Abstract
In the dynamic landscape of Industry 4.0, optimizing industrial processes like injection molding for enhanced efficiency and quality is paramount. This study endeavors to address this imperative of developing an automatic defect detection system, crucial for achieving intelligent process control and meeting evolving market demands while ensuring cost-effectiveness and sustainability. Utilizing transfer learning, this paper aims to explore and to verify the efficacy of three deep learning architectures—InceptionV2, ResNet50, and Inception-Resnet—unexplored in previous research on defect detection in the field of injection molding. The methodology adopted includes two essential steps to achieve the desired objective. The first step consists of training and testing the three RCNN architectures retained on a small data set after having determined the best values of three hyperparameters considered—learning rate, momentum, and number of iterations—allowing the obtaining of a better detection accuracy. The second step consists of improving the architecture of the best model obtained—here Inception v2—by using its last version v3, to consider and tune the values of additional hyperparameters—Solver and Bach size—and to use a training large dataset after adding other parts and proceeding with the different data augmentation techniques. Compared to the results obtained with the pre-trained models explored in previous studies such as MobileNet-SSD, VGG16, and YOLO, the results obtained during this study and particularly after the improvements undertaken are considered very satisfactory with an accuracy of 92%, a recall of 100%, and an F1 score of 90%. The results obtained during the test in the real world of production in the presence of external factors were very promising and revealed the importance of the luminosity factor as an influence on the precision and speed of detection.
Similar content being viewed by others
Data availability
The data used in this article is available on request (image dataset, python code, detection trained model, etc.).
References
el Ghadoui M, Mouchtachi A, Majdoul R (2023) A hybrid optimization approach for intelligent manufacturing in plastic injection molding by using artificial neural network and genetic algorithm. Sci Rep 13(1):1. https://doi.org/10.1038/s41598-023-48679-0
Chen JC, Guo G, Wang W-N (2020) Artificial neural network-based online defect detection system with in-mold temperature and pressure sensors for high precision injection molding. Int J Adv Manuf Technol 110(7–8):2023–2033. https://doi.org/10.1007/s00170-020-06011-4
Fu G et al (2019) A deep-learning-based approach for fast and robust steel surface defects classification. Opt Lasers Eng 121:397–405. https://doi.org/10.1016/j.optlaseng.2019.05.005
el Ghadoui M, Mouchtachi A, Majdoul R (2023) Intelligent surface roughness measurement using deep learning and computer vision: a promising approach for manufacturing quality control. Int J Adv Manuf Technol 129(7):3261–3268. https://doi.org/10.1007/s00170-023-12457-z
Kim G, Choi JG, Ku M, Cho H, Lim S (2021) A multimodal deep learning-based fault detection model for a plastic injection molding process. IEEE Access 9:132455–132467. https://doi.org/10.1109/ACCESS.2021.3115665
Zhu Y et al (2022) Faster-RCNN based intelligent detection and localization of dental caries. Displays 74:102201. https://doi.org/10.1016/j.displa.2022.102201
Li K, Wang X, Ji L (2019) Application of multi-scale feature fusion and deep learning in detection of steel strip surface defect. 2019 International Conference on Artificial Intelligence and Advanced Manufacturing (AIAM). pp 656–661. https://doi.org/10.1109/AIAM48774.2019.00136
Sun X, Gu J, Tang S, Li J (2018) Research progress of visual inspection technology of steel products—a review. Appl Sci 811:11. https://doi.org/10.3390/app8112195
Sun X, Gu J, Huang R, Zou R, Giron Palomares B (2019) Surface defects recognition of wheel hub based on improved faster R-CNN. Electronics 8(5):5. https://doi.org/10.3390/electronics8050481
Hoang D-T, Kang H-J (2019) A survey on deep learning based bearing fault diagnosis. Neurocomputing 335:327–335. https://doi.org/10.1016/j.neucom.2018.06.078
Czimmermann T et al (2020) Visual-based defect detection and classification approaches for industrial applications—a survey. Sensors 20(5):5. https://doi.org/10.3390/s20051459
Bang H-T, Park S, Jeon H (2020) Defect identification in composite materials via thermography and deep learning techniques. Compos Struct 246:112405. https://doi.org/10.1016/j.compstruct.2020.112405
Zhang W et al (2019) RCNN-based foreign object detection for securing power transmission lines (RCNN4SPTL). Procedia Computer Science 147:331–337. https://doi.org/10.1016/j.procs.2019.01.232
Hu H, Tang C, Shi C, Qian Y (2023) Detection of residual feed in aquaculture using YOLO and Mask RCNN. Aquacult Eng 100:102304. https://doi.org/10.1016/j.aquaeng.2022.102304
Kiran BR, Thomas DM, Parakkal R (2018) An overview of deep learning based methods for unsupervised and semi-supervised anomaly detection in videos. J Imaging 4(2):2. https://doi.org/10.3390/jimaging4020036
T. Schlegl, P. Seeböck, S. M. Waldstein, U. Schmidt-Erfurth, and G. Langs (2017) “Unsupervised anomaly detection with generative adversarial networks to guide marker discovery.” arXiv. https://doi.org/10.48550/arXiv.1703.05921.
Song H, Jiang Z, Men A, Yang B (2017) A hybrid semi-supervised anomaly detection model for high-dimensional data. Comput Intell Neurosci 2017:8501683. https://doi.org/10.1155/2017/8501683
L. Ruff et al., “Deep semi-supervised anomaly detection.” arXiv, Feb. 14, 2020. Accessed: Sep. 21, 2023. [Online]. Available: http://arxiv.org/abs/1906.02694
Sharma V, Mir RN (2022) Saliency guided faster-RCNN (SGFr-RCNN) model for object detection and recognition. Journal of King Saud University - Computer and Information Sciences 34(5):1687–1699. https://doi.org/10.1016/j.jksuci.2019.09.012
Bhatt PM et al (2021) Image-based surface defect detection using deep learning: a review. J Comput Inf Sci Eng 21:040801. https://doi.org/10.1115/1.4049535
Xu X, Zheng H, Guo Z, Wu X, Zheng Z (2019) SDD-CNN: small data-driven convolution neural networks for subtle roller defect inspection. Appl Sci 9(7):7. https://doi.org/10.3390/app9071364
Khumaidi A, Yuniarno EM, Purnomo MH (2017) Welding defect classification based on convolution neural network (CNN) and Gaussian kernel. 2017 International Seminar on Intelligent Technology and Its Applications (ISITIA). pp 261–265. https://doi.org/10.1109/ISITIA.2017.8124091
Li Y, Huang H, Xie Q, Yao L, Chen Q (2018) Research on a surface defect detection algorithm based on MobileNet-SSD. Appl Sci 8(9):1678. https://doi.org/10.3390/app8091678
Hu Z, Yin Z, Qin L, Xu F (2022) A novel method of fault diagnosis for injection molding systems based on improved VGG16 and machine vision. Sustainability 14(21):21. https://doi.org/10.3390/su142114280
Zhang Y et al (2022) Automated vision-based inspection of mould and part quality in soft tooling injection moulding using imaging and deep learning. CIRP Ann 71(1):429–432. https://doi.org/10.1016/j.cirp.2022.04.022
Liang H, Chen J, Xie W, Yu X, Wu W (2022) Defect detection of injection-molded parts based on improved-YOLOv5. J Phys: Conf Ser 2390(1):012049. https://doi.org/10.1088/1742-6596/2390/1/012049
Ha H, Jeong J (2021) CNN-based defect inspection for injection molding using edge computing and industrial IoT systems. App Sci 11(14):14. https://doi.org/10.3390/app11146378
Liu J, Guo F, Gao H, Li M, Zhang Y, Zhou H (2021) Defect detection of injection molding products on small datasets using transfer learning. J Manuf Process 70:400–413. https://doi.org/10.1016/j.jmapro.2021.08.034
Lockner Y, Hopmann C, Zhao W (2022) Transfer learning with artificial neural networks between injection molding processes and different polymer materials. J Manuf Process 73:395–408. https://doi.org/10.1016/j.jmapro.2021.11.014
Staar B, Lütjen M, Freitag M (2019) Anomaly detection with convolutional neural networks for industrial surface inspection. Procedia CIRP 79:484–489. https://doi.org/10.1016/j.procir.2019.02.123
Tercan H, Guajardo A, Heinisch J, Thiele T, Hopmann C, Meisen T (2018) Transfer-learning: bridging the gap between real and simulation data for machine learning in injection molding. Procedia CIRP 72:185–190. https://doi.org/10.1016/j.procir.2018.03.087
Onu P, Mbohwa C (2021) Industry 4.0 opportunities in manufacturing SMEs: sustainability outlook. Mater Today: Proc 44:1925–1930. https://doi.org/10.1016/j.matpr.2020.12.095
Author information
Authors and Affiliations
Contributions
This paper is mainly the work of the author Mr. EL Ghadoui Mohamed under the supervision of two other people who are the thesis director Mr. Mouchtachi Ahmed and his co-director Mr. Majdoul Radouane who participated in the supervision, orientation, discussion and revision of the work.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A1
Appendix A1
Main characteristics of the faster R-CNN architecture | |
---|---|
Inception v2 | 1. Architecture type: faster R-CNN 2. Backbone feature extractor: Inception v2 3. Number of classes: 4 (indicating the number of classes the model is trained to detect) 4. Input image resizing: Resizes input images to keep aspect ratio with a minimum dimension of 600 pixels and a maximum dimension of 800 pixels 5. Anchor generation: Generates anchor boxes using grid anchor generator with specified scales and aspect ratios 6. Localization and objectness loss weights: Specifies weights for the localization and objectness loss in the first stage 7. NMS (non-maximum suppression) thresholds: Sets score and IoU thresholds for NMS during inference 8. Box predictor: Utilizes mask R-CNN box predictor with optional dropout and fully connected layers 9. Post-processing: Applies batch NMS and softmax score conversion for post-processing 10. Optimizer: Uses Adam optimizer with manual step learning rate scheduling and momentum optimization 11. Data augmentation: Includes random horizontal flip as a data augmentation option 12. Maximum number of boxes: Limits the maximum number of predicted boxes to 50 per image during training 13. Evaluation metrics: Uses COCO detection metrics for evaluation with 53 examples in the evaluation dataset |
ResNet-50 | 1. Architecture type: Faster R-CNN 2. Backbone feature extractor: ResNet-50 (v1) 3. Number of classes: 4 (indicating the number of classes the model is trained to detect) 4. Input image resizing: Resizes input images to keep aspect ratio with a minimum dimension of 600 pixels and a maximum dimension of 800 pixels 5. Anchor generation: Generates anchor boxes using grid anchor generator with specified scales and aspect ratios 6. Localization and objectness loss weights: Specifies weights for the localization and objectness loss in the first stage 7. NMS (non-maximum suppression) thresholds: Sets score and IoU thresholds for NMS during inference 8. Box predictor: Utilizes mask R-CNN box predictor with optional dropout and fully connected layers 9. Post-processing: applies batch NMS and softmax score conversion for post-processing 10. Optimizer: Uses momentum optimizer with manual step learning rate scheduling and momentum optimization 11. Data augmentation: Includes random horizontal flip as a data augmentation option 12. Maximum number of boxes: Limits the maximum number of predicted boxes to 50 per image during training 13. Evaluation metrics: Uses COCO detection metrics for evaluation with 53 examples in the evaluation dataset |
Inception ResNet v2 | 1. Architecture type: Faster R-CNN 2. Backbone feature extractor: Inception ResNet v2 3. Number of classes: 4 (indicating the number of classes the model is trained to detect) 4. Input image resizing: Resizes input images to keep aspect ratio with a minimum dimension of 400 pixels and a maximum dimension of 800 pixels 5. Anchor generation: Generates anchor boxes using grid anchor generator with specified scales and aspect ratios 6. Atrous rate: Specifies the atrous rate used in the first stage 7. Localization and objectness loss weights: Specifies weights for the localization and objectness loss in the first stage 8. NMS (non-maximum suppression) thresholds: Sets score and IoU thresholds for NMS during inference 9. Box predictor: Utilizes Mask R-CNN box predictor with optional dropout and fully connected layers 10. Post-processing: Applies batch NMS and softmax score conversion for post-processing 11. Optimizer: Uses momentum optimizer with manual step learning rate scheduling and momentum optimization 12. Data augmentation: Includes random horizontal flip as a data augmentation option 13. Maximum number of boxes: Limits the maximum number of predicted boxes to 50 per image during inference 14. Evaluation metrics: Uses COCO detection metrics for evaluation with 53 examples in the evaluation dataset |
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
EL Ghadoui, M., Mouchtachi, A. & Majdoul, R. Smart defect detection using transfer learning in injection molding: a comparative exploration study of deep learning architectures. Int J Adv Manuf Technol (2024). https://doi.org/10.1007/s00170-024-13768-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00170-024-13768-5