Acerta Blog Q3 2019 Round-Up
Three months go by quick. In that time, I, your humble Content Manager, have had the chance to work with some incredibly bright people. It was intimidating at first: coming into work every morning and seeing whiteboards covered in arcane equations and incomprehensible diagrams. I knew what we did in principle when I joined Acerta—we’re using AI to help automakers get products to market faster and with fewer defects—but I certainly didn’t understand what that actually involves.
Over the past three months, I’ve had crash courses in automotive manufacturing, cloud migration, data quality, fleet maintenance, machine learning, software internships, and statistical process control, among other things. Suffice it to say, I’ve learned a lot and I’m still only scratching the surface. Fortunately, my colleagues at Acerta have been more than obliging, tolerating my attempts to encapsulate subjects they’ve been studying for years with glib analogies.
(It’s no coincidence that the first post I worked on compares machine learning models to dogs in a dog show.)
We’ve covered a lot of heavy topics in Q3—as well as a few lighter ones. This week’s post takes a look back at the twelve blog posts we’ve published from July to September in 2019.
The Biggest Challenge in Machine Learning is Other People
So, there I am, only a couple weeks into a brand new job and I find out that my first blog post will be written in collaboration with Senior Data Scientist, Mahmoud Salem, on the topic of…communication?
It certainly took me by surprise to start with such a non-technical subject, but it didn’t take long for Mahmoud to convince me that keeping everyone involved in a machine learning project on the same page isn’t nearly as simple as it sounds. I learned that Mahmoud’s background in computer engineering and graduate work in machine learning has made him uniquely capable of acting as a bridge between various groups at Acerta, including Data Science, Implementation, Infrastructure and UI.
My biggest contribution was the aforementioned analogy with the Westminster Dog Show.
Beyond Predictive Maintenance
Oh boy, only my second blog post and it involves working directly with my new boss, Acerta CEO, Greta Cutulenco. Happily, she has been great to work with, and I’m not just saying that because I know she’ll read this post. (Hi Greta!)
Our topic was predictive maintenance, and more specifically why Acerta’s machine learning platform for manufacturing—LinePulse—is better described as a tool for predictive analytics. The distinction between predictive maintenance and predictive analytics is a subtle one (my favorite kind) and you can see my attempts to grasp it via the analogies peppered throughout the post. In the end, the simplest way to understand the difference is that predictive maintenance enhances operational efficiency, while predictive analytics enhances efficiency and product quality.
AI in the Automotive Industry: An Insider Perspective
Okay, blog post no. 3 and I finally have a topic where I have some experience. Granted, it’s not the decade’s worth of working in automotive plants that our Account Executive Luke Richard has, but I spent enough time in my previous role as Managing Editor at engineering.com to be fairly comfortable with industrial lingo. So, of course, Luke ends up talking about machine learning.
Actually, hearing about it from someone who first encountered AI in the auto industry (rather than through software or data science) really helped me understand Acerta’s focus on automotive, why automakers may be hesitant to adopt machine learning, and what it can potentially do for the industry as a whole. Luke also came up with one of my favorite similes: AI is like a torque wrench…
Migrating from AWS to Azure: What We Learned
By this point, I felt like I was getting a handle on things at Acerta. Then I sat down with Acerta DevOps Engineer, Matthew Van Boxtel, Director of Engineering, Prashant Raaghav, and Intern Alex Liu to discuss Acerta’s cloud migration from Amazon Web Services (AWS) to Microsoft Azure. This was easily the most technically complex topic covered on the blog thus far, and you can tell how hard I was trying to wrap my head around it by the fact that the post starts out with an analogy right off the bat.
We intended the post for a technical audience, and that’s abundantly clear if you read through it. There are comparisons between AWS and Azure in terms of asset access management, resource groups, support for machine learning, and permission structures. In the end, we put together a step-by-step guide intended to help others looking to migrate from AWS to Azure.
The Automotive Data Challenge: Enter the Colosseum
I’d written about technical topics and collaborated with Acerta’s c-suite, but working with Acerta CTO, Jean-Christophe Petkovich, brought both of those together. I’d watched his presentation about Acerta’s auto-retraining framework at the company all-hands meeting, but there’s a huge gulf between understanding the content of a presentation and being able to explain it in a few thousand words. Anyone who’s taken copious notes during a conference talk, only to pull them out a month later and find them indecipherable should be able to relate.
Fortunately, our CTO’s combination of patience and articulacy more than made up for my pronounced lack of expertise. The basic idea is easy enough to grasp: manufacturing data changes over time, and machine learning needs to account for that—the best model for predictions today may not be the best a month from now. That’s the problem Acerta’s auto-retraining framework is designed to solve, and the way it works is pretty cool.
Artificial Intelligence & Industry 4.0: 5 Manufacturing Applications for AI
Hey, I wrote this one!
This was the first topic I felt capable of covering without needing to defer to anyone else at Acerta (although I still did). As much as I despise the word ‘listicle’ I’ll admit to having written a few at my old job. As with my previous work, I endeavored to do more than the bare minimum of listing of three to ten items with enough filler to look legit.
This post gave me a chance to do some research on industry trends and clarify where Acerta fits in. Did you know that the majority of automakers believe AI will be highly important if not critical to their success in manufacturing, but fewer than half of them are planning to increase their spending on AI by more than 10%?
Software Internship Advice from Acerta: What We Learned as Interns
Now this was a post where I definitely needed input from my coworkers. I was technically an intern for three months when I began my career in the private sector, but as a lowly humanities student I never had access to co-op programs, so this whole “interning” thing was quite foreign to me.
Sitting down with Alex, Sofia, Mary and Yifan helped me appreciate what it takes to become a software engineer, and the importance of thinking about your career strategically from an early stage. I couldn’t help but feel some resentment for the feckless, indolent pedant I was at their age. Come to think of it, maybe there’s a reason you don’t see co-op jobs for philosophers…
What Makes Manufacturing Data Different?
Any conversation that ends up filling a whiteboard is a good one, as far as I’m concerned, and that’s exactly what happened working with Senior Data Scientist, Sergey Strelnikov. I had a minimal grasp of what manufacturing data involved, but Sergey introduced me to a subject about which I knew even less: the stock market. It doesn’t take much of a leap to recognize that—especially as the volume of manufacturing data increases—manufacturing data and stock market data have a lot in common.
What’s really interesting, though, are the differences between the two. Consider data stationarity—the extent to which the statistical properties of a time series remain constant over time. While manufacturing data may appear to be non-stationary on the surface, if you break something like an end-of-line (EOL) test into phases, you’ll find stationarity in each segment. Understanding the contrast between financial and manufacturing data really helps illuminate what makes the challenges for machine learning in manufacturing unique.
Driving the Future of Mobility with Machine Learning
This was Acerta’s first guest-post and we couldn’t have asked for a better external contributor. Darren Coil, Director of Business Strategy at Microsoft, is working to bring intelligent supply chain technologies (such as ours) to the automotive industry. He’s had a long and varied career, manufacturing everything from microprocessors to carnival rides, and it’s given him a unique perspective on Industry 4.0.
In the post, Darren discusses what he sees as the biggest challenge for manufacturers: extracting all and only the relevant information from your organization in order to make business decisions. He also offers up his own take on Industry 4.0 as a stepping to something greater: the closed-loop supply chain.
Meet the Man Who Built Our Table
Our first guest post was followed up with another external collaboration, this time with engineer and automotive polymath, Lorin Maran. We’d wanted to talk about Lorin and his work for Acerta ever since we moved into our new office and he delivered the stunning engine coffee table he built. Now the chance had finally arrived.
In the course of our conversation, Lorin talked about building an electric Porsche, riding the Tail of the Dragon and, of course, building a coffee table for Acerta. If you’ve been working your way through the Acerta blog posts from Q3, this one should make for a nice change of pace.
Machine Learning & The Last 5%
One thing I learned working at a company that’s growing quickly is that you aren’t the newest person for very long. Nathan Lai joined Acerta as a Sales Engineer not long after I started, bringing a shocking amount of automotive manufacturing knowledge with him. I asked Nate to review Luke’s post before we sat down to discuss his perspective on AI and the automotive industry.
In the end, Nate’s post offers similar sentiments as Luke’s but from a somewhat different angle. While Luke worked primarily in testing, Nathan was charged with overseeing production. These are two very different worlds, despite being in the same industry, and the challenges particular to working on automotive assembly lines come through loud and clear in Nate’s writing.
The Difference Between Machine Learning & SPC (and Why it Matters)
A very fitting final entry on our list, this blog post, written in collaboration with Nathan Lai and Jean-Christophe Petkovich brings together many of the ideas and themes touched upon in previous posts. There’s confusion surrounding the relationship between conventional approaches to quality, such as statistical process control (SPC), and relatively new methods, such as machine learning.
In my view, that confusion is indicative of the ongoing transition from the third industrial revolution (characterized by automation and computerization) to the fourth (characterized by connectivity and digitalization). This post carefully pulls the concepts of SPC and machine learning apart in order to demonstrate that the two methods are most definitely not in competition. A better way to think of the relationship between them is that machine learning is the natural next step for SPC.
And that wraps up Q3 2019 for the Acerta blog!
For More Information...
Check out other blog posts or read our case studies.