Gaizka Erkizia

Blending poly(3-hydroxybutyrate-co-3-hydroxyvalerate) (PHBV) with thermoplastic polyurethane (TPU) improves toughness but compromises compostability above a critical TPU content. To our knowledge, this is the first three-dimensional, voxel-based Monte Carlo (MC) framework for simulating the disintegration of PHBV/TPU blends under ISO 20200 composting conditions. It couples Monod-type microbial growth, a cardinal-temperature correction, biomass and moisture transport, and a blend morphology generated by random sequential adsorption (RSA) and calibrated against scanning electron microscopy (SEM) data. The transition between complete and incomplete disintegration is described by partitioning the TPU phase into reachable and trapped fractions derived from the connectivity of the generated TPU network. Calibrated against eight compositions (100/0 to 0/100; 58 °C; 41 days) with 22 adjusted parameters, the model yields a root-mean-square error (RMSE) of 3.97% over all 40 fitted points and reproduces complete disintegration up to 40 wt% TPU, the incomplete disintegration of the 50/50 blend and the minimal degradation of neat TPU. In a hold-out test, the compositions withheld from calibration (90/10 and 70/30) were predicted with an RMSE of 3.23%. Sensitivity and identifiability analyses reveal a composition-dependent parameter hierarchy, and virtual temperature experiments illustrate the thermal behaviour implied by the model assumptions.

1. Engineering education struggles to maintain students’ engagement in foundational courses with abstract content (Hwang and Chen 2023; Streveler and Menekse 2017). Engineering Graphics is a prime ...

Purpose: The aim of this in vivo study was to evaluate and compare the accuracy (trueness and precision) of three portable facial scanning methods: a professional handheld structured‐light scanner (P‐HS), a low‐cost handheld structured‐light scanner (LC‐HS), and a smartphone with a depth‐sensing camera (SP). Materials and Methods: Twelve adult volunteers were recruited. Seventeen facial soft‐tissue landmarks were marked with adhesive fiducials, and 12 inter‐landmark distances were measured. Ten repeated facial scans were acquired from each participant with each method, plus reference scans from an industrial reference scanner. The relative error (RE) was calculated, and trueness and precision were expressed as the mean (Mean‐RE) and standard deviation (SD‐RE) of REs per participant, distance, and method. The methods were compared using linear mixed‐effects models, with pairwise comparisons adjusted by the Holm method ( α = 0.05). Results: Statistically significant differences were found in the trueness and precision. The P‐HS was the most accurate. The P‐HS and SP showed comparable precision, about twice that of the LC‐HS, which was the least precise and tended to overestimate. Despite a near‐zero net trueness, the SP produced the largest deviations at individual distances. Differences between the trueness and precision of P‐HS and SP were not statistically significant. Conclusions: The facial scanning method significantly affected accuracy. The P‐HS best suits high‐fidelity workflows; the LC‐HS remained within clinically acceptable limits in other studies for less demanding tasks; and the SP was least accurate where point‐to‐point fidelity is required. Method selection should follow the accuracy each workflow requires.

The disintegration time of poly(3-hydroxybutyrate-co-3-hydroxyvalerate) (PHBV) in compost varies by more than an order of magnitude for the same polymer, and standardised tests quantify this variability without explaining it. Here we present a three-dimensional, spatially resolved Monte Carlo model of PHBV film disintegration, in which the polymer is discretised into a voxel lattice derived from the specimen geometry and eroded via local Monod kinetics. A single reference kinetic set is calibrated once against a benchmark dataset (RMSE = 2.27 pp) and held fixed; each compost is then described by one physically interpretable parameter, the initial density of degrading inoculum, whose fitted range quantifies the microbial-activity variability the standards leave unexplained. With the kinetics fixed, the model reproduces an independent thermophilic dataset by varying only this parameter and the geometry (RMSE = 3.05 pp), and passes two a priori tests with the parameter transferred, not fitted: a 400 µm film from the literature (RMSE = 14.1 pp) and a 50 µm film characterised here (RMSE = 7.01 pp). Together these span two ISO standards and an eightfold range of thickness (50–400 µm), bracketing the 200 µm calibration point. A further dataset, obtained in undiluted real compost rather than a diluted synthetic substrate, is reported as a documented exception requiring its own kinetic set (RMSE = 5.24 pp). Because geometry is an explicit input, the model predicts how thickness sets the disintegration window and resolves the spatial fragmentation pattern, beyond the reach of empirical fits.

Functional dimensioning represents a critical competency in first-year engineering education, requiring students to transition from geometric description to functional reasoning. This quasi-experimental study (N=240) evaluated the technical efficacy and motivational impact of Generative Artificial Intelligence in teaching functional dimensioning and ISO 286 fit specification. A comparative design was employed, with an experimental group (n=170) using a structured four-phase prompting framework with a large language model, and a control group (n=70) following traditional instruction. Both groups analysed mechanical assemblies across eight 2-hour sessions. Academic performance was assessed via a blind-graded technical rubric, and motivation through validated questionnaires measuring self-efficacy, intrinsic motivation, perceived competence and engagement. Results indicate that Artificial Intelligence enhances conceptual understanding of functional dimensioning (85% vs. 72% critical dimension identification) while showing mixed calculation accuracy contingent on validation behaviours (systematic validators: 90% vs. minimal validators: 65%). The experimental group demonstrated elevated self-efficacy (d=0.79) though correlation with performance was modest (r=0.30). Findings suggest that properly scaffolded Artificial Intelligence integration reduces affective barriers while requiring explicit instruction in structured prompting and critical validation practices.

The transition to Industry 4.0 requires engineering graduates to possess high levels of autonomy and self-regulation. However, implementing Active Learning (AL) strategies in first-year courses often faces significant student resistance. This paper contributes empirical evidence to document and contextualise the 'autonomy gap' observed during the 2024-2025 academic year in a first-year engineering design course. Through a mixed-method approach comprising diagnostic tests and perception surveys, we evaluate student readiness for Flipped Classroom methodologies. Results indicate a marked preference for traditional passive instruction and printed materials, despite the availability of comprehensive digital resources. These findings suggest that while students acknowledge the importance of soft skills, they lack the necessary self-regulation strategies to thrive in AL environments without substantial scaffolding. The study concludes by recommending transitional pedagogical frameworks to gradually build student autonomy and engagement.