Experimental study on creep properties prediction of reed bales based on SVR and MLP

Reed has high lignin content, wide distribution and low cost. It is an ideal raw material for replacing wood in the paper industry. Reeds are rich in resources, but the density of reeds is low, leading to high transportation and storage costs. This paper aims to study the compression process of reeds and the creep behaviour of compressed reeds, and provide theoretical guidance for the reed compressor management, bundling equipment and the stability of compressed reed bales.

We have established a multi-layer perceptron network prediction model for the creep characteristics of reeds, and the prediction rate R² of this model is greater than 0.997. The constitutive equation, constitutive coefficient and creep quaternary model of the reed creep process were established by using the prediction model. The creep behaviour of the reed bale is positively correlated with the initial maximum compressive stress (σ₀). During the creep of the reed, the elastic power and the viscous resistance restrict each other. The results show that the proportion of elastic strain in the initial stage is the largest, and gradually decreases to 99.19% over time. The viscoelastic strain increases rapidly with time, then slowly increases, and finally stabilizes to 0.69%, while the plastic strain accounts for the proportion of the total strain. The specific gravity of the reed increases linearly with the increase of creep time, and finally accounts for 0.39%, indicating that as time increases, the damage of the reed's own structure gradually increases.

We studied the relationship between the strain and time of the reed and the strain and creep behaviour of the reed bag under different holding forces under constant force. It is proved that the multi-layer perceptron network is better than the support vector machine regression in predicting the characteristics of reed materials. The three stages of elasticity, viscoelasticity and plasticity in the process of reed creep are analysed in detail. This article opens up a new way for using machine learning methods to predict the mechanical properties of materials. The proposed prediction model provides new ideas for the characterization of material characteristics.