Identification of pear tissue mechanical parameters by integrating machine learning and optimization methods

Area of Science:

• Food Science and Technology
• Biomechanical Engineering
• Computational Modeling

Background:

• Mechanical properties of fruit tissue are vital for food processing, but traditional testing is destructive and slow.
• Existing methods struggle to capture the complex mechanical response of fruit tissues under load.

Purpose of the Study:

• To develop a computationally efficient model for identifying material parameters of fruit tissue.
• To accurately describe the viscoelastic behavior of pear tissue using a hybrid approach.

Main Methods:

• Combined experimental analysis, finite element analysis (FEA), and a hybrid machine learning (ML) model.
• Developed a surrogate architecture with convolutional layers and attention-based bidirectional LSTM for rapid mapping.
• Utilized an enhanced genetic algorithm for parameter identification.

Main Results:

• The surrogate model achieved a minimum mean absolute percentage error of 0.05.
• Force-displacement plots showed less than 0.08 error compared to experimental data.
• The model accurately predicted the mechanical behavior of Huangguan pears.

Conclusions:

• The proposed technique offers a computationally efficient and experimentally economical solution for determining fruit tissue material parameters.
• This approach reduces the need for extensive physical testing while maintaining predictive accuracy.
• Potential applications include designing better food protection equipment to minimize storage losses.