This paper compares the performance of Deep Learning (DL) and multi-template matching (MTM) models for detecting small defects. DL models extract distinguishing features of objects but require a large dataset of images. In contrast, alternative computer vision techniques like MTM need a relatively small dataset. The lack of large datasets for small metal-surface defects has inhibited the adoption of automation in small-defect detection in remanufacturing settings. This motivated this preliminary study to compare template-based approaches, like MTM, with feature-based approaches, such as DL models, for small-defect detection on an initial laboratory and remanufacturing industry dataset. This study used You Only Look Once v5 (YOLOv5) as the DL model and compared its performance against the MTM model for small-crack detection. The findings of our preliminary investigation are as follows: (i) YOLOv5 demonstrated higher performance than MTM in detecting small cracks; (ii) an extra-large variant of YOLOv5 outperformed a small-size variant; (iii) the size and object variety of the data are crucial in achieving robust pre-trained weights for use in transfer learning; and (iv) enhanced image resolution contributes to precise object detection.

Global plastic waste generation exceeds 460 million tons per year, while recycling rates remain below 10%. Innovative solutions are needed for recycling plastic waste into valuable products. This study employs experimental data and two methods to evaluate economic analysis of fluid catalytic cracking (FCC) of plastic-derived pyrolysis oil: physics-based and machine learning (ML)-based methods to model the effects of reactions in the FCC unit. FCC offers many advantages compared to hydrocracking: e.g., it operates under lower pressure (implying lower equipment costs), eliminates the use of hydrogen, and runs with reduced operating costs. The physics-based model was used to determine reaction stoichiometries, ensuring chemical plausibility, while a Decision Tree regressor was employed to predict product yields based on the cycle number and cycle time, a proxy for catalyst deactivation that captures time-dependent experimental observations. Both models were integrated surrogate models within a full process model in BioSTEAM to conduct a technoeconomic assessment (TEA) and a life-cycle assessment (LCA) with uncertainty analysis. The research also highlights the production of valuable chemicals, specifically light olefins such as ethylene and propylene and aromatic compounds such as benzene, toluene, and xylenes (BTX). The ML-based framework estimated a minimum selling price (MSP) for naphtha of approximately $1.38/kg under input uncertainties, while the physics-based approach estimated an MSP of $1.76/kg. These results are within 22% of each other, which is within the expected range of ± 30% for the preliminary TEA estimates. These findings suggest that ML-based approaches can be an effective substitute for physics-based models when there is limited understanding of the underlying chemical mechanisms. Sensitivity analysis captured the impacts of varying the catalyst cycle number and times. Cycle numbers between 1 and 4 and cycle times of up to 20 min resulted in MSPs ranging between 1 and 2 $/kg. This variation captures the impacts of catalyst deactivation in FCC systems, supporting the need for optimizing the catalyst performance.