Toward Fault Detection in Industrial Welding Processes with Deep Learning and Data Augmentation

TL;DR

Object detection algorithms from the TensorFlow object detection API are used and adapted to the use case of Laser Beam Welding quality control using transfer learning, finding that moderate scaling of the dataset via image augmentation leads to improvements in intersection over union and recall, whereas high levels of augmentation and scaling may lead to deterioration of results.

Abstract

With the rise of deep learning models in the field of computer vision, new possibilities for their application in industrial processes proves to return great benefits. Nevertheless, the actual fit of machine learning for highly standardised industrial processes is still under debate. This paper addresses the challenges on the industrial realization of the AI tools, considering the use case of Laser Beam Welding quality control as an example. We use object detection algorithms from the TensorFlow object detection API and adapt them to our use case using transfer learning. The baseline models we develop are used as benchmarks and evaluated and compared to models that undergo dataset scaling and hyperparameter tuning. We find that moderate scaling of the dataset via image augmentation leads to improvements in intersection over union (IoU) and recall, whereas high levels of augmentation and scaling may lead to deterioration of results. Finally, we put our results into perspective of the underlying use case and evaluate their fit.

Figures (6)

• Figure 1: A Sample image data from the camera module at the LBW station. The components are marked.
• Figure 2: The mAP values of the models displayed in a single plot. The SSD Mobilenet model trained on the multiple scales of the original dataset have been compared against the mAP values with various IOUs. The model performances tend to increase with the data augmentation till a certain level and then reduces. The Y-axis shows the mAP.
• Figure 3: The mAR values of the models displayed in a single plot, comparing the recall values of the fine-tuned models with the baseline model. The models tend to improve their performances till a level of augmentation and then deteriorate. The Y-axis shows mAR value.
• Figure 4: The mAP and mAR values of the developed models in detecting the larger pores. The precision and recall values of the models observed to increase along with the level of augmentation till a factor of $\times 6$. The Y-axis shows the values of mAP and mAR.
• Figure 5: The mAP and mAR values of the developed models in detecting the medium sized pores. The best performing model has been identified at a scaling factor of $\times6$. The y-axis shows the values of mAP and mAR.