Problem
Injection molded components are manufactured from plastics with uncertain material properties that experience large variations in shrinkage due to the post-molding cooling process. Traditionally, components such as this plastic step have accounted for this using large factors of safety, particularly in allowable stress design. In many situations, this can be over-conservative and result in products that use more material and are more expensive to manufacture than is necessary.

Solution
Uncertainty quantification is a field of data science / machine learning (ML) that can be combined with finite element analysis (FEA) to propagate uncertainty from controlled or uncontrolled sources of variation to product objectives and constraints. This enables risk pertaining to constraints to be quantified and managed while assessing or optimising objectives such as cost. This can lead to significant improvements in product performance, greater confidence in the durability of designs while reducing cost.

Result
The Youngs’s modulus and wall thickness were approximated as Gaussian random variables. By constructing a surrogate model, the uncertainty in both properties were propagated to predictions of the stress and displacement constraints. For a specified load limit, the optimum wall thickness and material were computed to minimise product cost subject to a 90% confidence that the maximum stress would not exceed yield and plastic collapse would be avoided. A 30% reduction in plastic volume was achieved, resulting in significant cost savings while ensuring the chair was strong enough for the required design load.