Non-Destructive High-Performance Techniques (NDTs) are the basis of efficient Pavement Management Systems (PMSs) considering their high accuracy, reliability, speed of execution, high coverage, and non-invasiveness. Generally, each of these techniques is appropriate for evaluating a specific aspect of infrastructures, and several NDT-based surveys are required to obtain a comprehensive assessment of an asset condition. In order to identify potential non-linear correlation among different surveys, and with the aim of reducing the number of on-site inspections, this paper presents a novel methodology based on Deep Neural Networks (DNNs) for integrating products derived by three on-ground NDTs. Specifically, we aim to estimate appropriately the International Roughness Index (IRI) detected by Laser Profiler (LaP) of 93 road sections of 100 meters in length by integrating the outcomes of surveys performed by Falling Weight Deflectometer (FWD) and Ground Penetrating Radar (GPR). Moreover, several environmental parameters concerning climate and rainfall information have been considered. Therefore, different architectures of DNNs optimized by the Levenberg-Marquardt backpropagation algorithm have been trained (by using 70% of samples), validated (15% of samples), tested (15% of samples), and their performance evaluated. Outcomes reveal that DNNs allow recognizing efficiently hidden patterns between NDT-based surveys and made the integration possible and reliable. In particular, a DNN architecture composed of two hidden layers containing 23 and 12 artificial neurons, respectively, shows a Correlation Coefficient (R) of 0.902 for the training phase, 0.872 for the validation phase, and 0.861 for the test phase. Supported by these findings, we have re-calibrated the DNN including all road sections, reaching an R parameter of 0.872 and a Mean Square Error (MSE) of 0.203 mm/km for its predictions. Road authorities could consider DNNs for integrating appropriately NDT-based surveys, optimizing their monitoring plans, and improving their inspection activities.

Predicting international roughness index by deep neural networks with Levenberg-Marquardt backpropagation learning algorithm

Pietro Leandri;Massimo Losa
2021

Abstract

Non-Destructive High-Performance Techniques (NDTs) are the basis of efficient Pavement Management Systems (PMSs) considering their high accuracy, reliability, speed of execution, high coverage, and non-invasiveness. Generally, each of these techniques is appropriate for evaluating a specific aspect of infrastructures, and several NDT-based surveys are required to obtain a comprehensive assessment of an asset condition. In order to identify potential non-linear correlation among different surveys, and with the aim of reducing the number of on-site inspections, this paper presents a novel methodology based on Deep Neural Networks (DNNs) for integrating products derived by three on-ground NDTs. Specifically, we aim to estimate appropriately the International Roughness Index (IRI) detected by Laser Profiler (LaP) of 93 road sections of 100 meters in length by integrating the outcomes of surveys performed by Falling Weight Deflectometer (FWD) and Ground Penetrating Radar (GPR). Moreover, several environmental parameters concerning climate and rainfall information have been considered. Therefore, different architectures of DNNs optimized by the Levenberg-Marquardt backpropagation algorithm have been trained (by using 70% of samples), validated (15% of samples), tested (15% of samples), and their performance evaluated. Outcomes reveal that DNNs allow recognizing efficiently hidden patterns between NDT-based surveys and made the integration possible and reliable. In particular, a DNN architecture composed of two hidden layers containing 23 and 12 artificial neurons, respectively, shows a Correlation Coefficient (R) of 0.902 for the training phase, 0.872 for the validation phase, and 0.861 for the test phase. Supported by these findings, we have re-calibrated the DNN including all road sections, reaching an R parameter of 0.872 and a Mean Square Error (MSE) of 0.203 mm/km for its predictions. Road authorities could consider DNNs for integrating appropriately NDT-based surveys, optimizing their monitoring plans, and improving their inspection activities.
978-151064570-7
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11568/1141145
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact