Cylindrical Grinding and Artifical Neural Networks

Why the interest in neural networks? Can't a simple computer/PLC program and a CNC machine do the job without all the AI hassle?

The PLC/computer and CNC machine will do do work what you instruct.The AI and ANN will do like adaptive automation . ANN is having a potentioal to do the work like bio robot acting to the Scenario.

The PLC/computer and CNC machine will do do work what you instruct.The AI and ANN will do like adaptive automation . ANN is having a potentioal to do the work like bio robot acting to the Scenario.

Hi,

Vermin is right, no need nor usefulness for ANN.

Cylindrical grinding shall be cylindrical with high accuracy: what is your goal.

I have a cylindrical grinding machine from Studer, Switzerland (built 1986) that achieves without problems well below 04.µm out of roundness.

The operator is the bigger problem: no good centering, dirt in the clamping devices, non ideal mounting and setting of conus-angle, not good dressing and truing of grinding wheel will affect your result.

So for a good approach you will need measurements during ongoing machining of:

actual diameter of the workpiece and roundness,

conicity,

wheel out of roundness,

wheel sharpness,

flushing and cooling conditions.

And vibration detection and isolation.

And motor(s) diagnostics and heat-flux-management.

If you go to the extremes of well below 0.5 µm then you need also a separation of spindle error from wheel error.

Without these inputs your system is a blind system and with these inputs a straightforward compensation will do the job. We have done most of this.

There is only one real reason to go fuzzy: confuse your customers what is going on in the machine/process.

RHABE

Dear Mr.RHABE.

Artificial Neural Network will help you in quality,will avoid the rework,it decide the cutting parameters hence you woun't required to calculate it.

the steps involed in it is

1.Decide the input parameters and it steps(Speed,Feed and Depth of cut)

2.Measure the output parameter(Surface finish,Roundness error,Cylindricity and many)

3.Feed the input and output parameters in Artificial Neural Network and get the ANN output.

4.Compare the Your measured output (Measured) and ANN predicted Output parameter find the error between them.

5.The close error(Minimum error) will tell you that the ANN is behave like the you grinding machine.

6.Form the front end

7.Connect the ANN file with it.

8.Connect the computer with your grinding machine which has ANN software.

9.With help of front end feed the desired output requirement.

10.The software will give the cutting parameter.

11.The cutting parameter is interpreted and taken as input commend of the machine will work automatically.

12.you will get the desired output of the product.with in the quality limits.

13.you further dig it you will know the power of ANN.

take care

bye

In principle ok, I know the benefits of NNs.

But there are extreme difficulties in getting involved at grinding.

The status of the grinding disk can change within seconds and this to detect???

What about slow changes: buildup of grinding wheel waviness?

What about ultraslow changes: main-spindle wear?

What about operating errors: bad center, bad balance, wrong wheel?

All this can be compensated deterministically. If this no longer helps then think about forgotten or unknown inputs. Only if all this is done then NNs may helpa bit further.

But to reach the decision point when to add an NN you will have 10 to 100 nanometer accuracy in regular work.

RHABE

RHABE: The status of the grinding disk can change within seconds and this to detect???

Ckdinakar:This process of machining is under standarded conditions.

RHABE:What about slow changes: buildup of grinding wheel waviness?

Ckdinakar:there will not be slow changes.Builup can be avoided by using coolent.

RHABE:What about ultraslow changes: main-spindle wear?

Ckdinakar:of what? of machine or product.if machine ofcource you will do the maintanance.if product where comes the wear?

RHABE:What about operating errors: bad center, bad balance, wrong wheel?

Ckdinakar:How could be possible yar ,then your machine is out of state.(bad center,bad balance).For product grinding you suppose to know the selection of wheel.

the thumb rule is hard wheel soft materials and hard materials soft wheel

carborundum wheel and dimond wheel for aluminium mild steel grinding

alumnia wheel for hard steel.

RHABE:All this can be compensated deterministically. If this no longer helps then think about forgotten or unknown inputs. Only if all this is done then NNs may helpa bit further.

Ckdinakar:what are the unknown input? please. so many factors are influnced in grinding vibration,coolent ect.

RHABE:But to reach the decision point when to add an NN you will have 10 to 100 nanometer accuracy in regular work.

Ckdinakar:definitely we can do it yar belive it.

I think the point here is that all of the parameters mentioned can be monitored by conventional programming and conventional equipment. Even changes to the parameters can be made on the fly using conventional software.

Seems like NN stuff (implementing and deploying it) is better suited for other situations. And for that matter, why a NN - what about an expert system?

Expert system will have collection of data and won't do the adoption to the situation. ANN can do.