https://www.nist.gov/sites/default/files/documents/el/isd/sbm/SMPE-ProgramDescription_2013.pdf

Thanks for sharing this file @HimadriNathSaha, it is very interesting. Measuring science is indeed a key enabler of SMS. :-)

Going a little deeper into such systems engine, how could we describe a robust predictive model, for instance, for smart machining targetting energy efficiency, surface quality or machining time. What are the abilities required by us research engineers to make those robust models? Is machine learning enough? Whatelse must be considered in terms of modelling knowledge?

One example of criterion 

Hi Lorena,

Very interesting question. It is not straightforward to guarantee "robustness" of predictive models when you are dealing with nonlinear, time-variant process in presence of uncertainty and noise. You may address "robustness" a priori by designing a good set of models with key steps such as inputs-outputs system characterization (systemic analysis), experimental identification, filtering, signal processing, cross correlation analysis..,,  and finally using an appropriate tool to obtain/produce the predictive model for example using machine learning. Here you may find a good set of AI and soft-computing. In relation to manufacturing and predictive models, we have conducted some works:

G. Beruvides, F. Castaño, R. Quiza, and R. E. Haber, "Surface roughness modeling and optimization of tungsten-copper alloys in micro-milling processes," Measurement: Journal of the International Measurement Confederation, Article vol. 86, pp. 246-252, 2016.

G. Beruvides, R. Quiza, R. Del Toro, and R. E. Haber, "Sensoring systems and signal analysis to monitor tool wear in microdrilling operations on a sintered tungsten-copper composite material," Sensors and Actuators, A: Physical, Article vol. 199, pp. 165-175, 2013.

F. Castaño, R. M. Del Toro, G. Beruvides, and R. E. Haber, "Application of hybrid incremental modeling for predicting surface roughness in micromachining processes," in IEEE SSCI 2014 - 2014 IEEE Symposium Series on Computational Intelligence - CIES 2014: 2014 IEEE Symposium on Computational Intelligence for Engineering Solutions, Proceedings, 2015, pp. 54-59.

You may find in Spanish:

G. Beruvides-López, R. Quiza-Sardiñas, R. Haber-Guerra, and R. Del Toro-Matamoros, "Features extraction from signals for indirect tool condition monitoring in microdrilling," Dyna (Spain), Article vol. 88, no. 4, pp. 405-413, 2013.

The second important issue is to conduct a posteriori analysis of the predictive model to study the robustness and reliability of the model using different techniques. But the engineering viewpoint, machine learning is not enough, before and after you have several important steps.

BR,

Rodolfo

Hi @Rodolfo E. Harber, thank you for your reply and for sharing these related research work =) I'll read these carefully, it may help me in drawing a robustness criteria set for predictive models applied to SMS in regard to machining.

I have done some similar work using AI to model the Ra and implement into a schematic for supervision control schematic to achieve the desired surface roughness in slot milling processes. It worked well in MATLAB/Simulink environment. I'm presenting this paper at the 47th Computer and Industrial Engineering Conference in Portugal, in October. I'm afraid I cannot share it before the conference, but I can forward to you afterwards.

In regard to your sentence "The second important issue is to conduct a posteriori analysis of the predictive model to study the robustness and reliability of the model using different techniques" - which techniques would you include (besides common statistical, R^2, RMSE,...)?

Thank you

Best regards,

Lorena

Hi Lorena,

You can perform the analysis of the model robustness based on heuristic techniques. You can use gradient-free techniques in simulation models for automatically creating test inputs. With these test inputs you can guarantee by simulations bounded input- bounded output (BIBO) robustness.

Regards,

Rodolfo

Hi Rodolfo,

Interesting, I'll take a look at these.

Also, I'm trying to compare the performance of predictive models developed applying different techniques (non-linear regression, ANN and neuro-fuzzy). I tried to check the 'consistency/amplitude' of predictions using a randomised sine-wave input signal, within the model bounds, to check the predictive profile, which oscillates more, etc. Is something like the gradient-free technique? - Please see some of my results in the attached file.

As I couldn't find similar analysis (maybe due to wrong search keywords), I wasn't sure whether I could use that as a criterion for robustness analysis or not.

Thank you

Best regards,

Lorena

Hi Lorena,

I think that it could be the way but...you have to take care with "randomised sine-wave input signal", you have to use an appropriate seed for the generation. Secondly, the frequency of sine wave. If you are dealing with dynamic models you have to carefully select the frequency or frequencies of the signal.

Best regards,

Rodolfo

Hi Rodolfo,

Thank you very much for your help :-)

Best regards,

Lorena