In the age of digital manufacturing, efficient realization of high-performing designs into high-quality products through advanced manufacturing remains a challenge. One of the critical reasons for this challenge is the need for more design knowledge at the manufacturers' end, which severely hinders manufacturers' exploration and exploitation of innovative designs. Manufacturers often have no choice but to adopt designer-optimized designs without the ability to make enhancements. However, the advent of data-driven reverse engineering based on generative modeling has shown promise for addressing this issue. As industries increasingly adopt digital manufacturing strategies, integrating AI-assisted design generation and optimization tools could enable efficient production outcomes. By leveraging large datasets and powerful machine learning techniques, manufacturers can now explore the vast design space in an efficient, automated manner, enabling them to drive innovation by uncovering novel design opportunities. The expanded design knowledge gained through generative modeling allows manufacturers to move beyond just implementing others' designs and truly innovate themselves to create higher-quality and more optimized products. Nevertheless, while generative models have been widely explored for two-dimensional (2D) design and optimization, generative modeling of three-dimensional (3D) shapes still remains under-explored. Although a variety of 3D shape generative models have been developed, the resolutions and surface qualities of the generated designs can hardly meet manufacturing requirements. To bridge this gap, this paper proposes a generative denoising diffusion model (DDM) to generate new voxel-represented 3D designs, where the model is trained with historical 3D design data available to manufacturers. This research is important, not only because manufacturers' design knowledge can be improved, but also because design exploration can be expanded and expedited when more feasible designs can be generated faster to pursue designs with improved manufacturability, higher feasibility, and optimized performance. DDMs have been extensively explored to synthesize 2D pixel images, and also have achieved state-of-the-art generation quality. Motivated by the capabilities of DDMs in generating high-quality images, we adapt the application of DDMs from 2D space to 3D space and develop a voxel-DDM for high-quality 3D topology generation. Moreover, we develop a surrogate model for efficient and gradient-based evaluation of the manufacturability of 3D designs via additive manufacturing. We further integrate the surrogate model as a guidance module into the voxel-DDM to steer the denoising process toward generating new designs with optimized manufacturability. To showcase the proposed guided voxel-DDM and validate its effectiveness, we implement it through a real-world manufacturing case study to design a Microbial Fuel Cell (MFC) anode structure that enables an efficient, reliable, and high-quality manufacturing process. This research can help manufacturers overcome their lack of design knowledge in design space exploration and lead to high-quality designs for the manufacturing process in pursuit of optimized manufacturability.

Abstract The use of local natural and recycled feedstock is promising for sustainable construction. However, unlike versatile engineered bricks, natural and recycled feedstock involves design challenges due to their stochastic, sequential, and heterogeneous nature. For example, the practical use of stone masonry is limited, as it still relies on human experts with holistic domain knowledge to determine the sequential organization of natural stones with different sizes/shapes. Reinforcement learning (RL) is expected to address such design challenges, as it allows artificial intelligence (AI) agents to autonomously learn design policy, that is, identifying the best design decision at each time step. As a proof-of-concept RL framework for design automation involving heterogeneous feedstock, a stone masonry design framework is presented. The proposed framework is founded upon a virtual design environment, MasonTris, inspired by the analogy between stone masonry and Tetris. MasonTris provides a Tetris-like virtual environment combined with a finite element analysis (FEA), where AI agents learn effective design policies without human intervention. Also, a new data collection policy, almost-greedy policy, is designed to address the sparsity of feasible designs for faster/stable learning. As computation bottleneck occurs when parallel agents evaluate designs with different complexities, a modification of the RL framework is proposed that FEA is held until training data are retrieved for training. The feasibility and adaptability of the proposed framework are demonstrated by continuously improving stone masonry design policy in simplified design problems. The framework can be generalizable to different natural and recycled feedstock by incorporating more realistic assumptions, opening opportunities in design automation for sustainability.

Both artificial intelligence (AI) and domain knowledge from human experts play an important role in manufacturing decision making. Smart manufacturing emphasizes a fully automated data-driven decision-making; however, the AI incubation process involves human experts to enhance AI systems by integrating domain knowledge for modeling, data collection and annotation, and feature extraction. Such an AI incubation process not only enhances the domain knowledge discovery but also improves the interpretability and trustworthiness of AI methods. In this article, we focus on the knowledge transfer from human experts to a supervised learning problem by learning domain knowledge as interpretable features and rules, which can be used to construct rule-based systems to support manufacturing decision making, such as process modeling and quality inspection. Although many advanced statistical and machine learning methods have shown promising modeling accuracy and efficiency, rule-based systems are still highly preferred and widely adopted due to their interpretability for human experts to comprehend. However, most of the existing rule-based systems are constructed based on deterministic human-crafted rules, whose parameters, such as thresholds of decision rules, are suboptimal. Yet the machine learning methods, such as tree models or neural networks, can learn a decision rule based structure without much interpretation or agreement with domain knowledge. Therefore, the traditional machine learning models and human experts’ domain knowledge cannot be directly improved by learning from data. In this research, we propose an interpretable neural network (INN) model with a center-adjustable sigmoid activation function to efficiently optimize the rule-based systems. Using the rule-based system from domain knowledge to regulate the INN architecture not only improves the prediction accuracy with optimized parameters but also ensures the interpretability by adopting the interpretable rule-based systems from domain knowledge. The proposed INN will be effective for supervised learning problems when rule-based systems are available. The merits of the INN model are demonstrated via a simulation study and a real case study in the quality modeling of a semiconductor manufacturing process. The source code of this work is hosted here: https://github.com/XiaoyuChenUofL/Interpretable-Neural-Network .

A portable seawater desalination system would be highly desirable to solve water challenges in rural areas and disaster situations. While many reverse osmosis-based portable desalination systems are already available commercially, they are not adequate for providing reliable drinking water in remote locations due to the requirement of high-pressure pumping and repeated maintenance. We demonstrate a field-deployable desalination system with multistage electromembrane processes, composed of two-stage ion concentration polarization and one-stage electrodialysis, to convert brackish water and seawater to drinkable water. A data-driven predictive model is used to optimize the multistage configuration, and the model predictions show good agreement with the experimental results. The portable system desalinates brackish water and seawater (2.5-45 g/L) into drinkable water (defined by WHO guideline), with the energy consumptions of 0.4-4 (brackish water) and 15.6-26.6 W h/L (seawater), respectively. In addition, the process can also reduce suspended solids by at least a factor of 10 from the source water, resulting in crystal clear water (<1 NTU) even from the source water with turbidity higher than 30 NTU (i.e., cloudy seawater by the tide). We built a fully integrated prototype (controller, pumps, and battery) packaged into a portable unit (42 × 33.5 × 19 cm3, 9.25 kg, and 0.33 L/h production rate) controlled by a smartphone, tested for battery-powered field operation. The demonstrated portable desalination system is unprecedented in size, efficiency, and operational flexibility. Therefore, it could address unique water challenges in remote, resource-limited regions of the world.

Semantics-based approaches – founded on the idea of explicitly encoding meaning separately from the data or the application code – are being applied to manufacturing, for example, to enable early manufacturability feedback. These approaches rely on formal, i.e., computer-interpretable, knowledge and rules along with the context or semantics, which facilitates the reuse and sharing of the knowledge via semantic web technologies. On the other hand, manufacturing knowledge has been maintained primarily in the form of unstructured English text. It is considered impractical for engineers to author accurate, formal, and structured manufacturing rules. However, previous efforts on extracting semantics from unstructured text in manufacturing have mainly focused on basic concept names and hierarchies for ontology creation, rather than extracting complex manufacturing rules. In this context, this paper focuses on the development of a semantics-based framework for acquiring formal manufacturing rules from English text, such as those written in manufacturing handbooks, by guiding standard Natural Language Processing (NLP) techniques with formal manufacturing knowledge (i.e., controlled vocabulary and domain ontology). Specifically, this paper studies the problem of rule extraction in the manufacturing domain, proposes the formal rule extraction framework, and demonstrates its feasibility. From the dataset of 133 sentences with a manufacturing rule, the proposed framework was able to extract correct rules from approximately 57% of the sentences. This paper also demonstrates the extensibility of the framework. Specifically, the framework was initially developed using the three sections of a manufacturing handbook, including milling, metal stamping, and die-casting sections, and could be successfully applied to the rest of the book after just updating the formal manufacturing knowledge to cover the other sections. This paper provides meaningful results in terms of formalization, thus will contribute to the development, sharing, and reuse of formal manufacturing knowledge that includes complex manufacturing rules.

In Industry 4.0, smart manufacturing is facing its next stage, cybermanufacturing, founded upon advanced communication, computation, and control infrastructure. Cybermanufacturing will unleash the potential of multi-modal manufacturing data, and provide a new perspective called computation service, as a part of service-oriented architecture (SOA), where on-demand computation requests throughout manufacturing operations are seamlessly satisfied by data analytics and machine learning. However, the complexity of information technology infrastructure leads to fundamental challenges in modeling and analysis under cybermanufacturing, ranging from information-poor datasets to a lack of reproducibility of analytical studies. Nevertheless, existing reviews have focused on the overall architecture of cybermanufacturing/SOA or its technical components (e.g., communication protocol), rather than the potential bottleneck of computation service with respect to modeling and analysis. In this paper, we review the fundamental challenges with respect to modeling and analysis in cybermanufacturing. Then, we introduce the existing efforts in computation pipeline recommendation, which aims at identifying an optimal sequence of method options for data analytics/machine learning without time-consuming trial-and-error. We envision computation pipeline recommendation as a promising research field to address the fundamental challenges in cybermanufacturing. We also expect that computation pipeline recommendation can be a driving force to flexible and resilient manufacturing operations in the post-COVID-19 industry.

Additive manufacturing (AM) is considered as a key to personalized product realization as it provides great design flexibility. As the flexibility radically expands the design space, current design space exploration methods for personalized geometric designs become time-consuming due to the use of physically based computer simulations (e.g., finite element analysis or computational fluid dynamics). This poses a significant challenge in design for an efficient personalized product realization cycle, which imposes a tight computation cost constraint to timely respond to every new requirement. To address the challenge, we propose a cost-efficient data-driven design space exploration method for personalized geometric design in AM, enabling feasible design regions under the computation constraint. Specifically, the proposed method adopts surrogate modeling of efficient voxel model-based design rules to identify feasible design regions considering both manufacturability and personalized needs. Since design rules take much less time for evaluation than physically based simulations, the proposed method can contribute to timely providing feasible design regions for an efficient personalized product realization cycle. Moreover, we develop a cost-based experimental design for surrogate modeling, which enables the evaluation of additional design points to provide more precise feasible design regions under the computation cost constraint. The merits of the proposed method are elaborated via additively manufactured microbial fuel cell (MFC) anode design.

Finite element analysis (FEA) has been widely adopted to identify potential defects in additive manufacturing (AM) processes. For personalized product realization, it is necessary to validate a number of heterogeneous product and process designs before or during manufacturing by using FEA. Multi-fidelity FEA simulations can be readily implemented with different capabilities in terms of simulation accuracy. However, due to its complexity, high-fidelity FEA simulation is time-consuming and decreases the efficiency of product realization in AM, while low-fidelity FEA simulation has fast computation speed yet limited capability. Hence, our objective is to improve the capability of FEA by providing an efficient data-driven model. In this research, a Gaussian process-constrained general path model is proposed to approximate the high-fidelity FEA simulation results based on low-fidelity results voxel-by-voxel. The proposed model quantifies the heterogeneous discrepancies between low- and high-fidelity FEA simulation results by incorporating the product design information (e.g., Cartesian coordinates of deposition sequence) and process design information from inputs of FEA simulation (e.g., input heat). Therefore, it enables the validation of new product and process designs based on the simulation results with the desired capability in a timely manner. The advantages of the proposed method are illustrated by FEA simulations of the fused deposition modeling (FDM) process with two levels of fidelity (i.e., low- and high-fidelity).

The value of (semi-)automatically recognizing the meaningful regions on a Computer Aided Design (CAD) model is well documented and has been traditionally studied as Geometric feature recognition. However, the effective tagging of the regions, especially tagging with contextually meaningful (i.e. semantic) keywords, has been elusive. In this paper, we focus on the non-geometric aspects of the tagging problem and present a semantic tagging framework. Specifically, given the contextually augmented instances created from the unidentified regions on a CAD model, the framework tags the instances with semantic keywords defined in the semantic model. This will provide the direct link between a CAD model and semantic models, and thus allow high-level reasoning to support design and manufacturing tasks. The feasibility of the approach has been verified by applying the developed framework to tag the regions on turbine blade models. This paper concludes with the future challenges of the approach.

Manufacturing knowledge is maintained primarily in the unstructured text in industry. To facilitate the reuse of the knowledge, previous efforts have utilized Natural Language Processing (NLP) to classify manufacturing documents or to extract structured knowledge (e.g. ontology) from manufacturing text. On the other hand, extracting more complex knowledge, such as manufacturing rule, has not been feasible in a practical scenario, as standard NLP techniques cannot address the input text that needs validation. Specifically, if the input text contains the information irrelevant to the rule-definition or semantically invalid expression, standard NLP techniques cannot selectively derive precise information for the extraction of the desired formal manufacturing rule. To address the gap, we developed the feedback generation method based on Constraint-based Modeling (CBM) coupled with NLP and domain ontology, designed to support formal manufacturing rule extraction. Specifically, the developed method identifies the necessity of input text validation based on the predefined constraints and provides the relevant feedback to help the user modify the input text, so that the desired rule can be extracted. We proved the feasibility of the method by extending the previously implemented formal rule extraction framework. The effectiveness of the method is demonstrated by enabling the extraction of correct manufacturing rules from all the cases that need input text validation, about 30% of the dataset, after modifying the input text based on the feedback. We expect the feedback generation method will contribute to the adoption of semantics-based technology in the manufacturing field, by facilitating precise knowledge acquisition from manufacturing-related documents in a practical scenario.

Manufacturing companies maintain manufacturing knowledge primarily as unstructured text. To facilitate formal use of such knowledge, previous efforts have utilized natural language processing (NLP) to classify manufacturing documents or extract manufacturing concepts/relations. However, extracting more complex knowledge, such as manufacturing rules, has been evasive due to the lack of methods to resolve ambiguities. Specifically, standard NLP techniques do not address domain-specific ambiguities that are due to manufacturing-specific meanings implicit in the text. To address this important gap, we propose an ambiguity resolution method that utilizes domain ontology as the mechanism to incorporate the domain context. We demonstrate its feasibility by extending our previously implemented manufacturing rule extraction framework. The effectiveness of the method is demonstrated by resolving all the domain-specific ambiguities in the dataset and an improvement in correct detection of rules to 70% (increased by about 13%). We expect that this work will contribute to the adoption of semantics-based technology in manufacturing field, by enabling the extraction of precise formal knowledge from text.

Formal ontology and rule-based approaches founded on semantic technologies have been proposed as powerful mechanisms to enable early manufacturability feedback. A fundamental unresolved problem in this context is that all manufacturing knowledge is encoded in unstructured text and there are no reliable methods to automatically convert it to formal ontologies and rules. It is impractical for engineers to write accurate domain rules in a structured semantic languages such as Web Ontology Language (OWL) or Semantic Application Design Language (SADL). Previous efforts in manufacturing research that have targeted extraction of OWL ontologies from text have focused on basic concept names and hierarchies. This paper presents a semantics-based framework for acquiring more complex manufacturing knowledge, primarily rules, in a semantically-usable form from unstructured English text such as those written in manufacturing handbooks. The approach starts with existing domain knowledge in the form of OWL ontologies and applies natural language processing techniques to extract dependencies between different words in the text that contains the rule. Domain specific triples capturing each rule are then extracted from each dependency graph. Finally, new computer-interpretable rules are composed from the triples. The feasibility of the framework has been evaluated by automatically and accurately generating rules for manufacturability from a manufacturing handbook. The paper also documents the cases that result in ambiguous results. Analysis of the results shows that the proposed framework can be extended to extract domain ontologies which forms part of the ongoing work that also focuses on addressing challenges to automate different steps and improve the reliability of the system.