Agree with your comments Sarah on several points.
First the original point, I think, is about needing a process for evaluating all these IIOT/predictive options out there. That is tricky but I also think some of this may take care of itself in time. There will be a few winners in this market from everyone who is trying to enter it. My approach so far has been to only look at something that will merge with what predictive systems I already have in place. I am not going to destroy a good system for a sales pitch of promises that are not deliverable yet.
Also agree on RCM. The philosophical principals are very sound to apply in making maintenance & reliability strategy decisions but executing the actual process on thousands of assets is rarely practical. This is why we need maintenance and reliability professionals in companies as knowledgeable consultants to guide organizations to better decision making. We do need at least street knowledge leaders in VP positions to support continued progress in all these areas and to empower those professionals in the organization to drive positive change.
Original Message:
Sent: 12-19-2019 09:41 AM
From: Henry Kocevar
Subject: Machine learning
HI Sarah,
I love simple, you know me I am simple.
Having worked with implementing predictive technologies RCM decision logic is at the core of any implementation. Is installing a sensor or capturing an already installed sensors data worth doing. Is IR/VIB/Ultrasound/MCA/Motion Amplification or the next tech to hit the field an applicable and effective task if implemented to mitigate the failure. Does it provide value to implement, monitor and analyze?
I've seen various offerings for analyzing and displaying the data, the decision is what makes sense, is it cost effective and meet your needs/wants. Face it we and our companies don't always purchase just what meets our needs, sometimes we want the "Bells and Whistles". That's why I don't drive a basic work truck with crank windows and vinyl seats.
I started analyzing data as Watchstander on equipment many years ago recording and analyzing readings on a log sheet. I finished my watchstanding career monitoring a computerized control board and reviewing/analyzing the hourly data capture sheet. It was frustrating that 20 years ago the organization did not have the capability to capture that data and trend it for me, because the capability was there but not affordable. Back then it was a manual effort to put the data in Excel and look for and set up scripts to identify trends.
Data capture and analytics is now affordable, machine learning is improving as more data is captured. The platform utilized to capture, analyze and display your data dependent on what you can afford and works for you.
I started out with log sheets and graph paper and finished up with data transfers to Excel. I envy the tools that are available now and think a decision assistant to outline the capabilities of the various offerings is needed and would be well received. Maybe something for Plant Services or another independent magazine or organization like SMRP to support.
This article basically sums up my comments, I wasn't a submariner but my experiences were similar.
Plant Services Article A-Navy-Perspective-On-Predictive-Maintenance
------------------------------
Hank Kocevar,CMRP
Consultant
Guardian Technical Services
hkocevar@guardiantech.org
Original Message:
Sent: 12-18-2019 08:50 AM
From: Sarah Lukens
Subject: Machine learning
You are talking about a different topic.
I was talking about something simple - determining different criteria to evaluate predictive analytics software by, and then using it to review different options. Creating something like the following decision map:
------------------------------
Sarah Lukens
GE Digital
Roanoke VA
Original Message:
Sent: 12-17-2019 10:41 AM
From: Richard Lamb
Subject: Machine learning
I thank Sarah for advising that the discussion get beyond the lurking desire to sell one's brilliant products and services. I agree that we need some type of decision framework for deciding if, how much, and what and which offerings best fit the plant case.
The framework should follow the basic decision principle of RCM for applying a condition-based maintenance task (AKA, PdM). It is, "is the task to detect whether a failure is occurring or about to occur technically feasible and worth doing?"
Worth will bring us to the decision for where, what and how much. This is because worth is decided by the threshold and ceiling for what would entail an effective CBM solution. In other words, how far into predictive analytics, asset performance management or data-based machine learning make sense-is worth it?
The decision framework should be designed to find the threshold and ceiling of worth as a function of system components. Unfortunately for the purveyors of the grand and glorious, the dull and practicable will likely win out many more times than not.
Original Message------
This is an interesting thread because it straddles the fine line between sharing information and selling products and services. At the same time, I believe there is a need for a "consumer reports" for predictive analytics, asset performance management, or any data-based machine learning solution software out there, so industry practitioners can make informed decisions beyond a brilliant sounding sales pitch.
Perhaps we should use this thread to brainstorm different evaluation criteria for different predictive maintenance solutions, and then start to review different solutions in the context of this criteria? The reviews may come out interesting because this board has members that are both vendors and users.
To start that off, some evaluation criteria I can think of off the top of my head are:
- Relevant use cases - there is no one magic predictive analytic modeling solution. Some software will be stronger for different use cases than others, and this could span failure mode characterizations, equipment catalogues, industrial application, etc.
- Verticals proven effective - such as oil and gas, aviation, manufacturing, mining, etc. What industries is this software been proven to be used in effectively?
- Integration with various "stuff" (need to work out what the different "stuff" are) - but I mean, a model working in isolation is not practical - a solution needs to integrate with work processes, asset strategies that are in place, different other software platforms in use, etc. so that alerts can be timely and the right actions can be taken in the right way at the right time
- Deployment options - such as on-prem or in the cloud (or both), is there a managed services options, etc.
- ...
What do you think?
------------------------------
Sarah Lukens
GE Digital
Roanoke VA
------------------------------