To ensure high quality castings, the ability to accurately quantify an as-cast surface's characteristics is of vital importance to the foundry industry. In addition, recent advancements in non-contact measurement systems have provided new opportunities to quantify a casting's surface beyond traditional roughness measurements. However, in the realm of non-contact measurement systems, there are numerous methodologies and metrics for evaluating and quantifying a surface. This paper investigates the critical surface features and approaches necessary for effective surface quality classification. More specifically, this study compares the accuracy of spatial statistical, modern digital processing techniques (e.g., convolution neural network) to classify as-cast surfaces. Through this comparison, this paper aims to develop a methodology to provide the most reliable results to industry. The findings will help guide the foundry industry in adopting the most appropriate techniques for surface quality assessment, ultimately enhancing product quality.

In the foundry industry, the quality of a cast part is often assessed by its surface roughness. This paper proposes a comprehensive framework for predicting surface roughness, in real-time, using digital images. The key to this framework is that the model used to perform these predictions is constructed offline from both digital images and point cloud data. This model, which integrates multilinear principal component analysis for feature extraction with convolutional neural networks for prediction, is used online with only the need for image data. For this approach to be viable in industry, it must be accurate across a wide range of surface types. Therefore, these predictions models are developed from two disparate surfaces, namely, sandpaper and C-9 comparator plates. To assess the robustness of the approach to multiple surfaces, two predictive modelling strategies are compared; (1) One model that created from both surface types and (2) Two individual models created separately for each surface type. This approach demonstrates the potential for integrating multi-sensor data to enhance surface quality assessments in the foundry industry without sacrificing cycle times.

The increasing risk and impact of cyber-physical attacks on the manufacturing industry are driven by vulnerabilities in industrial control systems and digital manufacturing technologies. These attacks can be strategically engineered to modify product dimensions while evading detection by current quality control (QC) tools. With the rise of cyber-physical attacks, new malicious process shift types are possible. Although QC studies have examined transient shifts from assignable causes, they have not specifically addressed intelligently designed attacks, which can dynamically alter both shift magnitude and duration, causing severe damage. This study aims to evaluate the performance of five control charting schemes in detecting intelligently designed cyber-physical transient attacks. Three types of transient attacks were developed and investigated: (1) alternating shifts, (2) constant shifts, and (3) trending shifts. These attacks were designed from an attacker's perspective, exploiting control chart vulnerabilities in manufacturing processes by leveraging knowledge of statistical process control methods. The performance of the control charting schemes was assessed using the average run length (ARL), average number of successful attacks (ANSAs), average quality loss (AQL), and detection probability (DP). The design of these attacks considers three critical factors: the magnitude (delta), the duration (l$l$), and the frequency or time between attacks (TBA). Our findings indicate that schemes relying solely on the exponentially weighted moving average EWMAX${\mathrm{EWMA}}_{X}$ chart are particularly vulnerable to alternating attacks, recording the highest ARL, ANSA, and AQL compared to other schemes. This vulnerability arises from the inherent characteristics of the EWMAX${\mathrm{EWMA}}_{X}$ chart, specifically its memory. In contrast, schemes that incorporate the X$X$ and/or EWMAX2${\mathrm{EWMA}}_{{X}<^>2}$ charts performed significantly better when subjected to alternating attacks. Additionally, constant shifts resulted in a lower ARL for schemes using EWMAX${\mathrm{EWMA}}_X$ and/or EWMAX2${\mathrm{EWMA}}_{{X}<^>2}$ charts for the delta values considered in this paper. For trending attacks, the EWMAX${\mathrm{EWMA}}_X$ scheme exhibited the lowest ANSA, while the X$X$ chart scheme demonstrated the highest. Overall, the combination of EWMAX${\mathrm{EWMA}}_X$ and EWMAX2${\mathrm{EWMA}}_{{X}<^>2}$ showed improved sensitivity and detection time across different attacks.

The ability to assess the surface quality quickly and accurately is of immense importance in manufacturing system. Modern metrology system along with machine learning is great at classification but requires more time. Traditionally accessing surface roughness is a time-consuming process. The progress in manufacturing technology necessitates improved approaches for quality control, specifically in the categorization of surface roughness, which has a substantial impact on the performance of materials. This research study introduces a novel method for classifying surface roughness by combining image data and point cloud data to create a comprehensive model. It then compares the performance of this model with a model that just relies on image data. A comprehensive analysis is conducted in this study, where image and point cloud data is collected and analysed. Multilinear principal component analysis (MPCA) along with random forest classifier is employed to create a model that classifies the surface texture. The primary goal is to showcase the enhanced precision and comprehensive understanding offered by the fused data model compared to the model that solely relies on images.Furthermore, the work presents a pragmatic approach for developing this enhanced model offline and applying it online in real-time production environments, with a particular focus on using only image data. This strategy is in line with the objectives of Industry 4.0, which seeks to achieve more intelligent and data-driven manufacturing processes. Subsequent investigations will prioritize expanding the model’s suitability to various manufacturing settings, particularly highlighting its capacity to ensure quality in manufacturing lines through the utilization of images.

This paper introduces a thorough approach for classifying refractory coatings used on chemically bonded sand according to their thickness, which is essential for monitoring mold and core coatings in foundries. The method combines feature extraction through vectorized principal component analysis (VPCA) with classification modeling using a machine learning algorithm. The study examines five different scenarios, which involve the utilization of raw axial, radial, and temperature data, as well as the use of scalar properties. Additionally, the study involves extracting features from the first two approaches and training on the complete dataset. An assessment of performance is carried out, showcasing the strong ability to classify accurately across all levels of coating thickness. In addition, Hotelling's T-squared statistics are used to identify changes in the process, offering valuable information about the structure and distinctiveness of the data classes. This study demonstrates the efficacy of feature extraction methods and machine learning algorithms in accurately categorizing coating thicknesses, providing practical solutions for applications in the foundry industry. This systematic methodology not only improves the comprehensibility and effectiveness of classification models but also offers vital understanding into process monitoring and identification of abnormalities within intricate datasets.

With recent technological advancements, production systems have become more susceptible to cyber‐physical attacks. Such attacks can be intelligently designed to both alter product dimensions and avoid detection by current Quality Control (QC) tools. The objective of this work is to assess the performance of random sampling strategies for control charting in detecting cyber‐physical attacks. More specifically, the methodology adopts traditional random sampling approaches used for inspection and applies them to univariate control charts. Different random sampling strategies are discussed and their performances under varying attack scenarios are evaluated. In addition, new control chart performance metrics were developed for a more in‐depth analysis, to guide practitioners in assessing how well these control charts increase their robustness to cyber‐physical attacks.

Background: Surgical site infections (SSIs) are a significant health care problem as they can cause increased medical costs and increased morbidity and mortality. Assessing a patient's preoperative risk factors can improve risk stratification and help guide the surgical decision-making process. Previous efforts to use pre-operative risk factors to predict the occurrence of SSIs have relied upon traditional statistical modeling approaches. The aim of this paper is to develop and validate, using state-of-the-art machine learning (ML) approaches, classification models for the occurrence of SSI to improve upon previous models.Methods: In this work, using the American College of Surgeons' National Surgical Quality Improvement Pro-gram (ACS NSQIP) database, the performances (eg prediction accuracy) of 7 different ML approaches (Logistic Regression (LR), Naive Bayesian (NB), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Deep Neural Network (DNN)) were compared. The performance of these models was evaluated using the area under the curve, accuracy, precision, sensitivity, and F1-score metrics.Results: Overall, 2,882,526 surgical procedures were identified in the study for the SSI predictive models' development. The results indicate that the DNN model offers the best predictive performance with 10-fold compared to the other 6 approaches considered (area under the curve = 0.8518, accuracy = 0.8518, preci-sion = 0.8517, sensitivity = 0.8527, F1-score = 0.8518). Emergency case surgeries, American Society of Anes-thesiologists (ASA) Index of 4 (ASA_4), BMI, Vascular surgeries, and general surgeries were most significant influencing features towards developing an SSI.Conclusions: Equally important is that the commonly used LR approach for SSI prediction displayed medio-cre performance. The results are encouraging as they suggest that the prediction performance for SSIs can be improved using modern ML approaches.& COPY; 2022 Association for Professionals in Infection Control and Epidemiology, Inc. Published by Elsevier Inc. All rights reserved.

As the collection and use of high-density (HD) spatial datasets has increased, the Statistical Process Control research community has strived to develop effective and efficient control charting techniques for these datasets. In general, these research efforts propose new control charting techniques and evaluate their abilities to detect different shift types. However, these works typically considered only conventional shift types, such as mean and variance shifts, which only account for a portion of the shift types that can manifest themselves in HD spatial datasets. In essence, advanced mathematical approaches are being developed for use with state-of-the-art measurement systems but assess their performance with traditional shift types developed for univariate statistics. This may hinder the effectiveness of these approaches in practice, as real-world systems may experience shift types other than (or in addition to) those addressed in the literature. The goal of this paper is to understand the ability of previously proposed control charting techniques to detect these unexplored shift types. This goal is accomplished through a simulation study that considers five different control charting techniques, identified from both the spatial statistics and spatial scan statistics literatures. The performances of these control charts are assessed against previously unexplored HD spatial dataset shift types. The results indicate that many control charting approaches were highly sensitive to variety of shift types. This suggests significant promise in the use of these approaches in systems that are susceptible to a wide variety of shift types, including shift types they were not specifically designed to detect.

Exponential Weighted Moving Average (EWMA) control charts have been widely used in Statistical Process Control (SPC) to detect small and persistent process shifts. In theory, EWMA control limits monotonically increase over time to account for the continual growth of the EWMA statistic’s variance. However, these control limits are often assumed constant and are set to their respective asymptotic limits to simplify the process of applying and analyzing EWMA control charts. One-sided EWMA charts are often implemented when it is only desirable to detect shifts in a specific direction. When using one-sided EWMA charts, reflecting boundaries (resets) can be used to prevent the statistic from drifting too far from the chart’s control limit, which can delay shift detection. There have been several research efforts into designing and studying the performance of one-sided EWMA charts with reflecting boundaries. However, these efforts have maintained the constant control limit assumption. When implementing a reflecting boundary, the EWMA statistic’s variance is constantly being reset to zero, which may significantly affect the constant control limit assumption’s validity. The focus of this paper is to understand behavior of the one-sided EWMA control charts with constant and time-varying limit assumption through simulation studies.

Machine vision system has been widely used for detecting surface defects in a manufacturing process. This paper proposes a new framework for image monitoring with control charts that aims to improve the overall performance in detecting a variety of surface‐related process shifts. Specifically, multiple images of the same object are acquired under different capturing parameters. To illustrate this framework, this paper considers two multi‐image control chart approaches: (1) fusing multiple images together with multilinear principal component analysis and monitoring with a single‐image control chart and (2) using a combined single‐image control chart. A two‐image simulation study, based upon real cross‐correlated images, is accomplished to compare the performances of these two multi‐image control chart approaches to a traditional single‐image control charts. The results indicate that multi‐image control charts outperform single‐image control charts when multiple shifts are considered. Another two‐image simulation study investigates the effect of cross‐correlations to multi‐image control charts’ performances. This study indicates that the two multi‐image control charts have similar performances at low cross‐correlation but the fused‐image control chart outperforms the combined single‐image control chart at high cross‐correlation. In addition, a case study with real images is conducted to demonstrate the proposed multi‐image monitoring framework's ability in detecting shifts.

Recent advancements in measurement systems have brought new opportunities to enhance the performance of quality control (QC) systems in modern manufacturing. Digital cameras and optical scanners are among these advanced measurement systems that are used for automated surface inspection. They can represent an entire product’s surface with high-density (HD) data in the forms of digital images and point clouds, respectively. Although both measurement systems provide HD data, their datasets are fundamental different and contain different information regarding a part’s surface. Extensive research efforts have been conducted to develop QC tools for each of these datasets individually; however, little research has focused on taking advantage of both point clouds and digital images simultaneously. To fully take advantage of information from both datasets, and more importantly their spatial cross correlation, this paper aims to use fused image/point cloud datasets to advance the capability of QC systems. A key challenge in incorporating both datasets is that the costs of acquiring data from these measurement systems differ drastically, making online monitoring using fused datasets less appealing. To overcome this challenge, a novel off-line/on-line hybrid monitoring scheme is proposed. The effectiveness of this proposed hybrid monitoring scheme is demonstrated with an additive manufacturing case study.

With the latest advances in computer and networking technologies, the threat of cyber-physical attacks against manufacturing systems is growing. Unlike traditional cyber-attacks, cyber-physical attacks are not limited to intellectual property theft and affect the physical world, which could be devastating to manufacturing, if they are undetected. Relying on traditional quality control to defend against these malicious attacks, manufacturers can choose to either closely monitor a large number of potential quality characteristics or only monitor a specific subset of the characteristics. However, the former choice may be impractical when a large number of potential characteristics exists, whereas the latter might be susceptible to an intelligently designed attack that targets unmonitored characteristics. Therefore, a novel random variable-selection approach that is both resilient to malicious cyber-physical attacks and sensitive to shifts over a small subset of characteristics is proposed in this work. Such an approach is based upon random sampling strategies when using multivariate Hotelling T (2) control charts. To assess its usefulness, the proposed approach was compared to an established variable-selection method, using a simplified cost model. The obtained results show that the proposed approach is both cost-effective and well-suited for industrial applications where the number of quality characteristics to monitor is quite significant.

In-process machining data (e.g., cutting forces and vibrations) have been typically collected and structured as time-referenced measurements (i.e., time-series data) and utilized in this structure to develop statistical data models used in process monitoring and control methods. This paper argues that a time-only-referenced representation overlooks the 3D nature of the physical process generating the data, and that machining data can be represented alternatively as functions of the tool-workpiece relative position resulting in a spatial point cloud data structure. High-density measurements of such spatially refenced data could be highly correlated to surrounding measurements, resulting in spatial correlation structures that could be of physical meaning and value to preserve and leverage. Using a simulated data study, this paper shows that preserving the spatial correlation structure of the data clearly improves the relative modeling performance when utilizing machining data point clouds versus the traditional time-referenced data structure. Specifically, this simulation study investigated the hypothesis that “considering the Gaussian process model class, the best model among all possible models developed using the spatial point cloud data structure has smaller/equal modeling and prediction errors compared to the best model among all possible models developed using the time-referenced data structure.” While this investigation was limited to considering the case of stationary isotropic processes, it demonstrated that the performance gap was relatively large. This encourages further investigations using real-world data to better understand the types of spatial correlations that exist in machining data and the specific machining regimes and process variables that would benefit the most from the spatial point cloud representation of the data.

With advances in computer and networking technologies, along with the increasing dependency on interconnected cyber-physical components, the threat of cyber-physical attacks against manufacturing is on the rise. As opposed to traditional cyber-attacks, cyber-physical attacks go beyond intellectual property theft and can affect the physical world. In manufacturing, such attacks can result in changed product designs, manipulated manufacturing equipment, and altered final products. For over a century, manufacturing systems have relied heavily on Quality Control (QC) systems to ensure stable processes and product integrity. However, previous research has suggested that current QC tools could be exploited by an adversary, making it difficult or even impossible to detect attacks. Unfortunately, there has been little to no effort to identify/understand opportunities where QC tools could be exploited, which is an essential step toward developing new cyber-security solutions for manufacturing. In response, this paper establishes a systematic approach to effectively categorize QC tool vulnerabilities. Furthermore, to highlight the importance of this research to the manufacturing community, the negative effects of exploiting QC tools by cyber-physical attacks are demonstrated in this paper. Finally, best practices and guidelines for better cyber-physical security in manufacturing are also presented.

As sensor and measurement technologies advance, there is a continual need to adapt and develop new Statistical Process Control (SPC) techniques to effectively and efficiently take advantage of these new datasets. Currently high-density noncontact measurement technologies, such as 3D laser scanners, are being implemented in industry to rapidly collect point clouds consisting of millions of data points to represent a manufactured parts' surface. For their potential to be realized, SPC methods capable of handling these datasets need to be developed. This paper presents an approach for performing SPC using high-density point clouds. The proposed approach is based on transforming the high-dimensional point clouds into Non-Uniform Rational Basis Spline (NURBS) surfaces. The control parameters for these NURBS surfaces are then monitored using a surface monitoring technique. In this paper point clouds are simulated to determine the performance of the proposed approach under varying fault scenarios.

In manufacturing, advanced measurement systems (e.g., 3D laser scanners) are continually being incorporated into modern quality control (QC) systems to provide high-density (HD) data. A significant amount of research efforts has been placed in the development of QC tools, such as Phase I and II statistical process control (SPC) approaches using HD data. However, the effectiveness of SPC tools highly depends on measurement system adequacy. The study of the quality and adequacy of a measurement system, known as measurement capability or Phase 0 in SPC applications, is a prerequisite to implementing any SPC tool; which has mostly been neglected for HD data. This paper proposes a holistic Gauge study approach for HD data obtained from 3D laser scanners (e.g. point clouds) by using spatial statistics data models. The main objectives of this work are two-fold: 1) Study how to analyse the repeatability and reproducibility of a point cloud and 2) Quantify the uncertainty associated with a point cloud under different factors involved in acquiring point clouds from 3D laser scanners.

With the latest developments in networking and internet technology, cyber-physical attacks are becoming more frequent across a wide range of industries, including manufacturing. Cyber-physical attacks, as the name implies, affect the physical components of the system and go beyond just intellectual property theft. Attacks of this nature in manufacturing could include destroying equipment, altering product designs, or modifying manufacturing processes. Due to the costly consequences of successful cyber-physical attacks in manufacturing going undetected, a literature review of the cyber-physical security efforts in manufacturing is presented in this work. Such a literature review allows determining the current state of the research efforts in this field and highlighting future research areas that need more focus.

Industry 4.0 and its related technologies (e.g., embedded sensing, internet-of-things, and cyber-physical systems) are promising a paradigm shift in manufacturing automation. However, with a continual increase in device interconnectivity, securing these systems becomes crucial. As these systems evolve, opportunities for cyberattacks extend to include attacks that can physically alter parts (Product-Oriented C2P attacks). Fortunately, since these cyber-physical attacks affect the physical world, there exists potential to detect an attack through its physical manifestation. Typically, in manufacturing, quality control (QC) systems are used to detect quality losses or deviations from nominal. This paper proposes that QC tools can be adapted to act as physical detection layers as part of a defense-in-depth strategy (common IT security strategy) that increases the difficulty/cost required for a successful attack. However, effectively designing physical detection layers requires understanding the extent to which attacks can (and cannot) be designed to avoid detection. In response, this paper proposes a machining specific attack design scheme and an attack design designation system (ADDS) that provides the structure to populate a wide variety of potential attacks. To illustrate the importance of applying a defense-in-depth strategy for machining, a case study is conducted with several realistic attacks against an example machining process that collects in-situ process data. Within this case study, the proposed ADDS is employed to systematically describe how these attacks could be designed to avoid detection. Finally, through this exploration, this paper shows how employing process-domain knowledge to understand the effects of Product-Oriented attacks on process physics can further aid in detection layer designs.

Recent measurement system developments have brought new opportunities to enhance the performance of quality control (QC) systems in manufacturing. Digital cameras and 3D optical scanners are among the advanced measurement systems that can represent an entire product's surface. These data-rich environments have been widely used in automated surface defect detection. However, despite the fact that datasets from both digital cameras and 3D optical scanners can both represent a surface, the datasets are fundamental different and contain different information regarding a part's surface. Extensive research efforts have been conducted on developing QC tools for each of these datasets individually, very few researches were focused on approaches to take advantage of both 3D point clouds (obtained from technologies such as 3D optical scanners) and digital images (obtained from digital cameras) at the same time. This paper proposes a hybrid approach for surface inspection of additive manufacturing parts through Multilinear Principal Component Analysis (MPCA). This proposed approach is demonstrated with additively manufactured parts, which shows the advantage of combining this information for surface classification.

With recent advancements in computer and network technologies, cyber-physical systems have become more susceptible to cyber-attacks, with production systems being no exception. Unlike traditional information technology systems, cyber-physical systems are not limited to attacks aimed solely at intellectual property theft, but include attacks that maliciously affect the physical world. In manufacturing, cyber-physical attacks can destroy equipment, force dimensional product changes, or alter a product's mechanical characteristics. The manufacturing industry often relies on modern quality control (QC) systems to protect against quality losses, such as those that can occur from an attack. However, cyber-physical attacks can still be designed to avoid detection by traditional QC methods, which suggests a strong need for new and more robust QC tools. As a first step toward the development of new QC tools, an attack taxonomy to better understand the relationships between QC systems, manufacturing systems, and cyber-physical attacks is proposed in this paper. The proposed taxonomy is developed from a quality control perspective and accounts for the attacker's view point through considering four attack design consideration layers, each of which is required to successfully implement an attack. In addition, a detailed example of the proposed taxonomy layers being applied to a realistic production system is included in this paper.