Machine learning strategy to improve impact strength for PP/cellulose composites via selection of biomass fillers

ABSTRACT Lignocellulosic materials have inherent complexities and natural nanoarchitectures, such as various chemical constituents in wood cell walls, structural factors such as fillers, surface properties, and variations in production. Recently, the development of lignocellulosic filler-reinforced polymer composites has attracted increasing attention due to their potential in various industries, which are recognized for environmental sustainability and impressive mechanical properties. The growing demand for these composites comes with increased complexity regarding their specifications. Conventional trial-and-error methods to achieve desired properties are time-intensive and costly, posing challenges to efficient production. Addressing these issues, our research employs a data-driven approach to streamline the development of lignocellulosic composites. In this study, we developed a machine learning (ML)-assisted prediction model for the impact energy of the lignocellulosic filler-reinforced polypropylene (PP) composites. Firstly, we focused on the influence of natural supramolecular structures in biomass fillers, where the Fourier transform infrared spectra and the specific surface area are used, on the mechanical properties of the PP composites. Subsequently, the effectiveness of the ML model was verified by selecting and preparing promising composites. This model demonstrated sufficient accuracy for predicting the impact energy of the PP composites. In essence, this approach streamlines selecting wood species, saving valuable time.


Data for impact prediction models
The data for the characterization of 14 types of lignocellulosic fibrillated fillers with different mechanical preparations, including dry pulverization and subsequent wet disk milling for four and ten passes (Table S1), and the tensile and Izod impact properties of PP/lignocellulosic fibrillated filler composites (Table S2) were based on the results obtained from the Development of Technologies for Manufacturing Processes of Chemicals Derived from Inedible Plants Project, commissioned by the New Energy and Industrial Technology Development Organization (NEDO), Japan (No. JPNP13006).

Demonstration for higher impact energy composites
Fibrillation of lignocellulosic fillers Wood flour (particle size <0.2 mm) was used as the raw material for preparation of LCNFs.
At this time, the distance between the grindstones was narrowed from the initial contact distance to 150 μm.This fibrillation treatment was repeated up to 4 times to obtain aqueous LCNF suspensions (Wet-DM-4).

Preparation of polymer composites
A composite material was prepared as previously reported procedure [S1].A mixture of powdered PP (iPP homopolymer; NOVATEC MA3, Japan Polypropylene Co., Ltd.) and MAPP (Kayabrid 006PP, Kayaku Akzo Co., Ltd.) was used as the matrix.Wet LCNF (PP /LCNF = 90:5 w/w%) was subjected to solid-state shearing with PP powder at room temperature to 100 °C and rotation speed of 60 rpm using a Labo Plastomill R30 (Toyo Seiki Seisaku-sho, Ltd., Japan) for 30 minutes, followed by oven drying.Subsequently, it was combined with MAPP for melt mixing (PP with LCNF/MAPP = 95:5 w/w%) at 170°C and 30 rpm using a twin-screw extruder (4C150 Laboplastomill, 2D15W screw, Toyo Seiki Seisaku-sho, Japan) and cut into pellets.The pellets were molded using an injection molding machine (Babyplast 6/10P, Cronoplasto, S. L., Spain) at an extrusion molding temperature of 180°C and an injection molding temperature of 190°C.The injection pressure was 13 MPa and the mold temperature was the same as room temperature (about 25°C).

Izod impact test
The composite plates were stored in a room with a constant temperature of 22.5°C and a constant humidity of 30% for over 1 week, before the measurement.To minimize errors, each test group consisted of at least three specimens, and the average of the test results was calculated as the measured strength.The dimensions of the specimens tested were 3.8x9.6x58mm.The notched Izod impact strengths of the injection-molded composites were measured using a universal impact tester (No. 258-D; Yasuda Seiki Seisakusyo, Ltd., Japan).Notches were fabricated with a radius of 0.25 mm and a depth of 2 mm.

Filler
No.

Figure S2 .
Figure S2.Loading plot representing the contribution of the original variables for the principal components PC1 to PC5.As variables that contribute most are plotted around the borders of the plot, the inset shows a typical ATR-IR spectra, which was set up with smoothing and 2nd deviation.

Table S2 .
Tensile and Izod impact properties of PP/nanofibrilled cellulose composites.

Table S3 .
Wood species for discovering promising composite materials with high izod impact energy.