A review of physics-informed and data-driven approaches for manufacturing process optimization in polymer matrix composites

As one of the most influential industries in public health and the global economy, the pharmaceutical industry is facing multiple challenges in drug research, development and manufacturing. With recent developments in artificial intelligence and machine learning, data-driven modeling methods and techniques have enabled fast and accurate modeling for drug molecular design, retrosynthetic analysis, chemical reaction outcome prediction, manufacturing process optimization, and many other aspects in the pharmaceutical industry. This article provides a review of data-driven methods applied in pharmaceutical processes, based on the mathematical and algorithmic principles behind the modeling methods. Different statistical tools, such as multivariate tools, Bayesian inferences, and machine learning approaches, i.e., unsupervised learning, supervised learning (including deep learning) and reinforcement learning, are presented. Various applications in the pharmaceutical processes, as well as the connections from statistics and machine learning methods, are discussed in the narrative procedures of introducing different types of data-driven models. Afterwards, two case studies, including dynamic reaction data modeling and catalyst-kinetics prediction of cross-coupling reactions, are presented to illustrate the power and advantages of different data-driven models. We also discussed current challenges and future perspectives of data-driven modeling methods, emphasizing the integration of data-driven and mechanistic models, as well as multi-scale modeling.

Indirect measurement of high grid strip densities over Nyquist sampling rate based on the moiré pattern analysis for quality assurance in grid manufacturing

Topography becomes more important in quality assurance for manufacturing. This regards the characterization of macro-sized to medium optics and the optimization of a manufacturing process using a coherence scanning interferometer, e.g., the measurement of touch points that directly relate to the polishing process and how to better optimize the manufacturing. Profiles or simple flatness measurements are not sufficient and areal measurements by line profiler or microscope-based systems are time consuming. Quality assurance in manufacturing is cost sensitive, especially, when a 100% control and optimization is needed. A macro lens optical metrology tool can overcome these disadvantages. We investigated the use of a large area coherence scanning interferometer for manufacturing, including measurements. Depending on the tolerances of parameter and the given cycle-time, solutions are presented on how to provide important measurement feedback to the manufacturing process.

One of significant challenges in the metallic additive manufacturing (AM) is the presence of many sources of uncertainty that leads to variability in microstructure and properties of AM parts. Consequently, it is extremely challenging to repeat the manufacturing of a high-quality product in mass production. A trial-and-error approach usually needs to be employed to attain a product with high quality. To achieve a comprehensive uncertainty quantification (UQ) study of AM processes, we present a physics-informed data-driven modeling framework, in which multi-level data-driven surrogate models are constructed based on extensive computational data obtained by multi-scale multi-physical AM models. It starts with computationally inexpensive metamodels, followed by experimental calibration of as-built metamodels and then efficient UQ analysis of AM process. For illustration purpose, this study specifically uses the thermal level of AM process as an example, by choosing the temperature field and melt pool as quantity of interest. We have clearly showed the surrogate modeling in the presence of high-dimensional response (e.g. temperature field) during AM process, and illustrated the parameter calibration and model correction of an as-built surrogate model for reliable uncertainty quantification. The experimental calibration especially takes advantage of the high-quality AM benchmark data from National Institute of Standards and Technology (NIST). This study demonstrates the potential of the proposed data-driven UQ framework for efficiently investigating uncertainty propagation from process parameters to material microstructures, and then to macro-level mechanical properties through a combination of advanced AM multi-physics simulations, data-driven surrogate modeling and experimental calibration.

Integrated data-driven modeling and experimental optimization of granular hydrogel matrices

Data-driven modeling of process, structure and property in additive manufacturing: A review and future directions

Nowadays, there are more ships equipped with on-board monitoring systems and vessels with many sensors are becoming a standard. Available measurements can be employed for system optimization which play an important role in the vessel power plant configuration. The improvements in the power system design can be based on theoretical (modeling based on physical and empirical laws — for simulation purposes, for example [1]) or data-driven modeling (machine learning, statistical approach [2]). The data-driven models can be supportive confirming theoretical assumptions or simplifications from simulations. They are also helpful to understand the real systems, including vessel dynamic behavior and interactions. Therefore, the combination of simulation and data-driven modeling will be beneficial by identifying relationships that help explain unidentified variations. This approach is recommended when aiming for a more reliable tool for design optimization and to overcome the limitation of the simulation models that all system properties and dynamic effects must be known beforehand. The scope in this work is to present a potential synergy between the simulation and the machine learning approach. A data-driven method can be complementary to a model based on physical and empirical laws. This is shown in the example of power plant model connected with a thruster and vessel model to simulate the typical transit scenario and the data-driven model. The paper proposes a simultaneous analysis of the theoretical and machine learning models to predict the vessel power/speed and study the complex systems’ interactions in more detail, which are essential while exploring the system behavior.

The RTM/LRI‐like processes have gained popularity in the preparation of fibre‐reinforced polymer matrix composites because of their high efficiency, low pollution and good reproducibility. With the industry willing to produce more and more complex shape parts with a high fiber volume fraction, void formation is still an open problem in term of prediction. Furthermore, simulation of infusion processes with high permeability draining mediums and very low permeability fabrics is an issue. In this paper, we propose to apply the Constrained Natural Element Method to the resolution of both Stokes and Brinkman equations with flow front progression.

This paper reviews the state of the art in applying uncertainty quantification (UQ) methods to additive manufacturing (AM). Physics-based as well as data-driven models are increasingly being developed and refined in order to support process optimization and control objectives in AM, in particular to maximize the quality and minimize the variability of the AM product. However, before using these models for decision-making, a fundamental question that needs to be answered is to what degree the models can be trusted, and consider the various uncertainty sources that affect their prediction. UQ in AM is not trivial because of the complex multiphysics, multiscale phenomena in the AM process. This article reviews the literature on UQ methodologies focusing on model uncertainty, discusses the corresponding activities of calibration, verification, and validation, and examines their applications reported in the AM literature. The extension of current UQ methodologies to additive manufacturing needs to address multiphysics, multiscale interactions, increasing presence of data-driven models, high cost of manufacturing, and complexity of measurements. The activities that need to be undertaken in order to implement verification, calibration, and validation for AM are discussed. Literature on using the results of UQ activities toward AM process optimization and control (thus supporting maximization of quality and minimization of variability) is also reviewed. Future research needs both in terms of UQ and decision-making in AM are outlined.

Design of hybrid PEEK composite with improved tribo-mechanical properties for biomedical applications – A machine learning approach

Resin film infusion (RFI) has been found to be a cost-effective technique for the fabrication of complex shaped composite parts for primary structural applications. Dry textile preforms are infiltrated, consolidated, and cured in a single step, eliminating the labor to lay-up prepreg tape. The large number of processing variables and the complex material behavior during infiltration and cure make experimental optimization of the RFI process extremely inefficient. The objective of this work was to develop and verify a three-dimensional model to simulate the RFI process. For a specified pressure and temperature cure cycle the code can predict resin pressure, viscosity and degree of cure, flow front progression, and temperature distribution in the preform and tooling components. The model was divided into submodels which describe resin flow, heat transfer, and resin kinetics. A finite element/control volume approach was used to model the flow of the resin through the preform. Boundary conditions include specified pressure, specified flow rate, and vents. A finite element formulation of the transient heat conduction equation was used to model the heat transfer. Thermal boundary conditions include either specified temperature or convection. The code was designed to be modular so the flow problem could be solved alone, or coupled with the thermal problem. The problems are solved sequentially in a quasi-steady state fashion. Non-isothermal experiments with a reactive resin were conducted to verify the thermal module and the resin model. A two blade stiffened panel, fitted with sensors, was manufactured using the RFI process. Thermocouples were used to measure the temperatures, and FDEM (dielectric) sensors and pressure transducers were used to monitor the flow front progression. Model predictions and experimental results were found to be in close agreement for the temperatures and flow front progression. The predicted and measured infiltration times matched to within 12%, and the temperatures to within 5%.

Geosynthetic clay liner (GCLs) combine geological materials with synthetic fabrics and require a new approach to conformance and quality control testing. GCLs have been used in various sealing applications over the last 10 years and have found an increased use in landfill applications. Therefore during the June 1993 ASTM meetings, a GCL subcommittee, D 35.04 was created with Mr. Larry Well of CH2M Hill, Portland, OR. serving as subcommittee chairman. The subcommittee comprises 6 task groups: • Physical properties • Manufacturing quality control and assurance (MQC/MQA), • Logistics, • Endurance, • Hydraulic properties, and • Mechanical properties. As an example, the Manufacturing QC/QA task group has currently developed a standard of practice for manufacturing quality control of GCLs, and methods for determining a swell index value for clay and the fluid loss of the clays. A standard for manufacturing quality assurance is currently in development. In the past quality control and quality assurance programs were developed by various manufacturers resulting in QC/QA programs which benefit the manufacturers products and not necessarily the project requirements. Another problem is the fact that the construction quality control and quality assurance (CQC/CQA) plan is often written in isolation by different people from those writing the MQC plan [1]. This paper will discuss the state-of-the-art of quality control and quality assurance for GCL testing to help to maximize GCL performance.

Machine learning (ML) has been leveraged to tackle a diverse range of tasks in almost all branches of nuclear engineering. Many of the successes in ML applications can be attributed to the recent performance breakthroughs in deep learning and the growing availability of computational power, data, and easy-to-use ML libraries. However, these empirical successes have often outpaced our formal understanding of the ML algorithms. An important but underrated area is the uncertainty quantification (UQ) of ML. ML-based models are subject to approximation uncertainty when they are used to make predictions due to sources including but not limited to, data noise, data coverage, extrapolation, imperfect model architecture, and the stochastic training process. The goal of this paper is to clearly explain and illustrate the importance of the UQ of ML. We elucidate the differences in the basic concepts of UQ of physics-based models and data-driven ML models. Various sources of uncertainties in physical modeling and data-driven modeling are discussed, demonstrated, and compared. We also present and demonstrate a few techniques to quantify the ML prediction uncertainties, including Monte Carlo dropout, deep ensemble, Bayesian neural networks, Gaussian processes, and conformal prediction. Finally, we discuss the need for building a verification, validation, and UQ framework to establish ML credibility.

Optimization of a thermal manufacturing process: Drawing of optical fibers

Electroless plating is a chemical process commonly used in semiconductor manufacturing that deposits a uniform metal coating onto a substrate through an auto-catalytic reduction reaction without using electricity. High quality deposition hinges on the precise control of bath composition and operating conditions, which in turn requires accurate and accessible monitoring of the bath chemical concentration. While chemical analyzers present an effective monitoring technique, they are generally limited by high costs, complexity, and maintenance. Time-delayed measurements also impose challenges on real-time operations, and any reduction to the monitoring lag would be highly beneficial. We introduce a data-driven machine learning (ML) approach able to achieve fast concentration predictions directly from operating conditions, and able to compute the prediction uncertainty that is valuable for subsequent robust control and optimization and for improving transparency and trustworthiness of ML tools in manufacturing settings. Notably, our ML procedure overcomes challenges of asynchronous time-series training data with missing values, and where available data are often noisy and sparse. Our ML approach begins with a data preprocessing step to handle asynchronous measurements and missing values by engineering features rooted in the underlying physical processes. Notably, a systematic feature selection is performed to down select features that are most correlated with the prediction target, thereby reduce model overfitting. We then compare a number of regression model architectures for capturing the sequential data relationships, including linear models, random forest, extreme gradient boosting, and fully connected, long short-term memory, and transformer neural networks. We quantify the model uncertainty by training them in a Bayesian manner using scalable Stein variational gradient descent, to compute the posterior probability distributions conditioned on the training data that reflect the uncertainty in the models induced by the quality and quantity of the available observations. We demonstrate the overall ML approach on a real-world dataset from a semiconductor manufacturing plant that exhibits the aforementioned data challenges. The best performing models exhibited 1, 5, 12% accuracy (defined as achieving within a 3% margin) improvements on predicting the metal ion, alkali, reductant concentrations, respectively, over a naive feature extraction technique. It also achieved 13, 10, 32% accuracy improvements over an autoregressive integrated moving average (ARIMA) baseline. These results show the approach’s effectiveness in providing reliable predictions with quantified uncertainty.