2024 Volume 15 Issue 1 Pages 1-7
Local exhaust ventilators (FHs) are used as exposure control devices for chemical substances in experimental laboratories, and their performance is regularly checked through statutory inspections. Despite the necessity for daily appropriate use of FHs as exposure control devices, current guidance is limited to general warnings with no existing method to evaluate their usage effectiveness. Therefore, in this study, a model was constructed to replicate the evaluation the good or bad usage of FHs from their photos by training a deep learning model based on expert assessments of FH usage using photos. The results showed that the convolutional neural network model judged the FHs condition with an accuracy of 93%, and the mechanical evaluation reproduced the expert's evaluation numerically. The model also visualized the basis on which the numerical evaluation was calculated from the photos, and it became clear that there is no specific area that consistently serves as the basis of judgment for all FHs, and that there are characteristics in the area of judgment basis for each FH. The similarity between the areas used for judgment extracted by the model and the check items used in actual research fields suggests that the basis for judgment of this expert’s evaluation complies with the guidelines. The objective condition evaluation focused on in this study is expected to be developed as a tool to support autonomous safety management that eliminates subjectivity and assumptions.
