Datasets:

aeyxen
/

stem-diagrams

file_name stringlengths 56 64	arxiv_id stringlengths 12 12	field stringclasses 5 values	page int64 1 40	method stringclasses 2 values	bbox stringlengths 16 20	p_diagram float64 0.5 0.99	caption stringlengths 1 800 ⌀	source_arxiv_url stringlengths 34 34	license stringclasses 1 value
images/manufacturing_engineering/2405.09941v1_p001_vector_01.png	2405.09941v1	manufacturing_engineering	1	vector	[85, 176, 527, 352]	0.8725	null	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p006_vector_02.png	2405.09941v1	manufacturing_engineering	6	vector	[85, 85, 527, 262]	0.8803	Figure 1: Brief illustrative representation of the presented approach. where σf and σs are the Cauchy stresses applied by the fluid and the solid respectively, v is the fluid velocity and ˙u is the solid velocity. In mesh-based methods for the fluid problem, the superposition of the two wet interfaces translates into a...	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p015_vector_03.png	2405.09941v1	manufacturing_engineering	15	vector	[163, 78, 449, 313]	0.6311	Figure 2: Fluid and solid ROM components inference of the two ROMs ˆF and ˆS, since, in our case, using POD as our encoder-decoder implies that ES(DS(ur)) = ur. However, the full force field must be recovered at each iteration (of the reduced fixed-point problem) because the relaxation used in line 9, together with an ...	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p016_vector_04.png	2405.09941v1	manufacturing_engineering	16	vector	[163, 78, 449, 317]	0.6809	Figure 3: Fluid and solid ROM components - Hybrid variant doing so, we ensure that the fluid ROM ˆF maintains a high enough fidelity so that the provided initial guess does indeed help the FSI converge faster, especially since the newest information, from the latest time steps and iterations will be used.	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p017_vector_05.png	2405.09941v1	manufacturing_engineering	17	vector	[41, 133, 571, 639]	0.8695	Figure 4: Global FSI coupling scheme with the ROM-FOM coupling and data-driven adaptive predictors. 15	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p027_image_06.png	2405.09941v1	manufacturing_engineering	27	image	[117, 282, 263, 426]	0.625	Figure 15: The SROM prediction evaluated on the solid nonlinear behaviour in example 2. In (a) we see the loads applied on the deformed solid at t = 4.7s, and (b) and (d) show the yy component of the Green Lagrange strain at the same time step, comparing the ROM-FOM and FOM-FOM solutions. In (c) the nonlinear behaviour...	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.09941v1_p030_image_07.png	2405.09941v1	manufacturing_engineering	30	image	[85, 85, 527, 329]	0.7368	Figure 19: Example 3 - Test case schematic explanation and dimensions. imposed on the y−and y+ faces and a zero pressure is imposed on the outlet. The discretization uses 437039 elements with 84988 nodes. This flow setting corresponds to a Reynolds number Re = 225 based on the maximum inlet velocity and the length of t...	https://arxiv.org/abs/2405.09941v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p003_image_01.png	2405.10135v1	manufacturing_engineering	3	image	[38, 81, 558, 237]	0.822	Figure 1: Overall approach is to identify most unique and informative MVEs for subsequent physics evaluation and surrogate training. Hypothesis is that more efficient training may be performed if MVEs are chosen using an appropriate design criteria. CNN activations can be related to localized features. For instance, a ...	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p005_image_02.png	2405.10135v1	manufacturing_engineering	5	image	[38, 81, 289, 307]	0.6135	Figure 2: VAE schematic for extracting localized MVE features. stochastic materials problem. Perhaps the closest related task occurs in the field of texture analysis (not crystal- lographic texture but image texture) [42]. Interestingly there is a similar implementation, focused on the extrac- tion of statistical featu...	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p006_image_03.png	2405.10135v1	manufacturing_engineering	6	image	[38, 81, 558, 301]	0.9344	Figure 3: Self-supervised approach for 3D MVE feature extraction and network training. Subsampling of MVEs enables self-supervised learning and, critically, encourages the learning of statistical descriptors.	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p006_image_04.png	2405.10135v1	manufacturing_engineering	6	image	[307, 401, 558, 652]	0.6949	Figure 5: Maximum projection design criteria ensures good spread- ing in all possible subspace projections. This ensures that even when unknown unimportant features are present the resulting design still exhibits desirable space-filling properties in the effective lower di- mensional space.	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p006_image_05.png	2405.10135v1	manufacturing_engineering	6	image	[38, 458, 289, 577]	0.8385	Figure 4: Novelty of the feature extraction procedure is that it intrin- sically operates on image statistics via construction of the network. Mean and variance of spatial-orientation feature maps are combined with volume averaged orientation features prior to passing through the final MLP. This encourages the network ...	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.10135v1_p007_image_06.png	2405.10135v1	manufacturing_engineering	7	image	[38, 81, 289, 285]	0.5877	Figure 6: Example designs created from a three dimensional candi- date data set consisting of 1000 points from Unif(−5, 5) and 500 points from N(0, 1). The three dimensional design and one two- dimensional projection is shown. where D is the constructed design (collection of x’s) and d is the Euclidean distance. Here w...	https://arxiv.org/abs/2405.10135v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p005_image_01.png	2405.11596v1	manufacturing_engineering	5	image	[85, 254, 510, 620]	0.9478	Fig. 1. Schematic representation of proposed methodology for developing lattice materials a) bioinspired nesting orders (NOs) mimicking the architecture of cortical bone osteons, b) nesting orientations (NORs) inspired by the fractals and golden spiral, c) three four-fold axis of rotational symmetry (T4FAS) to ensure c...	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p006_image_02.png	2405.11596v1	manufacturing_engineering	6	image	[85, 339, 510, 485]	0.9029	Fig. 2. a) NOs and NORs within nested circles, b) X-cross struts on XY, YZ, and XZ planes of respective NOs and NORs, and c) 3D view of X-cross struts within the nested base lattice	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p008_image_03.png	2405.11596v1	manufacturing_engineering	8	image	[85, 246, 510, 450]	0.9091	Fig. 3. An overview of the steps involved in the design of a unit cell geometry of XNLSs. 2.3. Bio-inspired 3D XNLSs	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p009_image_04.png	2405.11596v1	manufacturing_engineering	9	image	[164, 72, 431, 725]	0.7877	Fig. 4. Mono-XNLSs considering different NOs and NORs: a) XNOOs b) XNBSs and c) XNFSs	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p010_image_05.png	2405.11596v1	manufacturing_engineering	10	image	[156, 291, 439, 569]	0.7588	Fig. 5. Bi-XNLSs considering N0 and N1: a) XNOOs b) XNBSs and c) XNFSs The NOs, N0 and N1 along with their respective NORs, θ0 and θ1, are considered to generate	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p028_image_06.png	2405.11596v1	manufacturing_engineering	28	image	[136, 370, 477, 630]	0.8435	Fig. 15. The transition of anisotropic behavior from Tension/Compression Dominance (blue portion of arrow) to Isotropic to Shear Dominance (orange portion of arrow).	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11596v1_p038_image_07.png	2405.11596v1	manufacturing_engineering	38	image	[156, 505, 439, 659]	0.5411	Fig. S1.1. Meshed solid cube In order to validate the numerical homogenization scheme, we employed a solid cube with a	https://arxiv.org/abs/2405.11596v1	CC-BY-4.0
images/manufacturing_engineering/2405.11599v1_p006_image_01.png	2405.11599v1	manufacturing_engineering	6	image	[125, 74, 327, 235]	0.6277	Fig. 3 Ruddlesden-Popper (RP) faults: (A) shows RP-faults in a nucleated crystal. (B) illustrates the co-existing simulations of the RP structure alongside a homogeneous mixture of ions. structures’ formation, we carried out co-existing simulations of quasi-2D Cs2PbBr4 seeded structure alongside a homogeneous mixture o...	https://arxiv.org/abs/2405.11599v1	CC-BY-4.0
images/manufacturing_engineering/2405.11601v1_p002_vector_01.png	2405.11601v1	manufacturing_engineering	2	vector	[108, 440, 540, 593]	0.5233	II. PRE-REQUSITE KNOWLEDGE A. CLOUD COMPUTING	https://arxiv.org/abs/2405.11601v1	CC-BY-4.0
images/manufacturing_engineering/2405.11601v1_p004_image_02.png	2405.11601v1	manufacturing_engineering	4	image	[146, 296, 518, 432]	0.879	To visualize network traffics using cloud services, the following approach has been implemented:	https://arxiv.org/abs/2405.11601v1	CC-BY-4.0
images/manufacturing_engineering/2405.11601v1_p011_image_03.png	2405.11601v1	manufacturing_engineering	11	image	[74, 298, 542, 504]	0.8648	Figure 6. Proposed architecture using AWS (Amazon Web Services) Managed service AWS serverless capabilities such as S3, Athena, QuickSight, and SageMaker are critical for modern enterprises to modernize an analytics workflow [18]. The proposed architecture that uses the AWS managed services are Glue to prepare data for...	https://arxiv.org/abs/2405.11601v1	CC-BY-4.0
images/manufacturing_engineering/2405.11868v1_p004_vector_01.png	2405.11868v1	manufacturing_engineering	4	vector	[46, 85, 441, 597]	0.6513	Fig. 1. An overview of the taxonomy for existing GCL models. J. ACM, Vol. 1, No. 1, Article . Publication date: May 2024.	https://arxiv.org/abs/2405.11868v1	CC-BY-4.0
images/manufacturing_engineering/2405.11868v1_p007_vector_02.png	2405.11868v1	manufacturing_engineering	7	vector	[46, 85, 440, 246]	0.9628	Fig. 2. The general framework of graph contrastive learning (GCL). A contrastive method can be determined by defining its data augmentation strategy to generate different views, contrastive mode for the alignment between instances at different scales, and corresponding different contrastive optimization strategies. 3.1...	https://arxiv.org/abs/2405.11868v1	CC-BY-4.0
images/manufacturing_engineering/2405.11868v1_p016_vector_03.png	2405.11868v1	manufacturing_engineering	16	vector	[105, 85, 381, 224]	0.9434	Fig. 3. GCL in graph weakly supervised learning. 4 GRAPH CONTRASTIVE LEARNING FOR DATA-EFFICIENT LEARNING	https://arxiv.org/abs/2405.11868v1	CC-BY-4.0
images/manufacturing_engineering/2405.11868v1_p019_image_04.png	2405.11868v1	manufacturing_engineering	19	image	[132, 82, 354, 216]	0.929	Fig. 4. GCL in graph transfer learning. where 1{·} is the indicator function, h𝑠𝑜	https://arxiv.org/abs/2405.11868v1	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p004_vector_01.png	2405.11895v2	manufacturing_engineering	4	vector	[36, 0, 612, 404]	0.8138	(1) Physical Production Line: This module includes two	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p005_vector_02.png	2405.11895v2	manufacturing_engineering	5	vector	[44, 0, 612, 415]	0.8181	IV. QUALITY PREDICTION AND PARAMETER	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p006_vector_03.png	2405.11895v2	manufacturing_engineering	6	vector	[74, 0, 612, 574]	0.8657	It should be noted that all the process parameters and quality indicators are normalized using the maximum-minimum nor- malization method into the range of [0, 1], which handles the potential problem of vast difference between the quantities of process parameters and those of the quality indicators obtained by heteroge...	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p007_vector_04.png	2405.11895v2	manufacturing_engineering	7	vector	[39, 2, 502, 340]	0.7721	B(t) = BN(H(t)) = (H(t) −µ)/σ, (3) where B(t) represents the result of batch normalization at time t, H(t) is the output of dilated convolution following (2), and BN(·) represents the batch normalization operation. µ and σ are the mean value and the standard deviation of the current batch input, respectively.	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p010_vector_05.png	2405.11895v2	manufacturing_engineering	10	vector	[35, 26, 306, 409]	0.8832	(1) At the IoT gateway/edge, we deploy a Kepware-based	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p011_vector_06.png	2405.11895v2	manufacturing_engineering	11	vector	[66, 6, 590, 388]	0.8741	D. Generalizability of the DT Framework	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11895v2_p015_image_07.png	2405.11895v2	manufacturing_engineering	15	image	[57, 58, 292, 218]	0.6557	Fig. 11: The the DT production line on the client GUI. (a) Panel of production process states. (b) Panel of process quality predictions. (c) Panel of recommended parameters after optimization. (d) Weights of the influence of process parameters on quality indicators. (e) Process switching board. (f) Real-time visualizat...	https://arxiv.org/abs/2405.11895v2	CC-BY-4.0
images/manufacturing_engineering/2405.11960v1_p004_vector_01.png	2405.11960v1	manufacturing_engineering	4	vector	[119, 72, 493, 312]	0.9566	Figure 2: Predictive maintenance expert system. The main contribution of this work is the improvement of the existing classification module by means of unsupervised AD techniques (AD module). in the training set are acquired is one day. In order to build the data set, daily information about all the events from a machi...	https://arxiv.org/abs/2405.11960v1	CC-BY-4.0
images/manufacturing_engineering/2405.11960v1_p008_vector_02.png	2405.11960v1	manufacturing_engineering	8	vector	[95, 72, 517, 258]	0.7788	Figure 3: Anomaly detection streaming procedure for on-line auditing of the classifer. Input	https://arxiv.org/abs/2405.11960v1	CC-BY-4.0
images/manufacturing_engineering/2405.11960v1_p008_vector_03.png	2405.11960v1	manufacturing_engineering	8	vector	[212, 302, 400, 537]	0.8762	Figure 4: Proposed architecture for the AD module. The performance metric used to compare the existing classifier (baseline) with the proposed methods was the F1 score, which is a robust metric that combines both precision and recall performances:	https://arxiv.org/abs/2405.11960v1	CC-BY-4.0
images/manufacturing_engineering/2405.12382v2_p007_vector_01.png	2405.12382v2	manufacturing_engineering	7	vector	[109, 74, 461, 272]	0.7165	Figure 1 Diagram of the design for the qubit reservoir network. roughly 100 time steps afterward. Further details about these tasks and the simulation parameters are given in Supplementary Information S.3. The data and code used to obtain our results can be found in a figshare repository [30].	https://arxiv.org/abs/2405.12382v2	CC-BY-4.0
images/manufacturing_engineering/2405.12382v2_p010_vector_02.png	2405.12382v2	manufacturing_engineering	10	vector	[134, 74, 486, 272]	0.877	Figure 3 Diagram of the design for the stochastic optical network. a matrix-vector multiplier and joined with another laser array whose intensities are proportional to the input uk. This applies the linear part of the reservoir action to produce zk = Axk+Buk. Then, each laser in the array is measured with a single phot...	https://arxiv.org/abs/2405.12382v2	CC-BY-4.0
images/manufacturing_engineering/2405.12633v1_p003_image_01.png	2405.12633v1	manufacturing_engineering	3	image	[141, 116, 389, 527]	0.9581	Fig. 1. Diagram of the main operation of the recognition algorithm. The landscape of attendance tracking has long grappled with multifaceted challenges stemming from conventional methods, including verbal roll call, fingerprint scanning, smart cards, and self-registration systems. Each of these methods presents its own...	https://arxiv.org/abs/2405.12633v1	CC-BY-4.0
images/manufacturing_engineering/2405.12633v1_p013_image_02.png	2405.12633v1	manufacturing_engineering	13	image	[145, 346, 393, 540]	0.7465	Fig. 6. Face Recognition algorithm utilizing Haar Cascade for Attendance. 4.3.1. Capturing frames and recognizing faces To capture frames from a live video stream, a specific frame is chosen and pre-processed. We utilize the Haar Cascade face detection algorithm, a pre-trained face detection algorithm from OpenCV2 that...	https://arxiv.org/abs/2405.12633v1	CC-BY-4.0
images/manufacturing_engineering/2405.12633v1_p017_image_03.png	2405.12633v1	manufacturing_engineering	17	image	[165, 402, 396, 551]	0.7614	Fig. 10. The LBPH Method. The grayscale image is processed by taking a 3x3 pixel portion of the image as shown in Fig. 10. Each pixel’s intensity is represented in a matrix format. The threshold value for the center pixel is taken as the reference, and the intensity values of the neighboring pixels are compared with it...	https://arxiv.org/abs/2405.12633v1	CC-BY-4.0
images/manufacturing_engineering/2405.12633v1_p018_image_04.png	2405.12633v1	manufacturing_engineering	18	image	[108, 164, 454, 269]	0.7586	Fig. 11. Trainer.yml is utilized in the recognition process. 4.3.6. The trainer To teach the OpenCV Recognizer to recognize faces in a real-time video stream, the next step was to train the model by executing Algorithm 3. This algorithm relocated all face images from the dataset folder to the trainer.	https://arxiv.org/abs/2405.12633v1	CC-BY-4.0
images/manufacturing_engineering/2405.12962v1_p004_image_01.png	2405.12962v1	manufacturing_engineering	4	image	[137, 72, 612, 344]	0.6829	3.2 System performance metrics	https://arxiv.org/abs/2405.12962v1	CC-BY-4.0
images/manufacturing_engineering/2405.12962v1_p005_image_02.png	2405.12962v1	manufacturing_engineering	5	image	[187, 524, 585, 748]	0.822	null	https://arxiv.org/abs/2405.12962v1	CC-BY-4.0
images/manufacturing_engineering/2405.13468v1_p007_image_01.png	2405.13468v1	manufacturing_engineering	7	image	[428, 91, 520, 477]	0.7217	Fig. 2: Schematic of the C3PO (left) and C-LANDO (right) ar- chitectures, showing the different layers and sizes of the input, the dilation of this input at different layers, and finally the out- put format. The top grey block represents the input, with dimen- sions printed next to the arrow (e.g., for C3PO: 20 velocit...	https://arxiv.org/abs/2405.13468v1	CC-BY-4.0
images/manufacturing_engineering/2405.13512v1_p003_vector_01.png	2405.13512v1	manufacturing_engineering	3	vector	[328, 449, 547, 558]	0.7516	Fig. 2. Overall approach: optimizer interacting with process model Baeuerle et al. [2] have presented a model for both for the dispensing and the packaging processes. The input of the first model is the dispense path. This input path is shown on the left side of the process model in Figure 2. It is parameterized with a...	https://arxiv.org/abs/2405.13512v1	CC-BY-4.0
images/manufacturing_engineering/2405.13512v1_p003_vector_02.png	2405.13512v1	manufacturing_engineering	3	vector	[65, 96, 284, 191]	0.7576	Fig. 1. Area coverage types during Coverage Path Planning problem settings. Most approaches consider a constant path width. The above approaches can be clustered according to the underlying area coverage behavior for the cases, which are shown in Figure 1. Almost all of the classic coverage path planning approaches con...	https://arxiv.org/abs/2405.13512v1	CC-BY-4.0
images/manufacturing_engineering/2405.13512v1_p005_vector_03.png	2405.13512v1	manufacturing_engineering	5	vector	[81, 63, 520, 361]	0.8842	Fig. 3. Detailed workflow with intermediate in- and outputs area and wcomp,tab being the weighting factor for the taboo zones.	https://arxiv.org/abs/2405.13512v1	CC-BY-4.0
images/manufacturing_engineering/2405.13604v1_p006_image_01.png	2405.13604v1	manufacturing_engineering	6	image	[166, 312, 475, 648]	0.7158	Figure 2: Skills composition by the BTs backchaining. The BTs backchaining has some similarities to backward search algorithms from automated planning. Its working principal is shown in Figure 2. A goal or a set of goals is presented as a sequence composition of conditions (C1, C2, C3). The priorities go in the sequenc...	https://arxiv.org/abs/2405.13604v1	CC-BY-4.0
images/manufacturing_engineering/2405.13604v1_p008_image_02.png	2405.13604v1	manufacturing_engineering	8	image	[157, 72, 484, 282]	0.9417	Figure 4: Snowflake topology for the execution of BTs as distributed components. conditions remain fulfilled, then the composite structure again results in a consistent protocol. Composed is this way, computation processes form a so-called snowflake topology without any cycles.	https://arxiv.org/abs/2405.13604v1	CC-BY-4.0
images/manufacturing_engineering/2405.13604v1_p009_image_03.png	2405.13604v1	manufacturing_engineering	9	image	[58, 72, 526, 334]	0.8161	Figure 5: Implementation of the framework in the 4DIAC. operators). Thanks to the self-similarity property of BTs this interface and these composition rules are applied identically on all the levels of control hierarchy.	https://arxiv.org/abs/2405.13604v1	CC-BY-4.0
images/manufacturing_engineering/2405.13702v1_p012_vector_01.png	2405.13702v1	manufacturing_engineering	12	vector	[68, 257, 531, 595]	0.5744	Fig. 4 Photoluminescence (PL) spectra and energy level model. a Normalized PL intensity graph under 367 nm excitation from 78 K to 98 K of 750 ℃ and 790 ℃ sample. b Energy level model though	https://arxiv.org/abs/2405.13702v1	CC-BY-4.0
images/manufacturing_engineering/2405.14058v2_p002_vector_01.png	2405.14058v2	manufacturing_engineering	2	vector	[312, 523, 568, 633]	0.6703	III. RELATED WORK	https://arxiv.org/abs/2405.14058v2	CC-BY-4.0
images/manufacturing_engineering/2405.14058v2_p013_vector_02.png	2405.14058v2	manufacturing_engineering	13	vector	[175, 417, 437, 565]	0.5076	Fig. 5: A scheme of the DNN controller architecture for the 2D docking benchmark. Given an input state of the system, x = [xt,yt, ˙xt, ˙yt]T , the DNN outputs the forces [Fx,Fy], which are converted using dynamics f to produce the next state. The states generated in this trajectory pertain to the deputy spacecraft, whi...	https://arxiv.org/abs/2405.14058v2	CC-BY-4.0
images/manufacturing_engineering/2405.14058v2_p016_image_03.png	2405.14058v2	manufacturing_engineering	16	image	[253, 59, 359, 560]	0.6043	Fig. 8: Visualized architecture of the redesigned neural network controller. The visualization of the retrained deep learning model reflects the structure and the data flow of the model, its layers and how they are connected. Each box represents a layer in the neural network. The upper and lower parts of the box indica...	https://arxiv.org/abs/2405.14058v2	CC-BY-4.0
images/manufacturing_engineering/2405.14505v1_p004_image_01.png	2405.14505v1	manufacturing_engineering	4	image	[81, 61, 495, 428]	0.9328	FIGURE1: System architecture.	https://arxiv.org/abs/2405.14505v1	CC-BY-4.0
images/manufacturing_engineering/2405.14505v1_p009_image_02.png	2405.14505v1	manufacturing_engineering	9	image	[44, 302, 270, 529]	0.6116	(b) SVC similarity confusion matrix.	https://arxiv.org/abs/2405.14505v1	CC-BY-4.0
images/manufacturing_engineering/2405.14505v1_p009_image_03.png	2405.14505v1	manufacturing_engineering	9	image	[44, 61, 270, 287]	0.6356	(a) RF similarity confusion matrix.	https://arxiv.org/abs/2405.14505v1	CC-BY-4.0
images/manufacturing_engineering/2405.14548v2_p004_vector_01.png	2405.14548v2	manufacturing_engineering	4	vector	[71, 71, 541, 157]	0.9437	Figure 1: Coupling between the flow and transport simulator (IC-FERST) and the geochemical simulator (PHREEQC). where u is the Darcy velocity and µ is the viscosity. K is the permeability tensor, ρ is the density, p is the pressure, ∇z is the gravity direction, and g is the gravitational acceleration.	https://arxiv.org/abs/2405.14548v2	CC-BY-4.0
images/manufacturing_engineering/2405.14548v2_p006_vector_02.png	2405.14548v2	manufacturing_engineering	6	vector	[118, 71, 494, 208]	0.8507	Figure 2: Schematic of the cation exchange in a small portion of the porous space. The outflow from the schematic will be the inflow in the next portion of the domain in the next time step. Inputs Outputs	https://arxiv.org/abs/2405.14548v2	CC-BY-4.0
images/manufacturing_engineering/2405.14548v2_p006_vector_03.png	2405.14548v2	manufacturing_engineering	6	vector	[165, 257, 447, 421]	0.8926	Figure 3: Inputs and outputs of the geochemical reaction in the cation exchange problem. The reaction is performed for each grid cell at each time step. 8 cm	https://arxiv.org/abs/2405.14548v2	CC-BY-4.0
images/manufacturing_engineering/2405.14548v2_p007_vector_04.png	2405.14548v2	manufacturing_engineering	7	vector	[71, 489, 541, 576]	0.9463	Figure 6: Coupling between the flow and transport simulator (IC-FERST) and the machine learning surrogate. The reaction calculations need to be performed for each grid cell of the discretized domain and for each iteration of the coupling procedure. Here, since we use a sequential non-iterative approach and consider tha...	https://arxiv.org/abs/2405.14548v2	CC-BY-4.0
images/manufacturing_engineering/2405.14548v2_p017_image_05.png	2405.14548v2	manufacturing_engineering	17	image	[71, 568, 541, 654]	0.9112	Figure 16: Coupling between the flow and transport simulator (IC-FERST) and the machine learning surrogate. We add a post-processing step to guarantee mass/charge balance. 17	https://arxiv.org/abs/2405.14548v2	CC-BY-4.0
images/manufacturing_engineering/2405.14552v1_p002_image_01.png	2405.14552v1	manufacturing_engineering	2	image	[95, 563, 517, 700]	0.7768	Figure 1: IOLWS Safety PDU based on [8, 9]. 2	https://arxiv.org/abs/2405.14552v1	CC-BY-4.0
images/manufacturing_engineering/2405.14552v1_p003_image_02.png	2405.14552v1	manufacturing_engineering	3	image	[95, 558, 517, 689]	0.8662	Figure 2: a) Modular sensor-2-cloud automation topology, based on [6]. b) Simplified scenario comparison of two wireless cells and one roaming (FS-)W-Device. 3	https://arxiv.org/abs/2405.14552v1	CC-BY-4.0
images/manufacturing_engineering/2405.14552v1_p004_image_03.png	2405.14552v1	manufacturing_engineering	4	image	[95, 72, 517, 239]	0.8336	Figure 3: Measurement setup, comprising of two (FS-)W-Masters and one (FS-)W-Device. Table 1: Results for IOLW and IOLWS roaming connection times.	https://arxiv.org/abs/2405.14552v1	CC-BY-4.0
images/manufacturing_engineering/2405.16084v1_p002_image_01.png	2405.16084v1	manufacturing_engineering	2	image	[54, 423, 306, 514]	0.8242	Fig. 3. A diagram of the manipulator showing the tendon routing paths, and the pan and tilt directions for the proximal and distal modules. [11] A rectangular enclosure at the distal end of the tube, shown in fig. 4, contains the driving pulleys and actuators, the Arduino Mega control board, a servo control and power i...	https://arxiv.org/abs/2405.16084v1	CC-BY-4.0
images/manufacturing_engineering/2405.16084v1_p003_image_02.png	2405.16084v1	manufacturing_engineering	3	image	[335, 50, 536, 273]	0.9068	Fig. 6. The architecture of the control software for the complete macro- micro system. state of the arm to determine its trajectory in upcoming time frames.	https://arxiv.org/abs/2405.16084v1	CC-BY-4.0
images/manufacturing_engineering/2405.16084v1_p003_image_03.png	2405.16084v1	manufacturing_engineering	3	image	[76, 50, 277, 159]	0.6837	Fig. 5. The system hardware components and their interfaces. III. ROBOTIC SYSTEM ARCHITECTURE	https://arxiv.org/abs/2405.16084v1	CC-BY-4.0
images/manufacturing_engineering/2405.16084v1_p005_image_04.png	2405.16084v1	manufacturing_engineering	5	image	[101, 100, 252, 205]	0.782	Fig. 9. A diagram of the Aurora electromagnetic tracking system, depicting the tracking needle and the coordinate axes of the field generator [19]. The tracker measured the displacement of a sensor needle fixed to the micro manipulator’s end effector while it was guided through varying trajectories using the Touch styl...	https://arxiv.org/abs/2405.16084v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p001_vector_01.png	2405.16183v1	manufacturing_engineering	1	vector	[319, 173, 612, 512]	0.9168	largely because machine learning has the potential to expe- dite prediction processes by effectively leveraging existing datasets (Ladick`y et al., 2015; Kochkov et al., 2021; Pichi et al., 2024). Moreover, when trained with actual measured data, these models can achieve enhanced prediction accuracy (Lam et al., 2023)....	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p006_vector_02.png	2405.16183v1	manufacturing_engineering	6	vector	[55, 67, 287, 197]	0.6547	time t, to facilitate the temporal smoothness. This form dif- fers from that proposed in Brandstetter et al. (2022) because we need this operation to be linear for conservation.	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p007_image_03.png	2405.16183v1	manufacturing_engineering	7	image	[91, 51, 524, 359]	0.6211	4.2. Navier–Stokes Equations with Mixture	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p013_vector_04.png	2405.16183v1	manufacturing_engineering	13	vector	[104, 67, 493, 286]	0.6765	Figure 8. Geometry and variables used to construct FVM, focusing on the i-th and j-th cells. There are extensive variations of the interpolation, such as the semi-Lagrangian method and Lax–Wendroff method (Lax & Wendroff, 1960). The gradient in the direction of dij := xj −xi for the diffusion term can be computed as	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p022_image_05.png	2405.16183v1	manufacturing_engineering	22	image	[90, 471, 523, 779]	0.7142	null	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p027_image_06.png	2405.16183v1	manufacturing_engineering	27	image	[0, 103, 534, 540]	0.5004	• A substantial part of the error of FluxGNN is mainly due to the pressure. It seems that pressure starts to be unstable earlier than velocity and volume fraction, implying that instability of pressure would be key to establishing a more stable model. MP-PDE has a lower loss than FluxGNN at t 8.0, but these outputs are...	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16183v1_p029_image_07.png	2405.16183v1	manufacturing_engineering	29	image	[0, 232, 534, 668]	0.5522	29	https://arxiv.org/abs/2405.16183v1	CC-BY-4.0
images/manufacturing_engineering/2405.16729v2_p002_image_01.png	2405.16729v2	manufacturing_engineering	2	image	[73, 55, 536, 203]	0.9191	Fig. 1. (a) System diagram showing the general connections between instruments and turbulence ﬂow through the optical path, (b) View of the lab hallway with four turbulence generators marked . in classifying received FSO data transmitted through various turbulent channels. We also present details of the ML approach and...	https://arxiv.org/abs/2405.16729v2	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p001_vector_01.png	2405.17582v1	manufacturing_engineering	1	vector	[71, 158, 524, 749]	0.531	null	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p002_image_03.png	2405.17582v1	manufacturing_engineering	2	image	[323, 329, 515, 465]	0.8197	A branch of machine learning – Source: Internet	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p002_vector_02.png	2405.17582v1	manufacturing_engineering	2	vector	[72, 59, 280, 307]	0.6368	2. Related Works	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p003_vector_04.png	2405.17582v1	manufacturing_engineering	3	vector	[315, 59, 523, 502]	0.6	Neural network model RNN, source: internet	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p003_vector_05.png	2405.17582v1	manufacturing_engineering	3	vector	[315, 605, 523, 744]	0.7258	null	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p004_image_07.png	2405.17582v1	manufacturing_engineering	4	image	[94, 255, 258, 435]	0.5405	With:	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p004_vector_06.png	2405.17582v1	manufacturing_engineering	4	vector	[72, 59, 280, 249]	0.713	The recursive neural network activity model	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p004_vector_08.png	2405.17582v1	manufacturing_engineering	4	vector	[72, 541, 280, 623]	0.562	Methods	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17582v1_p005_image_09.png	2405.17582v1	manufacturing_engineering	5	image	[79, 284, 287, 376]	0.6562	Testing results from the temperature forecasting model with RNN:	https://arxiv.org/abs/2405.17582v1	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p003_vector_01.png	2405.17631v3	manufacturing_engineering	3	vector	[113, 181, 499, 315]	0.757	Figure 1: An AI agent for closed-loop experiment design. (a) Conventional Bayesian optimization approach for experiment design involves training a machine learning model in every experimental round, scoring all perturbations and defining an acquisition function for selecting genes to perturb in the next round. (b) Over...	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p003_vector_02.png	2405.17631v3	manufacturing_engineering	3	vector	[113, 98, 500, 163]	0.8677	Figure 1: An AI agent for closed-loop experiment design. (a) Conventional Bayesian optimization approach for experiment design involves training a machine learning model in every experimental round, scoring all perturbations and defining an acquisition function for selecting genes to perturb in the next round. (b) Over...	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p016_vector_03.png	2405.17631v3	manufacturing_engineering	16	vector	[108, 124, 504, 397]	0.5178	null	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p017_vector_04.png	2405.17631v3	manufacturing_engineering	17	vector	[108, 135, 504, 467]	0.6817	C ALGORITHM FOR BIODISCOVERYAGENT	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p018_vector_05.png	2405.17631v3	manufacturing_engineering	18	vector	[301, 263, 496, 363]	0.7371	Figure 4: Gene selection strategy: (a) The space of genes that can be tested in a given experiment is constrained by expeirmental limitations. BioDiscoveryAgent can take a few tries to select genes within this limited space. (b) A common error is repeating previously tested genes. (c) Often this will result in the agen...	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p018_vector_06.png	2405.17631v3	manufacturing_engineering	18	vector	[300, 378, 480, 478]	0.6466	Figure 4: Gene selection strategy: (a) The space of genes that can be tested in a given experiment is constrained by expeirmental limitations. BioDiscoveryAgent can take a few tries to select genes within this limited space. (b) A common error is repeating previously tested genes. (c) Often this will result in the agen...	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p019_vector_07.png	2405.17631v3	manufacturing_engineering	19	vector	[123, 137, 467, 527]	0.6734	Figure 5: BioDiscoveryAgent workflow with all tools over a single experimental round. Prompts and agent responses have been summarized. See Appendix G for full trace. (a) The input to the agent is the description of the problem. (b) In case of the literature search tool, the LLM first determines appropriate search term...	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p020_vector_08.png	2405.17631v3	manufacturing_engineering	20	vector	[144, 350, 504, 579]	0.5985	In addition to the above prompt, the critic agent was also provided with a list of all genes that were tested in the previous rounds along with genes that were identified as hits.	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17631v3_p027_vector_09.png	2405.17631v3	manufacturing_engineering	27	vector	[140, 128, 454, 313]	0.5329	Figure 7: Percentage of new genes predicted by Claude Haiku only when using tools Ha −Hn that are also predicted by Claude Sonnet with no-tools (Sn), where Hn is the set of genes predicted by Claude Haiku with no-tools and Ha is the set of genes predicted by Claude Haiku with all-tools Avg. Cost per	https://arxiv.org/abs/2405.17631v3	CC-BY-4.0
images/manufacturing_engineering/2405.17636v1_p001_image_01.png	2405.17636v1	manufacturing_engineering	1	image	[313, 157, 558, 326]	0.6781	Fig. 1: An overview of the used the sensorized CM with the OFDR- based SSA. Figure also shows the SSA structure modeled as a com- posite beam. cannot completely address this poor accuracy and may increase the wavelength shifts interfering with each other, especially during high curvature bending [8], [9].	https://arxiv.org/abs/2405.17636v1	CC-BY-4.0
images/manufacturing_engineering/2405.17636v1_p002_image_02.png	2405.17636v1	manufacturing_engineering	2	image	[54, 50, 299, 188]	0.6628	Fig. 2: Fabrication setup used to mount the OFDR fiber on a flat NiTi wire. and a single OFDR fiber. We also demonstrated the calibration procedure for this sensor. However, in that study, we did not evaluate the SS performance of this SSA inside a CM and in different bending situations. Therefore, in this paper and as...	https://arxiv.org/abs/2405.17636v1	CC-BY-4.0

End of preview. Expand in Data Studio

STEM Diagrams

30,325 technical diagrams (block diagrams, schematics, flowcharts, architectures) extracted from arXiv papers across six engineering fields, each with a source attribution and a quality score. Built by an LLM-curated pipeline and used to show that a small frozen-feature classifier can replace the paid LLM labeling gate.

Paper: Distilling an LLM Diagram-Curation Pipeline into Local Classifiers (Adnan Abbasi, Thothica, 2026)
Code: https://github.com/adoistic/stem-diagrams
Fields: manufacturing, robotics & automation, telecommunications, utilities & power systems, semiconductor engineering

What's here

Path	Contents
`images/{field}/{name}.png`	30,325 diagram images, foldered by field
`metadata.csv`	one row per image: `file_name`, `arxiv_id`, `field`, `page`, `method`, `bbox`, `p_diagram`, `caption`, `source_arxiv_url`, `license`
`releases/stem-diagrams-v1.zip`	the earlier 2,000-diagram v1 release
`extras/`	label exports (`labels.csv`, xlsx), gallery manifest

Source PDFs are not redistributed

Only the individual diagram figures are published here, with per-image source attribution. The full arXiv papers are not hosted (copyright; they are freely available from arXiv). To rebuild the source PDFs locally from the arxiv_id column, use fetch_source_pdfs.py.

Licensing

Labels and metadata are released under CC BY 4.0. Diagram images are figures from arXiv papers under their original licenses; the source arxiv_id is retained per record so provenance is always recoverable.

Citation

@misc{abbasi2026stemdiagrams,
  title  = {Distilling an LLM Diagram-Curation Pipeline into Local Classifiers},
  author = {Adnan Abbasi},
  year   = {2026},
  note   = {Thothica. https://github.com/adoistic/stem-diagrams}
}

Downloads last month: 28