Chenhui Shao

Additive manufacturing (AM) continues to transform modern manufacturing by enabling flexible, on-demand production of complex geometries across diverse industries. Fused filament fabrication (FFF) has extended AM to laboratories, classrooms, and small production environments, but this accessibility shifts process-planning responsibility to users who may lack manufacturing expertise. A syntactically valid slicer profile can still encode thermally or geometrically harmful settings, and subtle G-code edits can alter extrusion, cooling, or adhesion before a print begins. Pre-print G-code screening catches accidental or adversarial machine-program errors before material or machine time is wasted. This paper proposes LLM-ADAM as a generalizable LLM framework for pre-print anomaly detection in AM. The framework decomposes the task into three roles: Extractor-LLM maps a G-code file to a structured process-parameter schema; Reference-LLM converts printer and material documentation into aligned operating ranges; and Judge-LLM interprets a deterministic deviation table and G-code evidence to decide whether a part is non-defective or belongs to an anomaly class. Printers, materials, and LLM backbones are interchangeable test conditions, not fixed assumptions. We evaluate the framework on an N=200 FFF G-code corpus spanning two desktop printer families, two materials, and five classes including non-defective, under-extrusion, over-extrusion, warping, and stringing. The best framework configuration reaches 87.5% accuracy, compared with 59.5% for the strongest engineered single-LLM baseline. The results show that structured decomposition, rather than backbone strength alone, is the dominant source of improvement, with defect classes identified at or near ceiling for leading configurations while residual errors concentrate on conservative false alarms for non-defective samples.

Additive manufacturing (AM) continues to transform modern manufacturing by enabling flexible, on-demand production of complex geometries across diverse industries. Fused filament fabrication (FFF) has extended AM to laboratories, classrooms, and small production environments, but this accessibility shifts process-planning responsibility to users who may lack manufacturing expertise. A syntactically valid slicer profile can still encode thermally or geometrically harmful settings, and subtle G-code edits can alter extrusion, cooling, or adhesion before a print begins. Pre-print G-code screening catches accidental or adversarial machine-program errors before material or machine time is wasted. This paper proposes LLM-ADAM as a generalizable LLM framework for pre-print anomaly detection in AM. The framework decomposes the task into three roles: Extractor-LLM maps a G-code file to a structured process-parameter schema; Reference-LLM converts printer and material documentation into aligned operating ranges; and Judge-LLM interprets a deterministic deviation table and G-code evidence to decide whether a part is non-defective or belongs to an anomaly class. Printers, materials, and LLM backbones are interchangeable test conditions, not fixed assumptions. We evaluate the framework on an N=200 FFF G-code corpus spanning two desktop printer families, two materials, and five classes including non-defective, under-extrusion, over-extrusion, warping, and stringing. The best framework configuration reaches 87.5% accuracy, compared with 59.5% for the strongest engineered single-LLM baseline. The results show that structured decomposition, rather than backbone strength alone, is the dominant source of improvement, with defect classes identified at or near ceiling for leading configurations while residual errors concentrate on conservative false alarms for non-defective samples.

Machine learning, particularly deep learning, is transforming industrial quality inspection. Yet, training robust machine learning models typically requires large volumes of high-quality labeled data, which are expensive, time-consuming, and labor-intensive to obtain in manufacturing. Moreover, defective samples are intrinsically rare, leading to severe class imbalance that degrades model performance. These data constraints hinder the adoption of machine learning-based quality inspection methods in real production environments. Synthetic data generation (SDG) offers a promising solution by enabling the creation of large, balanced, and fully annotated datasets in an efficient, cost-effective, and scalable manner. This paper presents a hybrid SDG framework that integrates simulation-based rendering, domain randomization, and real background compositing to enable annotation-free computer vision-based industrial part inspection and zero-shot deployment on real parts. In our implementation, the SDG pipeline generates 12,960 labeled, balanced images in one hour, demonstrating its efficiency and scalability in addressing manufacturing data bottlenecks. A two-stage architecture utilizing a YOLOv8n backbone for object detection and MobileNetV3-small for quality classification is trained exclusively on synthetic data and evaluated on 300 real industrial parts. The proposed approach achieves a mean Average Precision (mAP) of 0.995 at an intersection-over-union threshold of 0.5 (mAP@0.5) for object detection and 96% overall accuracy with 90.5% balanced accuracy for quality classification, indicating robust performance under severe class imbalance. Compared with few-shot baselines trained on limited real data, the SDG approach improves balanced accuracy from 50% to 91%. These results demonstrate that the proposed method enables annotation-free, scalable, and robust quality inspection for real-world manufacturing applications.

Ultrasonic metal welding (UMW) is widely used in industrial applications but is sensitive to tool wear, surface contamination, and material variability, which can lead to unexpected process faults and unsatisfactory weld quality. Conventional monitoring systems typically rely on supervised learning models that assume all fault types are known in advance, limiting their ability to handle previously unseen process faults. To address this challenge, this paper proposes an adaptive condition monitoring approach that enables unknown fault detection and few-shot continual learning for UMW. Unknown faults are detected by analyzing hidden-layer representations of a multilayer perceptron and leveraging a statistical thresholding strategy. Once detected, the samples from unknown fault types are incorporated into the existing model through a continual learning procedure that selectively updates only the final layers of the network, which enables the model to recognize new fault types while preserving knowledge of existing classes. To accelerate the labeling process, cosine similarity transformation combined with a clustering algorithm groups similar unknown samples, thereby reducing manual labeling effort. Experimental results using a multi-sensor UMW dataset demonstrate that the proposed method achieves 96% accuracy in detecting unseen fault conditions while maintaining reliable classification of known classes. After incorporating a new fault type using only five labeled samples, the updated model achieves 98% testing classification accuracy. These results demonstrate that the proposed approach enables adaptive monitoring with minimal retraining cost and time. The proposed approach provides a scalable solution for continual learning in condition monitoring where new process conditions may constantly emerge over time and is extensible to other manufacturing processes.

Ultrasonic metal welding (UMW) is widely used in industrial applications but is sensitive to tool wear, surface contamination, and material variability, which can lead to unexpected process faults and unsatisfactory weld quality. Conventional monitoring systems typically rely on supervised learning models that assume all fault types are known in advance, limiting their ability to handle previously unseen process faults. To address this challenge, this paper proposes an adaptive condition monitoring approach that enables unknown fault detection and few-shot continual learning for UMW. Unknown faults are detected by analyzing hidden-layer representations of a multilayer perceptron and leveraging a statistical thresholding strategy. Once detected, the samples from unknown fault types are incorporated into the existing model through a continual learning procedure that selectively updates only the final layers of the network, which enables the model to recognize new fault types while preserving knowledge of existing classes. To accelerate the labeling process, cosine similarity transformation combined with a clustering algorithm groups similar unknown samples, thereby reducing manual labeling effort. Experimental results using a multi-sensor UMW dataset demonstrate that the proposed method achieves 96% accuracy in detecting unseen fault conditions while maintaining reliable classification of known classes. After incorporating a new fault type using only five labeled samples, the updated model achieves 98% testing classification accuracy. These results demonstrate that the proposed approach enables adaptive monitoring with minimal retraining cost and time. The proposed approach provides a scalable solution for continual learning in condition monitoring where new process conditions may constantly emerge over time and is extensible to other manufacturing processes.

Repeated machining passes (i.e., continuous toolpaths) are common in CNC manufacturing, including multi-level machining of prismatic parts and iso-contour passes in contour machining. They present an opportunity to exploit pass-to-pass learning to improve productivity without sacrificing quality through feedrate optimization. Traditional iterative learning methods provide a means to exploit pass-to-pass learning for quality improvements, but they are not well-suited to feedrate optimization because the reference trajectories change as the feedrate increases. In the authors’ prior work, learning-based feedrate optimization was demonstrated for repeated machining along identical toolpaths. This paper extends that concept to the more challenging case of similar but non-identical cutting paths, as encountered in contour machining. A pass-to-pass learning strategy is proposed in which corresponding sections of non-identical iso-contour passes are identified using a contour-matching method based on geometric similarity. Bayesian linear regression models are then used to learn and predict contour error and spindle torque across passes, with uncertainty explicitly quantified through credible intervals. These predictions are embedded in a window-based feedrate optimization framework solved via sequential linear programming, enabling feedrate maximization subject to kinematic, contour-error, and spindle-torque constraints. The proposed approach is experimentally validated on a 3-axis desktop CNC milling machine through multiple 2.5D contour machining case studies. Results show that the method can rapidly approach near-optimal feedrates after only a few passes, culminating in up to 16.4% increase in productivity compared to an equivalent learning-based feedrate optimization approach for identical toolpaths.

Two-photon lithography (TPL) is a sophisticated additive manufacturing technology for creating three-dimensional (3D) micro- and nano-structures. Maintaining the health of TPL systems is critical for ensuring consistent fabrication quality. Current maintenance practices often rely on experience rather than informed monitoring of machine health, resulting in either untimely maintenance that causes machine downtime and poor-quality fabrication, or unnecessary maintenance that leads to inefficiencies and avoidable downtime. To address this gap, this paper presents three methods for TPL machine health monitoring through analyzing the geometric changes in manufactured parts. Through integrating physics-guided data-driven predictive models for structure geometry with statistical approaches, the proposed methods are able to handle increasingly complex scenarios featuring different levels of generalizability. A comprehensive experimental dataset that encompasses six process parameter combinations and six structure dimensions under two machine health conditions was collected to evaluate the effectiveness of the proposed approaches. Across all test scenarios, the approaches are shown to achieve high accuracies, demonstrating excellent effectiveness, robustness, and generalizability. These results represent a significant step toward condition-based maintenance for TPL systems.

Two-photon lithography (TPL) is a sophisticated additive manufacturing technology for creating three-dimensional (3D) micro- and nano-structures. Maintaining the health of TPL systems is critical for ensuring consistent fabrication quality. Current maintenance practices often rely on experience rather than informed monitoring of machine health, resulting in either untimely maintenance that causes machine downtime and poor-quality fabrication, or unnecessary maintenance that leads to inefficiencies and avoidable downtime. To address this gap, this paper presents three methods for TPL machine health monitoring through analyzing the geometric changes in manufactured parts. Through integrating physics-guided data-driven predictive models for structure geometry with statistical approaches, the proposed methods are able to handle increasingly complex scenarios featuring different levels of generalizability. A comprehensive experimental dataset that encompasses six process parameter combinations and six structure dimensions under two machine health conditions was collected to evaluate the effectiveness of the proposed approaches. Across all test scenarios, the approaches are shown to achieve high accuracies, demonstrating excellent effectiveness, robustness, and generalizability. These results represent a significant step toward condition-based maintenance for TPL systems.

Surrogate modeling is an essential data-driven technique for quantifying relationships between input variables and system responses in manufacturing and engineering systems. Two major challenges limit its effectiveness: (1) large data requirements for learning complex nonlinear relationships, and (2) heterogeneous data collected from sources with varying fidelity levels. Multi-task learning (MTL) addresses the first challenge by enabling information sharing across related processes, while multi-fidelity modeling addresses the second by accounting for fidelity-dependent uncertainty. However, existing approaches typically address these challenges separately, and no unified framework simultaneously leverages inter-task similarity and fidelity-dependent data characteristics. This paper develops a novel hierarchical multi-task multi-fidelity (H-MT-MF) framework for Gaussian process-based surrogate modeling. The proposed framework decomposes each task's response into a task-specific global trend and a residual local variability component that is jointly learned across tasks using a hierarchical Bayesian formulation. The framework accommodates an arbitrary number of tasks, design points, and fidelity levels while providing predictive uncertainty quantification. We demonstrate the effectiveness of the proposed method using a 1D synthetic example and a real-world engine surface shape prediction case study. Compared to (1) a state-of-the-art MTL model that does not account for fidelity information and (2) a stochastic kriging model that learns tasks independently, the proposed approach improves prediction accuracy by up to 19% and 23%, respectively. The H-MT-MF framework provides a general and extensible solution for surrogate modeling in manufacturing systems characterized by heterogeneous data sources.

Surrogate modeling is an essential data-driven technique for quantifying relationships between input variables and system responses in manufacturing and engineering systems. Two major challenges limit its effectiveness: (1) large data requirements for learning complex nonlinear relationships, and (2) heterogeneous data collected from sources with varying fidelity levels. Multi-task learning (MTL) addresses the first challenge by enabling information sharing across related processes, while multi-fidelity modeling addresses the second by accounting for fidelity-dependent uncertainty. However, existing approaches typically address these challenges separately, and no unified framework simultaneously leverages inter-task similarity and fidelity-dependent data characteristics. This paper develops a novel hierarchical multi-task multi-fidelity (H-MT-MF) framework for Gaussian process-based surrogate modeling. The proposed framework decomposes each task's response into a task-specific global trend and a residual local variability component that is jointly learned across tasks using a hierarchical Bayesian formulation. The framework accommodates an arbitrary number of tasks, design points, and fidelity levels while providing predictive uncertainty quantification. We demonstrate the effectiveness of the proposed method using a 1D synthetic example and a real-world engine surface shape prediction case study. Compared to (1) a state-of-the-art MTL model that does not account for fidelity information and (2) a stochastic kriging model that learns tasks independently, the proposed approach improves prediction accuracy by up to 19% and 23%, respectively. The H-MT-MF framework provides a general and extensible solution for surrogate modeling in manufacturing systems characterized by heterogeneous data sources.

Ultrasonic metal welding (UMW) is a solid-state joining technology with widespread industrial applications. However, the weld quality in UMW is highly sensitive to process disturbances such as tool degradation and surface contamination. To address this challenge, this paper presents an integrated learning, monitoring, and control (LMC) system to improve process robustness and weld quality in UMW. The proposed system integrates in-situ sensing, online process monitoring, and within-cycle process adjustment to automatically compensate for process disturbances. Extensive experiments involving 700 welds with varied acting time, pressure adjustments, and contamination levels, are carried out to thoroughly evaluate the effectiveness of the LMC system. It is shown that the proposed method significantly and consistently outperforms the existing controller. Specifically, the weld success rate is increased from 0% to 92% under 20% surface contamination, and from 6% to 72% under 10% surface contamination. Furthermore, a response surface model is developed to quantify the causal relationships between control inputs (i.e., acting time and pressure increase amount) and the resulting weld success rate, which enables efficient optimization of control parameters. Overall, the proposed LMC approach improves the UMW process robustness and weld quality, demonstrating strong potential for industrial-scale implementation. To the best of our knowledge, this study represents one of the first integrated LMC systems developed for UMW.

Ultrasonic metal welding (UMW) is a solid-state joining process widely used in industrial applications. However, its sensitivity to tool wear, surface contamination, and material variability presents persistent challenges for ensuring weld quality. Existing online monitoring systems often emphasize predictive accuracy while neglecting practical constraints such as hardware cost, data acquisition rate, and computational latency. To overcome this gap, this paper develops a systematic framework for cost- and time-efficient sensor and feature selection in UMW monitoring. The proposed method integrates signal decomposition, feature importance analysis, cost-aware genetic algorithm optimization, and a separability-analysis-based adaptation mechanism to identify an optimal subset of sensors, features, and time segments that balance predictive accuracy with resource efficiency. Extensive case studies using a multi-sensor data acquisition system demonstrate that the framework achieves high monitoring accuracy in both weld quality prediction and mixed tool and sample surface condition classification while reducing the feature pool by 96.8%–99.4%. Even under reduced sampling frequency (6.25 kHz) and shortened time windows (0.3 s), the model maintains strong predictive performance. Furthermore, the separability-analysis-based adaptation accurately recognizes new fault types using only three samples, reducing retraining data requirements by 90%. Overall, the proposed framework provides a new, scalable solution for cost- and time-efficient UMW monitoring and establishes a foundation for adaptive, lightweight monitoring systems applicable to other manufacturing processes.

Two-photon lithography (TPL) is an advanced additive manufacturing (AM) technique for fabricating high-precision micro-structures. While computer vision (CV) is proofed for automated quality control, existing models are often static, rendering them ineffective in dynamic manufacturing environments. These models typically cannot detect new, unseen defect classes, be efficiently updated from scarce data, or adapt to new part geometries. To address this gap, this paper presents an adaptive CV framework for the entire life-cycle of quality model maintenance. The proposed framework is built upon a same, scale-robust backbone model and integrates three key methodologies: (1) a statistical hypothesis testing framework based on Linear Discriminant Analysis (LDA) for novelty detection, (2) a two-stage, rehearsal-based strategy for few-shot incremental learning, and (3) a few-shot Domain-Adversarial Neural Network (DANN) for few-shot domain adaptation. The framework was evaluated on a TPL dataset featuring hemisphere as source domain and cube as target domain structures, with each domain categorized into good, minor damaged, and damaged quality classes. The hypothesis testing method successfully identified new class batches with 99-100% accuracy. The incremental learning method integrated a new class to 92% accuracy using only K=20 samples. The domain adaptation model bridged the severe domain gap, achieving 96.19% accuracy on the target domain using only K=5 shots. These results demonstrate a robust and data-efficient solution for deploying and maintaining CV models in evolving production scenarios.

Two-photon lithography (TPL) is an advanced additive manufacturing (AM) technique for fabricating high-precision micro-structures. While computer vision (CV) is proofed for automated quality control, existing models are often static, rendering them ineffective in dynamic manufacturing environments. These models typically cannot detect new, unseen defect classes, be efficiently updated from scarce data, or adapt to new part geometries. To address this gap, this paper presents an adaptive CV framework for the entire life-cycle of quality model maintenance. The proposed framework is built upon a same, scale-robust backbone model and integrates three key methodologies: (1) a statistical hypothesis testing framework based on Linear Discriminant Analysis (LDA) for novelty detection, (2) a two-stage, rehearsal-based strategy for few-shot incremental learning, and (3) a few-shot Domain-Adversarial Neural Network (DANN) for few-shot domain adaptation. The framework was evaluated on a TPL dataset featuring hemisphere as source domain and cube as target domain structures, with each domain categorized into good, minor damaged, and damaged quality classes. The hypothesis testing method successfully identified new class batches with 99-100% accuracy. The incremental learning method integrated a new class to 92% accuracy using only K=20 samples. The domain adaptation model bridged the severe domain gap, achieving 96.19% accuracy on the target domain using only K=5 shots. These results demonstrate a robust and data-efficient solution for deploying and maintaining CV models in evolving production scenarios.

Near-field perception is essential for the safe operation of autonomous mobile robots (AMRs) in manufacturing environments. Conventional ranging sensors such as light detection and ranging (LiDAR) and ultrasonic devices provide broad situational awareness but often fail to detect small objects near the robot base. To address this limitation, this paper presents a three-tier near-field perception framework. The first approach employs light-discontinuity detection, which projects a laser stripe across the near-field zone and identifies interruptions in the stripe to perform fast, binary cutoff sensing for obstacle presence. The second approach utilizes light-displacement measurement to estimate object height by analyzing the geometric displacement of a projected stripe in the camera image, which provides quantitative obstacle height information with minimal computational overhead. The third approach employs a computer vision-based object detection model on embedded AI hardware to classify objects, enabling semantic perception and context-aware safety decisions. All methods are implemented on a Raspberry Pi 5 system, achieving real-time performance at 25 or 50 frames per second. Experimental evaluation and comparative analysis demonstrate that the proposed hierarchy balances precision, computation, and cost, thereby providing a scalable perception solution for enabling safe operations of AMRs in manufacturing environments.

Cyberphysical attacks on the digital backbone of Additive Manufacturing (AM) can compromise the printed part’s functionality. They can alter features in the digital geometry to introduce geometric defects (e.g., missing fillets) or alter process parameters to create local defects (e.g., voids). Addressing the downtime, waste, and quality deterioration associated with existing solutions requires operational resilience, i.e., rapid elimination or disruption of defect formation (to retain part function) without production stoppage or part disposal (to retain yield). This need is unmet due to the inherently unpredictable nature of attack-induced alterations, lack of access to the original geometric model for identification of altered geometric features, and in-process imposition of unknown process dynamics via attack-driven alteration of real-time-uncontrolled (or exogenous) parameters. This work establishes the above-mentioned operational resilience for the first time by creating two Digital Twins (DT). The Geometric DT (Geo-DT) is based on a unique physical-field-driven soft sensor and topology optimization method. The Process Digital Twin (Pro-DT) combines local defect quantification with a novel Reinforcement Learning formulation and training method. The importance of these methodological advances and the scalability of our approach are examined on a real AM testbed. It is shown that Geo-DT can correct geometric defects without access to the original digital geometry or explicit knowledge of attack-altered geometric features. Further, Pro-DT can accelerate real-time disruption of local defects despite attack-driven imposition of unknown process dynamics. We discuss how our framework goes beyond the contemporary focus on pre-attack security and in-attack detection towards resilience for AM and beyond.

Paper drying is responsible for over two-thirds of energy consumption in the U.S. pulp and paper industry, presenting significant potential for energy savings through optimization of process parameters. Current approaches often assume fixed operating conditions, neglecting dynamic ambient and process variations that limit achievable savings and real-world applicability. To this end, we develop a physics-based simulation environment for a paper machine dryer section and propose a reinforcement learning (RL) framework to minimize overall energy consumption by optimizing drying process parameters under diverse operating conditions. To mitigate overdrying and numerical instabilities caused by suboptimal local RL actions, we introduce Reinforcement Learning-Guided Beam Search (RLGBS), which explores multiple action sequences in parallel using beam search. Instead of making step-by-step decisions, RLGBS prioritizes solutions based on cumulative probability, reducing the impact of individual suboptimal actions. Experiments demonstrate that RLGBS achieves consistent energy savings under unseen operating conditions not encountered during training, outperforming conventional RL methods. While validated in drying optimization, this framework is broadly applicable to other RL-based industrial process control problems.

Abstract Corrosion prediction plays a vital role in material design and optimization but often depends on expensive high-fidelity simulations or experiments. Gaussian process (GP)-based surrogate models have been used to reduce computational costs. However, they require large amounts of high-fidelity training data for accurate predictions. This challenge becomes more significant when incorporating anode microstructure to enhance simulation accuracy, as it increases computational complexity. Multi-task learning (MTL) provides a promising approach to enhance prediction efficiency by leveraging data from multiple material systems with similar parameter configurations. This study presents a hybrid GP-based MTL framework to predict mass loss due to corrosion based on geometric and environmental parameters, including crevice gap, electrolyte conductivity, and temperature. The model is applied to four different material systems: (1) mild steel/AE44, (2) mild steel/AZ31B, (3) UNS 17400/AA7075-T6, and (4) mild steel/AA7075-T6, using finite element analysis (FEA) simulation data for training and testing. High-fidelity data is generated by incorporating anode microstructure, while low-fidelity data is obtained without it. Results demonstrate that the proposed MTL model outperforms single-task learning approaches under data-scarce conditions. Especially, prediction accuracy for high-fidelity cases improves as additional low-fidelity training data is incorporated. These findings highlight the potential of MTL to enhance the accuracy and efficiency of corrosion modeling in multi-material systems.

Fused Deposition Modeling (FDM) is a widely used additive manufacturing (AM) technique valued for its flexibility and cost-efficiency, with applications in a variety of industries including healthcare and aerospace. Recent technological developments have made affordable FDM machines accessible and encouraged adoption among diverse users. However, the design, planning, and production process in FDM require specialized interdisciplinary knowledge. Managing the complex parameters and resolving print defects in FDM remain challenging. These technical complexities form the most critical barrier preventing individuals without technical backgrounds and even professional engineers without training in other domains from participating in AM design and manufacturing. Large Language Models (LLMs), with their advanced capabilities in text and code processing, offer the potential for addressing these challenges in FDM. However, existing research on LLM applications in this field is limited, typically focusing on specific use cases without providing comprehensive evaluations across multiple models and tasks. To this end, we introduce FDM-Bench, a benchmark dataset designed to evaluate LLMs on FDM-specific tasks. FDM-Bench enables a thorough assessment by including user queries across various experience levels and G-code samples that represent a range of anomalies. We evaluate two closed-source models (GPT-4o and Claude 3.5 Sonnet) and two open-source models (Llama-3.1-70B and Llama-3.1-405B) on FDM-Bench. A panel of FDM experts assess the models’ responses to user queries in detail. Results indicate that closed-source models generally outperform open-source models in G-code anomaly detection, whereas Llama-3.1-405B demonstrates a slight advantage over other models in responding to user queries. These findings underscore FDM-Bench’s potential as a foundational tool for advancing research on LLM capabilities in FDM.

Fruit drying is widely used in food manufacturing to reduce product moisture, ensure product safety, and extend product shelf life. Accurately predicting final moisture content (MC) is critically needed for quality control of drying processes. State-of-the-art methods can build deterministic relationships between process parameters and MC, but cannot adequately account for inherent process variabilities that are ubiquitous in fruit drying. To address this gap, this paper presents a novel multi-modal data fusion framework to effectively fuse two modalities of data: tabular data (process parameters) and high-dimensional image data (images of dried apple slices) to enable accurate MC prediction. The proposed modeling architecture permits flexible adjustment of information portion from tabular and image data modalities. Experimental validation shows that the multi-modal approach improves predictive accuracy substantially compared to state-of-the-art methods. The proposed method reduces root-mean-squared errors by 19.3%, 24.2%, and 15.2% over tabular-only, image-only, and standard tabular-image fusion models, respectively. Furthermore, it is demonstrated that our method is robust in varied tabular-image ratios and capable of effectively capturing inherent small-scale process variabilities. The proposed framework is extensible to a variety of other drying technologies.

Two-photon lithography (TPL) is an advanced additive manufacturing technique capable of fabricating micro- and nano-scale three-dimensional structures with high precision. Ensuring structural accuracy and part quality is essential for guaranteeing the functionality of TPL-fabricated structures. Current quality control practice heavily relies on manual inspection, which is time-consuming and does not permit timely decision-making. To address this challenge, this paper develops a computer vision-based end-to-end quality classification framework for TPL-produced parts. The proposed framework consists of two main modules—image segmentation and quality classification. The segmentation module uses Gaussian blur preprocessing with Hough Transform to automatically identify and locate structures from a large image. The quality classification module uses convolutional neural network as a basis model architecture and provides novel semi-supervised and unsupervised learning capabilities, which are critically needed to address data scarcity challenges in scalable industrial scenarios. A rich experimental dataset comprising 1,598 individual structures, four quality conditions, and six structural design sizes is utilized to thoroughly evaluate the performance of the developed models. For multi-class classification tasks, the supervised and semi-supervised learning models achieve accuracies of 93.3% and 88.4%, respectively. When tested on a binary classification task distinguishing between ”good” and ”damaged” structures, the unsupervised learning model achieves an accuracy of 94.8%. These results demonstrate the performance and generalizability of the proposed framework across diverse scenarios.

Abstract Tool condition monitoring (TCM) is a critical maintenance task in industrial-scale ultrasonic metal welding (UMW). UMW tools, consisting of a horn and an anvil, experience geometric changes over time, which negatively affect the joint quality and introduce significant process variations. Conventional indirect TCM methods have achieved high accuracy in classification tasks; however, they cannot provide detailed insights into the geometric changes of tools. In contrast, using high-resolution 3D metrology, direct TCM methods can characterize and monitor tool surface degradation more comprehensively, but such measurements are usually expensive and time-consuming to acquire, which prevent their widespread use on the factory floor. To address these challenges, this article presents a novel, cost-effective optical imaging system that uses optical images of resin cast replicas for fine-scale characterization of tool surface degradation in UMW. Furthermore, computer vision (CV) algorithms are developed to process and analyze 2D optical images to support a series of maintenance decision-making tasks. Based on a U-Net architecture, a segmentation network identifies the tip regions most prone to wear, while a two-stage model further reconstructs 3D height maps, all from 2D optical images. Additionally, a convolutional neural network provides end-to-end predictions of aggregated knurl-level geometric features. Experimental results demonstrate that the CV models accurately capture geometric tool wear features at both knurl level and tool level from optical images, offering an efficient, effective, and interpretable alternative to high-resolution 3D measurements for fine-scale TCM in UMW.

Abstract In-field Additive Manufacturing (AM) is distinguished from in-factory AM via imposition of extraneous disturbances that induce unknown and unpredictable in-process variations in externalities, i.e., process parameters which are typically assumed as fixed during printing. Real-time mitigation of the resulting part defects is critical for retaining function and delivery of parts, both of which are critical for in-field AM. But the state-of-the-art either needs prior or real-time knowledge of the externality variation, which is often unavailable, or renders the part unusable due to infeasibly slow defect mitigation. This work addresses this issue via a novel Conditional Reinforcement Learning (ConRL) approach for real-time data-driven defect mitigation. ConRL is integrated with existing real-time defect detection methods to create a smart manufacturing framework and is validated for a Fused Filament Fabrication testbed. Orders of magnitude increase in defect mitigation speed within one control step is demonstrated. Further, it is possible to address untrained-for and unknown externality variations that nonlinearly affect the defect dynamics without offline or online retraining of the as-learnt policy. The importance of conditionality, defect quantification, and access to machine firmware, is discussed in light of the experimental observations.

Food drying is essential for food production, extending shelf life, and reducing transportation costs. Accurate real-time forecasting of drying readiness is crucial for minimizing energy consumption, improving productivity, and ensuring product quality. However, this remains challenging due to the dynamic nature of drying, limited data availability, and the lack of effective predictive analytical methods. To address this gap, we propose an end-to-end multi-modal data fusion framework that integrates in-situ video data with process parameters for real-time food drying readiness forecasting. Our approach leverages a new encoder-decoder architecture with modality-specific encoders and a transformer-based decoder to effectively extract features while preserving the unique structure of each modality. We apply our approach to sugar cookie drying, where time-to-ready is predicted at each timestamp. Experimental results demonstrate that our model achieves an average prediction error of only 15 seconds, outperforming state-of-the-art data fusion methods by 65.69% and a video-only model by 11.30%. Additionally, our model balances prediction accuracy, model size, and computational efficiency, making it well-suited for heterogenous industrial datasets. The proposed model is extensible to various other industrial modality fusion tasks for online decision-making.

In-field Additive Manufacturing (AM) is exposed to irregular variations in process conditions (externalities) that affect defect dynamics. These externalities are invariable in conventional in-factory AM. Stoppage-free and real-time mitigation of part defects induced by these externality variations is necessary for timely delivery of quality parts in in-field AM. But existing solutions either require explicit knowledge of externality variations which is typically unavailable or they render the part unusable due to infeasibly slow defect mitigation. This work addresses this issue by establishing a novel Conditional Reinforcement Learning (ConRL) approach for rapid and real-time data-driven defect mitigation based on an implicit consideration of externality variations. Validation within a smart manufacturing pipeline on a Fused Filament Fabrication testbed reveals the unprecedented ability to mitigate defects at 10× greater speed via a single control action and within the same printed line. A hitherto unreported degree of scalability is observed, i.e., it is possible to mitigate defects induced by untrained-for, unknown and unmeasured externality variations without any retraining of the policy. The results also reveal new insight into the significance of conditionality in ConRL and of real-time defect quantification. The implications for wider adoption of ConRL to other in-field AM processes is discussed. • Addressed real-time defect mitigation under extraneous disturbances in in-field AM • Created new Conditional Reinforcement Learning with implicit disturbance inclusion • Enhanced speed of real-time defect mitigation by orders of magnitude • Mitigation is generalizable without knowledge of disturbances or retraining. • Broke knowledge-speed barrier in current solutions for such defect mitigation

The rise in popularity of two-photon polymerization (TPP) as an additive manufacturing technique has impacted many areas of science and engineering, particularly those related to biomedical applications. Compared with other fabrication methods used for biomedical applications, TPP offers 3D, nanometer-scale fabrication dexterity (free-form). Moreover, the existence of turnkey commercial systems has increased accessibility. In this review, we discuss the diversity of biomedical applications that have benefited from the unique features of TPP. We also present the state of the art in approaches for patterning and reading 3D TPP-fabricated structures. The reading process influences the fidelity for both in situ and ex situ characterization methods. We also review efforts to leverage machine learning to facilitate process control for TPP. Finally, we conclude with a discussion of both the current challenges and exciting opportunities for biomedical applications that lie ahead for this intriguing and emerging technology.

Existing approaches to enforcing design constraints in Reinforcement Learning (RL) applications often rely on training-time penalties in the reward function or training/inference-time invalid action masking, but these methods either cannot be modified after training, or are limited in the types of constraints that can be implemented. To address this limitation, we propose Reinforcement Learning Constrained Beam Search (RLCBS) for inference-time refinement in combinatorial optimization problems. This method respects flexible, inference-time constraints that support exclusion of invalid actions and forced inclusion of desired actions, and employs beam search to maximize sequence probability for more sensible constraint incorporation. RLCBS is extensible to RL-based planning and optimization problems that do not require real-time solution, and we apply the method to optimize process parameters for a novel modular testbed for paper drying. An RL agent is trained to minimize energy consumption across varying machine speed levels by generating optimal dryer module and air supply temperature configurations. Our results demonstrate that RLCBS outperforms NSGA-II under complex design constraints on drying module configurations at inference-time, while providing a 2.58-fold or higher speed improvement.

Fused Deposition Modeling (FDM) is a widely used additive manufacturing (AM) technique valued for its flexibility and cost-efficiency, with applications in a variety of industries including healthcare and aerospace. Recent developments have made affordable FDM machines accessible and encouraged adoption among diverse users. However, the design, planning, and production process in FDM require specialized interdisciplinary knowledge. Managing the complex parameters and resolving print defects in FDM remain challenging. These technical complexities form the most critical barrier preventing individuals without technical backgrounds and even professional engineers without training in other domains from participating in AM design and manufacturing. Large Language Models (LLMs), with their advanced capabilities in text and code processing, offer the potential for addressing these challenges in FDM. However, existing research on LLM applications in this field is limited, typically focusing on specific use cases without providing comprehensive evaluations across multiple models and tasks. To this end, we introduce FDM-Bench, a benchmark dataset designed to evaluate LLMs on FDM-specific tasks. FDM-Bench enables a thorough assessment by including user queries across various experience levels and G-code samples that represent a range of anomalies. We evaluate two closed-source models (GPT-4o and Claude 3.5 Sonnet) and two open-source models (Llama-3.1-70B and Llama-3.1-405B) on FDM-Bench. A panel of FDM experts assess the models' responses to user queries in detail. Results indicate that closed-source models generally outperform open-source models in G-code anomaly detection, whereas Llama-3.1-405B demonstrates a slight advantage over other models in responding to user queries. These findings underscore FDM-Bench's potential as a foundational tool for advancing research on LLM capabilities in FDM.

Ultrasonic metal welding (UMW) is a key joining technology with widespread industrial applications. Condition monitoring (CM) capabilities are critically needed in UMW applications because process anomalies significantly deteriorate the joining quality. Recently, machine learning models emerged as a promising tool for CM in many manufacturing applications due to their ability to learn complex patterns. Yet, the successful deployment of these models requires substantial training data that may be expensive and time-consuming to collect. Additionally, many existing machine learning models lack generalizability and cannot be directly applied to new process configurations (i.e., domains). Such issues may be potentially alleviated by pooling data across manufacturers, but data sharing raises critical data privacy concerns. To address these challenges, this paper presents a Federated Transfer Learning with Task Personalization (FTL-TP) framework that provides domain generalization capabilities in distributed learning while ensuring data privacy. By effectively learning a unified representation from feature space, FTL-TP can adapt CM models for clients working on similar tasks, thereby enhancing their overall adaptability and performance jointly. To demonstrate the effectiveness of FTL-TP, we investigate two distinct UMW CM tasks, tool condition monitoring and workpiece surface condition classification. Compared with state-of-the-art FL algorithms, FTL-TP achieves a 5.35%--8.08% improvement of accuracy in CM in new target domains. FTL-TP is also shown to perform excellently in challenging scenarios involving unbalanced data distributions and limited client fractions. Furthermore, by implementing the FTL-TP method on an edge-cloud architecture, we show that this method is both viable and efficient in practice. The FTL-TP framework is readily extensible to various other manufacturing applications.

Additive manufacturing (AM) has the unique capability to produce parts with complex three-dimensional structures with features that are internal to the part and hidden from view. Such internal features are difficult to inspect and may have defects that affect the function of a part. X-ray computed tomography (CT) is one of the only methods for nondestructive inspection (NDI) of the interior of AM parts. This paper explores how machine learning (ML) methods can be used to analyze CT scans of AM parts to automatically identify the presence of defects in geometric features. We designed a nozzle part with internal three-dimensional (3D) channels and introduced five different synthetic defects whose presence could affect the nozzle performance. A resin-based AM process fabricated 155 nozzle parts that included both defect-free parts and parts with synthetic defects. CT scans were collected for each part and processed into 68,510 image cross sections. The extracted images were used to train an ML model based on the ResNet34 architecture. The model can automatically identify and classify defects in individual CT slice images with over 98% accuracy. High model accuracy is possible with training on as few as 30 parts. The research demonstrates the potential of ML methods to automatically identify hidden defects and qualify AM parts using X-ray CT scans.

Geometric integrity directly impacts the functionality, reliability, and safety of final manufactured products, making the qualification of parts based on measurements of their geometry a fundamental quality control activity in modern manufacturing. Recent advancements in three-dimensional (3D) metrology technologies have enabled fine-scale inspection of geometric integrity characterized by dimensional accuracy, surface quality, and shape conformity. However, the widespread adoption of high-resolution 3D metrology in manufacturing faces some significant challenges posed by the inherent data structures of 3D point clouds such as high dimensionality, unstructured nature, and sparsity in defective regions. To address these challenges, this paper first creates a “MFGNet-gear” dataset, which is a scalable and comprehensive benchmark dataset comprising 12 part designs with four quality classes for each design. Subsequently, we develop a deep learning model adapted from the PointNet++ architecture to enable automated, end-to-end analysis of 3D point clouds. The model can be configured for different decision-making tasks including part design classification and multi-class geometric defect detection. Implementations of the proposed model on the “MFGNet-gear” dataset achieve accuracies up to 100% in classifying gear designs and up to 85% in four-class quality inspection. Additionally, we systematically investigate the impacts of measurement resolution and precision on the classification performance through a series of case studies. The obtained results highlight the potential of using deep learning methods for automated analysis of 3D point clouds for a variety of quality control tasks beyond gear manufacturing. This study also proposes future research directions, including the development of new deep learning architectures specifically designed for manufacturing 3D point clouds and strategies for adaptive measurement planning.

Online tool condition monitoring (TCM) is a pivotal capability in many manufacturing applications including ultrasonic metal welding (UMW). Effective and efficient TCM can facilitate predictive maintenance, improve product quality, and enhance productivity. Existing online TCM systems based on conventional machine learning models often require a large amount of labeled data, the collection of which is cost-prohibitive, time-consuming, and labor-intensive. Such models fail to satisfy the requirements of cost-effectiveness and agility posed by modern, reconfigurable UMW systems. As such, data-efficient TCM methods with excellent generalization ability are of vital importance. To this end, we develop a novel similarity-based meta-representation learning (SMRL) method for domain generalization. SMRL effectively learns high-level meta-knowledge that is shared among different welding scenarios or domains. Therefore, the model trained in source domains can be generalized to other domains without access to labeled data in the training phase. Case studies are performed using four welding scenarios with varied welding materials and welding parameters. It is demonstrated that the proposed method is superior to the baseline methods, including neural network, hierarchical neural network, model-agnostic meta-learning (MAML), and hierarchical MAML. Compared with the baseline methods, SMRL offers an average improvement of 15.44%–31.62% in TCM accuracy. These results show that SMRL is readily applicable to industrial applications to enable cost-effective and data-efficient TCM.

A crucial part of quality control in additive manufacturing (AM) is the decision to accept or reject parts based on their dimensional accuracy. Machine learning (ML) models can learn nuanced relationships between process parameters and resulting geometries; however, obtaining large quantities of supervised learning data to train ML models incurs non-trivial costs. Relevant measurement data may be available at other factories or stations within the same factory but cannot be pooled together for conventional centralized learning (CL) due to privacy constraints. Here, we propose federated learning (FL) that uses private data and information from distributed sources to train ML models that predict part dimensions and inform part qualification, thereby allowing collaborative model building without compromising privacy. We manufacture and measure 405 parts having three different designs and distribute them across three to 45 factories. The performance of FL is evaluated when factories produce similar parts, parts of different designs, and parts of different qualities. When factories produce similar parts or parts of different designs, FL predicts part geometry within 10 µm of the CL benchmark and within 15 µm of the process capability limit, with as few as nine parts at each factory. FL also outperforms individual learning where factories train on their own data by up to 96% in geometry prediction, thereby providing a win-win paradigm for privacy-preserving collaborative learning. When factories produce parts of different qualities, FL outperforms individual factories only when they produce less than 15 parts locally. We show that FL performance is influenced by two factors: the quantity of data in the federation and the statistical homogeneity of data across participating factories. Overall, this research demonstrates the promise of FL for privacy-preserving, data-efficient quality control and offers key insights to design data federations in scalable AM production.