search

NewCodeAlg / DiffusionForFaultDiagnosis

MIT License
5 stars 0 forks source link

open source code #1

Open ShiyuCHEN2003 opened 2 months ago

ShiyuCHEN2003 commented 2 months ago

Hi, I've seen your article on TII has been published in May, when will the source code be open please?

NewCodeAlg commented 2 months ago

Thank you for your reminder. The network model code has been uploaded now.

ShiyuCHEN2003 commented 2 months ago

Thank you so much!!! I still have a small question, in the paper mentioned case1: when using CWRU dataset, there are 400 segments for each type, and the length of each segment is 512, but the official dataset such as X186_DE_time variable has only 122,136 data points, which is less than 512*400, is this part to do the overlap operation to extend it to 400 segments?

NewCodeAlg commented 2 months ago

Yes, your understanding is correct. We used overlapping sliding windows

ShiyuCHEN2003 commented 1 month ago

hello, I'm sorry that I have some problems again. When I used the original dataset and the generated dataset for fault diagnosis, the accuracy rate was relatively low. After I printed out the wrong samples, I found that it was difficult to distinguish the three fault types of ball fault with a diameter of 7, 14 and 21, resulting in a low accuracy rate. This problem will occur even if I use all the original dataset (no data is generated). I use Lenet and Alexnet for classifiers. Do you have any suggestions？ 1721129416793

NewCodeAlg commented 1 month ago

Please confirm that the sample image you saved is in PNG format. If using JPG format, there may be compression issues. For classification networks, you can use ResNet.

ShiyuCHEN2003 commented 1 month ago

Thank you for your prompt response. I have noted that the PNG format is being utilized. I am planning to experiment with the ResNet-18 network for this project. hahaha~