News & Events

Transformer Image Black And White

Valuable Insights with Condition Monitoring

How I found professional asset management guidance from an unsuspecting source

Author: Wesley Suplit | Product Development Manager at SDMyers

It’s clear. Artificial Intelligence agrees that condition-based monitoring can drive valuable insights for electrical assets that reduce costs, improve uptime, and reduce emergency repairs—I think I just got ahead of myself—let me back up.

ChatGPT is an Artificial Intelligence tool created by Open AI. On November 30, 2022, it was released to the public, and by December 4th, it had one million users. The meteoric rise has continued. This year ChatGPT has littered media outlets that have reported on its technical features, human-like language aptitude, and inaccuracies (in some hilariously comical instances). These reports were so impressive that I had to create a user account and experience it for myself. Let me share with you what I learned.

Of course, my explanation about ChatGPT would be second-rate compared to “it” explaining itself:

ChatGPT: ChatGPT is an advanced natural language processing (NLP) model developed by OpenAI, using deep learning techniques such as Transformer architecture. It has been pre-trained on a massive amount of text data from the internet, including books, articles, and websites, to understand the nuances of human language.

At its core, ChatGPT is essentially a machine learning model that has learned to predict the next word in a given text sequence. It does this by analyzing patterns and relationships in the data it has been trained on, and then uses these insights to generate new text that is similar in tone and style to the input it receives.

Okay, stop right there. Let me re-phrase the question.

WESLEY: How can I explain ChatGPT to my two-year old?

ChatGPT: ChatGPT is like a very smart robot that can talk to you and answer your questions. It has been programmed to know a lot of things about the world and can help you learn new things. It's kind of like having a really clever friend who can talk to you whenever you want, and it never gets tired of answering your questions.

Much better. I was surprised that the responses are written as if someone is typing them for you to read. I’m mildly impressed that it can correctly and quickly construct sentences, but let’s see what it says about slightly more complex matters, like the reliability topic of the month at SDMyers: condition monitoring.

WESLEY: There is the idea that maintenance professionals and asset managers can save money using a condition based monitoring approach, what do you think about that idea?

ChatGPT: Condition-based monitoring (CBM) is an approach that aims to monitor the health of assets or equipment and detect potential failures or malfunctions before they occur. By using sensors and data analysis techniques, maintenance professionals and asset managers can make informed decisions about when to perform maintenance, repair or replacement of assets, rather than relying on pre-scheduled maintenance activities or waiting for equipment to break down.

Yes, it wrote that out—right in front of my eyes. I was shocked. It may have used different language than I would have to describe condition-based monitoring, but the concepts are spot on. To my surprise, it went on…

ChatGPT: CBM can be an effective approach for reducing maintenance costs and increasing asset availability. By detecting and addressing potential issues before they cause major problems, maintenance professionals can avoid costly downtime and emergency repairs. Additionally, CBM can extend the lifespan of assets by identifying and correcting minor issues before they cause more serious damage.

However, the success of CBM largely depends on the quality of the data being collected and the effectiveness of the data analysis techniques. In order to implement an effective CBM program, maintenance professionals and asset managers need to carefully select the right sensors and tools, ensure that they are calibrated and functioning properly, and use appropriate analytical methods to interpret the data.

Overall, CBM is a promising approach that has the potential to save money and increase
asset availability, but it requires careful planning, implementation, and ongoing maintenance to be effective.

At this point, I’m shaking my head in disbelief. Condition Based Monitoring—a topic and tool I have had to learn to implement, monitor, manage, and report on- is so simply explained by this program.

Okay, let’s get more specific and see if ChatGPT can apply these concepts to condition-based electrical equipment monitoring, specifically transformers and substation equipment. Let’s see if it was trained on the “articles” and “books” related to this specificity. Also, let’s introduce some subjectivity—asking for its opinion on the success of Condition based monitoring.

WESLEY: Has this maintenance approach seen success in electrical assets, specifically thinking about substations and power transformers?

ChatGPT: Yes, condition-based monitoring (CBM) has been widely adopted in the electrical power industry, particularly for monitoring substation equipment and power transformers. CBM is particularly useful for these assets because they are critical to power system operation and failure can lead to significant downtime and costly repairs.

In the case of power transformers, CBM can be used to monitor key parameters such as oil level, temperature, and dissolved gas analysis (DGA) to detect potential issues and predict when maintenance or replacement is required. By monitoring DGA, for example, maintenance professionals can detect abnormal levels of gases that are indicative of potential faults, such as overheating or insulation breakdown, before they lead to major failures. This enables maintenance professionals to schedule maintenance or replacement before significant damage occurs.

Again, I’m truly impressed—it correctly mentioned arguably the number one most widely used diagnostic method/ tool in transformer condition-based monitoring, which is dissolved gas analysis (DGA)—both manual sampling and online monitoring.

I’m re-reading its response to find inaccuracies, but I’m on my third read-through and can’t find a mistake—maybe an omission, but not a mistake. These ideas that ChatGPT touches on are the basis of our daily work to test, monitor and report diagnostic recommendations for the tens of thousands of transformers and substation equipment we service. The testing performed is used as a diagnostic tool to determine if a transformer has any failure modes present—thermal, electrical, dielectric, or external faults. Based on the results from our laboratory or the data from online monitoring, we can more precisely diagnose what is occurring deep within the transformer—often invisible to the naked eye.

It’s quite apparent I am not dealing with a novice on these topics, so let’s test its humility and professionalism.

WESLEY: Would you consider yourself a trusted source on condition based monitoring? You seem to be fairly insightful on these matters.

ChatGPT: As an Al language model, I have been trained on a diverse range of texts and have access to a wide variety of knowledge on condition-based monitoring and other related topics. While I can provide information and insights based on my training, I cannot claim to be a trusted source in the same way that a human expert or organization specializing in CBM can.

However, I am designed to provide accurate and helpful information to the best of my abilities, based on the information I have been trained on. I can provide a general understanding of CBM and its potential benefits and drawbacks, and I can offer guidance on how CBM can be implemented effectively.

It's important to note that while I can provide useful insights, my responses should not be considered a substitute for professional advice or expertise. It's always a good idea to seek the advice of experts in the field before making any significant decisions or taking any actions related to CBM.

Honestly, this was my “get back to reality” moment. People trust people. Of course, those people are supported by machines, processes, technology, and decision frameworks, but ultimately it’s the people we trust.

Each day, SDMyers monitors, manages, and services hundreds of transformers. Some may have a clean bill of health. Others may need a planned maintenance (PM) modification, while others require remediation or repair. We are fortunate to be supported by technology and a rich history of people—starting with Stan Myers, founder of SDMyers, collaborating with each unique customer and helping make decisions for a more reliable and safe electric power system.

We have a few different next steps that you can use to enhance your current CBM program:

  • Update your equipment list by including equipment not in your database and excluding equipment in the boneyard or decommissioned.
  • Modify your inspection and testing frequency or scope to emphasize equipment with a higher criticality or generally in worse condition.
  • Introduce an online monitor for those pieces of equipment that would be considered “mission critical” or need to be “nursed” until a planned outage can occur for replacement.
  • Attempt to view equipment data in a new way to gain fresh insights, like a health index score, professional interpretation, or IEEE model, etc.
  • Clearly define follow-up actions based on testing and monitoring results.

Each one of the options above can help refine your CBM program, which, as confirmed by ChatGPT, can decrease costs from unplanned outages but also improve equipment uptime and reliability.

Now, let’s have some fun with our new ChatGPT friend.

WESLEY: Can you make a joke about condition based monitoring and transformers?

ChatGPT: Sure, here's a joke about condition based monitoring and transformers:

Why did the transformer keep failing even though it was being monitored with condition- based monitoring?

Because it was just too "transformer-mational" for its own good!

Good attempt, but I’m not laughing. Let’s try again.

WESLEY: What about just a joke about transformers?

ChatGPT: Certainly! Here's a joke about transformers:

Why did the transformer break up with his girlfriend?

Because he felt like he wasn't "grounded" enough for her!

Classic – electrical joke. Here, this seems fun.

WESLEY: How would you write that joke in binary for a computer to understand?

ChatGPT: Sure, here's how the joke might be written in binary code:

01010111 01101000 01111001 00100000 01100100 01101001 01100100 00100000 01110100 01101000 01100101 00100000 01110100 01110010 01100001 01101110 01110011 01100110 01101111 01110010 01101101 01100101 01110010 00100010 00100000 01100010 01110010 01100101 01100001 01101011 00100000 01110101 01110000 00100000 01110111 01101001 01110100 0110100000100000 01101000 01101001 01110011 00100000 01100111 01101001 01110010 01101100 01100110 01110010 01101001 01100101 01101110 01100100 00111111

To translate it back to text, you can use a binary to text converter, which would give you:

"Why did the transformer" break up with his girlfriend?"

I’m not done with this joke yet.

WESLEY: Can you use that joke as an idea for a short stand up skit with a few characters?

ChatGPT: Sure, I'd be happy to give you an idea for a short stand-up skit based on the transformer joke!

The skit could feature two characters: a transformer who is feeling down about his recent breakup, and his supportive friend, who tries to cheer him up with some puns and jokes.

Here's a possible script for the skit:

[The transformer is sitting on a stool on stage, looking sad. His friend walks on stage.]

Friend: Hey, buddy, what's wrong?

Transformer: [sighs] I just broke up with my girlfriend.

Friend: Oh no, I'm sorry to hear that. Was she a robot too?

Transformer: Yeah, she was a power converter. But she said I wasn't grounded enough for her.

Friend: [laughs] That's a good one! You know what they say, there are plenty of transformers in the sea.

Transformer: [smiles a little] Thanks, that's kind of you to say.

Friend: And hey, you're a transformer, you can always change your voltage and start anew!

Transformer: [laughs] I guess you're right. Thanks for cheering me up, pal.

Friend: Anytime, my friend. Now let's go recharge our batteries and hit the town!

[They exit the stage together, with the transformer looking a little more hopeful.]

March 01, 2023
Please wait while logging in.