Predictive Maintenance Done Right: Q&A with a Tetra Pak Data Scientist

Predictive Maintenance Done Right: Q&A with a Tetra Pak Data Scientist

Discover more about the Altius approach to predictive maintenance here 

Predictive maintenance is a key discussion area for manufacturers when it comes to ways data can improve machine efficiency and productivity. McKinsey Digital says that businesses, on average, spend 80% of their time responding to maintenance issues that rise rather than preventing them. And in 2019, 50% of UK logistics companies said predictive maintenance is of high or very high relevance 

That’s why we caught up with data scientist Noah Schellenberg to learn more about his experience with predictive maintenance. He works at Tetra Pak, the world’s leading food processing and packaging solutions company on a mission to “Protect What’s Good”. Producing over 500 million packages per day requires machines that operate effectively, which is where a robust predictive maintenance makes a difference.   

Tell us about predictive maintenance and its importance in manufacturing.    

Predictive maintenance estimates when machines require maintenance, and also helps prevent costly, unexpected equipment failuresAt Tetra Pak, we use predictive maintenance on our filling, processing and distribution equipment 

When thinking about which machines to prioritise, you’ll find the most value in putting sensors on ones that operate at a higher utilisationMachines that do not operate around-the-clock have more opportunity for preventive servicing, hence they have lower priority.      

When getting started, it’s important to understand the utilisation of your equipment, as there is no one size fits all approach.  

Speaking of getting started, what advice do you have for companies to get going?  

It begins with putting sensors on machines and using the power of the Internet of Things (IoT) to collect data, but eventually you need to find value from the data. To do this, you’ll need a proper data acquisition strategy in place with quality and governance and then eventually use data science techniques.  

One key point to remember is that installing sensors and collecting data is not a quick fix. It is a journey, one which can take several years. To accelerate any development, it is key to use the technical expertise of a cloud partner such as Microsoft that has sensors and experts in IoT data capture  

The movement to the Microsoft Azure cloud has been a huge factor in all of this, which we went all in on a few years back. The cloud allows you to scale infinitely, while on premise data is complicated and discouraging to work with.  

What teams do you need involved in successful predictive maintenance?  

First off, you need to foster a collaborative culture at your company between business, engineering and data science. You also cannot underestimate the need for a robust DevOps team. You’ll need business engagement, data science ownership of models, and constant monitoring, and if you can’t find this in house, partners like Altius provide these capabilities.  

It can be helpful to have a reliability engineer involved who understands failure frequencies and can use past data on maintenance actions: which parts failed and what were the downtimes. Doing this can really set yourself up for success.  

Remember, parts have to fail to collect the data needed for predictive maintenance and to subsequently perform a Machine Learning model. We have a rule-of-thumb to have a labelled data set of at least 60 points for a particular failure mode before we engage our data scientists. 

What advice do you have for working with data scientists? 

Don’t distract your data scientists with huge sets of data that don’t add value to the business. You want the data to be actionable and high impact. Additionally, data scientists should work closely with subject matter experts to find context and get the most out of the data. 

As a data scientist myself, I chose Tetra Pak because it is all in on the Microsoft Azure cloud. The cloud means way less constraints which makes data science much more exciting. I would always suggest fellow data scientists choose a cloud-first organisation where possible! 

Lastly, how does predictive maintenance help manufacturers in certain industries meet spikes in demand during a pandemic? 

In times like this, the value of predictive maintenance can go up for businesses with spikes in demand. There is no opportunity for downtime, so companies that have chosen predictive maintenance may have found more value during the pandemic when you have to push your machines to the limit.  

Discover more about the Altius approach to predictive maintenance here.