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

The mechanical properties of fruit tissue play a crucial role in various food processing activities, including sorting, packaging, and transportation. Most traditional testing processes are destructive, time-consuming, and cannot capture the mechanical response under loading conditions. The present research proposes a parameter identification model that combines experimental analysis, finite element analysis, and a hybrid machine learning model to describe the viscoelastic behaviour of pear tissue. To obtain a rapid mapping from the intrinsic material constants to mechanical behavior, we develop a feasible surrogate architecture. This framework incorporates convolutional layers to extract spatial features, along with an attention-based bidirectional LSTM, which ensures a high-fidelity approximation of mechanical behavior. An enhanced genetic algorithm is suggested as a solution to the identification problem. The framework is tested through compression tests on Huangguan pears. The results indicate that the mean absolute percentage error of the surrogate model can reach a minimum of 0.05, and the force-displacement plots show an error of less than 0.08 compared to experimental data. The suggested technique minimizes the need for extensive physical testing while preserving predictive accuracy, providing a computationally efficient and experimentally economical solution for determining the material parameters of fruit tissues. It has the potential to assist in designing and optimizing food protection equipment for transporting fresh fruits, thereby reducing storage losses.