Injection-molded plastic parts defect dataset: raw images, segmentation masks, and quantitative defect metrics under varying process parameters

Injection-Molded Polypropylene Parts: Image Dataset with Segmented Defects and CAD Geometry Overview This dataset contains a collection of images of polypropylene (PP) parts manufactured by plastic injection molding under controlled variations of six process parameters. The dataset is intended to support research in computer vision, deep learning, automated defect detection, explainable artificial intelligence, and process–quality correlation studies within Industry 4.0 and smart manufacturing environments. For each manufactured part, the dataset includes: Raw high-resolution images of the injected parts. Pixel-level segmentation masks of visible surface defects. CAD geometry of the reference injected part and sprue system. Injection molding process parameters used during production. Quantitative defect areas associated with each parameter configuration. The dataset provides: A complete raw image dataset. A labeled dataset split into training, validation, and testing subsets. COCO-format annotations for segmentation tasks. A structured CSV dataset containing process parameters and quantified defects. A CAD model of the molded geometry and feeding system. Associated publication. This dataset originates from the experimental campaign reported in the following peer-reviewed article, which should be consulted for the full description of the inspection setup, defect quantification methodology, and modeling approach: Sánchez-Calleja, I., Ferrero-Guillén, R., Martínez-Gutiérrez, A., Díez-González, J., & Perez, H. (in press). Explainable Closed-Loop Injection Molding Parameter Tuning via Robotic Inspection and Quantitative Multidefect Modeling. DOI: [to be added upon publication]. Companion dataset. Complementary experimental process data on the same injected parts — including injection process parameters, individual and average part weights, energy consumption, and cycle time — is published in a separate Zenodo record associated with the following peer-reviewed article: Companion Zenodo. DOI: https://doi.org/10.5281/zenodo.20309380 Companion Zenodo Associated publication: Pérez-González, J., Sánchez-Calleja, I., Fernández-Gorgojo, A., Ferrero-Guillén, R., Martínez-Gutiérrez, A., & Díez-González, J. (in press). Explainability-Driven Robust Multi-Objective Optimization for Plastic Injection Molding Parameter Adjustment. Journal of Intelligent Manufacturing. DOI: [to be added upon publication]. Users of this dataset are encouraged to read both articles first to understand the context, scope, and intended use of the data, and to consult the companion dataset when process parameter information is required. Experimental Setup Item Description Material SABIC 579S homopolymer polypropylene Manufacturing process Plastic injection molding Mold type Two-cavity multicavity mold Parts per injection cycle Two injected parts CAD geometry Included (units in mm) CAD contents Injected part geometry and sprue Dataset type Experimental manufacturing dataset The experiments were carried out using a two-cavity multicavity mold producing two parts per injection cycle. The included CAD model describes both the injected geometry and the sprue system, enabling geometric interpretation, simulation studies, and process-analysis applications. Dataset Contents Images_dataset.rar Dataset_labeled.zip Quantitative_defects_and_process_parameters.csv Mold_geometry.step README.md File Description Images_dataset.rar Raw high-resolution images of the injected parts grouped by batch identifier Dataset_labeled.zip Images split into train/validation/test subsets together with segmentation masks and COCO-format annotations Quantitative_defects_and_process_parameters.csv Dataset containing injection molding parameters, process viability labels, and quantified defect areas Mold_geometry.step CAD model of the injected geometry and sprue system README.md Documentation of the dataset structure and variables Experimental Design The complete dataset contains 1000 parameter configurations (batches) corresponding to varying injection parameters. For each batch, 5 injection shots were performed, and since the mold has two cavities, each shot produces 2 parts. This results in 10 inspected parts per batch, each photographed twice, yielding 20 images per batch. Only feasible injection parameters that provided inspectionable parts were considered, yielding a total of 14620 pictures. The 1000 batches were created following two strategies: The first subset contains 768 parameter configurations generated using a structured DoE grid based on the main injection molding parameters. The second subset contains 232 additional configurations selected using a maximin criterion to increase parameter-space diversity and reduce dependency on a perfectly regular parameter mesh. A Design of Experiments (DoE) methodology was used to systematically vary six injection molding process parameters within the following ranges: Parameter Description Range of Values Step Value* Tinj Injection temperature [180-240] ºC 20ºC tinj Injection time [0.5 - 2] s 0.5 s Pinj Injection pressure [5 - 35] bar 10 bar Bp Back pressure [15 - 30] bar 15 bar Ph Holding pressure [5 - 35] bar 15 bar th Holding time [4 - 6] s 2 s *Step Values corresponding to the first 768 batches. Defect Classes The dataset includes annotations for the following injection molding defect classes: Air traps Cold slugs Cold spots Flash Flow marks Short shots Silver streaks Voids Weld lines Wrinkles Raw Images Dataset The file Images_dataset.rar contains raw images with a resolution of 6000 × 4000 pixels. Images are organized into folders named according to the corresponding batch identifier. Labeled Dataset The file Dataset_labeled.zip contains the labeled dataset prepared for image segmentation tasks. The dataset is divided into training, validation, and testing subsets and includes COCO-format annotations. Folder Structure Folder Description Contents annotations COCO-format annotation files instance_train.json, instance_val.json, instance_test.json train Training images .jpg files val Validation images .jpg files test Test images .jpg files Injection Parameters and Quantified Defects Dataset The file Quantitative_defects_and_process_parameters.csv contains one row per injected part. Each row includes: Batch identifier Part identifier Injection identifier Process viability label Injection molding parameters Quantified defect areas for each defect class Example: variables values for the second part of the total 10 parts produced in the batch (2/10); of the third injection (3/5); and of the fourth batch (4/1000) would contain: batch: 4 part_number: 2 injection_number: 3 CSV Structure Column Description Unit / Values batch Injection batch identifier int part_number Part identifier within batch int injection_number Injection identifier within batch int Tinj Injection temperature °C tinj Injection time s Pinj Injection pressure bar Bp Back pressure bar Ph Holding pressure bar th Holding time s Air_Trap Air trap defect area mm² Cold_Slug Cold slug defect area mm² Cold_Spot Cold spot defect area mm² Flash Flash defect area mm² Flow_Mark Flow mark defect area mm² Short_Shot Short shot defect area mm² Silver_Streak Silver streak defect area mm² Void Void defect area mm² Weld_Line Weld line defect area mm² Wrinkle Wrinkle defect area mm² CAD Geometry The dataset includes a CAD file describing the injected geometry and sprue system. Item Description Geometry type Injection-molded polypropylene parts Included geometry Injected parts and sprue Mold configuration Two-cavity multicavity mold Production output Two parts per injection cycle Purpose Geometry documentation, simulation support, and process interpretation The CAD geometry enables the relation between experimental process data and the physical characteristics of the molded component and the feeding system. Intended Applications This dataset is suitable for: Training and benchmarking deep learning models for defect detection and segmentation. Developing explainable AI models for manufacturing quality analysis. Studying correlations between process parameters and defect generation. Developing process monitoring and quality-control strategies. Research in Industry 4.0 and smart manufacturing. Educational use in manufacturing, machine vision, and AI courses. License This dataset is released under the Creative Commons Attribution 4.0 International License. CC BY 4.0 Contact Corresponding author: Iván Sánchez-CallejaEmail: [e-mail]: 0009-0009-5321-230X Funding This work was partially supported by the Spanish Research Agency (AEI) under grant number PID2023-153047OB100, by the Department of Education of the Regional Government of Castile and León, and the Universidad de León. The author Iván Sánchez-Calleja acknowledges funding for doctoral studies of the University of León.