AI and Machine Learning

Hi all,

I am a DSP and data scientist working with solutions to vibration and ultra sound diagnostic using data modelling, machine learning and deep learning. I have been in this industry for a couple of years and it is hard to identify the feelings of the users of our application. Sometimes they seem very suspicious about what the software is saying, other times they feel pretty happy about the results. 

I'm worried about it because as I don´t have direct contact with the plants engineers, so that my only source of feedback is what the project manager says. And the PM is also a computer scientist. So, there is some kind of pressure to implement the automation in the plants.

So, this questions if for reliability engeneers around:

- Have you ever had contact with AI solution (or other data-driven approach) for asset maintenence?

- From your experience, what are positive and negative points?

- How do you feel about it? Do you trust it? Do you think it is very imature? Do you think there is no way to reflect engeneer's experience in the models?

- Do you have any suggestions or doubts I can help with?

Thank you very much, people. Any detail, please contact me on:

paularomerolopes@gmail.com

I have not heard of any good experience feedback. Large investments, unreliable systems, unpredicted machinery faults (faulty alarms on the other hand), overload computers, system shutdowns, wireless difficulties, unsuitable results interpretation etc.... Sounds good for some managers without real field experience but I am 100% sure those systems cannot replace quality periodic measurements and engineer's hand close to the machine.

Hi Becar, Thanks for your answer.

So, how would you convince that manager that a new data-driven technology is not needed or not good enough to help the engineers ? 

I failed convincing them, now they are probably remembering my words. The biggest problem I see is managers being influenced by commercial staff talking about Industry 4.0 in a way like going from Windows 8 to Windows 10! 

It is true that industry 4.0 is a very buzz word that is not being applied in the right pace, however - as windows 10 over windows 8 -  it will prove its value at some point in the future.

To be honest, the most used feature of our solution is just the real time vibration trends. The people do not look at the diagnostics given from the model. In my opnion, the ideal use case would be, not to replace quality periodic checks, but optimize them by focusing in the most important assets.

Other benefit I see is the fact that the knowledge is not centralized in one person or group. My goal, for example, has always been to put the knowledge and experience from the engeneers inside the AI model through the feature extraction step. That is not easy at all.

And you are totaly right about the technological cons that we are still facing: communications and costs of computing are bottlenecks, but I also see lots of investments been done on that. In near future it will not be a blocker anymore.

Sometimes I think more than investments in technology, the companies should invest in changing to a more digital culture for all stakeholders. 

Anyway, from what I understand you cannot see the data driven technology helping your day to day work, right?

I work on periodic vibration diagnostics and see the method as invaluable compared to some mentioned AI solutions. I get goosebumps when I hear about AI vibration diagnostics. That's just me, wait for other's view.

Becar hit the nail on the head when he wrote about managers without field experience. When you combine that factor with people in general having a penchant for "chasing shiny objects" (e.g. Thisthing 5.4, Thatthing 6.0) the problem gets worse.

The key is getting "management" to first identify what they want the system they are thinking about "to do" as opposed to what they want it "to be" (e.g. Nextthing 5.0).  This doesn't guarantee success but it can sure help stop the shiny object chasing.  Once the "to do" question is properly answered, then its a matter of finding what technologies will make that so and getting them behind it. 

Hi Paula,

It's nice talking about innovations and improvements.

I have some experience about predictive maintenance driven by ruled based (you can called it machine learning) on Emerson Machinery Health Software. For me, it was really help although it had some limitation and need a lot of improvement.

I think AI/machine learning is the future of condition monitoring or predictive maintenance technique.

Regard,

Wirana

When ai-based predictive maintenance solutions have several benefits and costs savings such as:

• Predictive maintenance is designed to tackle the core problem of maintenance, i.e. Fix the correct part of the asset, only when it is needed, and do so at the correct time.
• Can reduce maintenance costs by 5-10%.
• Minimizes the time that normally goes into planned maintenance by 20-50%.
• Increases production efficiency and significantly improve asset performance.
• Mitigates ecological footprint by gauging and controlling the consumption of energy.
• Boosts operations, work production and enhances employee safety.
• Has the potential to boost asset uptime and availability by a whopping 10-20%.

You just posted an ad for your product..and the answer is still the same. AI Industry 4.0 stuff still isn't good enough to replace periodic measurements by a competent inspector and then the visual inspection element also.

very interesting question and rich dialogue. From my side , as electric utility engineer there is so much talk and focus on smart grid, and industry 4.0.

for data Analytics , there is big project for that depending on sensory data and ERP system database, and the use cases are promising.

for predictive maintenance /condition monitoring, the management asked for complete AI based system especially for underground cables which fail suddenly .

in my view , we are still in premature state and whatever claims by Vendors, it is not so accurate and cannot replace engineer experience.

in my view AI /ML or Analytics should be toolbox tool for engineers and not replacing them , and i think this one psychological factor which make most O&M engineers dislike AI/ML or Robotics in general.

Also the complicated nature of AI and it is huge mathematics burden , make discussion and argument somehow difficult even for engineers level.

This post was originally from 2019.  Change in the AI world has been rapid.  We now have ChatGPT, Bing AI, and more being released every month.  I am wondering if you are still in this space Paula, and what is the latest?  Personally, I think the idea of AI analyzing machine data is much needed.  My worry is how we give it enough background to do it's job?  How do we share data without opening access to proprietary information?  Many companies still haven't made the switch to keep their machine information and IT systems separate.  Until that is complete, we can't even contemplate using AI in an organization.  I have used ChatGPT, and it just didn't have enough source material yet, to be useful in reliability.  (An example, ChatGPT attributed some of Ricky Smith's work to me; I assume not because it found a link, but because I was asking and Ricky has done some powerful stuff.)  When we are talking about vibration data for example, where would the source data come from? Would the software providers have a way to pull that data into their data archives? If so, how do companies opt out and still use the software?  (I'm sure Alexa is listening to me, even though I've opted out every way I know how.)  We need to have traceability of where information came from and anonymity for provided data.  How is this solved in the condition-based monitoring space?

I provided some quick feedback on this some time ago on Linkedin and suffice to say, there are some definite benefits to AI performing vibration analysis for well-established indicators for imbalance, alignment, oil whirl, etc... However, anomalies such as corrosion, fatigue, harmonic resonances, and especially process induced problems requires experience or the feel of the deck plates, if you will, to objectively assess risk and provide the intuition for corrective action. Someday, you may be able to simply ask a computer, what's wrong with this turbine but NOT in this generation and probably the next. Seasoned, experienced, vibration specialists will remain a critical resource.to ensure world class reliability for rotating equipment for a long time. 

Hi Paul, great discussion! Let me add a few real life examples:

1. Stress corrosion cracking of 5 stage 25,000 hp, centrifugal air compressor impellers - no indications of a vibration anomaly until impeller flew apart destroying entire machine, resulting in nearly a 1-year outage. High efficiency filtration and periodic inspection required in addition to vibration analysis.
2. Natural harmonic resonance of gear components in sync with RPM - no indication of problem until tooth broke off high speed pinion on 5 stage oxygen compressor as was the case for 4 more pinions over the next 5 years. Modal analysis finally revealed pinion, bull gear, and casing all had a natural resonance exactly 6X RPM.  Redesigned entire gearbox. 
3. Process induced surge - 30,000 hp, axial-radial, compressor failed and found all axial blades broken and separated from rotor with no indication of vibration anomaly in monthly data. Downtime 6 months. Root cause; surge control insufficient. 

There are many other examples where monthly vibration analysis, automated or not, would have prevented these catastrophic failures but with experience and skill PM's can be optimized to prevent them which AI can't do. At least not for the foreseeable future. 

Cheers!