the IRI. The results indicate that pavement age and higher traffic loads significantly increase roughness, while thicker pavements contribute to smoother surfaces. Climatic factors such as temperature and precipitation showed varying impacts depending on the regional conditions. The developed models provide a powerful tool for predicting pavement roughness, enabling more accurate maintenance planning and resource allocation. The findings highlight the necessity of tailoring pavement management practices to specific environmental and traffic conditions to enhance road quality and longevity. This research offers a comprehensive framework for understanding and predicting pavement performance, with implications for infrastructure management both locally and worldwide.

This paper analyses some of the most widely used indices for the characterization of longitudinal roughness. Measured longitudinal profiles available in the Long-Term Pavement Performance (LTPP) program database were used. A total of 207 pavement sections, in 42 States of the USA, were used in these analyses. Using a suitable software, the International Roughness Index (IRI), the Standard Deviation of longitudinal roughness (σ) and the Root Mean Square of Vertical Acceleration (RMSVA) with different base lengths were computed for every measured longitudinal pavement profile. Linear regression analyses between the IRI and σ statistics produced a very high correlation (R2=0.93). Multivariate regression between IRI and the RMSVA also showed a high correlation (R2=0.96) for base lengths of 1.0 m and 3.5 m. On the other hand, when the RMSVA and σ statistics were analyzed, it was observed that all regressions including a RMSVA with the base length of 1.5m presented a nearly perfect correlation (R2≅1.00). Probabilistic analyses were also carried out using the First Order Second Moment (FOSM) method applied to the regression models obtained between the IRI × RMSVA and σ × RMSVA. Thus, it was possible to determine the relative contribution of different base lengths used in the RMSVA statistic on each roughness index. It was concluded that the base length of 1.5 m contributed in average with 96% of roughness values computed for σ statistic, whereas the base length of 1.0 m was responsible for 74.5% of computed IRI values.

The International Roughness Index (IRI) is calculated from elevation profiles acquired by high-speed profilers or laser scanners, but these raw data often contain measurement noise and extraneous wavelength components that can degrade the accuracy of IRI calculations. Existing filtering methods expose a limitation in removing IRI-insensitive wavelength components. Thus, this paper proposes a Gaussian filtering algorithm based on the Nyquist sampling theorem to remove IRI-insensitive components of the longitudinal profile. The proposed approach first adaptively determines Gaussian template lengths according to sampling intervals, and then incorporates a boundary padding strategy to ensure processing stability. The proposed method enables precise wavelength selection within the IRI-sensitive band of 1.3–29.4 m while maintaining computational efficiency. The method was validated using the Paris–Lille dataset and the U.S. Long-Term Pavement Performance (LTPP) program dataset. The filtered profiles were evaluated by Power Spectral Density (PSD), and IRI values were calculated and compared with those obtained by conventional profile filtering methods. The results show that the proposed method is effective in removing the non-sensitive components of IRI and obtaining highly accurate IRI values. Compared with the standard IRI provided by the LTPP dataset, mean absolute error of the IRI values from the proposed method reaches 0.051 m/km, and mean relative error is less than 4%. These findings indicate that the proposed method improves the reliability of IRI calculation.

The international roughness index (IRI) is an important indicator reflecting the roughness of the road surface. Pavement roughness performance is affected by many factors such as pavement surface disease, pavement service life and natural environment. In view of this, this paper aims to study an IRI prediction model based on the random forest algorithm of multi-feature analysis. The mechanism of influencing factors of pavement roughness was analyzed to extract IRI and its influencing factors data in the seven major states of the United States for nearly 26 years from the long term pavement performance (LTPP) database, and then an asphalt pavement roughness dataset was established. Furthermore, through the feature selection based on the linear Pearson correlation coefficient, feature heat map and the nonlinear random forest, XGBoost (extreme Gradient Boosting) algorithm, the factors that affect the road surface roughness were analyzed. Finally, in order to effectively fit the IRI through the pavement surface roughness influencing factors, a random forest model based on grid search optimization was constructed. Optimized random forest was compared with the traditional method KNN, decision tree and random forest to verify their performance on the pavement roughness data set. Experiments show that the random forest prediction model based on grid search optimization has higher prediction accuracy, and the fitting goodness R² is as high as 98.6%.

The intricacies and instability of introducing cryogenic propellants into the combustion system have piqued the curiosity of scientists studying the procedure. The latest innovation is utilizing data-driven machine learning and deep learning approaches to gain deeper insights into the related difficulties. However, the current work serves as a baseline for future research because relatively few studies have used data-driven methodologies to assess the temperature of liquid fuel injections in combustion systems. The performance of Linear Regression (LR), Random Forest (RF), Extra Trees Regressor (ETR), Polynomial Regression (PR), Support Vector Regressor (SVR), Decision Tree Regressor (DTR), Gradient Boost Regressor (GBR), XGB Regressor (XGBoost), AdaBoost Regressor (ABR), K-Neighbors Regressor (KNR), Long-Short Term Memory (LSTM), Bi-LSTM (Bi-directional Long-Short Term Memory) has all been investigated in this study. The study also suggested a Fully Connected Neural Network (FCNN) to examine its performance and paired it with an Extra Tree Regressor (ETR). The coupled FCNN and Extra Tree Regressor outperform the other algorithms with a Mean Square Error (MSE) of 0.0000005062, Root Mean Square Error (RMSE) of 0.00071148, Mean Absolute Error (MAE) of 0.00020672, and R-squared (R2) value of 0.99998689. Linear Regression, Polynomial Regression, and Support Vector Regressor are found to be the least-performing algorithms. The current work uses machine learning and deep learning methods to make data-driven decisions for liquid fuel injection in the combustion system.

Road roughness is the result of a chain of distress mechanisms, and it combines the effects of various modes of pavement deterioration. The common measure of roughness is the international roughness index (IRI). The IRI represents the functional performance of road pavements and serves as an indicator of the structural performance. However, the use of the IRI in setting priorities for maintenance and rehabilitation programs has some flaws. The IRI cannot detect the mode and rate of pavement deterioration that contribute to the surface roughness. In addition, the IRI may be a poor measure of truck ride and the dynamic wheel loads generated by the interaction of heavy vehicles and rough road surfaces. Roughness spectrum can be divided into certain wavebands, with each waveband corresponding to a certain mode of pavement deterioration. The power spectral density analysis of road profiles is used to view the distribution of the various wavelengths within the profile and to show the contribution of each waveband to roughness. Waveband analysis can help overcome the shortcomings of the IRI, and it provides information that helps road authorities in planning rehabilitation programs and managing the road network.

Pavement performance assessment and prediction are crucial for efficient infrastructure management and strategic planning of maintenance activities. Conventional techniques are insufficient and lack the efficiency and flexibility required for modern transportation networks. This study proposes a groundbreaking integrated approach that merges machine learning (ML) classification techniques with Geographical Information Systems (GIS) to evaluate road conditions using the Pavement Condition Index (PCI) and the International Roughness Index (IRI). Given that IRI data collection is more straightforward and cost-effective than gathering pavement distress data, this study aims to classify the IRI of flexible pavements to estimate PCI models using advanced ML algorithms (Artificial Neural Network (ANN), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), Decision Trees (DT), and Random Forest (RF)) and accurately determine pavement conditions. This research gathered 1042 data points using a smartphone application, TotalPave, to measure the IRI values for the (Nizwa–Muscat) and (Muscat–Nizwa) routes in the Sultanate of Oman. It meticulously applied feature selection techniques to identify the pavement parameters significantly impacting pavement performance. The research then spatially visualized and analyzed the results to determine the critical pavement sections. Among the ML models, RF demonstrated outstanding performance with an accuracy rate of 99.9% and an F1-score of 99. SVM has the lowest accuracy of 85.8% and an F1-score of 40.3. A comprehensive assessment comprising a confusion matrix, uncertainty analysis, box and whisker plot, and noise sensitivity provides in-depth insights into the reliability and consistency of predictions. The ML and GIS methods revolutionized the way transportation agencies interpret and implement their findings. The proposed framework is not merely a tool, but a transformative solution that facilitates the formulation of proactive maintenance strategies and optimizes resource utilization by providing a scalable and intelligent decision-support tool designed specifically for pavement management systems (PMS).

While asphalt-concrete layer properties are more sensitive towards temperature variations, the subgrade stiffness have proven to be quite sensitive to moisture content fluctuations. When embankments are needed during the pavement construction, the soil must be compacted at a particular moisture-density condition to reach homogeneous properties. Due to construction variability, the as-compacted conditions might vary from the required specifications. This is of particular interest when dealing with compacted fine grained subgrade materials due to the influence of soil permeability in the seasonal moisture content fluctuation. The main objective of this study is the evaluation of the structural capacity and pavement performance due to the variability associated with the as-compacted moisture and density conditions, for different climatic regions. Three types of materials ranging from silt to clay were used in the analysis. Weather information from three locations was collected to represent different climate regimes. The pavement performance was analyzed using the Mechanistic-Empirical Pavement Design Guide (MEPDG) procedure by using the response of the pavement to three major distresses: fatigue cracking, rutting and thermal cracking. Two different levels of input accuracy were compared: Level 2, which uses simple index material and compaction properties and Level 3, which uses just the simple index material properties to estimate the resilient modulus. Results indicated that the soil compacted to optimum moisture content predicted less amount of distress when compared to the wet or dry optimum conditions for the low plasticity soils; whereas the predicted distresses increase with the increase in the as-compacted moisture content for the high plasticity soils. When compared with ratio of optimum condition, international roughness index (IRI) had a maximum increase of 5% in Chicago and Atlanta region exhibited a fatigue cracking increase of 40% while Phoenix area showed a lot of rutting increase of 30%. The difference in the distress results obtained for Level 2 and Level 3 analyses were not significant for the soil compacted at optimum condition, but the dry and wet of optimum conditions resulted in significant larger differences that depended on the location or climate where the pavement is constructed.

Calibration of mechanics-based pavement predictive framework for top-down cracking performance of flexible pavement considering wheel wander effect

The first step in the road handling effort is to survey to get an accurate road condition value to take correct action in implementing maintenance. As pavement performance is known to be probabilistic, various levels of uncertainty must be assumed. Modern pavement management methods are ineffective without an effective model to predict pavement performance. Discrete-time Markov chains are the most widely used probabilistic models, and examples from different countries worldwide can be found in pavement management systems. This research aims to predict the value of road conditions during maintenance and compare road assessments with real conditions during road maintenance using the IRI, SDI, and PCI methods using the Markov process. The analysis method used is to collect secondary data from related departments and carry out direct data collection in the field to obtain condition values based on IRI, SDI, and PCI to forecast by making a pavement condition prediction model based on the Markov process and then assessing road conditions by comparing the three index values with the slightest deviation value. The analysis showed that the average value of road conditions with the IRI indicator is 4.45, which is moderate, and the most negligible difference between the probability distribution of pavement condition prediction modeling and the actual survey results was the IRI (International Roughness Index) method. This model is closest to the actual conditions during implementation, with a difference value of 5.7%.

Flexible pavements are considered an essential element of transportation infrastructure. So, evaluations of flexible pavement performance are necessary for the proper management of transportation infrastructure. Pavement condition index (PCI) and international roughness index (IRI) are common indices applied to evaluate pavement surface conditions. However, the pavement condition surveys to calculate PCI are costly and time-consuming as compared to IRI. This article focuses on developing regression models that predict PCI from IRI. Eighty-three flexible pavement sections, with section length equal to 250 m, were selected in Al-Diwaniyah, Iraq, to develop PCI-IRI relationships. In terms of the quantity and severity of each observed distress, the pavement condition surveys were conducted by actually walking through all the sections. Using these data, PCI was calculated utilizing Micro PAVER software. Dynatest Road Surface Profiler (RSP) was used to collect IRI data of all the sections. Using the SPSS software, linear and nonlinear regressions have been used for developing two models between PCI and IRI based on the collected data. These models have the coefficients of determination (R2) equal to 0.715 and 0.722 for linear and quadratic models. Finally, the results indicate the linear and quadratic models are acceptable to predict PCI from IRI directly.

Daily land surface temperature (LST) forecasting is of great significance for application in climate-related, agricultural, eco-environmental, or industrial studies. Hybrid data-driven prediction models using Ensemble Empirical Mode Composition (EEMD) coupled with Machine Learning (ML) algorithms are useful for achieving these purposes because they can reduce the difficulty of modeling, require less history data, are easy to develop, and are less complex than physical models. In this article, a computationally simple, less data-intensive, fast and efficient novel hybrid data-driven model called the EEMD Long Short-Term Memory (LSTM) neural network, namely EEMD-LSTM, is proposed to reduce the difficulty of modeling and to improve prediction accuracy. The daily LST data series from the Mapoling and Zhijiang stations in the Dongting Lake basin, central south China, from 1 January 2014 to 31 December 2016 is used as a case study. The EEMD is firstly employed to decompose the original daily LST data series into many Intrinsic Mode Functions (IMFs) and a single residue item. Then, the Partial Autocorrelation Function (PACF) is used to obtain the number of input data sample points for LSTM models. Next, the LSTM models are constructed to predict the decompositions. All the predicted results of the decompositions are aggregated as the final daily LST. Finally, the prediction performance of the hybrid EEMD-LSTM model is assessed in terms of the Mean Square Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Pearson Correlation Coefficient (CC) and Nash-Sutcliffe Coefficient of Efficiency (NSCE). To validate the hybrid data-driven model, the hybrid EEMD-LSTM model is compared with the Recurrent Neural Network (RNN), LSTM and Empirical Mode Decomposition (EMD) coupled with RNN, EMD-LSTM and EEMD-RNN models, and their comparison results demonstrate that the hybrid EEMD-LSTM model performs better than the other five models. The scatterplots of the predicted results of the six models versus the original daily LST data series show that the hybrid EEMD-LSTM model is superior to the other five models. It is concluded that the proposed hybrid EEMD-LSTM model in this study is a suitable tool for temperature forecasting.