Accelerating Research With Predictions to Outperform Mechanical Performances in Elastomer-Sealing Products
The customer is an industry pioneer in the elastomer-sealing products space. Prior to engaging with Polymerize, they had been searching for an artificial intelligence solution to address their need to continue new product development while minimising trial-and-error. Their goals consist of ensuring maximum accuracy by using optimal combinations and concentrations of its ingredients, thereby securing high efficacy and quality formulations.
- Number of employees: 1000+
- Industry: Seal Manufacturing, Elastomers
- Location: Japan, USA
- Cutting down trial-and-error-based experimental efforts by 50%
- Providing key insights in optimal combinations and concentrations of ingredients to be utilised for future product development
- Significantly improved mechanical performance of the end product
In recent years, there has been a rising trend in using sealing products in variedly harsh environments. From HPHT steam and sour gases in the petrochemical process to plasma processes in the semiconductor industry, the standard of mechanical performance of sealing products has risen. Their success majorly rests on the factored ratios and concentrations of constituting ingredients used in the formulations.
The customer needed to minimise the compression set whilst ensuring that the elongation at the breaking point was maintained to optimise the seal's performance at high temperatures. After three months of experimenting as per usual, the customer sought out Polymerize for a pilot.
The customer looked into engaging Polymerize and provided Polymerize with 25 sets of experiments with their corresponding outputs and the targeted property objectives. These experiments mainly consisted of changing the combinations and compositions of different carbon compounds, curing agents, and additives to achieve target properties.
Using a combination of:
- Test results from the provided experiments
- Constraint ranges of the constituent ingredients
- Pre-existing domain knowledge
Polymerize's forward prediction AI models made predictions based on the given ingredients that served as inputs, while our inverse models were simultaneously trained to recommend formulations based on input target properties. The result was a set of five experiments that were suggested to the customer to be performed and validated.
Within the very first iteration, the customer fulfilled their objectives by validating 4 out of the 5 recommended experiments. When validated against their best-performed results, they saw a stark increase in the performance of the compression sets while maintaining elongations in desired ranges.
Within just three weeks, the customer managed to obtain beyond their desired results; and even got the opportunity to experiment with different carbons to achieve their desired performance goals, significantly easing the overall supply chain related to the new product development.