Tim von Hahn

Always building. Always learning. Manufacturing - Data Science - ML

Beautiful Plots: The Violin

Music to my ears… err… eyes? The violin plot is a worthy tool for any data visualization tool box. Let’s build one, in Plotly, as we explore historic birth trends in the USA! You can run the Colab notebook here, or visit my github. One thing leads to the next… I read about the Apgar score (more on that in a future post) in Daniel Kahneman’s book Thinking, Fast and Slow.

Analyzing the Results - Advances in Condition Monitoring, Pt VII

We’ve trained the variational autoencoders, and in this post, we see how the models perform in anomaly detection. We check both the input and latent space for anomaly detection effectiveness. In the last post we built and trained a bunch of variational autoencoders to reconstruct milling machine signals. This is shown by steps 1 and 2 in the figure below. In this post, we’ll be demonstrating the last step in the random search loop by checking a trained VAE model for its anomaly detection performance (step 3).

Beautiful Plots: The Lollipop

The lollipop chart is great at visualizing differences in variables along a single axis. In this post, we create an elegant lollipop chart, in Matplotlib, to show the differences in model performance. You can run the Colab notebook here, or visit my github. As a data scientist, I am often looking for ways to explain results. It’s always fun, then, when I discover a type of data visualization that I was not familiar with.

Beautiful Plots: The Decision Boundary

How can we communicate complex concepts using data visualization tools? In this first post – in a series titled “Beautiful Plots” – we build an elegant chart demonstrating the decision boundary from a KNN classifier. “Brevity is the soul of wit” – Polonius in Hamlet Communicating ideas through plots and charts – the process of data visualization – is not always easy. Oftentimes, the ideas being communicated are complex, subtle, and deep.

Building a Variational Autoencoder - Advances in Condition Monitoring, Pt VI

In this post, we’ll explore the variational autoencoder (VAE) and see how we can build one for use on the UC Berkeley milling data set. A variational autoencoder is more expressive than a regular autoencoder, and this feature can be exploited for anomaly detection. We’ve explored the UC Berkeley milling data set – now it’s time for us to build some models! In part IV of this series, we discussed how an autoencoder can be used for anomaly detection.

Data Exploration - Advances in Condition Monitoring, Pt V

Data exploration is an important first step in any new data science problem. In this post we introduce a metal machining data set that we’ll use to test anomaly detection methods. We’ll explore the data set, see how it is structured, and do some data visualization. Let’s pretend again, like in Part II of this series, that you’re at a manufacturing company engaged in metal machining. However, this time you’re an engineer working at this company, and the CEO has now tasked you to develop a system to detect tool wear.

Anomaly Detection - Advances in Condition Monitoring, Pt IV

Machines fail, and anomaly detection can be used to identify failures that have never been seen before. In this post, we explore the use of a simple autoencoder and see how it can be used for anomaly detection in industrial environments. Equipment fails. And if you’ve ever worked in an industrial environment, you’ll know that equipment can fail in any number of weird and wonderful ways. Detecting the strange behaviour in the machinery, early enough, is one way to help prevent these costly failures.

The Case for Feature Engineering - Advances in Condition Monitoring, Pt III

Feature engineering can be powerful tool, but it doesn’t have to be complicated. Often, the simple solution is best. In part two of this series, we went over the differences between the feature engineering approach and the end-to-end deep learning approach. The feature engineering approach can be labor intensive. But, I don’t want to discourage you. Feature engineering can be a great choice for many applications within the condition monitoring space.

Data-Driven Methods - Advances in Condition Monitoring, Pt II

In part two, we give an overview of the two data-driven approaches in condition monitoring; that is, feature engineering and end-to-end deep learning. Which approach should you use? Well, it all depends… Imagine you’re the CEO of a manufacturing company. You make small components out of metal, manufactured on CNC machines. One important jobs you have is to improve the productivity of your operation. If you don’t, well, you’ll slowly see your company’s profits eaten away by the competition.