Spectrum estimation and processing of cutting force sensor signals for chip form monitoring and classification
Authors: Doriana D'Addona, Anupam Keshari, Roberto Teti, Alessandro De Maio
Abstract
Dealing with a sensor controlled longitudinal turning operation of carbon steel, yielding different chip forms; authors have presented advanced sensor signal analysis for chip form classification and its identification during turning operation. The cutting force sensor signals have been obtained through the spectrum estimation. For all cutting force signal and associated chip form, a set of features, corresponding to the characteristic parameters of the spectrum model, are obtained by linear predictive analysis. To classify the chip form, decision making and analysis of sensor signal features are performed using an unsupervised neural network methodology based on Kohonen maps.
| Attachment | Size |
|---|---|
| Doriana_anupam_spectrum Iproms08.wmv | 1.94 MB |
Dear Authors,
Thank you for this interesting paper. I have a few questions regarding the settings of the SOM which was used in this work:
1)What was the output grid size N x N?
2)How many epochs did you use for the training?(after the initial training did you perform any fine-tuning?)
3)What neighbourhood function was employed?
4)Did the neighbourhood function’s range decrease as the epochs went by? If so, what range function was used to accomplish this?
5)What was the learning rate parameter value?
Thank you,
Gonzalo.
Doriana M. D'Addona, Ph.D.
Dept. of Materials and Production Engineering
University of Naples Federico II
Tel.: +39 0817682336
fax: +39 0817682362
Dear Dr. Gonzalo,
thank you for your interest on our paper.
The simplest way to initialize and train a SOM is to use the function “som_make”: sM = som_make(sD).
This function both initializes and trains the map. The training is done is two phases: rough training with large (initial) neighborhood radius and large (initial) learning rate and fine tuning with small radius and learning rate. By default, linear initialization and batch training algorithms are used.
The som_make function selects map size and training parameters automatically.
The neighbourhood function was gaussian.
Doriana M. D'Addona, Ph.D.
Dept. of Materials and Production Engineering
University of Naples Federico II
Tel.: +39 0817682336
fax: +39 0817682362
Dear Co-chair,
thank you for your suggestion.
I have re-submitted an updated version of our paper including the Conclusion Section.
Dear Authors,
the paper is missing the conclusions section (section 6), can you please upload a new version with section 6 included so it can be ok for the proceedings later on.
Thank You,
Gonzalo (co-chair)