How to analyze a signal...

I have the following setup: On a grinding mill (think of a -very large 30' diameter or so- coffe-can turned on its side and filled about 20-30% full of marbles), I have a strain gauge in one of the bolts that holds the mill together. Connected to the strain gauge I have an analog signal transmitter that sends the signal off the mill to a receiver which converts the analog value to a digital byte value. The byte is sent serially (2400bps) to a computer where it is recorded. The rotation of the mill is about 10 - 12 rpm, resulting in about 1500 data points per revolution. Now, the tricky part...I need to analyze the data to relate the incoming signal to the charge level of the mill (how full of "marbles" it is). The charge level does not change rapidly (i.e. it wont be 25% in one revolution and 20% in the next...it takes minutes/hours to change). Currently I have about 6 sets of data from a lab-scale version of this setup (i.e. the mill is only a couple feet in diameter), at different charge levels, each set has a few thousand datapoints representing several revolutions of the mill (the lab scale mill turns much faster, about 30 rpm or so - so there are significantly less data points per revolution). I've been looking into using FFT's to analyze the data (especially interesting is the amount of time it spends under high-strain - i.e. in the strain gauge is "in the charge")...but I know little/nothing about FFTs other than I remember some vague mention of Fourier transforms in college - but that has been more than a decade ago. My question(s)...does this seem like an appropriate application of FFTs? Is there a good starting/newbie text for doing practical applications of FFTs? So far, I've just brought the data into Excel, used its Fourier transform, got my magnitude and phase data, plotted magnitude against bin...but I'm still not sure what I'm looking at. Thanks for any tips, help Dave Taylor Process Engineering Resources, Inc.

"David Taylor" <nospam:dave@processeng.com> wrote in message news:<g31La.62$On3.71014@news.uswest.net>... > I have the following setup: > > On a grinding mill (think of a -very large 30' diameter or so- coffe-can > turned on its side and filled about 20-30% full of marbles), I have a strain > gauge in one of the bolts that holds the mill together. Connected to the > strain gauge I have an analog signal transmitter that sends the signal off > the mill to a receiver which converts the analog value to a digital byte > value. The byte is sent serially (2400bps) to a computer where it is > recorded. The rotation of the mill is about 10 - 12 rpm, resulting in about > 1500 data points per revolution. > > Now, the tricky part...I need to analyze the data to relate the incoming > signal to the charge level of the mill (how full of "marbles" it is). The > charge level does not change rapidly (i.e. it wont be 25% in one revolution > and 20% in the next...it takes minutes/hours to change). > > Currently I have about 6 sets of data from a lab-scale version of this setup > (i.e. the mill is only a couple feet in diameter), at different charge > levels, each set has a few thousand datapoints representing several > revolutions of the mill (the lab scale mill turns much faster, about 30 rpm > or so - so there are significantly less data points per revolution). > > I've been looking into using FFT's to analyze the data (especially > interesting is the amount of time it spends under high-strain - i.e. in the > strain gauge is "in the charge")...but I know little/nothing about FFTs > other than I remember some vague mention of Fourier transforms in college - > but that has been more than a decade ago. My question(s)...does this seem > like an appropriate application of FFTs? As for using Discrete Fourier Transforms (DFTs), I think your setup seems to be good. You have relatively high sampling rates and long observation times, and that's good. It's usually easier to reduce these parameters than the other way around. Which means you have some freedom to experiment. Assuming that the measured strain actually reflects the contents level inside the mill, I would first record a reference data set. That is, I would fill the mill and let it run empty (or to the lowest safe operating level) and record the data along the way. Continuous operation can be important, as very narrow filters will induce start-up transients that last longer the narrower the filters are. Once you know what to expect from the filters, you can chop the data series into shorter, more manageable sequences. You don't want the data sequences so short that there are only start-up and end transients. My first attempt on analysis would be use a narrow-band low pass filter, and see how the DC and "almost DC" parts behaved. Assuming that the measured strain is related only to the contents level (i.e. that the engine does not start stressing the container or structure itself, as the grain levels reduce), I would expect to see a decrease in the "almost DC" components as the container empties. > Is there a good starting/newbie > text for doing practical applications of FFTs? So far, I've just brought the > data into Excel, used its Fourier transform, got my magnitude and phase > data, plotted magnitude against bin...but I'm still not sure what I'm > looking at. I don't know it first hand, but I have seen several people recommend Rick Lyons' "Understanding Digital Signal Processing". Rune > Thanks for any tips, help > > > Dave Taylor > Process Engineering Resources, Inc.

Rune Allnor wrote: > David Taylor wrote: ..... > > Is there a good starting/newbie > > text for doing practical applications of FFTs? So far, I've just brought the > > data into Excel, used its Fourier transform, got my magnitude and phase > > data, plotted magnitude against bin...but I'm still not sure what I'm > > looking at. > > I don't know it first hand, but I have seen several people recommend > Rick Lyons' "Understanding Digital Signal Processing". You can also try http://www.dspguide.com, which is pretty hands-on. Just ignore the lousy cover-photo on the opening page! Andor.

Andor wrote: > Rune Allnor wrote: > >>David Taylor wrote: > > .... > >>>Is there a good starting/newbie >>>text for doing practical applications of FFTs? So far, I've just brought > > the > >>>data into Excel, used its Fourier transform, got my magnitude and phase >>>data, plotted magnitude against bin...but I'm still not sure what I'm >>>looking at. >> >>I don't know it first hand, but I have seen several people recommend >>Rick Lyons' "Understanding Digital Signal Processing". > > > You can also try http://www.dspguide.com, which is pretty hands-on. Just > ignore the lousy cover-photo on the opening page! > > > Andor. How dare you say that. The only thing customers ever see of many of us, is our hand holding the thing we have laboured over in a publicity photo. Please allow us some tiny measure of recognition. :-) Regards, Steve

"Andor" <an2or@nospam.com> wrote in message news:<3efe00c4_2@news.tiscalinet.ch>... > Rune Allnor wrote: > > David Taylor wrote: > .... > > > Is there a good starting/newbie > > > text for doing practical applications of FFTs? So far, I've just brought > the > > > data into Excel, used its Fourier transform, got my magnitude and phase > > > data, plotted magnitude against bin...but I'm still not sure what I'm > > > looking at. > > > > I don't know it first hand, but I have seen several people recommend > > Rick Lyons' "Understanding Digital Signal Processing". > > You can also try http://www.dspguide.com, which is pretty hands-on. Just > ignore the lousy cover-photo on the opening page! Nah... I'd say that photo truly proves the book is hands-on... Rune

"Jerry Avins" <jya@ieee.org> wrote in message news:3EFCC8D2.DAA130E6@ieee.org... > David Taylor wrote: > > The best analysis technique depends on the kind of result you seek. Can > you give more detail about that? What we are looking for is a measurement of the total mill charge, which includes steel grinding balls, ore and water. Ideally we would know the charge level and the "toe" and "shoulder" positions...i.e. the position of the charge in the mill when it is rotating, but right now, we would be happy just to know the total charge level. > > Does the weight vary with the charge in a consistent way? With the slow > change you cite, a load cell with suitable low-pass filter might do if > so. The charge varies with ore feed rate and grinding ball addition rate. Most mills are "slug charged" meaning that they dump several tons of grinding balls in the mill at once and let them wear down until the next slug charge - more automated plants have a continuous slow feed of grinding balls. The ore is fed to the mills continuously from a conveyor belt. Unfortunately, most mills do not have load cells - though some newer ones do. But the charge level cannot be determined from the load cell value alone because the total charge is made of both ore (typically about 2800 kg/m3) and steel grinding media (7800 kg/m3) in ratios that are not known at any given point in time, i.e. amount of ore depends on feed rate and mill residence time and grinding media quantity is determined by the media wear rate which is affected by ore hardness, mill speed, operating conditions, etc etc. > > What do you grind in your ball mill? In power plants, big ones are used > to pulverize coal. The banks of them I remember in the Ravenswood > (Astoria, NY) plant were driven by 50 HP motors. What is being ground depends on the operation...typically we deal with copper, iron and gold ores. We are most interested in performing this analysis on SAG mills currently (semi-autogenous grinding), which usually have diameters of 30 - 40 feet, and motors with several thousand HP. > > Jerry > > Jetty > -- > Engineering is the art of making what you want from things you can get. > ��

David Taylor wrote: > ... > > What is being ground depends on the operation...typically we deal with > copper, iron and gold ores. We are most interested in performing this > analysis on SAG mills currently (semi-autogenous grinding), which usually > have diameters of 30 - 40 feet, and motors with several thousand HP. > Before the application of spectrometers to Bessemer converter plumes, there were a few individuals with slightly deviant visual pigments who could see when the conversion was finished, and they got high salaries. If there's hope that the spectrum you'd get from Fourier analysis of the mill's vibration carries useful information, then I would guess that some mill operators can know something useful by listening to them. If that's the case, whatever they hear can probably be quantified. Whoever (or whatever group) succeeds will need the operators' practical knowhow, an awareness of different ways to extract information from sampled data, good tools, and a keen "eye". That platitude is about all I can offer. You have a difficult and interesting problem, and I don't know enough about it to ask reasonable questions. Good luck. Jerry -- Engineering is the art of making what you want from things you can get. ��

Rune Allnor wrote: > > "Andor" <an2or@nospam.com> wrote in message news:<3efe00c4_2@news.tiscalinet.ch>... > > Rune Allnor wrote: > > > David Taylor wrote: > > .... > > > > Is there a good starting/newbie > > > > text for doing practical applications of FFTs? So far, I've just brought > > the > > > > data into Excel, used its Fourier transform, got my magnitude and phase > > > > data, plotted magnitude against bin...but I'm still not sure what I'm > > > > looking at. > > > > > > I don't know it first hand, but I have seen several people recommend > > > Rick Lyons' "Understanding Digital Signal Processing". > > > > You can also try http://www.dspguide.com, which is pretty hands-on. Just > > ignore the lousy cover-photo on the opening page! > > Nah... I'd say that photo truly proves the book is hands-on... > > Rune It's a good book. I got tired of reading it on line, so I bought a copy. I like Lyons' book better, but they're different. For an engineer like David Taylor (related to the eponym of the towing basin?), who wants to do DSP but not break new ground, they're collectively excellent. Jerry -- Engineering is the art of making what you want from things you can get. ��

Hi David A couple of bits of information you didn't provde. You said previously, that you did an FFT on your data, but you didn't say what the length of the data window that you processed. Also the fact that you are asking the question here implies that nothing in particular appeared to be different between an FFT of data collected from a fully loaded drum to a lightly loaded one. But you didn't expressly state that computing the difference between the 2 FFT's produced nothing. I would guess, that if the amount of load and the composition of the material in your mill produces different frequency components in your data, then they will be at fairly low frequencies. So you may need a much longer time frame for your data to discern any subtle differences. Also if the frequencies of interest have long cycles relative to your sampling rate it may be a good idea to decimate the data. For example, if you downsample the data by a factor of four (with the right filters its sort of like converting your 8-bit data to 32-bit) you will get much better resolution and S/N ratio at the frequencies that are of interest. -jim David Taylor wrote: > > "Jerry Avins" <jya@ieee.org> wrote in message > news:3EFCC8D2.DAA130E6@ieee.org... > > David Taylor wrote: > > > > The best analysis technique depends on the kind of result you seek. Can > > you give more detail about that? > > What we are looking for is a measurement of the total mill charge, which > includes steel grinding balls, ore and water. Ideally we would know the > charge level and the "toe" and "shoulder" positions...i.e. the position of > the charge in the mill when it is rotating, but right now, we would be happy > just to know the total charge level. > > > > > Does the weight vary with the charge in a consistent way? With the slow > > change you cite, a load cell with suitable low-pass filter might do if > > so. > > The charge varies with ore feed rate and grinding ball addition rate. Most > mills are "slug charged" meaning that they dump several tons of grinding > balls in the mill at once and let them wear down until the next slug > charge - more automated plants have a continuous slow feed of grinding > balls. The ore is fed to the mills continuously from a conveyor belt. > Unfortunately, most mills do not have load cells - though some newer ones > do. But the charge level cannot be determined from the load cell value alone > because the total charge is made of both ore (typically about 2800 kg/m3) > and steel grinding media (7800 kg/m3) in ratios that are not known at any > given point in time, i.e. amount of ore depends on feed rate and mill > residence time and grinding media quantity is determined by the media wear > rate which is affected by ore hardness, mill speed, operating conditions, > etc etc. > > > > > What do you grind in your ball mill? In power plants, big ones are used > > to pulverize coal. The banks of them I remember in the Ravenswood > > (Astoria, NY) plant were driven by 50 HP motors. > > What is being ground depends on the operation...typically we deal with > copper, iron and gold ores. We are most interested in performing this > analysis on SAG mills currently (semi-autogenous grinding), which usually > have diameters of 30 - 40 feet, and motors with several thousand HP. > > > > > Jerry > > > > Jetty > > -- > > Engineering is the art of making what you want from things you can get. > > �� -----= Posted via Newsfeeds.Com, Uncensored Usenet News =----- http://www.newsfeeds.com - The #1 Newsgroup Service in the World! -----== Over 80,000 Newsgroups - 16 Different Servers! =-----

David, An easy thing to overlook is that the samples you get may not be suitably bandlimited for processing. If I recall correctly, you have about 1500 samples per revolution. In order to avoid corrupting the FFT spectrum with aliases, you need to remove frequencies above 750 cycles per revolution before sampling and digitizing. (Practical systems need some margin.) For 12 RPM -- 1/5 RPS -- the cutoff needs to be as much below below 15 Hz as you can tolerate. At lower RPMs or coarser resolution, it needs to be lower. Many converters are AC coupled. At these low frequencies, that can be a problem. Jerry -- Engineering is the art of making what you want from things you can get. �� Jerry Avins wrote: > > David Taylor wrote: > > > ... > > > > What is being ground depends on the operation...typically we deal with > > copper, iron and gold ores. We are most interested in performing this > > analysis on SAG mills currently (semi-autogenous grinding), which usually > > have diameters of 30 - 40 feet, and motors with several thousand HP. > > > Before the application of spectrometers to Bessemer converter plumes, > there were a few individuals with slightly deviant visual pigments who > could see when the conversion was finished, and they got high salaries. > If there's hope that the spectrum you'd get from Fourier analysis of the > mill's vibration carries useful information, then I would guess that > some mill operators can know something useful by listening to them. If > that's the case, whatever they hear can probably be quantified. Whoever > (or whatever group) succeeds will need the operators' practical knowhow, > an awareness of different ways to extract information from sampled data, > good tools, and a keen "eye". That platitude is about all I can offer. > You have a difficult and interesting problem, and I don't know enough > about it to ask reasonable questions. Good luck. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > ��

On Mon, 30 Jun 2003 13:18:46 -0400, Jerry Avins <jya@ieee.org> wrote: (snipped) > >It's a good book. I got tired of reading it on line, so I bought a copy. >I like Lyons' book better, but they're different. For an engineer like >David Taylor (related to the eponym of the towing basin?), who wants to >do DSP but not break new ground, they're collectively excellent. > >Jerry Hi Jer, I agree with you. My book has a little more math (for those folk who like a little math), but Smith's book covers SOooo many topics! Whew. He did a good job. I saw an advertisement for another Smith DSP book. The new title is: "Digital Signal Processing: A Practical Guide for Engineers and Scientists". (Publisher is Elsevier Science, and they're different from whoever publisher Smith's first book.) I wonder how different Smith's 2nd book is from his first book. I'm workin' on a 2nd edition of my book. WHAT A PAIN! See Ya, [-Rick-]