Uh, hello everyone! My name is Mansell, and I am from the Bruno University of Technology. Today, I will be discussing my submission to the student conference regarding the exploration of GPT capabilities for creating measuring applications in LabVIEW.
Before I begin, let me introduce the content of my presentation. I will start by explaining my motives for conducting this research. Then, I will discuss the desired application I want to build using GPT. After that, I will delve into general prompts and their success rates. Following that, I will focus on detailed prompts. Finally, I will conclude the presentation with the results and conclusions.
So, why did I conduct this research? One day, while teaching my students how to program in LabVIEW, I wondered if my colleagues who teach text-based programming languages have encountered similar problems with plagiarism. I realized that GPT could be used to generate code and potentially create problems. This led me to question the capabilities of GPT in creating programs in visual-based languages like LabVIEW.
To begin my experiment, I needed to create a specification for the application I wanted to build using GPT. I took inspiration from a project we assign to our students, which involves creating a simple logging application. This application measures voltage input and logs the data into a file. The application is based on an event-driven state machine, with important states including initialization, measurement, data analysis, and stopping the measurement.
The specification required me to create an application for measuring voltage input and analyzing the input. The input consisted of two squared signals, simulating a simplified incremental rotary motion sensor analysis. The goal was to determine the direction of the rotor’s rotation. I turned to GPT for help in creating this application.
I started with general prompts, asking GPT to help me create a LabVIEW app for acquiring voltage input from two analog inputs. However, the responses were not useful as the generated program’s picture was unavailable. I tried multiple times, but the results were the same. GPT was unable to generate a program picture.
Next, I asked GPT to help me build only the measurement part of the application using a simple state machine template. Again, the responses were promising at first, but the advice to use a state machine structure that did not exist in LabVIEW was a failure. GPT could not provide a concrete solution.
I then moved on to more detailed prompts, asking GPT how to compute the frequency from a square signal using various LabVIEW functions and techniques. The responses were more usable, suggesting the use of zero-crossing detection, fast Fourier transform, and period measurement. However, GPT did not provide concrete instructions on how to connect these functions in LabVIEW.
I continued with detailed prompts, asking GPT about custom voltage level crossing detection with hysteresis. While GPT suggested using non-existing functions like hysteresis threshold, it led me to discover the preshaped detector function, which could be used for this purpose.
In conclusion, GPT was unable to help me create the LabVIEW application on its own. It provided some keywords and ideas, but the advice was not precise enough for a developer without experience. GPT’s responses were not as helpful as the LabVIEW help documentation or searching in Google. However, GPT can be useful for those who are new to programming or need a starting point. It can help users find keywords and understand concepts, but it falls short in providing concrete programming solutions.
Thank you for your attention. If you have any questions, I will gladly answer them.
