Real-world Manufacturing Research Articles

PCB Defect Classification with Data Augmentation-Based Ensemble Method for Sustainable Smart Manufacturing

In the rapidly evolving field of printed circuit board (PCB) manufacturing, automated optical inspection (AOI) systems play a critical role but often face challenges such as computational inefficiencies, high costs, and limited defect data. To address these issues, we propose an ensemble methodology that combines lightweight models with custom data augmentation techniques to enhance defect classification accuracy in real-time production environments. Our approach mitigates overfitting in small datasets by generating diverse models through advanced data augmentation and employing feature-specific validation strategies. These models are integrated into an ensemble framework, achieving complementary results that improve classification accuracy while reducing computational overhead. We validate the proposed method using two datasets: the general classification dataset CIFAR-10 and an on-site real-world PCB dataset. With our approach, the average accuracy on CIFAR-10 improved from 97.6% to 98.2%, and the accuracy on the PCB dataset increased from 81% to 89%. These results demonstrate the method’s effectiveness in addressing data scarcity and computational challenges in real-world manufacturing scenarios. By improving quality control and reducing waste, our method optimizes production processes and contributes to sustainability through cost savings and environmental benefits. The proposed methodology is versatile, scalable, and applicable to a range of defect classification tasks beyond PCB manufacturing, making it a robust solution for modern production systems.

Enhancing the productivity of mould shop using continuous improvement tools and simulation

In today’s competitive manufacturing environment, enhancing productivity and operational efficiency is crucial for companies to meet the rising customer demands and expectations driven by recent technological advancements. This study presents a case of productivity enhancement in mould shop manufacturing, which was experiencing bottlenecks and non-value-added (NVA) activities, leading to inefficiencies that limit the throughput. A Discrete Event Simulation (DES) model is developed for the mould shop considering the major components such as Hub, Mainframe, and Planet Carrier. Bottleneck resources are identified from the DES model. The validation of the simulation results is carried out using Pareto analysis. Root Cause Analysis (RCA) and Five Whys (5-Why) analysis are used to brainstorm the causes that lead to lower productivity. Time-reduction strategies using the Theory of Inventive Problem Solving (TRIZ) improved the throughput. The improvements are incorporated into the DES model, and updated results are obtained. As a result, throughput for the hub improved by 60% for the hub, 40% for the mainframe, and 55.88% for the planet carrier. These results show how integrating continuous improvement tools with simulation models enables streamlining operations and substantially increases productivity. The scenario reveals that Lean principles successfully overcome real-world manufacturing problems.

A Q-learning-based multi-population algorithm for multi-objective distributed heterogeneous assembly no-idle flowshop scheduling with batch delivery

Distributed heterogeneous factory environments have become the mainstream in real-world manufacturing enterprises. Scheduling the assembly no-idle flowshops in such distributed heterogeneous environments is significant for practitioners. This problem takes into account the heterogeneity between different factories and the batch delivery process. To minimize inventory and tardiness costs of this new problem, a novel Q-learning-based multi-population multi-objective evolutionary algorithm is proposed. The algorithm incorporates five problem-specific properties to design the solution representation and calculate objectives. To obtain high-quality initial dual populations, a collaborative initialization technique combining four NEH heuristics and a completely random approach is proposed. To enhance algorithm’s search efficiency, a Q-learning-based selection method is developed to coordinate four job-related and four product-related operators. The dual-cooperation strategy is introduced to improve the global search capability. Experimental results demonstrate the effectiveness of all the designed components, indicating that the proposed algorithm outperforms five existing algorithms. It is well-suited for addressing real-world distributed heterogeneous scheduling scenarios by the proposed algorithm.

Capacity Constraint Analysis Using Object Detection for Smart Manufacturing

The increasing adoption of Deep Learning (DL)-based Object Detection (OD) models in smart manufacturing has opened up new avenues for optimizing production processes. Traditional industries facing capacity constraints require noninvasive methods for in-depth operations analysis to optimize processes and increase revenue. In this study, we propose a novel framework for capacity constraint analysis that identifies bottlenecks in production facilities and conducts cycle time studies using an end-to-end pipeline. This pipeline employs a Convolutional Neural Network (CNN)-based OD model to accurately identify potential objects on the production floor, followed by a CNN-based tracker to monitor their lifecycle in each workstation. The extracted metadata are further processed through the proposed framework. Our analysis of a real-world manufacturing facility over six months revealed that the bottleneck station operated at only 73.1% productivity, falling to less than 40% on certain days; additionally, the processing time of each item increased by 53% during certain weeks due to critical labor and materials shortages. These findings highlight significant opportunities for process optimization and efficiency improvements. The proposed pipeline can be extended to other production facilities where manual labor is used to assemble parts, and can be used to analyze and manage labor and materials over time as well as to conduct audits and improve overall yields, potentially transforming capacity management in smart manufacturing environments.

Variable Parameters t Chart: An Application in Automobile Engine Piston Rings Manufacturing Process

The [Formula: see text] type charts possess a drawback as they may falsely indicate out-of-control (OOC) situations when the process standard deviation is unstable. They are not robust against estimation errors or variations in the process standard deviation. To overcome this setback, the [Formula: see text] type charts were developed. This research introduces the variable parameters (VP) [Formula: see text] chart for effectively monitoring the process mean. This study encompasses the implementation, optimal designs and performance evaluation of the proposed VP [Formula: see text] chart. The average time to signal (ATS), standard deviation of the time to signal (SDTS) and expected average time to signal (EATS) criteria are used to assess and compare the performance of the VP [Formula: see text] chart with existing charts, such as Shewhart (SH) [Formula: see text], synthetic (Syn) [Formula: see text], side-sensitive group runs (SSGR) [Formula: see text], exponentially weighted moving average (EWMA) [Formula: see text] and variable sampling interval (VSI) EWMA [Formula: see text] charts. From extensive analysis, we observe that the VP [Formula: see text] chart outperforms the SH [Formula: see text], Syn [Formula: see text] and SSGR [Formula: see text] charts for all sizes of mean shifts. However, the EWMA [Formula: see text] and VSI EWMA [Formula: see text] charts exhibit better performance for small shifts. Additionally, to demonstrate the implementation and usage of the proposed VP [Formula: see text] chart, we present a real-world manufacturing example.

On-line recognition of mixture control chart patterns using hybrid CNN and LSTM for auto-correlated processes

Statistical process control (SPC), designed to detect and identify process disturbances, is an effective quality control method for ensuring process stability. Owing to the growing automation of the industry, the manufacturing process can be autocorrelated. Engineer process control (EPC) is typically used to address autocorrelation. However, process disturbances can be offset by feedback compensation, making control chart patterns (CCPs) difficult to identify. Most studies on control chart pattern recognition (CCPR) techniques are based on a single abnormal control chart pattern (CCP). However, a mixture of CCPs can occur during real-world manufacturing processes. With the development of intelligent manufacturing systems, early detection of abnormal concurrent CCPs has become an important issue. In this study, a hybrid model combining a convolutional neural network (CNN) and long short-term memory (LSTM) in an online detection system was used to recognize the concurrent CCPR problem in SPC-EPC processes. The results showed that the average accuracy of the deep learning CNN-LSTM method was 99.83%, which was significantly better than that of the machine learning method. In addition, the running time of the CNN-LSTM model was shortened. In a comparison of online monitoring between the deep learning method and the machine learning method, the CNN-LSTM model for an online monitoring system identified abnormal concurrent patterns faster. Therefore, the proposed CNN-LSTM online monitoring scheme can be applied confidently and successfully to identify mixture CCPs in an SPC-EPC process.

Anomaly detection in smart manufacturing: An Adaptive Adversarial Transformer-based model

In Industry 5.0, smart manufacturing brings additional intricacies and novel data processing challenges. Given the evolving nature of manufacturing processes and the inherent complexity of data, including noise and missing entries, achieving accurate anomaly detection becomes even more intricate. Conventional methods often miss nuanced anomalies, especially when dealing with high-dimensional, multivariate, non-stationary data. These data types are typical of smart manufacturing environments. Hence, many recent approaches have embraced deep learning to confront these challenges, making use of diverse attention mechanisms to acquire data representations. However, in manufacturing, where the dynamics of time series data change over time, methods relying solely on pointwise or pairwise representations often fall short. Thus, ensuring product quality and operational integrity calls for even more advanced methodologies. The deficiency lies in the capability of state-of-the-art models to effectively capture abnormal patterns while considering both local and global contextual information. This challenge is compounded by the rarity of anomalies, making it exceedingly challenging to establish substantial associations between individual abnormal points and the entire time series. To tackle these challenges, we introduce the “Adaptive Adversarial Transformer” as a novel deep learning technique that combines Transformer architecture with an anomaly attention mechanism and Adversarial Learning. Our Model effectively captures intricate temporal patterns, distinguishes normal and anomalous behaviors, and dynamically adjusts thresholds to align with the evolving dynamics of time-series data. Empirical validation on four benchmark datasets and three real-world manufacturing datasets demonstrates our model’s effectiveness compared to the state-of-the-art, as evidenced by the F1-Score.

Optimized XGBoost Model with Whale Optimization Algorithm for Detecting Anomalies in Manufacturing

Anomalies and defects in the manufacturing process hinder operating efficiency and product quality. The Whale Optimization Algorithm (WOA) optimizes the XGBoost model for better anomaly identification by iteratively refining hyperparameters. Experiments using real-world manufacturing datasets prove proposed model works. Comparing the proposed model to traditional anomaly detection methods shows its superior performance in industry patent concept. The optimized XGBoost model's interpretability and anomaly detection features are also discussed. In this paper, WOA is applied in this work to optimize hyperparameters of XGBoost, a robust gradient boosting technique for accurate anomaly detection in manufacturing systems. Optimized XGBoost gained 1.00 precision value, 0.9 recall value and 0.96 f1-score for class 0.0 and gained a 0.95 precision value, 1.00 recall value, and a 0.97 f1-score for class 1.0. The proposed model gained 0.993 Train Score and 0.964 Test Score. Our findings suggest that integrating XGBoost with the WOA may uncover manufacturing process irregularities. Optimization improves detection accuracy and provides a flexible and interpretable framework, helping modern industrial processes maintain quality and efficiency. This research encourages machine learning optimization for industrial patent applications, advancing anomaly detection methods.

THE INFLUENCE OF THE CUTTING PARAMETERS ON THE SURFACE ROUGHNESS AND VIBRATION SIGNAL IN THE TURNING PROCESS

The performance of the final part of the turning process, a topic of practical importance, heavily depends on the cutting parameters. Surface quality, a crucial product attribute in manufacturing, often tops consumer demands during machining due to its significant influence on product functionality. This paper assesses the correlation between the detected vibration signal, statistical parameters (Crest, Kurtosis and I-kaz3D coefficient) and surface roughness and provides valuable insights for practical applications. When the tool experiences vibrations during material removal, these oscillations leave a ripple effect on the surface of the workpiece. We aim to determine the impact of cutting parameters on surface roughness while turning 42CrMo4 steel using a carbide tool insert. Cutting parameters, such as spindle speed, feed, and depth of cut, were used. Signal processing is carried out using different techniques to identify the effect of the cutting parameters on vibration signals. Finally, we delve into the interaction effects between the cutting parameters, vibration signals and surface roughness, offering a comprehensive understanding of real-world manufacturing scenarios. Higher I-kaz3D coefficients correspond to higher surface roughness values. The I-kaz3D coefficient decreases as the surface roughness measurements decrease, indicating that the I-kaz3D technique can accurately indicate an increase or decrease in surface roughness. On the other hand, the cutting force Fc was the component most strongly correlated with surface roughness (Ra), yielding the best results across all indices (adjusted R²adj = 95.1%, er=.8 ± 2.3%).

Optimization of Cost-Based Hybrid Flowshop Scheduling Using Teaching-Learning-Based Optimization Algorithm

A cost-based hybrid flowshop scheduling (CHFS) combines flow shop and job shop elements, with cost considerations as a key indicator. CHFS is a complex combinatorial optimization challenge encountered in real-world manufacturing and production environments. This paper investigates the optimization of a CHFS problem using the Teaching Learning-Based Optimization (TLBO) algorithm. Effective CHFS is crucial for achieving production balance, reducing costs, and improving customer satisfaction. The authors formulate the CHFS scheduling problem and propose applying the TLBO algorithm to minimize total costs, including labor, energy, maintenance, and delay expenses. The performance of the TLBO technique is evaluated through computational experiments on various CHFS problem instances. The results demonstrate the effectiveness of the TLBO algorithm, which achieved the best results in 42% of the test cases, surpassing other algorithms like the Grey Wolf Optimizer and Particle Swarm Optimization. Additionally, the TLBO algorithm had the highest average performance ranking across the comparative algorithms. The study highlights the potential of the TLBO algorithm as an efficient optimization tool for complex manufacturing scheduling problems.

Dynamic Forces and Vibrations of the Planetary Gear-Set Radial Bearings with the Roller Dimension Deviation

Double planetary gear trains (DPGTs) have high loading capacity and transmission, making them indispensable in various heavy rotating machinery applications. However, the presence of planetary gear-set radial bearing roller dimension deviation (PBRDD) from real-world manufacturing discrepancies poses a significant challenge to the dynamics of the DPGTs. Previous dynamic models often overlooked the internal excitations of the planetary gear-set radial bearing, focusing only on the single planetary gear train dynamics, rendering them inadequate for studying the impact of the PBRDD on DPGTs. Therefore, this paper proposes a coupling method integrating the PBRDD model with the DPGT dynamic model, taking into consideration the planetary gear-set radial bearing’s internal excitations, to reveal the impact of PBRDD on DPGTs’ dynamics. Specifically, the changes in roller–ring contact force and roller–cage collision force induced by PBRDD are considered. Through comprehensive simulation and experimental analyses, the accuracy of the dynamic model is assessed and its efficacy in capturing the effects of PBRDD on DPGTs’ dynamics is validated. The results show that the DPGTs’ dynamics increment with the increment of PBRDD.

A Comprehensive examination of utilizing neutrosophic parameters in manufacturing industry through queueing approach

Queueing theory plays a pivotal role in analyzing the dynamics of manufacturing systems subject to random arrivals and service times. The M/G/1 queueing model, which represents a single-server queue with general distributions for inter-arrival and service times, is fundamental in this regard. However, traditional queueing models often fail to account for the inherent uncertainties encountered in real-world manufacturing scenarios. Neutrosophic theory provides a robust framework for modeling indeterminacy, ambiguity, and inconsistency in queueing parameters, thereby offering a more adaptable and precise depiction of system behavior. This research investigates the utilization of Neutrosophic sets in defining arrival rates, service times, and queue length distributions within the M/G/1 framework tailored to manufacturing settings. By employing numerical examples and comparative analyses, the study explores the effects of Neutrosophic parameters on key performance indicators such as mean waiting time, queue length distribution, and server utilization, considering distributions like Exponential, Erlang, and Deterministic. Additionally, it delves into the implications of integrating Neutrosophic parameters into queueing theory, providing valuable insights into improved decision-making and system optimization across various manufacturing operational contexts.

Exploring the interplay of costs and flow time in server allocation for flow lines with parallel non-identical machines

ABSTRACT The Server Allocation Problem is a significant challenge in manufacturing systems involving deciding the number and version of machines for each stage. This optimization problem is of utmost importance when there is a trade-off between minimizing investment costs and flow time while achieving target performance, such as a minimum throughput. Manufacturing systems often employ hybrid flow lines, with non-identical machines at each stage. This paper analyzes the trade-off between minimizing costs and flow time when allocating servers in hybrid flow lines by proposing a bi-criteria approach that uses an efficient pattern-based problem representation and a Variable Neighborhood Search solution algorithm. The objectives are to minimize total cost and flow time while ensuring a target throughput. A metric for comparing obtained solutions is proposed, providing insights for practical implementation. A comprehensive numerical analysis is conducted, to validate the proposed approach and demonstrate its applicability in real-world manufacturing settings.

Application of Generative Adversarial Networks in Semi-Annotated Defect Synthesis and Detection Under Limited Resources

Defect detection is a crucial technology that is extensively employed in the manufacturing industry to monitor and ensure the quality of output. Deep learning models have shown remarkable potential for defect detection. However, the success of these models heavily relies on voluminous training data. Collecting substantial amounts of defect data is challenging in practical settings, and the tedious process of pixel-level defect annotation further complicates the task. Among the common defects encountered in manufacturing, scratches are particularly significant. To address these challenges, this study proposes a two-phase generative adversarial network (GAN) approach for synthesizing defect images and generating semi-automatic pixel-wise labels for anomaly detection. The first phase primarily focuses on synthesizing images, while the second phase involves the pixel-wise labeling of the images. The synthesized paired images generated by the GANs serve as input to the semantic network. Notably, the proposed methodology requires only a few real defect samples for training and a small amount of annotated data, making it practical and computationally efficient for implementation in the manufacturing industry. Experimental results indicate the effectiveness of the proposed deep-learning solution in defect detection, specifically in identifying scratches on various textured and patterned surfaces. A notably high detection accuracy is achieved, validating the potential of the approach in real-world manufacturing scenarios.

Optimising distributed heterogeneous flowshop group scheduling arising from PCB mounting: integrating construction and improvement heuristics

ABSTRACT In real-world manufacturing systems for processing printed circuit boards (PCBs), the workshops integrating various flowline-based cells are common in large-scale enterprises. The scheduling problem within the systems is modelled as the distributed heterogeneous flowshop group scheduling problem (DHFGSP) in this study. Departing from practical requirements and considering the grouping characteristics among PCB components, we account for carryover sequence-dependent setup time (CSDST). Moreover, recognising the critical importance of just-in-time production in semiconductor manufacturing, the total tardiness, a previously unexplored objective within the DHFGSP context, is addressed. To tackle this problem, a mixed-integer linear programming (MILP) model, capable of obtaining optimal solutions for small-scale instances, is proposed. Due to the NP-hard nature of the problem, obtaining high-quality solutions in a reasonable time using the MILP model becomes challenging for large-scale instances. Therefore, a solution algorithm, comprising construction and improvement heuristics, is developed. Capitalising on the problem’s characteristics, the construction heuristic efficiently generates high-quality feasible solutions in a very short time. Built primarily around artificial bee colony (ABC) optimisation, the improvement heuristic can significantly further enhance solutions by integrating the collaborative and restart operators. Comprehensive experiments on instances of varying scales demonstrate the effectiveness of the proposed algorithm for the problem under investigation.

Atomic-Scale Investigation of Electromigration Behavior in Cu-to-Cu Joints at High Current Density

Three-dimensional integrated circuit (3D IC) packaging offers significant advantages, such as enhanced computational efficiency through greater integration and shorter interconnection distances. The effective interconnection of layers, particularly through redistribution layer (RDL) technology, is crucial for the successful operation of 3D ICs. While copper remains the preferred material for interconnections, often in conjunction with through-silicon via (TSV) technology for chip-to-chip links, the shrinking dimensions of components present challenges. In particular, the reduction in size may lead to elevated current densities in individual joints, giving rise to issues like electromigration (EM) and potential failure. EM-induced atomic migration may lead to the formation of voids in the interconnection. The accumulation of voids to a critical point may result in an open circuit in the joints, significantly impacting the conductor's reliability. This work delves into the micro-scale aspects of these challenges, complementing previous macroscopic investigations. We aim to elucidate the behavioral mechanisms through atomic-scale observations, employing an in-situ high-resolution transmission electron microscope (in-situ HRTEM) to observe atomic-scale images of EM. The insights gained from this work not only contribute to the academic understanding of industry challenges but hold potential applications in real-world manufacturing processes. Fig. 1. (a) Schematic diagram of TEM sample preparation. (b) Cross-sectional HRTEM image of copper bonding. (c, d) TEM images showing voids expansion at bonding interface. Figure 1

An incremental learning approach to dynamic parallel machine scheduling with sequence-dependent setups and machine eligibility restrictions

Minimizing the tardiness of a parallel machine scheduling problem has been actively studied in modern manufacturing systems. In particular, dynamic parallel machine scheduling problems (DPMSPs) have gained much attention since rescheduling is required to address unpredictable events such as unexpected job arrivals and machine breakdowns, which usually occur in real-world manufacturing systems. To deal with such unpredictable events, many researchers have employed deep supervised learning (SL). However, it is still challenging to solve a DPMSP since SL requires a large number of high-quality schedules for the training neural networks. In this paper, we propose an incremental learning-based scheduling method (ILS) in which neural networks (NNs) are periodically trained by utilizing schedules built from the updated NNs in previous training intervals. Furthermore, to perform training without reducing solution space within a dynamic environment, the proposed method is designed to consider the dependency between consecutive allocations on a machine for the sum of setup time and tardiness. To verify the effectiveness of the proposed method, extensive experiments are carried out in static and dynamic scheduling problems. The experiment results demonstrate that ILS outperforms the existing methods such as dispatching rules, metaheuristics, and DRL-based methods. Additionally, we show that the proposed input features are effective in appropriately selecting job-machine pairs for assigning jobs on eligible machines through deep Shapley additive explanations analysis.

Multi-objective optimization using improved NSGA-II for integrated process planning and scheduling problems in a machining job shop for large-size valve.

This paper studied an integrated process planning and scheduling problem from a machining workshop for large-size valves in a valve manufacturing plant. Large-size valves usually contain several key parts and are generally produced in small-series production. Almost all the parts need to be manufactured in the same workshop at the same time in the plant. Facilities have to handle various items in one order, including different models, sizes, and types. It is a classical NP-hard problem on a large scale. An improved NSGA-II algorithm is suggested to obtain satisfactory solutions for makespan and manufacturing costs, which involve large optimization parameters and interactions. A two-section encoding method and an inserting greedy decoding method are chosen to enable the algorithm. The dynamic population update strategy based on dynamic population update and the adaptive mutation technique depending on the population entropy changing rate are selected for enhancing both the solution quality and population diversity. The methodology was successfully implemented in a real-life case at a major valve machining workshop operated by Yuanda Valve Company in China. By taking into account realistic factors and restrictions that have been identified from a real-world manufacturing setting, this technique aids in bridging the knowledge gap between present IPPS research and practical valve production implementations.

An Integrated Process Planning and Scheduling problem solved from an adaptive multi-objective perspective

The planning and scheduling of processes are the main foundations of modern manufacturing systems, and their efficient integration represents one of the principal interests in operations research. Most of the reported literature in operations research formulates the optimization process considering a single objective. However, real-world manufacturing systems are influenced by several variables that must be considered in a complete process planning task. To solve this gap, this paper presents a computational study where the main variables of process planning are integrated: production times, production costs, and the Makespan of the system. To incorporate such variables, the problem is reformulated as a multi-objective optimization system by employing an Adapted Non-dominated Sorting Genetic Algorithm ii (ANSGA-ii) methodology. Under such an approach, the ANSGA-ii codifies the process planning elements into its structure to execute the operators for determining the best tradeoff between objectives. In this work, different cases of study are considered and compared with some of the most employed multi-objective approaches in the literature to demonstrate the performance and robustness of the proposed method. Statistical analysis corroborates that the proposed ANSGA-ii can generate competitive results outperforming the techniques employed in the experimental study.

Probabilistic physics-integrated neural differentiable modeling for isothermal chemical vapor infiltration process

Chemical vapor infiltration (CVI) is a widely adopted manufacturing technique used in producing carbon-carbon and carbon-silicon carbide composites. These materials are especially valued in the aerospace and automotive industries for their robust strength and lightweight characteristics. The densification process during CVI critically influences the final performance, quality, and consistency of these composite materials. Experimentally optimizing the CVI processes is challenging due to the long experimental time and large optimization space. To address these challenges, this work takes a modeling-centric approach. Due to the complexities and limited experimental data of the isothermal CVI densification process, we have developed a data-driven predictive model using the physics-integrated neural differentiable (PiNDiff) modeling framework. An uncertainty quantification feature has been embedded within the PiNDiff method, bolstering the model’s reliability and robustness. Through comprehensive numerical experiments involving both synthetic and real-world manufacturing data, the proposed method showcases its capability in modeling densification during the CVI process. This research highlights the potential of the PiNDiff framework as an instrumental tool for advancing our understanding, simulation, and optimization of the CVI manufacturing process, particularly when faced with sparse data and an incomplete description of the underlying physics.