About 60% of palatability comes from texture. Sensory evaluation is widely used but depends on subjective factors, while instrumental analysis is objective often disagrees with sensory results. Humans perceive texture through multi-point / multi-directional sensing and texture-dependent mastication, whereas conventional food texture analyzer records only one-axis force during vertical uniform compression independent of each texture. This large gap of experimental conditions is considered a major cause of the discrepancy. Increasing the amount of information obtained during compression and evaluating thousands of measurements could enable more detailed food texture evaluation.

In this study, a machine-learning–based approach was employed to improve the accuracy of texture evaluation to identify key factors contributing to the texture. Two types of automated compression systems were developed. A tooth-shaped plunger and a six-axis force/torque sensor were installed, and compared with the results obtained with conventional simple texture analyzer. Machine learning was applied to the obtained compression results, which is required for more complex results compared with conventional compression instruments. A larger lateral force component was observed with the tooth-shaped plunger than with the conventional disc-shaped plunger. This is likely due to increased lateral sliding during compression of the uneven rice-cracker surface, resulting in larger F _x and M _y values with the tooth-shaped plunger. More than 4,000 compression tests were performed, and two different rice crackers having similar textures were measured. Widely used analysis methods such as TPA (with PCA) could not distinguish the two crackers, whereas the machine learning based method showed higher classification accuracy. Among the six axes, the lateral force component (F_x) and the rotational moment (M_y) showed improved classification accuracy as the number of measurements increased. These axes reflect lateral movements of the tooth-shaped plunger, suggesting that grinding-type motion plays an important role in texture perception.

The machine-learning–based texture evaluation using a tooth-shaped plunger with a six-axis sensor outperformed conventional methods. The six-axis analysis indicates that grinding-type movements are key factors in detailed texture evaluation.

Conference Theme: Physical properties of food hydrocolloids for enhanced product

development