Machine Learning Times
Machine Learning Times
Three Best Practices for Unilever’s Global Analytics Initiatives
    This article from Morgan Vawter, Global Vice...
Getting Machine Learning Projects from Idea to Execution
 Originally published in Harvard Business Review Machine learning might...
Eric Siegel on Bloomberg Businessweek
  Listen to Eric Siegel, former Columbia University Professor,...
Effective Machine Learning Needs Leadership — Not AI Hype
 Originally published in BigThink, Feb 12, 2024.  Excerpted from The...

3 years ago
Wise Practitioner – Predictive Analytics Interview Series: Markus Larsson at Palo Alto Research Center (PARC)


By: Steven Ramirez, Conference Chair, Predictive Analytics World for Industry 4.0

In anticipation of his upcoming presentation at Predictive Analytics World for Industry 4.0 Livestream, May 24-28, 2021, we asked Markus Larsson, Head of Predictive Maintenance at Palo Alto Research Center (PARC), a few questions about their deployment of predictive analytics. Catch a glimpse of his presentation, The Next Frontier for IoT Predictive Maintenance, and see what’s in store at the PAW Industry 4.0 conference.

Q: In your work with predictive analytics, what behavior or outcome do your models predict?

My team focuses entirely on predicting machine faults. Unexpected manufacturing equipment failures cost factories millions of dollars each year, and increase risks, resulting in employee safety and environmental disasters. We are building our models to help prevent that.

Q: How does predictive analytics deliver value at your organization – what is one specific way in which it actively drives decisions or operations?

We use predictive analytics in a variety of ways at Xerox. For instance, we are delivering PdM solutions to customers today. Our prognostics technology runs on subway trains in Tokyo, and in manufacturing plants in Asia. We are also implementing our technology in Xerox chemical plants in upstate New York and Canada.

Q: Can you describe a quantitative result, such as the predictive lift of your model or the ROI of an analytics initiative?

When we build a prognostics model, our objective is always 90% or better true positive rate with negligible false negative rates, which we define in each case by working with the partner or customer. We have a strong track record of achieving these results and use a variety of modeling approaches to make sure we always get the best results possible. In terms of ROI, the good news is that the data for predictive maintenance ROI – the area we play in the most – is quite good. A recent market report on PdM from IoT Analytics showed that among survey respondents who were able to quantify ROI, not one reported negative ROI, and more than half reported a pay-back period of less than one year. This is in line with our experience.

Q: What surprising discovery or insight have you unearthed in your data?

I’d highlight two things:

  1. Data quality matters more than data quantity. We often come across partners who have done a great job of setting up data collection infrastructure, spent millions of dollars and years’ worth of time to collect terabytes of data, and it’s mostly not useful. It’s better to do it right and collect the appropriate signals that can power prediction models. We are often able to overcome this problem by using pre-built physics models and targeted instrumentation, augmented with machine learning approaches. This issue is something we see many struggles with when implementing a PdM solution.
  2. Technology isn’t enough. To realize the value of predictive analytics in industrial settings, manufacturers require much more than accurate predictions. The information needs to be used properly, which may require anything from appropriate levels of system integration to thoughtful user experience design, workflow integration and staff education and training. Otherwise a great prediction becomes that proverbial tree that fell in the forest when no one was around. Doing this right requires holistic or system-level thinking and, a solution-minded approach as opposed to a tech-first mindset.

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.

I’ll cover many of the items I have discussed above in more detail. I’ll share some cases where the data wasn’t sufficient to get great results, and how we overcame that by using physics models. I’ll also share some insights into how we plan for execution of a great PdM deployment. Tune in – I’m sure you’ll find it useful.


Don’t miss Markus’ presentation, The Next Frontier for IoT Predictive Maintenance at PAW Industry 4.0 on Thursday, May 27, 2021 from 10:20 AM to 11:05 AM. Click here to register for attendance.

By: Steven Ramirez, Conference Chair, Predictive Analytics World for Industry 4.0

Leave a Reply