63 research outputs found
Estimation de la couverture du scan 3D basée sur l’apprentissage automatique pour les applications de contrôle intelligent
No full textCette thèse vise à proposer une méthodologie pour prédire a priori la qualité des scans avec des configurations données, afin d'optimiser la planification des points de vue. Elle a d'abord exploré les métriques de qualité et a trouvé que certaines métriques, en particulier la couverture, qui indique la zone de la surface qui serait acquise lors d'une acquisition physique, peuvent être utilisées pour optimiser les configurations de scan. Ensuite, une nouvelle approche basée sur l'apprentissage automatique, 3DSCP-Net, a été proposée pour prédire la couverture a priori. Enfin, un nouveau pipeline utilisant la prédiction de la couverture a été proposé pour générer des nuages de points tels que scannés. Enfin, une résolution du View Planning Problem utilisant la prédiction de couverture et le contrôle du chevauchement a été formulée pour résoudre l'objectif de la thèse, à savoir l'identification de l'ensemble optimal de points de vue afin d'obtenir des acquisitions de haute qualité. Ce travail fournit non seulement un cadre robuste pour l'estimation de la qualité dans la numérisation 3D, mais explore également les applications de l'outil d'estimation de la qualité.This thesis aims to propose a methodology to predict a priori quality of scans with given configurations in order to optimize the viewpoints planning. This thesis has first explored the quality metrics and found that some metrics, specifically the coverage that indicates the area of the surface would be acquired within a physical acquisition, can be used to optimise the scan configurations. Then, a novel machine-learning-based approach, 3DSCP-Net, has been proposed to predict a priori coverage. Following, a new pipeline using the coverage prediction has been proposed to generate as-scan point cloud. In the end, the resolution of the View Planning Problem using coverage prediction and overlapping control has been formulated to solve the objective of the thesis, i.e. identifying the optimal set of viewpoints so as to obtain high-quality acquisitions. Experiments have validated that the proposed methodologies can achieve the objectives. This work not only provides a robust framework for quality estimation in 3D scanning but also explores the applications of the quality estimation tool. This research contributes to the broader goal of integrating artificial intelligence into the optimization of 3D acquisition workflows, supporting advancements in Industry 4.0 and digital twin technologies
Estimation de la couverture du scan 3D basée sur l’apprentissage automatique pour les applications de contrôle intelligent
No full textThis thesis aims to propose a methodology to predict a priori quality of scans with given configurations in order to optimize the viewpoints planning. This thesis has first explored the quality metrics and found that some metrics, specifically the coverage that indicates the area of the surface would be acquired within a physical acquisition, can be used to optimise the scan configurations. Then, a novel machine-learning-based approach, 3DSCP-Net, has been proposed to predict a priori coverage. Following, a new pipeline using the coverage prediction has been proposed to generate as-scan point cloud. In the end, the resolution of the View Planning Problem using coverage prediction and overlapping control has been formulated to solve the objective of the thesis, i.e. identifying the optimal set of viewpoints so as to obtain high-quality acquisitions. Experiments have validated that the proposed methodologies can achieve the objectives. This work not only provides a robust framework for quality estimation in 3D scanning but also explores the applications of the quality estimation tool. This research contributes to the broader goal of integrating artificial intelligence into the optimization of 3D acquisition workflows, supporting advancements in Industry 4.0 and digital twin technologies.Cette thèse vise à proposer une méthodologie pour prédire a priori la qualité des scans avec des configurations données, afin d'optimiser la planification des points de vue. Elle a d'abord exploré les métriques de qualité et a trouvé que certaines métriques, en particulier la couverture, qui indique la zone de la surface qui serait acquise lors d'une acquisition physique, peuvent être utilisées pour optimiser les configurations de scan. Ensuite, une nouvelle approche basée sur l'apprentissage automatique, 3DSCP-Net, a été proposée pour prédire la couverture a priori. Enfin, un nouveau pipeline utilisant la prédiction de la couverture a été proposé pour générer des nuages de points tels que scannés. Enfin, une résolution du View Planning Problem utilisant la prédiction de couverture et le contrôle du chevauchement a été formulée pour résoudre l'objectif de la thèse, à savoir l'identification de l'ensemble optimal de points de vue afin d'obtenir des acquisitions de haute qualité. Ce travail fournit non seulement un cadre robuste pour l'estimation de la qualité dans la numérisation 3D, mais explore également les applications de l'outil d'estimation de la qualité
Estimation de la couverture du scan 3D basée sur l’apprentissage automatique pour les applications de contrôle intelligent
No full textThis thesis aims to propose a methodology to predict a priori quality of scans with given configurations in order to optimize the viewpoints planning. This thesis has first explored the quality metrics and found that some metrics, specifically the coverage that indicates the area of the surface would be acquired within a physical acquisition, can be used to optimise the scan configurations. Then, a novel machine-learning-based approach, 3DSCP-Net, has been proposed to predict a priori coverage. Following, a new pipeline using the coverage prediction has been proposed to generate as-scan point cloud. In the end, the resolution of the View Planning Problem using coverage prediction and overlapping control has been formulated to solve the objective of the thesis, i.e. identifying the optimal set of viewpoints so as to obtain high-quality acquisitions. Experiments have validated that the proposed methodologies can achieve the objectives. This work not only provides a robust framework for quality estimation in 3D scanning but also explores the applications of the quality estimation tool. This research contributes to the broader goal of integrating artificial intelligence into the optimization of 3D acquisition workflows, supporting advancements in Industry 4.0 and digital twin technologies.Cette thèse vise à proposer une méthodologie pour prédire a priori la qualité des scans avec des configurations données, afin d'optimiser la planification des points de vue. Elle a d'abord exploré les métriques de qualité et a trouvé que certaines métriques, en particulier la couverture, qui indique la zone de la surface qui serait acquise lors d'une acquisition physique, peuvent être utilisées pour optimiser les configurations de scan. Ensuite, une nouvelle approche basée sur l'apprentissage automatique, 3DSCP-Net, a été proposée pour prédire la couverture a priori. Enfin, un nouveau pipeline utilisant la prédiction de la couverture a été proposé pour générer des nuages de points tels que scannés. Enfin, une résolution du View Planning Problem utilisant la prédiction de couverture et le contrôle du chevauchement a été formulée pour résoudre l'objectif de la thèse, à savoir l'identification de l'ensemble optimal de points de vue afin d'obtenir des acquisitions de haute qualité. Ce travail fournit non seulement un cadre robuste pour l'estimation de la qualité dans la numérisation 3D, mais explore également les applications de l'outil d'estimation de la qualité
Estimation de la couverture du scan 3D basée sur l’apprentissage automatique pour les applications de contrôle intelligent
No full textThis thesis aims to propose a methodology to predict a priori quality of scans with given configurations in order to optimize the viewpoints planning. This thesis has first explored the quality metrics and found that some metrics, specifically the coverage that indicates the area of the surface would be acquired within a physical acquisition, can be used to optimise the scan configurations. Then, a novel machine-learning-based approach, 3DSCP-Net, has been proposed to predict a priori coverage. Following, a new pipeline using the coverage prediction has been proposed to generate as-scan point cloud. In the end, the resolution of the View Planning Problem using coverage prediction and overlapping control has been formulated to solve the objective of the thesis, i.e. identifying the optimal set of viewpoints so as to obtain high-quality acquisitions. Experiments have validated that the proposed methodologies can achieve the objectives. This work not only provides a robust framework for quality estimation in 3D scanning but also explores the applications of the quality estimation tool. This research contributes to the broader goal of integrating artificial intelligence into the optimization of 3D acquisition workflows, supporting advancements in Industry 4.0 and digital twin technologies.Cette thèse vise à proposer une méthodologie pour prédire a priori la qualité des scans avec des configurations données, afin d'optimiser la planification des points de vue. Elle a d'abord exploré les métriques de qualité et a trouvé que certaines métriques, en particulier la couverture, qui indique la zone de la surface qui serait acquise lors d'une acquisition physique, peuvent être utilisées pour optimiser les configurations de scan. Ensuite, une nouvelle approche basée sur l'apprentissage automatique, 3DSCP-Net, a été proposée pour prédire la couverture a priori. Enfin, un nouveau pipeline utilisant la prédiction de la couverture a été proposé pour générer des nuages de points tels que scannés. Enfin, une résolution du View Planning Problem utilisant la prédiction de couverture et le contrôle du chevauchement a été formulée pour résoudre l'objectif de la thèse, à savoir l'identification de l'ensemble optimal de points de vue afin d'obtenir des acquisitions de haute qualité. Ce travail fournit non seulement un cadre robuste pour l'estimation de la qualité dans la numérisation 3D, mais explore également les applications de l'outil d'estimation de la qualité
The Study of Litter in Leymus chinensis Meadow
No full textLeymus chinensis meadow is a main type of grassland in the northeast of China and produces a high yield. Leymus chinensis is dominant species in the grassland and is a kind of high-quality grass with rich nutrients. Owing to over utilization for it, there is little litter in the grassland. It causes the degeneration of grassland. Litter plays a important role in keeping ecological balance of grassland. The decomposition and accumulation of litter are key link for guaranteeing material flow and have a great influence on restitution of grassland. The study on litter in grassland is still rare in China, particularly in meadow steppe. Li Jiazao (1981) studied the decomposition of cellulose in alpine meadow. Chen Zuozong (1982) reported the decay rate of litter and horse\u27s dung on typical grassland in Inner Mongolia. Liu Gengchang (1982) observed the accumulation of litter in Leymus chinensis grassland. In this work, the decomposition and accumulation of litter have been studied in Leymus chinensis meadow
Optimization Control for Orderly Charge and Discharge Control Strategy of Electric Vehicles Based on Reliable Index of Charging
No full text
A Computational Fluid Dynamics Modified Friction Factor and Leakage Model for an Improved Bulk-Flow Analysis of Labyrinth Gas Seals
No full textBulk-flow predictive models (BFM), though simple and fast, often fail to accurately predict the performance of gas labyrinth seals (LSs). In this work, a Computational Fluid Dynamics (CFD) analysis quantifies the effects of LS tip clearance (Cvr) and operating conditions on the circumferential direction friction factors (fvr, fvs) at the rotor and stator surfaces, as well as on the kinetic energy carry-over coefficient (��v1i) for mass flow prediction. A fourteen teeth on stator LS seal (L/D=0.29) with clearance Cvr=1/733 D is selected for analysis.
The analysis models the seal with a fine mesh of a few million nodes and a commercial CFD code calculates the flow field for the nominal operating conditions, includes wide changes in clearance, 80% to 200% of the nominal Cvr, shaft speed from 5 krpm to 15 krpm (58 m/s ~173 m/s), inlet pre-swirl velocity varying from 0% to 72% of rotor surface speed, a gas supply pressure ranging from 60 bar to 100 bar, and along with various discharge pressures producing a pressure ratio (PR = Pvout/Pvin) ranging from 0.40 to 0.85. The rotor surface friction factor fvr�� is independent of the changes in clearance (Cvr) or the inlet circumferential velocity pre-swirl ratio; whereas an increase in rotor speed or in pressure ratio (PR) decreases fr��. On the other hand, an increase in rotor speed, pressure ratio and inlet preswirl ratio decreases fs��, the stator friction factor. Besides, fvs�� increases with an increase in radial clearance. Further, fvr�� and fvs�� are only sensitive to the pressure ratio, but not to the magnitude of either the supply pressure or discharge pressure.
The kinetic energy carry-over coefficient (��v1i) increases with respect to the seal radial clearance (Cvr); whereas ��v^1i shows a parabolic correlation with the pressure ratio PR. ��v1i is only sensitive to PR, and not to the magnitude of either the supply pressure or the discharge pressure.
Furthermore, based on the CFD derived results, this work presents a modified friction factor model, Rem f n = (where Re is the flow Reynolds number)^1 , as well as a modified kinetic energy carry-over coefficient model, both quantifying the effect of seal geometry and operating conditions. An independent case analysis serves to validate the model; and the modified BFM does improve the prediction of the direct stiffness (maximum discrepancy decreases from 320% to 70%), direct damping (discrepancy decreases from 90% to 50%), and mass flow rate (discrepancy decreases from 14% to 2%).The above coefficients and flow agree well with both CFD and experimental results.
(Note: this dissertation is organized based on the author���s previous publications and reports during his PhD study; and the format follows American Society of Mechanical Engineers (ASME) journal publications format).
1 n = 0.079, m = -0.25 for the classical Blasius friction factor model, strictly valid for smooth surface pipeline
Soil Animal Composition and Distribution in the Leymus chinensis Grassland Region in Central Part of North-East of China
No full textSince 1990, we have investigated the soil animal composition and distribution in the Leymus chinensis (Aneuro chinensis) grassland region in central part of north-east of Chinn. Soil animals were studied in 20 plant communities in 4 habitats. A total of 15 797 soil animals belonging to 4 phyla, 6 classes, 28 orders and 86 families were obtained. Among them, 26 families and 22 species were new records in norlh-easl of China. The dominant groups were Formicidne, Actinedida, Oribatidn, lsotomidne, Rhnbditidae, Dorylainidae. Sparse woods of Ulmus, and the most animal groups (40) and Puccinellia chinapoesis and Heleochalis intersiIa communities the least (14). The biggest numbers of individuals were found in mixed grass communites and the smallest in Kochia sieversia11a and Puccinella chinampoesis communities. Both numbers of groups and of individuals decreased with increasing depth of soil. A number of environmental factors were used to compare data from different communities. In general 1-2 factors were dominant and varied from habitat to habitat
- …
