An Automatic Foreign Matter Detection and Sorting System for PVC Powder

Round 1

Reviewer 1 Report

An interesting article on automatic foreign matter detection system for PVC method. Some points to consider:

Line 5: word "and" is in the wrong place

Lines 119, 126, 560-561: rephrasing is needed

Figure 4 is very informative, but legend should be in the same page with the figure

Figure 10 should be before conclusions section

 

Author Response

Thank you very much for your comments and appreciation

Author Response File: Author Response.pdf

Reviewer 2 Report

From my point of view, this manuscript could be accepted to Applied Sciences. The content is succinctly described and contextualized with respect to previous and present theoretical background and empirical research on the topic. The research design, questions, hypotheses and methods are clearly stated. The arguments and discussion of findings are coherent, balanced and compelling. The results are clearly presented.  The conclusions are thoroughly supported by the results presented in the manuscript. The manuscript is well illustrated and interesting to read. The main process for PVC powder quality inspection utilizes magnetic rods and visual inspection by human eye in industry. However, the disadvantages of this approach are expensive, subjective, and inconsistent. Because the sizes of foreign objects are extremely small leading to be neglected, the inspection process may harm the vision of the operator. In the present study, an automatic defect detection system has been assembled and introduced for Polyvinyl chloride (PVC) powder. The average diameter for PVC powder is approximately 100 μm. The system hardware includes powder delivering device, sieving device, circular platform, image capture device and recycling device. The defect detection algorithm based on YOLOv4 was developed using CSPDarkNet53 as the backbone for feature extraction, spatial pyramid pooling (SPP) and path aggregation network (PAN) as the neck, and Yoloblock as the head. An automatic labeling algorithm has been developed on the digital image processing algorithm to save time in feature engineering. Several hyper-parameters have been employed to improve the efficiency of detection in the process of training YOLOv4. Taguchi method is utilized to optimize the performance of detection in which the mean average precision (mAP) to be the response. Results show that optimized YOLOv4 has a test mAP of 0.9385, comparing to 0.8653 and 0.7999 for naïve YOLOv4 and Faster RCNN, respectively. Also, there is no false alarm to the images without any foreign matter with optimized YOLOv4.

Following minor suggestions could be given:

- English language and style are fine, but some polishing from native speaker would be desirable. May be, MDPI could provide such services?

- Some perspectives for future research could be formulated in the conclusion section with appropriate citation of the references: e.g. application of other ML approaches (J. Phys. Chem. Lett. 2021, 12, 2017−2022); may be, some statistical analysis based on Hirshfeld surfaces (Polyhedron 139 (2018) 282–288).

Author Response

Thank you for your valuable comments. Your comments are included and The English has been polished. 

Author Response File: Author Response.pdf