Manufacturers of automotive rubber products grapple with variable input materials. These compound other rubber mixing process optimization issues in tire production. In combination, adjustable interdependencies and uncontrollable factors can lead to excess waste if not understood completely. In an industry feeling pressure to achieve more energy-efficient production, data-driven techniques can shed light on the road ahead.

Recently, a world-leading tire manufacturer engaged DataProphet to optimize its rubber mixing process. Technologically, this customer was seeking a data-to-value solution. On a strategic level, it aimed for peak operational efficiency insights to accelerate its digital transformation roadmap. In practice, this collaboration leveraged the core capabilities of functional manufacturing expertise and data science. Read on to learn how it helped position an automotive OEM for successful optimization in the digital manufacturing era.


Initial conversations with the tire manufacturing experts clarified immediate optimization targets. They revealed the need to address the rubber mixing process and quality results. Opportunities and challenges arose around optimizing the mixed compound product. 

Firstly, there was the proprietary nature of the tire manufacturing process. It added complexity to outsourcing target quality optimization. Secondly, success would mean simultaneously optimizing the six target quality metrics. Finally, the solution would need to account for weather and raw-material quality variability.

Collaborating with the manufacturer, DataProphet relished adapting its technology for high-quality rubber outcomes. Consequently, the team explored the predictive capabilities of its data-driven technology statically and without plant access to address the constraints. 

From here, the data scientists sought to deliver prescriptive optimization guidance for the manufacturing process. The recommendations aimed to improve six quality metrics on two lines (illustrated below in figure 1):


The ultimate goal was to determine peak process efficiency for the tire manufacturing process. However, DataProphet needed an adaptive approach that leveraged extensive data analysis of the six quality metrics without seeing the actual parameters.  

As a project team member and DataProphet data scientist, Renita Raidoo explains:

“Typically, with deployments, DataProphet consults with plant teams on the factory floor to get a dynamic understanding of their process from input A to result Z. This means, for example, we can learn optimal temperatures and unreasonable values. Our teams thus apply some process knowledge while we train models and find optimal plant states with sight on specific parameters.”

In this case, however, the data scientists did not have the luxury of inherent process knowledge to guide their decisions. Yet, by analyzing the data, they found a way to adapt their techniques.

“Preprocessing reduced the effects of outliers, and we identified appropriate imputation methods to handle missing data points,” said Raidoo.

Jumps in the data suggested production regime changes. Subsequently, digging into the regime changes yielded valuable optimization insight. The team observed that the plant operated in one band for specific periods. 

But after a couple of weeks, another period of operation in a different band could be detected across various parameters. This variability suggested that the plant’s peak operating efficiency varied with changing, uncontrollable conditions.


Let’s take a step back. 

Normally, DataProphet analyzes and models a factory’s historical process and quality data. This analysis reveals a plant’s best-of-best (BOB) operating region. Once DataProphet teams establish the BOB, they can suggest changes to plant control plans (See figure 2).

These adjustments guide the plant towards its holistic optimal.

In this use case, the targeted rubber manufacturing process presented an interesting test for the  PRESCRIBE system.

Rubber mixing is a critical phase in the tire manufacturing value chain. Additionally, it is particularly susceptible to uncontrollable factors over time. According to Raidoo:

“A manufacturer may produce in summer with a certain supplier for a period. In this instance, a BOB range for winter from a different supplier may not apply.” 

Therefore, the crux of the work was context discovery. 

“The aim was to prescribe for different BOB states and plant state shifts. So, we had to proceed based on uncontrollable values. These led us to find multiple BOBS.” (See figure 3)

Contextualizing these operating regimes crystallized optimals based on a particular plant state over time.


The team sent the initial prescriptions to the tire manufacturer’s in-house data scientists to interpret. Next, DataProphet’s data scientists consulted with the plant engineers to determine if the recommendations made sense. Happy with the prescriptions, the manufacturer was confident in the abilities of the PRESCRIBE system (See figure 1) to add value. 

Extending the solution, DataProphet’s team looked further into jumps in the production regimes. The deeper analysis determined other specific plant state contexts. These unique states apply to the rubber mixing process and their controllable and uncontrollable values. This deep dive confirmed that a global BOB might be relevant to one distribution but unattainable for different rubber production process contexts.


The manufacturer only previously targeted one metric for rubber mixing optimization. Conversely, the DataProphet team deployed PRESCRIBE to optimize all six target quality metrics simultaneously. Figure 4 shows the specific rubber mixing outcomes and their broader operational impact:

With this collaborative solution, plant engineers now have the insights to leverage their rubber manufacturing process knowledge with high-quality data. These insights are an opportunity to guide better decision-making — around tire production parameters, blending rules, raw material quality, weather, and substitute quality.


What are the next steps of the collaborative process?

Firstly, these use-case discoveries usher in a commercial partnership to scale the insights across other rubber manufacturing plants. Secondly, the partnership will examine the potential for various transfer learning methodologies and their applications to high-quality tire manufacturing. 

In particular, DataProphet and the global tire OEM will aim to leverage transferability with multi-product lines and process shifts in time. In conclusion, Raidoo acknowledges the great potential for this partnership:

“We are very excited to be part of the fast rollout journey with this tire manufacturing partner. As a group, we believe our insights from the use case can greatly impact the quality outcomes of this next phase at scale.”

Specifically, the team will take the models trained on quite an advanced tire manufacturing plant regarding data maturity. It will then apply that learned knowledge to a less mature plant that is yet to have six months of tire production data to train models on. Raidoo defines the digital roadmap:

“In partnership, we will leverage our AI and their rubber manufacturing expertise to find the best applications for transfer learning at different plants in this manufacturing space.”

%d bloggers like this: