Online Program

Using Deep Learning Models on Images to Predict Physical Stability Failure Modes of Prototype Liquid Consumer Products (309989)

Keywords: product stability, consumer products, deep learning, image analysis, Convolutional Neural Networks (CNN)

The enduring success of P&G brands relies on the quality of our products. Liquid formulated consumer products may experience physical stability problems (e.g. phase separation, crystallization, flocculation, etc.) over time. To screen prototype liquid formulations for potential physical stability problems, hundreds of thousands of images were collected over time. Manual grading of these images was tedious and time consuming. Deep learning model using Convolutional Neural Networks (CNN) and Variance of Laplacian method were used to predict poor image quality (e.g. off-centered or blurred images). A deep learning model using CNN and Transfer Learning was trained with over 95% test accuracy to predict physical stability failure modes. Our approaches can be reapplied to many liquid-based stability detection problems across consumer industry.