In the aftermath of earthquakes, structures can become unsafe and hazardous for humans to safely reside. Automated methods that detect structural damage can be invaluable for rapid inspections and faster recovery times. Deep neural networks (DNNs) have proven to be an effective means to classify damaged areas in images of structures but have limited generalizability due to the lack of large and diverse annotated datasets (e.g., variations in building properties like size, shape, color). Given a dataset of paired images of damaged and undamaged structures supervised deep learning methods could be employed, but such paired correspondences of images required for training are exceedingly difficult to acquire. Obtaining a variety of undamaged images, and a smaller set of damaged images is more viable. We present a novel application of deep learning for unpaired image-to-image translation between undamaged and damaged structures as a means of data augmentation to combat the lack of diverse data. Unpaired image-to-image translation is achieved using Cycle Consistent Adversarial Network (CCAN) architectures, which have the capability to translate images while retaining the geometric structure of an image. We explore the capability of the original CCAN architecture, and propose a new architecture for unpaired image-to-image translation (termed Eigen Integrated Generative Adversarial Network or EIGAN) that addresses shortcomings of the original architecture for our application. We create a new unpaired dataset to translate an image between domains of damaged and undamaged structures. The dataset created consists of a set of damaged and undamaged buildings from Mexico City affected by the 2017 Puebla earthquake. Qualitative and quantitative results of the various architectures are presented to better compare the quality of the translated images. A comparison is also done on the performance of DNNs trained to classify damaged structures using generated images. The results demonstrate that targeted image-to-image translation of undamaged to damaged structures is an effective means of data augmentation to improve network performance.
{"title":"IMAGE TO IMAGE TRANSLATION OF STRUCTURAL DAMAGE USING GENERATIVE ADVERSARIAL NETWORKS","authors":"Subin Varghese, Rebecca Wang, Vedhus Hoskere","doi":"10.12783/shm2021/36307","DOIUrl":"https://doi.org/10.12783/shm2021/36307","url":null,"abstract":"In the aftermath of earthquakes, structures can become unsafe and hazardous for humans to safely reside. Automated methods that detect structural damage can be invaluable for rapid inspections and faster recovery times. Deep neural networks (DNNs) have proven to be an effective means to classify damaged areas in images of structures but have limited generalizability due to the lack of large and diverse annotated datasets (e.g., variations in building properties like size, shape, color). Given a dataset of paired images of damaged and undamaged structures supervised deep learning methods could be employed, but such paired correspondences of images required for training are exceedingly difficult to acquire. Obtaining a variety of undamaged images, and a smaller set of damaged images is more viable. We present a novel application of deep learning for unpaired image-to-image translation between undamaged and damaged structures as a means of data augmentation to combat the lack of diverse data. Unpaired image-to-image translation is achieved using Cycle Consistent Adversarial Network (CCAN) architectures, which have the capability to translate images while retaining the geometric structure of an image. We explore the capability of the original CCAN architecture, and propose a new architecture for unpaired image-to-image translation (termed Eigen Integrated Generative Adversarial Network or EIGAN) that addresses shortcomings of the original architecture for our application. We create a new unpaired dataset to translate an image between domains of damaged and undamaged structures. The dataset created consists of a set of damaged and undamaged buildings from Mexico City affected by the 2017 Puebla earthquake. Qualitative and quantitative results of the various architectures are presented to better compare the quality of the translated images. A comparison is also done on the performance of DNNs trained to classify damaged structures using generated images. The results demonstrate that targeted image-to-image translation of undamaged to damaged structures is an effective means of data augmentation to improve network performance.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134377205","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Gaussian process regression is a powerful method for predicting states associated with uncertainty. A common application field is to predict damage states of structural systems. Recently, Gaussian processes became very popular as they deliver credible intervals for the predicted states. However, one major disadvantage of Gaussian processes is that they assume a normal distribution. This is not justified when the relevant variables can only assume positive values, such as crack lengths or damage states. This paper presents a way to bypass this problem by using warped Gaussian processes: We (1) transform the data with a warping function, (2) apply Gaussian process regression in the latent space, and (3) transform the results back by using the inverse of the warping function. The method is applied to a crack growth example. The paper shows how to integrate prior knowledge into warped Gaussian processes in order to increase prediction accuracy and that warped Gaussian processes lead to better and more plausible results.
{"title":"WARPED GAUSSIAN PROCESSES FOR PROGNOSTIC HEALTH MONITORING","authors":"Simon Pfingstl, Christian Braun, M. Zimmermann","doi":"10.12783/shm2021/36358","DOIUrl":"https://doi.org/10.12783/shm2021/36358","url":null,"abstract":"Gaussian process regression is a powerful method for predicting states associated with uncertainty. A common application field is to predict damage states of structural systems. Recently, Gaussian processes became very popular as they deliver credible intervals for the predicted states. However, one major disadvantage of Gaussian processes is that they assume a normal distribution. This is not justified when the relevant variables can only assume positive values, such as crack lengths or damage states. This paper presents a way to bypass this problem by using warped Gaussian processes: We (1) transform the data with a warping function, (2) apply Gaussian process regression in the latent space, and (3) transform the results back by using the inverse of the warping function. The method is applied to a crack growth example. The paper shows how to integrate prior knowledge into warped Gaussian processes in order to increase prediction accuracy and that warped Gaussian processes lead to better and more plausible results.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"103 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133007883","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this study, a novel method for online and real-time identification of dynamic systems is presented. This method is based on the newly introduced algorithm Physics Informed Neural Network (PINN). In order to find the dynamic characteristics of the system, sparse displacement measurements are fed to the Artificial Neural Network (ANN); By introducing the classic vibration equation of the system to the ANN as a physics constraint, the PINN estimates both dynamic characteristic and state of the system. The proposed framework is evaluated by several numerical studies with different system properties, noise levels, architecture, and training data. On that account, four structural systems are presented: (1) single-degree-of-freedom (SDOF) systems with different properties and noise levels, as basis model with an accurate analytical solution (2) a three-degree-of-freedom (3-DOF) system with both complete and sparse measurements, representing the structural model of the n-story shear frames (3) a simple supported beam subjected to an initial displacement with several NNs architecture and sensor numbers, and (4) a Pure Cubic Oscillator (PCO) as a nonlinear dynamic system. The results of the proposed platform for the PINN are compared to a mutual ANN in all cases to emphasize the superiority of the PINN in both determining the dynamic characteristics and state estimation of dynamic systems. In addition, the performance of both NNs is examined with different training data to ensure the resilience of the algorithm and affirm the role of the added criteria, physics constraint, in reducing the dependency on the training data. The proposed algorithm can accurately estimate the dynamic characteristics of different dynamic systems with sparse, noisy measurements; by means of the classic dynamic equations and smartly selection of the hidden layer numbers, the PINN will be a powerful predictive tool for the dynamic analysis in the absence of any prior knowledge of the dynamic systems.
{"title":"PHYSICS-INFORMED NEURAL NETWORK APPROACH FOR IDENTIFICATION OF DYNAMIC SYSTEMS","authors":"Sarvin Moradi, S. E. Azam, M. Mofid","doi":"10.12783/shm2021/36352","DOIUrl":"https://doi.org/10.12783/shm2021/36352","url":null,"abstract":"In this study, a novel method for online and real-time identification of dynamic systems is presented. This method is based on the newly introduced algorithm Physics Informed Neural Network (PINN). In order to find the dynamic characteristics of the system, sparse displacement measurements are fed to the Artificial Neural Network (ANN); By introducing the classic vibration equation of the system to the ANN as a physics constraint, the PINN estimates both dynamic characteristic and state of the system. The proposed framework is evaluated by several numerical studies with different system properties, noise levels, architecture, and training data. On that account, four structural systems are presented: (1) single-degree-of-freedom (SDOF) systems with different properties and noise levels, as basis model with an accurate analytical solution (2) a three-degree-of-freedom (3-DOF) system with both complete and sparse measurements, representing the structural model of the n-story shear frames (3) a simple supported beam subjected to an initial displacement with several NNs architecture and sensor numbers, and (4) a Pure Cubic Oscillator (PCO) as a nonlinear dynamic system. The results of the proposed platform for the PINN are compared to a mutual ANN in all cases to emphasize the superiority of the PINN in both determining the dynamic characteristics and state estimation of dynamic systems. In addition, the performance of both NNs is examined with different training data to ensure the resilience of the algorithm and affirm the role of the added criteria, physics constraint, in reducing the dependency on the training data. The proposed algorithm can accurately estimate the dynamic characteristics of different dynamic systems with sparse, noisy measurements; by means of the classic dynamic equations and smartly selection of the hidden layer numbers, the PINN will be a powerful predictive tool for the dynamic analysis in the absence of any prior knowledge of the dynamic systems.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122428207","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Timely recognition of rock fragments and their morphological sizes can help adjust excavation parameters during tunnel boring machine (TBM) tunneling. Traditional manual inspection highly relies on experiences and subjective judgments of human operators and conducting sieving tests is not real-time and energy-consuming. Rock fragments in real-world images are often observed against a dark background, distributed with high size diversity, complicatedly distributed, and blocked by each other. To solve these problems, this study proposes a novel instance segmentation-based method for on-site rock fragments recognition. The proposed instance segmentation model includes two subnetworks: object detection and semantic segmentation. The results show that 88% of rock fragments can be recognized, and the average recall and average IoU values reach 0.85 and 0.75 on the 15 test images, respectively. Besides, both small and large rock fragments can be recognized well. The predicted size distributions of the major and minor axis lengths of the rock fragments fit well with the ground-truth ones statistically. In conclusion, this study can provide both visual recognition and statistical results for the size distribution of on-site rock fragments.
{"title":"INSTANCE-SEGMENTATION-BASED DENSE ON-SITE ROCK FRAGMENT RECOGNITION DURING REAL-WORLD TUNNEL EXCAVATION","authors":"Xu Yang, Qiao Weidong, Li Hui","doi":"10.12783/shm2021/36320","DOIUrl":"https://doi.org/10.12783/shm2021/36320","url":null,"abstract":"Timely recognition of rock fragments and their morphological sizes can help adjust excavation parameters during tunnel boring machine (TBM) tunneling. Traditional manual inspection highly relies on experiences and subjective judgments of human operators and conducting sieving tests is not real-time and energy-consuming. Rock fragments in real-world images are often observed against a dark background, distributed with high size diversity, complicatedly distributed, and blocked by each other. To solve these problems, this study proposes a novel instance segmentation-based method for on-site rock fragments recognition. The proposed instance segmentation model includes two subnetworks: object detection and semantic segmentation. The results show that 88% of rock fragments can be recognized, and the average recall and average IoU values reach 0.85 and 0.75 on the 15 test images, respectively. Besides, both small and large rock fragments can be recognized well. The predicted size distributions of the major and minor axis lengths of the rock fragments fit well with the ground-truth ones statistically. In conclusion, this study can provide both visual recognition and statistical results for the size distribution of on-site rock fragments.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123918952","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Rail internal defects such as detail fracture and transverse fissure are among the leading causes of track-related railway accidents. Therefore, it is critical to develop effective rail defect inspection systems and data processing methods to prevent catastrophic accidents and derailments. This study developed an anomaly detection framework using deep autoencoder (DAE) for rail defect detection. And the team evaluated its performance based on data collected by a prototype passive acoustic rail inspection system. Autoencoder is a semi-supervised learning algorithm that identifies observations in a dataset that deviate significantly from the remaining data. First, the team performed data cleaning and transfer function reconstruction using a dataset collected at the Federal Railroad Administration’s Transportation Technology Center in Pueblo, Colorado. Then, handcrafted or knowledge-driven features were extracted from the transfer functions and fed into a statistical outlier analysis as the benchmark. Also, reconstructed transfer functions at clean rail segments were directly used as the input to train and validate the DAE algorithm. The results demonstrated the effectiveness of DAE for structural discontinuity detection and showed promise for rail flaw detection.
{"title":"HIGH-SPEED RAIL INSPECTION EXPLOITING AN ANOMALY DETECTION DATA PROCESSING APPROACH","authors":"Yuning Wu, Xuan Zhu, Jay Baillargeon","doi":"10.12783/shm2021/36302","DOIUrl":"https://doi.org/10.12783/shm2021/36302","url":null,"abstract":"Rail internal defects such as detail fracture and transverse fissure are among the leading causes of track-related railway accidents. Therefore, it is critical to develop effective rail defect inspection systems and data processing methods to prevent catastrophic accidents and derailments. This study developed an anomaly detection framework using deep autoencoder (DAE) for rail defect detection. And the team evaluated its performance based on data collected by a prototype passive acoustic rail inspection system. Autoencoder is a semi-supervised learning algorithm that identifies observations in a dataset that deviate significantly from the remaining data. First, the team performed data cleaning and transfer function reconstruction using a dataset collected at the Federal Railroad Administration’s Transportation Technology Center in Pueblo, Colorado. Then, handcrafted or knowledge-driven features were extracted from the transfer functions and fed into a statistical outlier analysis as the benchmark. Also, reconstructed transfer functions at clean rail segments were directly used as the input to train and validate the DAE algorithm. The results demonstrated the effectiveness of DAE for structural discontinuity detection and showed promise for rail flaw detection.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124209244","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Non-destructive testing of rail is an essential part of maintaining in-service rail tracks to avoid accidents. Conventional methods such as the traditional ultrasonic technique are relatively slow and cumbersome resulting in non-frequent monitoring. This study explores active and passive techniques for continuous and long range rail damage monitoring. Firstly, the experiment, simulation and challenges of the ultrasonic guided wave generated through surface-bonded piezoelectric transducer are studied. Due to the presence of numerable inseparable modes occurring in rail, the application of machine learning algorithms is explored. Classification of damage in rail head and severity of damage have been achieved using features derived from the signal. To map changes in features with respect to damage, various ML algorithms are trained, tested and compared. Among them, the k-nearest neighbour has been found to have the highest accuracy in classifying rail head damage, while the Gaussian process regression is best suited for determining damage severity. Trained algorithms are then tested with simulated and experiment of different damage sizes. Secondly, the application of acoustic emission in rail is investigated through simulation and pencil lead break source experiments. The behaviour of rail as waveguide and wide band of generating frequency are observed to be the challenges in determining the zone of AE source. Thus, to classify the zone of AE source, a deep learning algorithm based on continuous wavelet transform is presented. This method results in 88% accuracy in finding the AE source zone. The presented study then concluded with challenges in monitoring complex geometry such as rail and application of machine learning in monitoring.
{"title":"ACTIVE AND PASSIVE MONITORING OF RAIL THROUGH THE APPLICATION OF MACHINE LEARNING ALGORITHM","authors":"Harsh Mahajan, Sauvik Banerjee","doi":"10.12783/shm2021/36330","DOIUrl":"https://doi.org/10.12783/shm2021/36330","url":null,"abstract":"Non-destructive testing of rail is an essential part of maintaining in-service rail tracks to avoid accidents. Conventional methods such as the traditional ultrasonic technique are relatively slow and cumbersome resulting in non-frequent monitoring. This study explores active and passive techniques for continuous and long range rail damage monitoring. Firstly, the experiment, simulation and challenges of the ultrasonic guided wave generated through surface-bonded piezoelectric transducer are studied. Due to the presence of numerable inseparable modes occurring in rail, the application of machine learning algorithms is explored. Classification of damage in rail head and severity of damage have been achieved using features derived from the signal. To map changes in features with respect to damage, various ML algorithms are trained, tested and compared. Among them, the k-nearest neighbour has been found to have the highest accuracy in classifying rail head damage, while the Gaussian process regression is best suited for determining damage severity. Trained algorithms are then tested with simulated and experiment of different damage sizes. Secondly, the application of acoustic emission in rail is investigated through simulation and pencil lead break source experiments. The behaviour of rail as waveguide and wide band of generating frequency are observed to be the challenges in determining the zone of AE source. Thus, to classify the zone of AE source, a deep learning algorithm based on continuous wavelet transform is presented. This method results in 88% accuracy in finding the AE source zone. The presented study then concluded with challenges in monitoring complex geometry such as rail and application of machine learning in monitoring.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116071968","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The value of structural health monitoring (SHM) can be quantified as the difference in expected total life-cycle costs between two different maintenance planning strategies, one representing the standard means to assessment, namely intermittent visual inspections, and the other based on availability of continuous SHM data. We show how to quantify the value of vibration-based SHM conditional on a damage history over the structural lifetime. We showcase the analysis through application on a numerical benchmark model of a two-span bridge system subjected to gradual deterioration and sudden damages in the middle elastic support over its life-cycle, simulating the case of scour. The effect of environmental variability is included in the analysis by means of a stochastic model for the dependence of the Young’s modulus on temperature (E-T). The numerical investigations provide insights related to the effect of the temperature variability, as well as the visual inspections’ quality, on the value of SHM.
{"title":"QUANTIFYING THE VALUE OF VIBRATION-BASED STRUCTURAL HEALTH MONITORING CONSIDERING ENVIRONMENTAL VARIABILITY","authors":"A. Kamariotis, E. Chatzi, D. Štraub","doi":"10.12783/shm2021/36356","DOIUrl":"https://doi.org/10.12783/shm2021/36356","url":null,"abstract":"The value of structural health monitoring (SHM) can be quantified as the difference in expected total life-cycle costs between two different maintenance planning strategies, one representing the standard means to assessment, namely intermittent visual inspections, and the other based on availability of continuous SHM data. We show how to quantify the value of vibration-based SHM conditional on a damage history over the structural lifetime. We showcase the analysis through application on a numerical benchmark model of a two-span bridge system subjected to gradual deterioration and sudden damages in the middle elastic support over its life-cycle, simulating the case of scour. The effect of environmental variability is included in the analysis by means of a stochastic model for the dependence of the Young’s modulus on temperature (E-T). The numerical investigations provide insights related to the effect of the temperature variability, as well as the visual inspections’ quality, on the value of SHM.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126588160","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jasper Ryvers, M. Callewaert, M. Loccufier, Wim De Waele
A key vulnerability of offshore energy production facilities are submarine power cables. Monitoring a power cable during its entire lifetime is a strategy to minimize critical damages. Submarine power cables contain fiber optic cables that can be used for monitoring purposes. In this paper we show that a recent fiber optic sensing technique (CP-ΦOTDR) [11] is capable of detecting bending and impact events in a fiber optic cable. It does so with a limited reproducibility however, which for impact events cannot be solely explained by the variability of the impact force. There is a slight signature present for impact events: identical impact force events correlate more than events belonging to non-identical forces. This work is a first step in developing a monitoring tool to help assess the severity of power cable damages due to incident impacts and critical bending states.
{"title":"DETECTING BENDING AND IMPACT EVENTS IN A FIBER OPTIC CABLE USING DISTRIBUTED ACOUSTIC SENSING TO ASSESS POTENTIAL OFFSHORE POWER CABLE DAMGES","authors":"Jasper Ryvers, M. Callewaert, M. Loccufier, Wim De Waele","doi":"10.12783/shm2021/36349","DOIUrl":"https://doi.org/10.12783/shm2021/36349","url":null,"abstract":"A key vulnerability of offshore energy production facilities are submarine power cables. Monitoring a power cable during its entire lifetime is a strategy to minimize critical damages. Submarine power cables contain fiber optic cables that can be used for monitoring purposes. In this paper we show that a recent fiber optic sensing technique (CP-ΦOTDR) [11] is capable of detecting bending and impact events in a fiber optic cable. It does so with a limited reproducibility however, which for impact events cannot be solely explained by the variability of the impact force. There is a slight signature present for impact events: identical impact force events correlate more than events belonging to non-identical forces. This work is a first step in developing a monitoring tool to help assess the severity of power cable damages due to incident impacts and critical bending states.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131744587","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
During the in-service lifetime of an aircraft, the surface bond of any SHM transductor is submitted to thermal stresses, induced by aviation environmental conditions. In this paper, the influence of a disbond, whether willingly introduced or caused by thermal aging, on the ability of a PZT ultrasonic transducer to generate and receive Lamb waves is numerically and experimentally studied.
{"title":"STUDY OF THE EFFECTS OF THERMAL STRESS ON PIEZOELECTRIC SENSORS FOR THE STRUCTURAL HEALTH MONITORING","authors":"L. Gavérina, J. Roche, P. Beauchêne, F. Passilly","doi":"10.12783/shm2021/36353","DOIUrl":"https://doi.org/10.12783/shm2021/36353","url":null,"abstract":"During the in-service lifetime of an aircraft, the surface bond of any SHM transductor is submitted to thermal stresses, induced by aviation environmental conditions. In this paper, the influence of a disbond, whether willingly introduced or caused by thermal aging, on the ability of a PZT ultrasonic transducer to generate and receive Lamb waves is numerically and experimentally studied.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134490133","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yawei Feng, Yapeng Guo, Yi Zhuo, Hao Di, Jianfeng Wei, Shunlong Li
Rivet corrosion, which is a common disease of steel truss bridges, directly reflects the safety status of steel structures. The identification of rivet corrosion is critical to ensure the normal service of steel truss bridges. In practical engineering, the main detection method of rivet corrosion is manual visual inspection. However, this method has low efficiency and poses a threat to the personal safety. To address this issue, an intelligent identification method for rivet corrosion on steel truss bridges by a single shot detector (SSD) is proposed after obtaining the panoramic image of the bridge. The sub-images cut from the panoramic image are as the network’s input. Considering the small size of bridge rivets and low precision of small object detection of SSD, this study divides the panoramic image into sub-images of 100 × 100 pixels, and then uses bilinear interpolation to resize the sub-images into 300 × 300 pixels. To improve the robustness of the detection model, gaussian noise, random rotation and roll-over tricks are applied to the original dataset. The expanded dataset includes 600 labelling images, which is divided into training set (80%) and testing set (20%), including corroded rivets and normal rivets. The network is trained with transfer learning technique for 12000 iterations, with cross entropy loss for classification and smooth L1 loss for location. The confidence threshold in network inference is set to 0.6 considering the rivet space distribution to reduce false positives of corroded rivets. The qualitative and quantitative testing results show the accuracy of the proposed approach.
{"title":"INTELLIGENT IDENTIFICATION OF RIVET CORROSION ON STEEL TRUSS BRIDGE BY SINGLE-STAGE DETECTION NETWORK","authors":"Yawei Feng, Yapeng Guo, Yi Zhuo, Hao Di, Jianfeng Wei, Shunlong Li","doi":"10.12783/shm2021/36254","DOIUrl":"https://doi.org/10.12783/shm2021/36254","url":null,"abstract":"Rivet corrosion, which is a common disease of steel truss bridges, directly reflects the safety status of steel structures. The identification of rivet corrosion is critical to ensure the normal service of steel truss bridges. In practical engineering, the main detection method of rivet corrosion is manual visual inspection. However, this method has low efficiency and poses a threat to the personal safety. To address this issue, an intelligent identification method for rivet corrosion on steel truss bridges by a single shot detector (SSD) is proposed after obtaining the panoramic image of the bridge. The sub-images cut from the panoramic image are as the network’s input. Considering the small size of bridge rivets and low precision of small object detection of SSD, this study divides the panoramic image into sub-images of 100 × 100 pixels, and then uses bilinear interpolation to resize the sub-images into 300 × 300 pixels. To improve the robustness of the detection model, gaussian noise, random rotation and roll-over tricks are applied to the original dataset. The expanded dataset includes 600 labelling images, which is divided into training set (80%) and testing set (20%), including corroded rivets and normal rivets. The network is trained with transfer learning technique for 12000 iterations, with cross entropy loss for classification and smooth L1 loss for location. The confidence threshold in network inference is set to 0.6 considering the rivet space distribution to reduce false positives of corroded rivets. The qualitative and quantitative testing results show the accuracy of the proposed approach.","PeriodicalId":180083,"journal":{"name":"Proceedings of the 13th International Workshop on Structural Health Monitoring","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114467862","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：