Articles de revista
http://hdl.handle.net/2117/2260
2024-03-29T08:02:56Z
2024-03-29T08:02:56Z
Traffic management of multi-AGV systems by improved dynamic resource reservation
Verma, Parikshit
Olm Miras, Josep Maria
Suárez Feijóo, Raúl
http://hdl.handle.net/2117/404200
2024-03-18T01:54:08Z
2024-03-12T12:27:00Z
Traffic management of multi-AGV systems by improved dynamic resource reservation
Verma, Parikshit; Olm Miras, Josep Maria; Suárez Feijóo, Raúl
Automated guided vehicles (AGVs) are widely used for material handling in warehouses and automated production lines due to their high efficiency and low failure rate with respect to human operated load carriers. However, AGVs usually interact with each other because of the restricted capacity of the layout, and conflicts arise. Although many traffic scheduling algorithms have been proposed to address the AGV fleet control problem, most of them are inefficient for collision and deadlock avoidance in dynamic environments. This paper proposes an improved dynamic resource reservation (IDRR) based method which renders time-efficient task completion and deadlock-free movements of multiple AGVs in a manufacturing system. Unlike traditional approaches, most of which adopt a dynamic single agent reservation of the shared resource points and/or force path deviations, IDRR exploits dynamic multiple reservations of shared resource points. This is combined with a conflict detection and resolution method that accommodates the AGV motions when they meet at a resource point. Extensive, realistic simulation results demonstrate the feasibility and efficiency of the proposed collision and deadlock prevention method in productivity, travelled distance, and time completion of the assigned tasks. The proposal can be implemented on both central and local controllers.
2024-03-12T12:27:00Z
Verma, Parikshit
Olm Miras, Josep Maria
Suárez Feijóo, Raúl
Automated guided vehicles (AGVs) are widely used for material handling in warehouses and automated production lines due to their high efficiency and low failure rate with respect to human operated load carriers. However, AGVs usually interact with each other because of the restricted capacity of the layout, and conflicts arise. Although many traffic scheduling algorithms have been proposed to address the AGV fleet control problem, most of them are inefficient for collision and deadlock avoidance in dynamic environments. This paper proposes an improved dynamic resource reservation (IDRR) based method which renders time-efficient task completion and deadlock-free movements of multiple AGVs in a manufacturing system. Unlike traditional approaches, most of which adopt a dynamic single agent reservation of the shared resource points and/or force path deviations, IDRR exploits dynamic multiple reservations of shared resource points. This is combined with a conflict detection and resolution method that accommodates the AGV motions when they meet at a resource point. Extensive, realistic simulation results demonstrate the feasibility and efficiency of the proposed collision and deadlock prevention method in productivity, travelled distance, and time completion of the assigned tasks. The proposal can be implemented on both central and local controllers.
Robot Operating System (ROS)
Suárez Feijóo, Raúl
Rosell Gratacòs, Jan
Vinagre, Manuel
Cortes, Francesc
Ansuategui, Ander
Maurtua, Iñaki
Martin, Daniel
Guasch Petit, Antonio
Azpiazu, Jon
Serrano, Daniel
García, Néstor
http://hdl.handle.net/2117/402653
2024-02-25T05:36:59Z
2024-02-22T12:18:47Z
Robot Operating System (ROS)
Suárez Feijóo, Raúl; Rosell Gratacòs, Jan; Vinagre, Manuel; Cortes, Francesc; Ansuategui, Ander; Maurtua, Iñaki; Martin, Daniel; Guasch Petit, Antonio; Azpiazu, Jon; Serrano, Daniel; García, Néstor
La irrupción de sistemas robóticos en la industria está generando una creciente demanda de necesidades de comunicación entre los múltiples procesos involucrados. Las ‘fábricas inteligentes’ son ya un hecho tras la eclosión del 4.0; en ellas confluyen aplicaciones robóticas cuyo objetivo es conciso: sumar eficiencia a los procesos productivos.
2024-02-22T12:18:47Z
Suárez Feijóo, Raúl
Rosell Gratacòs, Jan
Vinagre, Manuel
Cortes, Francesc
Ansuategui, Ander
Maurtua, Iñaki
Martin, Daniel
Guasch Petit, Antonio
Azpiazu, Jon
Serrano, Daniel
García, Néstor
La irrupción de sistemas robóticos en la industria está generando una creciente demanda de necesidades de comunicación entre los múltiples procesos involucrados. Las ‘fábricas inteligentes’ son ya un hecho tras la eclosión del 4.0; en ellas confluyen aplicaciones robóticas cuyo objetivo es conciso: sumar eficiencia a los procesos productivos.
Classification of aluminum scrap by laser induced breakdown spectroscopy (LIBS) and RGB + D image fusion using deep learning approaches
Diaz-Romero, Dillam
Van den Eynde, Simon
Zaplana Agut, Isiah
Sterkens, Wouter
Goedemé, Toon
Peeters, Jef R.
http://hdl.handle.net/2117/383642
2023-02-19T20:01:19Z
2023-02-17T10:21:35Z
Classification of aluminum scrap by laser induced breakdown spectroscopy (LIBS) and RGB + D image fusion using deep learning approaches
Diaz-Romero, Dillam; Van den Eynde, Simon; Zaplana Agut, Isiah; Sterkens, Wouter; Goedemé, Toon; Peeters, Jef R.
Integrating multi-sensor systems to sort and monitor complex waste streams is one of the most recent innovations in the recycling industry. The complementary strengths of Laser-Induced Breakdown Spectroscopy (LIBS) and computer vision systems offer a novel multi-sensor solution for the complex task of sorting aluminum (Al) post-consumer scrap into alloy groups. This study presents two novel methods for fusing RGB and Depth images with LIBS using Deep Learning models. The first method is a single-output model that combines LIBS UNET and two DenseNets in a late fusion framework. The second method is a multiple-output model that uses the structure of the single-output model to enhance learning and avoid overfitting. In particular, the network has two outputs that enable the regularization of the individual sensors. A data set of 773 aluminum scrap pieces was created with two sets of ground truth-values, corresponding to the two envisaged sorting tasks, to train and evaluate the developed models. The first sorting task is separating Cast and Wrought (C&W) aluminum. The second is the division of the post-consumer aluminum scrap into three commercially interesting fractions. The single-output model performs best for separating C&W, with a Precision, Recall, and F1-score of 99%. The multiple-output model performs best for classifying the three selected commercial fractions, with a Precision, Recall, and F-score of 86%, 83%, and 84%, respectively. The presented data fusion method for LIBS and computer vision images encompasses the great potential for sorting post-consumer aluminum scrap. By sorting mixed post-consumer aluminum scrap in alloy groups, more wrought-to-wrought recycling can occur, and quality losses can be mitigated during recycling.
2023-02-17T10:21:35Z
Diaz-Romero, Dillam
Van den Eynde, Simon
Zaplana Agut, Isiah
Sterkens, Wouter
Goedemé, Toon
Peeters, Jef R.
Integrating multi-sensor systems to sort and monitor complex waste streams is one of the most recent innovations in the recycling industry. The complementary strengths of Laser-Induced Breakdown Spectroscopy (LIBS) and computer vision systems offer a novel multi-sensor solution for the complex task of sorting aluminum (Al) post-consumer scrap into alloy groups. This study presents two novel methods for fusing RGB and Depth images with LIBS using Deep Learning models. The first method is a single-output model that combines LIBS UNET and two DenseNets in a late fusion framework. The second method is a multiple-output model that uses the structure of the single-output model to enhance learning and avoid overfitting. In particular, the network has two outputs that enable the regularization of the individual sensors. A data set of 773 aluminum scrap pieces was created with two sets of ground truth-values, corresponding to the two envisaged sorting tasks, to train and evaluate the developed models. The first sorting task is separating Cast and Wrought (C&W) aluminum. The second is the division of the post-consumer aluminum scrap into three commercially interesting fractions. The single-output model performs best for separating C&W, with a Precision, Recall, and F1-score of 99%. The multiple-output model performs best for classifying the three selected commercial fractions, with a Precision, Recall, and F-score of 86%, 83%, and 84%, respectively. The presented data fusion method for LIBS and computer vision images encompasses the great potential for sorting post-consumer aluminum scrap. By sorting mixed post-consumer aluminum scrap in alloy groups, more wrought-to-wrought recycling can occur, and quality losses can be mitigated during recycling.
Deep learning regression for quantitative LIBS analysis
Van den Eynde, Simon
Diaz-Romero, Dillam
Zaplana Agut, Isiah
Peeters, Jef R.
http://hdl.handle.net/2117/383638
2023-02-17T09:30:22Z
2023-02-17T09:20:51Z
Deep learning regression for quantitative LIBS analysis
Van den Eynde, Simon; Diaz-Romero, Dillam; Zaplana Agut, Isiah; Peeters, Jef R.
One of the most promising innovation strategies for sorting and recycling post-consumer aluminium scrap is using quantitative Laser-Induced Breakdown Spectroscopy (LIBS) analysis. However, existing methods to estimate alloying element concentrations based on LIBS spectra, such as linear univariate regression and Machine Learning models, are still too limited in their performance to achieve the accuracy demanded by the industry. Therefore, this study presents novel Deep Learning approaches and compares their performance to those of traditional univariate regression and Machine Learning methods in terms of RMSE, MAE, and R2 value. For this evaluation, two sample sets of aluminium pieces are used: one containing 27 certified aluminium reference samples and the second containing 733 post-consumer scrap pieces for which the ground truth concentrations are determined by X-Ray Fluorescence (XRF). Adopting multiple loss functions, one for each element, has proven its significant value for the regression performance. It improves the results for all performance metrics in the Scrap Sample set, and the same is true for the Reference Sample set, except for the coefficient of determination of Fe, Mn and Mg. In addition, the proposed methodology considers the learning prioritisation problem to prevent that learning the concentration of the base element is prioritised over the alloying elements. Although the effect of excluding the base alloy aluminium from the learning is small and not always positive for the performance, demonstrating this effect is also considered valuable. Since the average RMSE on the prediction is just 0.02 wt% for Al and Si, and not more than 0.01 wt% for Fe, Cu, Mn, Mg, and Zn, the best-performing Deep Learning model shows promise for the future of LIBS in metal sorting applications.
2023-02-17T09:20:51Z
Van den Eynde, Simon
Diaz-Romero, Dillam
Zaplana Agut, Isiah
Peeters, Jef R.
One of the most promising innovation strategies for sorting and recycling post-consumer aluminium scrap is using quantitative Laser-Induced Breakdown Spectroscopy (LIBS) analysis. However, existing methods to estimate alloying element concentrations based on LIBS spectra, such as linear univariate regression and Machine Learning models, are still too limited in their performance to achieve the accuracy demanded by the industry. Therefore, this study presents novel Deep Learning approaches and compares their performance to those of traditional univariate regression and Machine Learning methods in terms of RMSE, MAE, and R2 value. For this evaluation, two sample sets of aluminium pieces are used: one containing 27 certified aluminium reference samples and the second containing 733 post-consumer scrap pieces for which the ground truth concentrations are determined by X-Ray Fluorescence (XRF). Adopting multiple loss functions, one for each element, has proven its significant value for the regression performance. It improves the results for all performance metrics in the Scrap Sample set, and the same is true for the Reference Sample set, except for the coefficient of determination of Fe, Mn and Mg. In addition, the proposed methodology considers the learning prioritisation problem to prevent that learning the concentration of the base element is prioritised over the alloying elements. Although the effect of excluding the base alloy aluminium from the learning is small and not always positive for the performance, demonstrating this effect is also considered valuable. Since the average RMSE on the prediction is just 0.02 wt% for Al and Si, and not more than 0.01 wt% for Fe, Cu, Mn, Mg, and Zn, the best-performing Deep Learning model shows promise for the future of LIBS in metal sorting applications.
Simultaneous mass estimation and class classification of scrap metals using deep learning
Diaz-Romero, Dillam
Van den Eynde, Simon
Sterkens, Wouter
Engelen, Bart
Zaplana Agut, Isiah
Dewulf, Wim
Goedemé, Toon
Peeters, Jef R.
http://hdl.handle.net/2117/381993
2023-02-05T19:14:28Z
2023-02-02T12:02:35Z
Simultaneous mass estimation and class classification of scrap metals using deep learning
Diaz-Romero, Dillam; Van den Eynde, Simon; Sterkens, Wouter; Engelen, Bart; Zaplana Agut, Isiah; Dewulf, Wim; Goedemé, Toon; Peeters, Jef R.
While deep learning has helped improve the performance of classification, object detection, and segmentation in recycling, its potential for mass prediction has not yet been explored. Therefore, this study proposes a system for mass prediction with and without feature extraction and selection, including principal component analysis (PCA). These feature extraction methods are evaluated on a combined Cast (C), Wrought (W) and Stainless Steel (SS) image dataset using state-of-the-art machine learning and deep learning algorithms for mass prediction. After that, the best mass prediction framework is combined with a DenseNet classifier, resulting in multiple outputs that perform both object classification and object mass prediction. The proposed architecture consists of a DenseNet neural network for classification and a backpropagation neural network (BPNN) for mass prediction, which uses up to 24 features extracted from depth images. The proposed method obtained 0.82 R2, 0.2 RMSE, and 0.28 MAE for the regression for mass prediction with a classification performance of 95% for the C&W test dataset using the DenseNet+BPNN+PCA model. The DenseNet+BPNN+None model without the selected feature (None) used for the CW&SS test data had a lower performance for both classification of 80% and the regression (0.71 R2, 0.31 RMSE, and 0.32 MAE). The presented method has the potential to improve the monitoring of the mass composition of waste streams and to optimize robotic and pneumatic sorting systems by providing a better understanding of the physical properties of the objects being sorted.
© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works
2023-02-02T12:02:35Z
Diaz-Romero, Dillam
Van den Eynde, Simon
Sterkens, Wouter
Engelen, Bart
Zaplana Agut, Isiah
Dewulf, Wim
Goedemé, Toon
Peeters, Jef R.
While deep learning has helped improve the performance of classification, object detection, and segmentation in recycling, its potential for mass prediction has not yet been explored. Therefore, this study proposes a system for mass prediction with and without feature extraction and selection, including principal component analysis (PCA). These feature extraction methods are evaluated on a combined Cast (C), Wrought (W) and Stainless Steel (SS) image dataset using state-of-the-art machine learning and deep learning algorithms for mass prediction. After that, the best mass prediction framework is combined with a DenseNet classifier, resulting in multiple outputs that perform both object classification and object mass prediction. The proposed architecture consists of a DenseNet neural network for classification and a backpropagation neural network (BPNN) for mass prediction, which uses up to 24 features extracted from depth images. The proposed method obtained 0.82 R2, 0.2 RMSE, and 0.28 MAE for the regression for mass prediction with a classification performance of 95% for the C&W test dataset using the DenseNet+BPNN+PCA model. The DenseNet+BPNN+None model without the selected feature (None) used for the CW&SS test data had a lower performance for both classification of 80% and the regression (0.71 R2, 0.31 RMSE, and 0.32 MAE). The presented method has the potential to improve the monitoring of the mass composition of waste streams and to optimize robotic and pneumatic sorting systems by providing a better understanding of the physical properties of the objects being sorted.
You Only Demanufacture Once (YODO): WEEE retrieval using unsupervised learning
Zhou, Chuangchuang
Sterkens, Wouter
Diaz-Romero, Dillam
Zaplana Agut, Isiah
Peeters, Jef R.
http://hdl.handle.net/2117/381535
2023-02-05T19:13:14Z
2023-01-31T12:55:35Z
You Only Demanufacture Once (YODO): WEEE retrieval using unsupervised learning
Zhou, Chuangchuang; Sterkens, Wouter; Diaz-Romero, Dillam; Zaplana Agut, Isiah; Peeters, Jef R.
Recent developments in robotic demanufacturing raise the potential to increase the cost-efficiency of recycling and recovering resources from Waste of Electrical and Electronic Equipment (WEEE). However, the industrial adoption of robotic demanufacturing for mixed WEEE streams requires tailored instructions for every product model. Considering the large variation in product models, it is not expected to be feasible in the coming decade to rely only on computer vision technologies to define the tailored instructions required for robust and time-efficient robotic demanufacturing. Therefore, the presented research developed a generic retrieval system named You Only Demanufacture Once (YODO) based on content-based image retrieval (CBIR) to identify the product model and retrieve product model-specific demanufacturing instructions. The system compares the visual features represented on a color image of the WEEE with a database of known descriptions representing previously imaged WEEE to find a match or to figure out whether the analyzed product model is new to the system. The performance of YODO is evaluated with a case study for laptop model identification, where a large dataset is created including 4089 images of a representative laptop waste stream. The results demonstrate a top-1 retrieval mean average precision (mAP) of 93.75%. After running YODO on 3600 laptops, the system learned 1079 unique product models, and the presented results show an 85% chance that the next laptop presented to the system is already registered in the database, allowing the retrieval of relevant information for robotic demanufacturing. This corroborates that a fast learning rate can be achieved, allowing a YODO system to support the robotic demanufacturing by making prior product-specific learnings available.
2023-01-31T12:55:35Z
Zhou, Chuangchuang
Sterkens, Wouter
Diaz-Romero, Dillam
Zaplana Agut, Isiah
Peeters, Jef R.
Recent developments in robotic demanufacturing raise the potential to increase the cost-efficiency of recycling and recovering resources from Waste of Electrical and Electronic Equipment (WEEE). However, the industrial adoption of robotic demanufacturing for mixed WEEE streams requires tailored instructions for every product model. Considering the large variation in product models, it is not expected to be feasible in the coming decade to rely only on computer vision technologies to define the tailored instructions required for robust and time-efficient robotic demanufacturing. Therefore, the presented research developed a generic retrieval system named You Only Demanufacture Once (YODO) based on content-based image retrieval (CBIR) to identify the product model and retrieve product model-specific demanufacturing instructions. The system compares the visual features represented on a color image of the WEEE with a database of known descriptions representing previously imaged WEEE to find a match or to figure out whether the analyzed product model is new to the system. The performance of YODO is evaluated with a case study for laptop model identification, where a large dataset is created including 4089 images of a representative laptop waste stream. The results demonstrate a top-1 retrieval mean average precision (mAP) of 93.75%. After running YODO on 3600 laptops, the system learned 1079 unique product models, and the presented results show an 85% chance that the next laptop presented to the system is already registered in the database, allowing the retrieval of relevant information for robotic demanufacturing. This corroborates that a fast learning rate can be achieved, allowing a YODO system to support the robotic demanufacturing by making prior product-specific learnings available.
Closed-form solutions for the inverse kinematics of serial robots using conformal geometric algebra
Zaplana Agut, Isiah
Hadfield, Hugo
Lasenby, Joan
http://hdl.handle.net/2117/381531
2023-02-05T19:15:14Z
2023-01-31T12:17:59Z
Closed-form solutions for the inverse kinematics of serial robots using conformal geometric algebra
Zaplana Agut, Isiah; Hadfield, Hugo; Lasenby, Joan
This work addresses the inverse kinematics of serial robots using conformal geometric algebra. Classical approaches include either the use of homogeneous matrices, which entails high computational cost and execution time, or the development of particular geometric strategies that cannot be generalized to arbitrary serial robots. In this work, we present a compact, elegant and intuitive formulation of robot kinematics based on conformal geometric algebra that provides a suitable framework for the closed-form resolution of the inverse kinematic problem for manipulators with a spherical wrist. For serial robots of this kind, the inverse kinematics problem can be split in two subproblems: the position and orientation problems. The latter is solved by appropriately splitting the rotor that defines the target orientation in three simpler rotors, while the former is solved by developing a geometric strategy for each combination of prismatic and revolute joints that forms the position part of the robot. Finally, the inverse kinematics of 7 DoF redundant manipulators with a spherical wrist is solved by extending the geometric solutions obtained in the non-redundant case.
2023-01-31T12:17:59Z
Zaplana Agut, Isiah
Hadfield, Hugo
Lasenby, Joan
This work addresses the inverse kinematics of serial robots using conformal geometric algebra. Classical approaches include either the use of homogeneous matrices, which entails high computational cost and execution time, or the development of particular geometric strategies that cannot be generalized to arbitrary serial robots. In this work, we present a compact, elegant and intuitive formulation of robot kinematics based on conformal geometric algebra that provides a suitable framework for the closed-form resolution of the inverse kinematic problem for manipulators with a spherical wrist. For serial robots of this kind, the inverse kinematics problem can be split in two subproblems: the position and orientation problems. The latter is solved by appropriately splitting the rotor that defines the target orientation in three simpler rotors, while the former is solved by developing a geometric strategy for each combination of prismatic and revolute joints that forms the position part of the robot. Finally, the inverse kinematics of 7 DoF redundant manipulators with a spherical wrist is solved by extending the geometric solutions obtained in the non-redundant case.
Singularities of serial robots: identification and distance computation using geometric algebra
Zaplana Agut, Isiah
Hadfield, Hugo
Lasenby, Joan
http://hdl.handle.net/2117/381329
2023-01-29T18:39:48Z
2023-01-27T10:02:20Z
Singularities of serial robots: identification and distance computation using geometric algebra
Zaplana Agut, Isiah; Hadfield, Hugo; Lasenby, Joan
The singularities of serial robotic manipulators are those configurations in which the robot loses the ability to move in at least one direction. Hence, their identification is fundamental to enhance the performance of current control and motion planning strategies. While classical approaches entail the computation of the determinant of either a 6×n or n×n matrix for an n-degrees-of-freedom serial robot, this work addresses a novel singularity identification method based on modelling the twists defined by the joint axes of the robot as vectors of the six-dimensional and three-dimensional geometric algebras. In particular, it consists of identifying which configurations cause the exterior product of these twists to vanish. In addition, since rotors represent rotations in geometric algebra, once these singularities have been identified, a distance function is defined in the configuration space C , such that its restriction to the set of singular configurations S allows us to compute the distance of any configuration to a given singularity. This distance function is used to enhance how the singularities are handled in three different scenarios, namely, motion planning, motion control and bilateral teleoperation.
2023-01-27T10:02:20Z
Zaplana Agut, Isiah
Hadfield, Hugo
Lasenby, Joan
The singularities of serial robotic manipulators are those configurations in which the robot loses the ability to move in at least one direction. Hence, their identification is fundamental to enhance the performance of current control and motion planning strategies. While classical approaches entail the computation of the determinant of either a 6×n or n×n matrix for an n-degrees-of-freedom serial robot, this work addresses a novel singularity identification method based on modelling the twists defined by the joint axes of the robot as vectors of the six-dimensional and three-dimensional geometric algebras. In particular, it consists of identifying which configurations cause the exterior product of these twists to vanish. In addition, since rotors represent rotations in geometric algebra, once these singularities have been identified, a distance function is defined in the configuration space C , such that its restriction to the set of singular configurations S allows us to compute the distance of any configuration to a given singularity. This distance function is used to enhance how the singularities are handled in three different scenarios, namely, motion planning, motion control and bilateral teleoperation.
Real-time classification of aluminum metal scrap with laser-induced breakdown spectroscopy using deep and other machine learning approaches
Diaz-Romero, Dillam
Van den Eynde, Simon
Sterkens, Wouter
Zaplana Agut, Isiah
Goedemé, Toon
Peeters, Jef R.
http://hdl.handle.net/2117/379521
2023-01-15T18:33:50Z
2023-01-09T12:05:28Z
Real-time classification of aluminum metal scrap with laser-induced breakdown spectroscopy using deep and other machine learning approaches
Diaz-Romero, Dillam; Van den Eynde, Simon; Sterkens, Wouter; Zaplana Agut, Isiah; Goedemé, Toon; Peeters, Jef R.
In the recycling industry, the use of deep spectral convolutional networks for the purpose of material classification and composition estimation is still limited, despite the great opportunities of these techniques. In this study, the use of Laser-Induced Breakdown Spectroscopy (LIBS), Machine Learning (ML), and Deep Learning (DL) for the three-way sorting of Aluminum (Al) is proposed. Two sample sets of Al scrap are used: one containing 733 pieces for pre-training and validation with a ground truth of X-Ray Fluorescence (XRF), and the second containing 210 pieces for testing for unknown compositions. The proposed method comprises a denoising system combined with a method that extracts 145 features from the raw LIBS spectra. Further, three ML algorithms are assessed to identify the best-performing one to classify unknown pieces of aluminum post-consumer scrap into three commercially interesting output classes. The classified pieces are weighed, melted, and analyzed using spark analysis. Finally, to optimize the best-performing ML system, three state-of-the-art denoising and three feature extraction networks are pre-trained for learning the baseline correction and the proposed feature extraction. Transfer Learning from the six pre-trained networks is applied to create and evaluate 24 end-to-end DL models to classify Al in real-time from >200 spectra simultaneously. The end-to-end DL scheme shows the advantages of learning and denoising the spectra, allowing the transfer of traditional spectral analysis knowledge and the proposed feature extraction into DL, where the network learns from the entire spectrum. The best results for ML and DL were obtained with Random Forest processing one spectrum in 150 ms and BPNN+GHOSTNET(Fine-tuning) processing 200 spectra in 9 ms, which achieved 0.80 Precision, 0.81 Recall, 0.80 F1-score, and 0.80 Precision, 0.79 Recall, 0.79 F1-score, respectively.
2023-01-09T12:05:28Z
Diaz-Romero, Dillam
Van den Eynde, Simon
Sterkens, Wouter
Zaplana Agut, Isiah
Goedemé, Toon
Peeters, Jef R.
In the recycling industry, the use of deep spectral convolutional networks for the purpose of material classification and composition estimation is still limited, despite the great opportunities of these techniques. In this study, the use of Laser-Induced Breakdown Spectroscopy (LIBS), Machine Learning (ML), and Deep Learning (DL) for the three-way sorting of Aluminum (Al) is proposed. Two sample sets of Al scrap are used: one containing 733 pieces for pre-training and validation with a ground truth of X-Ray Fluorescence (XRF), and the second containing 210 pieces for testing for unknown compositions. The proposed method comprises a denoising system combined with a method that extracts 145 features from the raw LIBS spectra. Further, three ML algorithms are assessed to identify the best-performing one to classify unknown pieces of aluminum post-consumer scrap into three commercially interesting output classes. The classified pieces are weighed, melted, and analyzed using spark analysis. Finally, to optimize the best-performing ML system, three state-of-the-art denoising and three feature extraction networks are pre-trained for learning the baseline correction and the proposed feature extraction. Transfer Learning from the six pre-trained networks is applied to create and evaluate 24 end-to-end DL models to classify Al in real-time from >200 spectra simultaneously. The end-to-end DL scheme shows the advantages of learning and denoising the spectra, allowing the transfer of traditional spectral analysis knowledge and the proposed feature extraction into DL, where the network learns from the entire spectrum. The best results for ML and DL were obtained with Random Forest processing one spectrum in 150 ms and BPNN+GHOSTNET(Fine-tuning) processing 200 spectra in 9 ms, which achieved 0.80 Precision, 0.81 Recall, 0.80 F1-score, and 0.80 Precision, 0.79 Recall, 0.79 F1-score, respectively.
Design and analysis of a fully actuated cable-driven joint for hyper-redundant robots with optimal cable routing
Guardiani, Paolo
Ludovico, Daniele
Pistone, Alessandro
Abidi, Haider
Lee, Jinoh
Caldwell, Darwin
Zaplana Agut, Isiah
Canali, Carlo
http://hdl.handle.net/2117/378966
2022-12-25T18:21:16Z
2022-12-20T11:17:09Z
Design and analysis of a fully actuated cable-driven joint for hyper-redundant robots with optimal cable routing
Guardiani, Paolo; Ludovico, Daniele; Pistone, Alessandro; Abidi, Haider; Lee, Jinoh; Caldwell, Darwin; Zaplana Agut, Isiah; Canali, Carlo
Cable-driven hyper-redundant robots have been adopted in many fields for accessing harsh and confined environments that may be inaccessible or dangerous for humans. The cable actuation strategy makes the robot hardware safer and increases the robot payload reducing its weight. In this paper, a novel design of a fully actuated cable-driven hyper-redundant robot has been proposed. This solution is a pulleyless design that decreases the mechanical complexity, allowing to have a compact arm diameter and avoid tension losses on the cables during the motion. Three different joint designs have been taken into account and experiments have been carried to study their performances.The kinematics for the n-joint robot has been formulated, and two cable routing optimization methods, based on a genetic algorithm, have been proposed and applied to a five-joint robot.
2022-12-20T11:17:09Z
Guardiani, Paolo
Ludovico, Daniele
Pistone, Alessandro
Abidi, Haider
Lee, Jinoh
Caldwell, Darwin
Zaplana Agut, Isiah
Canali, Carlo
Cable-driven hyper-redundant robots have been adopted in many fields for accessing harsh and confined environments that may be inaccessible or dangerous for humans. The cable actuation strategy makes the robot hardware safer and increases the robot payload reducing its weight. In this paper, a novel design of a fully actuated cable-driven hyper-redundant robot has been proposed. This solution is a pulleyless design that decreases the mechanical complexity, allowing to have a compact arm diameter and avoid tension losses on the cables during the motion. Three different joint designs have been taken into account and experiments have been carried to study their performances.The kinematics for the n-joint robot has been formulated, and two cable routing optimization methods, based on a genetic algorithm, have been proposed and applied to a five-joint robot.