CONTRIBUTIONS TO THE INVESTIGATION OF THE MECHANICAL BEHAVIOR OF ROSE HIP SEEDS UNDER COMPRESSION AND CREEP TESTS USING UNSUPERVISED MACHINE LEARNING TECHNIQUES
CONTRIBUȚII LA CERCETAREA COMPORTAMENTULUI MECANIC AL SEMINȚELOR DE MĂCEȘE PRIN TESTE DE COMPRESIUNE ȘI FLUAJ UTILIZÂND TEHNICI DE ÎNVĂȚARE AUTOMATĂ NESUPRAVEGHEATĂ
DOI : https://doi.org/10.35633/inmateh-79-95
Authors
Abstract
This study proposes an artificial intelligence-based method for the automatic identification of natural groups of mechanical behaviour in rose hip seeds (Rosa canina). The method combines the analysis of features extracted from force–displacement curves obtained during compression and creep tests with unsupervised machine learning techniques, namely Principal Component Analysis (PCA) and K-means clustering. This approach enables the identification of distinct mechanical patterns and the objective classification of samples according to their response to mechanical loading. Each cluster was analysed separately in terms of the experimentally determined mechanical parameters. For each group, the mean values of maximum compression force (Fmax), displacement at maximum force (xmax), deformation energy, and initial stiffness were calculated. The results revealed the existence of two main categories of mechanical behaviour. The first group was characterized by an average maximum force of approximately 141 N and a displacement at maximum force of 0.73 mm, indicating seeds with high mechanical strength. The second group exhibited an average maximum force of approximately 75 N and a displacement at maximum force of 0.85 mm, corresponding to seeds that were more deformable and less resistant to compression. To validate cluster separation and reduce the dimensionality of the dataset, Principal Component Analysis (PCA) was applied. The quality of clustering was assessed using the silhouette index, which indicated a satisfactory separation of samples according to their mechanical response. PCA revealed that the main differences between clusters were associated with maximum force, absorbed energy, and initial stiffness values, confirming the existence of distinct categories of mechanical behaviour within the analysed seed batch
Abstract in Romanian



