Supernova Award Category

The Problem

To ensure consistent product quality, the lab at the metallurgical complex in Sorel-Tracy, Quebec, creates and subsequently tests hundreds of sample parts each week (one part for each grade of powder) to ensure they meet specifications for hardness, transverse rupture strength, and various other physical properties. The lab results showed a slow but steady slippage of specifications for one grade of powder, which resulted in the plant receiving internal rejects – i.e. batches of powder that it wouldn't sell because they didn't meet the plant’s high standards for quality.

Marko Litalien, Superintendent of Process Development, believed the explanation for the slipping specification could be found somewhere within the RTFT databases. The challenge then became how to quickly and accurately search through the raw sensory data containing more than 100 variables of processing conditions, and the physical and chemical properties of every input and output. In the past, Litalien relied on an outside statistical analysis firm, or a desktop multivariate-analysis tool with a 15 variable capacity for answers, but RTFT needed a solution that offered simple, accurate data that showed them how and why this decline occurred.

The Solution

RTFT turned to Nutonian’s machine intelligence application, Eureqa, to automatically sort through the large amounts of data to isolate which variables were influencing the outcome, to what extent, and how the different process conditions played together to determine the final product. 45 minutes into the search, Eureqa's analytical model revealed a previously unknown link between changes in an aspect of chemical composition and changes in the product specification - a link RTFT believes is still largely unknown within their industry, which gives them a massive competitive advantage enabling them to consistently produce extremely high-quality powders every time.

The results

At first, the engineering team wasn’t immediately convinced that the implicated variable had a "real" effect, but after a couple days of crosschecking and searching through the metallurgical literature, the correlation held up.

The engineering team then went to work devising process modifications that would control the variable. After testing, these were implemented, and the specification returned to normal.

Eureqa had figured out the root cause of what was driving the quality issue: one of the specific input variables (that the team of PhDs had initially excluded from their analysis, because they were convinced it didn’t matter) was interacting with the others to drive down quality. Eureqa found this answer in less than an hour.

Eureqa may change how any and all manufacturers optimize nearly every part of their assembly line by being able to communicate what process conditions and inputs yield high-quality (and lower cost) products. This was previously an extremely time-intensive and inaccurate process. Eureqa has automated it entirely.

Metrics

Prior to using Eureqa to uncover the link between changes in an aspect of chemical composition and changes in the product specification, RTFT experienced a decline in quality, and lost product earnings.

Eureqa’s model transparency provided RTFT with clear signals on which variables are important and which are not, and they were also able to run normalization and other pretreatment of data within Eureqa. Engineers can now apply their domain expertise in the choice of mathematical "building blocks," thereby significantly increasing search efficiency.

Since implementing the process modifications, RTFT has improved quality assurance, and gained a consistent supply of product to customers.

Eureqa enabled Rio Tinto to figure out a problem that a team of a half dozen PhDs had been working on unsuccessfully for months. These batches of titanium powder cost $100K per week to produce. Whenever these batches failed, they’d have to reprocess the entire batch. Eureqa prevented Rio Tinto from ever dipping below its quality requirement, when it was consistently sharply decreasing and dangerously nearing the limit. 

The Technology

Nutonian’s machine intelligence application, Eureqa, acts like a virtual data scientist and automatically discovers patterns from raw data. Eureqa searches an infinite linear/non-linear equation space – iterating through billions of potential solutions per second – to present the end user with transparent analytical models that explain why and how their engineering processes fundamentally “work”, and what they can change to maximize their desired outcome.

Disruptive Factor

The experience has made Eureqa evangelists out of RTFT. The project managers recently presented on this case study, touting Eureqa's potential, which was warmly received by an audience of RTFT process plant managers, engineers, and technicians. Two R&D directors reported having particular applications already in mind.

RTFT process plant managers, engineers, and technicians now turn to Eureqa to sort through their data variables and deliver insight that leads to increased quality control and more efficient plant operations. Below are testimonials from the customer:

Litalien supplies a closing thought: "Knowing what we now know, we're going to use Eureqa whenever there's an opportunity. It's something special you have here. I'm convinced that if you give it good data, and the relevant variables are included, you're going to get a good answer."

Frédéric Benoit, a process engineer, adds: "We're living in an era where we have an enormous amount data that no one really understands. When I see software like this that can take that data and make sense of it, I think we're going in the right direction."

No other big data or data science technology could possibly uncover the complexity of relationships happening in Rio Tinto’s data like Eureqa did…let alone the speed at which Eureqa did it. Eureqa tells you why and how things happen, and what to change to get the best results – all automatically.

Shining Moment

Eureqa’s ability to test an infinite number of variables empowered plant engineers to uncover critical changes in an aspect of chemical composition and changes in the product specification that prior analyses (and analysts) had missed. This is because a normal statistical analysis wouldn’t have included that variable. RTFT only included it because they knew Eureqa could easily handle large numbers of variables.