Pub Date : 2024-10-30DOI: 10.1109/TIM.2024.3488156
Feng Zhu;Xi Chen;Qinqing Cai;Xiaohong Zhang
High-precision continuous position and attitude determination are the critical modules of mobile mapping and autonomous driving (AD). Research in the integration of Global Navigation Satellite System (GNSS) and strapdown inertial navigation system (SINS) has greatly enhanced the accuracy and robustness of position and attitude in different scenes. However, the complexity and variability of the real scenes are still challenging for the existing models, parameters, strategies, and algorithms (MPSA). It is worth noting that high-quality datasets are key to accelerating the research and development of MPSA, which has been proved in the computer vision (CV) fields represented by the ImageNet dataset. Unfortunately, current public datasets either do not provide the raw observations of GNSS and inertial measurement unit (IMUs) or are not collected in abundant scenes and moving platforms. Therefore, a large-scale diverse GNSS/SINS dataset, named SmartPNT-POS, is presented. This dataset covers rich real-world environments, such as open-sky and complex urban, and multiple moving platforms, such as aircraft, land vehicles, and ships. In addition, different types of IMUs, including those manufactured in Hexagon, iMAR Navigation GmbH, and Honeywell, are contained in SmartPNT-POS as well. Moreover, it provides ground truths in each group of data for users to analyze and evaluate their MPSA. Now, the dataset is publicly available through Kaggle, a data science community, and the website to obtain the dataset is provided in the text. There have been 30 sets of data published on the website up to the present, and comprehensive analyses have been made in this contribution for the position and attitude determination results obtained by different processing modes. More data will be collected for different environments and applications and published on the same website in the future.
{"title":"A Large-Scale Diverse GNSS/SINS Dataset: Construction, Publication, and Application","authors":"Feng Zhu;Xi Chen;Qinqing Cai;Xiaohong Zhang","doi":"10.1109/TIM.2024.3488156","DOIUrl":"https://doi.org/10.1109/TIM.2024.3488156","url":null,"abstract":"High-precision continuous position and attitude determination are the critical modules of mobile mapping and autonomous driving (AD). Research in the integration of Global Navigation Satellite System (GNSS) and strapdown inertial navigation system (SINS) has greatly enhanced the accuracy and robustness of position and attitude in different scenes. However, the complexity and variability of the real scenes are still challenging for the existing models, parameters, strategies, and algorithms (MPSA). It is worth noting that high-quality datasets are key to accelerating the research and development of MPSA, which has been proved in the computer vision (CV) fields represented by the ImageNet dataset. Unfortunately, current public datasets either do not provide the raw observations of GNSS and inertial measurement unit (IMUs) or are not collected in abundant scenes and moving platforms. Therefore, a large-scale diverse GNSS/SINS dataset, named SmartPNT-POS, is presented. This dataset covers rich real-world environments, such as open-sky and complex urban, and multiple moving platforms, such as aircraft, land vehicles, and ships. In addition, different types of IMUs, including those manufactured in Hexagon, iMAR Navigation GmbH, and Honeywell, are contained in SmartPNT-POS as well. Moreover, it provides ground truths in each group of data for users to analyze and evaluate their MPSA. Now, the dataset is publicly available through Kaggle, a data science community, and the website to obtain the dataset is provided in the text. There have been 30 sets of data published on the website up to the present, and comprehensive analyses have been made in this contribution for the position and attitude determination results obtained by different processing modes. More data will be collected for different environments and applications and published on the same website in the future.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"73 ","pages":"1-13"},"PeriodicalIF":5.6,"publicationDate":"2024-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142600446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-29DOI: 10.1109/TIM.2024.3485429
Zeyu Gu;Gang Tang;Jianwei Ma
The mechanical and electrical performance degradation of satellite components has a serious impact on imaging. How to perform high-precision reconstruction of the monitor data from compressive-sensing (CS) data (limited by online computing, storage and transmission) is often the first stage in the fault diagnosis of in-orbit satellites. In this article, a deep generative model named denoising diffusion probabilistic model (DDPM) is applied for the equipment monitor data reconstruction. The priors-assisted reconstruction method is useful for reducing reconstruction error and decreasing measurement/monitor cost. The reconstruction method mainly consists of unconditional generation transition from pre-trained DDPM noise matching network and conditional likelihood correction step toward downsampling data. An inverse time decay technique is embedded into step size strategy of gradient computation to ensure data consistency. As an unsupervised learning paradigm, the learned deep generative priors can be utilized for measurements with different compressive sampling ratio (CSR) like plug-and-play prior. Numerical experiments executed on control moment gyro (CMG) data and reciprocating refrigeration compressor (RRC) data validate the effectiveness of the new method, in comparison with conventional sparse prior methods and advanced deep learning reconstruction methods. Finally, we conduct out-of-distribution (OOD) generalization experiments on fault working condition, which demonstrates the DDPM priors-assisted data reconstruction method are suitable for different operating conditions.
{"title":"Compressive-Sensing Reconstruction for Satellite Monitor Data Using a Deep Generative Model","authors":"Zeyu Gu;Gang Tang;Jianwei Ma","doi":"10.1109/TIM.2024.3485429","DOIUrl":"https://doi.org/10.1109/TIM.2024.3485429","url":null,"abstract":"The mechanical and electrical performance degradation of satellite components has a serious impact on imaging. How to perform high-precision reconstruction of the monitor data from compressive-sensing (CS) data (limited by online computing, storage and transmission) is often the first stage in the fault diagnosis of in-orbit satellites. In this article, a deep generative model named denoising diffusion probabilistic model (DDPM) is applied for the equipment monitor data reconstruction. The priors-assisted reconstruction method is useful for reducing reconstruction error and decreasing measurement/monitor cost. The reconstruction method mainly consists of unconditional generation transition from pre-trained DDPM noise matching network and conditional likelihood correction step toward downsampling data. An inverse time decay technique is embedded into step size strategy of gradient computation to ensure data consistency. As an unsupervised learning paradigm, the learned deep generative priors can be utilized for measurements with different compressive sampling ratio (CSR) like plug-and-play prior. Numerical experiments executed on control moment gyro (CMG) data and reciprocating refrigeration compressor (RRC) data validate the effectiveness of the new method, in comparison with conventional sparse prior methods and advanced deep learning reconstruction methods. Finally, we conduct out-of-distribution (OOD) generalization experiments on fault working condition, which demonstrates the DDPM priors-assisted data reconstruction method are suitable for different operating conditions.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"73 ","pages":"1-15"},"PeriodicalIF":5.6,"publicationDate":"2024-10-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142598620","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-29DOI: 10.1109/TIM.2024.3485392
Ze-Qi Ding;Qiang Qin;Yi-Fan Zhang;Yan-Hui Lin
In prognostic and health management (PHM), predicting remaining useful life (RUL) and quantifying the uncertainties in predictions are necessary. This article proposes a Gaussian process (GP) autoregression-variational autoencoder (GPVAE) framework that can predict RUL based on degradation data, quantify predictive uncertainty, decompose this uncertainty into epistemic and aleatory types, and further quantify epistemic uncertainties on RUL-related features. Subsequently, uncertainty calibration is proposed to ensure that the quantified uncertainty matches the actual error of the model. The calibrated uncertainty is used for out-of-distribution (OOD) detection and active learning for the labeled and unlabeled data, which can improve the RUL prediction accuracy with limited computational resources and limited cost of degradation tests for obtaining RUL labels. The effectiveness of the proposed method is illustrated by the case study on lithium-ion batteries dataset.
{"title":"An Uncertainty Quantification and Calibration Framework for RUL Prediction and Accuracy Improvement","authors":"Ze-Qi Ding;Qiang Qin;Yi-Fan Zhang;Yan-Hui Lin","doi":"10.1109/TIM.2024.3485392","DOIUrl":"https://doi.org/10.1109/TIM.2024.3485392","url":null,"abstract":"In prognostic and health management (PHM), predicting remaining useful life (RUL) and quantifying the uncertainties in predictions are necessary. This article proposes a Gaussian process (GP) autoregression-variational autoencoder (GPVAE) framework that can predict RUL based on degradation data, quantify predictive uncertainty, decompose this uncertainty into epistemic and aleatory types, and further quantify epistemic uncertainties on RUL-related features. Subsequently, uncertainty calibration is proposed to ensure that the quantified uncertainty matches the actual error of the model. The calibrated uncertainty is used for out-of-distribution (OOD) detection and active learning for the labeled and unlabeled data, which can improve the RUL prediction accuracy with limited computational resources and limited cost of degradation tests for obtaining RUL labels. The effectiveness of the proposed method is illustrated by the case study on lithium-ion batteries dataset.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"73 ","pages":"1-13"},"PeriodicalIF":5.6,"publicationDate":"2024-10-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142598636","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This article presents an ultrasonic gas meter based on a scandium-doped aluminum nitride (Sc0.2Al0.8N) piezoelectric micromachined ultrasonic transducer (PMUT) array, where the characteristic dimension of each PMUT cell is only $600~mu text {m}$