Registered: 1442568538 Posts: 1,121
Reply with quote
Thought I would start a new topic and not take away too much from RGF's thread.
I'm of the slightly younger generation on here and a computer nerd at heart. I appreciate the expertise and the knowledge of who and what has gone before me, whilst at the same time am always open to what can be done differently in our area of expertise. Vibration analysis is the thread that binds us all I think I am fair in saying? But as a whole is something little innovation has been made in the last few years with the exception of cheaper wireless sensing with better battery life. Detection technology wise It seems to have been static the last 20 years with not much more done past Enveloping/peakvue/HDspm /GsE..I could go on.... Sure there are some expert systems about but they still can only follow rules we give them, they lack nuance and the human factor which I think is still required, even if to just function as a gatekeeper...final approval step. I notice in the last 3 years there has been a sudden explosion of those from outside the core discipline of Pdm throwing machine learning and neural networks etc to try and automate what we do with the correlation of large data sets from process and some sensing( I actually had someone tell me they could do it with 4-20ma vibration sensor...good luck to that I told him) I don't think that this approach alone will work satisfactorily, sure they have great marketing and they are mainly targeting managers who haven't a clue and bottom line don't care who are afraid to get left behind in the current IIOT/Indutrie4.0(No I didn't spell it wrong) deluge if they don't grasp at the first technology that drops in their inbox, coupled with an ageing experienced workforce leaving the system and the always present 'Cost cutting/Increase margin' Chestnut. I am all for some sort of automation to eliminate obvious low hanging fruit/telltale faults to assist the analyst but any of the 'Data in-Insights out' systems I have seen leave a lot to be desired and after a time they will have caused issues with false alarms and the like theyre relevance will be ignored. So Rant over, what are you guy's/girls(International womens day and all that) thoughts on this, have we reached a summit of what we can do? How can we leverage the powerful technology we didn't have 20 years ago and get more value out of the work we do with less manpower available though ever higher demands. Maybe I am being naively protectionist? Regards.
Registered: 1443074482 Posts: 394
Reply with quote
I remember this being spoken about on the old board many years ago. Mainly about AI and neural networks etc. There is an Australian company at the moment who sell wireless vibration sensors that they say have machine learning capabilities. I was told we were getting some of these to trial. https://www.movus.com.au
I am sure that was a promise years ago when one of our paper machines were built around the year 2000/2001 when they decided to buy a DLI analyser for that site while we were contacting to all the other sites for that company using CSi. Vibe-Rater will know the story better than me but I am sure DLI said that they would be able to go full automated eventually and that was nearly 20 years ago. Management people will love this sort of thing as all they are thinking is how they can decrease spend. If that means they can get rid of a vibe guy and rely on a computer to do that job then that will be a success for them.
Registered: 1442538912 Posts: 1,810
Reply with quote
What is not factored into the equation to this point, is the tremendous variability in the physical design of machines. This has at least 2 implications: each machine has somewhat its own personality, and measurement point accessibility varies wildly. Correctly calling a problem often requires intimate knowledge of the particular machine. This is why using 'data dogs' (not to put them down - it's just the common term) to collect data that is sent to an 'analyst' somewhere is - to be kind - a less than perfect system. Any type of remote analysis is going to be less than reliable, unless remote means out in the plant where you can walk to the machine if need be.
That said, some machines fail in a very predictable way. At one site I have a population of probably 100 large pumps I monitor. There are 3 - and only 3 - failure modes. Motor bearings fail with an increase in broadband noise (I trend peak acceleration in the TWF), pump bearings fail (same), or coupling element (rubber, split) starts to fail and "wad up" causes high 1x or 2x vibration. If I had remote sensors on those machines, there'd be almost no need for me to ever set foot in a pump room. Now that I have comfortable, bluetooth earbuds (that are also hearing protection), I wear them constantly and have come to rely on what I hear in the vibration signal. It's really one of the biggest innovations in my 30 years of analysis, and I bet less than 1 in 5 analyst/data collectors use them. How could we do that remotely? The answer, and the answer to what "tomorrow" will look like, is 5G. 5G will make IOT possible. We will finally be able to move tremendous amounts of data in real time. But the limiting factor of all the powerful technology we can imagine, is people who know what they are talking about. The same people with little practical knowledge who sit in a lab or office and design our crappy data collectors (and yes, even my 2140 is crappy compared to what it could - and should - be) will be designing these systems, and they don't know squat. But neither do the managers who will be making the decisions to install these crappy systems and eliminate the hands-on analyst/technician with 10, 20, or 30 years of experience. __________________ "The trend is your friend"