Data augmentation techniques for transfer learning improvement in drill wear classification using convolutional neural network

Abstract

This paper presents an improved method for recognizing the drill state on the basis of hole images drilled in a laminated chipboard, using convolutional neural network (CNN) and data augmentation techniques. Three classes were used to describe the drill state: red -- for drill that is worn out and should be replaced, yellow -- for state in which the system should send a warning to the operator, indicating that this element should be checked manually, and green -- denoting the drill that is still in good condition, which allows for further use in the production process. The presented method combines the advantages of transfer learning and data augmentation methods to improve the accuracy of the received evaluations. In contrast to the classical deep learning methods, transfer learning requires much smaller training data sets to achieve acceptable results. At the same time, data augmentation customized for drill wear recognition makes it possible to expand the original dataset and to improve the overall accuracy. The experiments performed have confirmed the suitability of the presented approach to accurate class recognition in the given problem, even while using a small original dataset.

Multiclass Image Classification Using GANs and CNN Based on Holes Drilled in Laminated Chipboard

“…Data-augmentation techniques work well in image-analysis problems, improving the results [ 19 , 20 ], which was also proved to be true in the case of drill wear recognition [ 6 ]. In this study, we chose to use the following operations on images for data augmentation: Colour to grayscale—in our dataset, colours were very similar to black and white and did not carry any relevant information that our model should learn.…”

“…In [ 5 , 6 , 7 ], it was shown that using only images of drilled holes and convolutional neural networks (CNN) can give satisfying results, and it is a much simpler solution than that based on multiple sensors. In [ 5 ], only an original set of 242 images was used.…”

“…Trying to build simple CNN on dataset of this size achieved poor accuracy (35%) and required the use of highly pretrained neural networks on the model data in order to achieve a high value of said metric (around 93%). On the other hand, in [ 6 , 8 ], the data were expanded by simple image operations and data-augmentation techniques, such as converting an image to gray-scale values, adjusting the brightness of an image, adjusting its contrast, changing the hue, adding Gaussian noise or salt and pepper noise, creating reflections, or rotating and scaling an image. It was shown that expanding the dataset improved the results.…”

“…In previous works, related to the data and the subject of our interest [ 5 , 6 ], a well-known AlexNet model was used to achieve accurate results. However, this model requires substantial memory resources, computing power, and training time.…”

The Review of New Scientific Developments in Drilling in Wood-Based Panels with Particular Emphasis on the Latest Research Trends in Drill Condition Monitoring

“…Generally, tool condition monitoring in the field of woodworking has also been popular for a long time [24][25][26]. Therefore, at the end of this introductory (and as concisely as possible) overview of the latest research trends, it is also worth noting the new and quite spectacular approach to drill condition monitoring in wood-based panels machining [27][28][29][30][31][32][33][34][35][36][37][38].…”