Equipment failure in paper mills represents a critical barrier to operational efficiency and the adoption of Industry 4.0 principles. To address this, a systematic literature review was conducted to identify the multifactorial determinants of such failures. A novel hybrid methodology was proposed, integrating the Functional Analysis Systems Technique (FAST), enhanced by Lean 5S (Sort “Seiri”, Set in Order “Seiton”, Shine “Seiso”, Standardize “Seiketsu”, Sustain “Shitsuke”) principles, to structure the qualitative data collection. The analysis was performed using a Pugh matrix, followed by a Principal Component Analysis (PCA) to extract knowledge systematically. This approach facilitated the development of a conceptual model for downtime causation. The PCA results indicate that two principal components collectively explain 58.5% of the observed variance in failure data. The f irst component was strongly correlated with maintenance practices and operational errors, while the second was associated with intrinsic equipment characteristics and their operating conditions. This data-driven modeling elucidates underlying correlations between disparate factors, providing a robust foundation for prioritizing targeted maintenance optimization actions. This research contributes to the field of industrial intelligence by demonstrating an original methodology for transforming qualitative systematic review data into a quantifiable analytical framework. The application of PCA to this corpus enables the identification of multidimensional interactions that are frequently overlooked in conventional analyses, thereby enriching root-cause failure analysis and informing strategic decision making for predictive maintenance. The identified factors underscore the imperative of a balanced integration between technical data and human factors for the successful digital transformation of production systems.

Material extrusion additive manufacturing (MEX-AM) has emerged as a transformative technology with the potential to redefine industrial production; however, persistent challenges remain regarding variability in part quality, the absence of robust in-process defect detection, and limited capacity for process optimization. To address these limitations, an integrated multi-sensor and machine learning (ML) framework was developed to enhance real-time monitoring and defect detection during MEX-AM. Data were acquired from thermocouples, accelerometers, and high-resolution cameras, and subsequently processed through a multi-sensor data fusion pipeline to ensure robustness against noise and variability. A Multi-Criteria Decision Analysis (MCDA) framework was employed to evaluate candidate ML algorithms based on accuracy, computational cost, and interpretability. Random Forest (RF) and Artificial Neural Network (ANN) models were identified as the most suitable approaches for MEX-AM applications. Validation experiments demonstrated a 92% success rate in corrective interventions, with a reduction of defective components by 38% compared with conventional monitoring methods. The integration of sensor fusion with advanced learning models provided improved predictive capability, enhanced process stability, and significant progress toward intelligent, self-optimizing manufacturing systems. The proposed methodology advances statistical quality control and reduces material waste while aligning with the objectives of Industry 4.0 and smart manufacturing. By demonstrating the efficacy of multi-sensor fusion and ML in real-world AM environments, this study highlights a pathway toward scalable, autonomous, and sustainable industrial production.

The proliferation of the Industrial Internet of Things (IIoT) has transformed energy distribution infrastructures through the deployment of smart metering networks, enhancing operational efficiency while concurrently expanding the attack surface for sophisticated cyber threats. In response, a wide range of Machine Learning (ML)–based Intrusion Detection and Prevention Systems (IDPS) have been proposed to safeguard these networks. In this study, a systematic review and comparative analysis were conducted across seven representative implementations targeting the Internet of Things (IoT), IIoT, fog computing, and smart metering contexts. Detection accuracies reported in these studies range from 90.00% to 99.95%, with models spanning clustering algorithms, Support Vector Machine (SVM), and Deep Neural Network (DNN) architectures. It was observed that hybrid Deep Learning (DL) models, particularly those combining the Convolutional Neural Network and the Long Short-Term Memory (CNN-LSTM), achieved the highest detection accuracy (99.95%), whereas unsupervised approaches such as K-means clustering yielded comparatively lower performance (93.33%). Datasets utilized included NSL-KDD, CICIDS2017, and proprietary smart metering traces. Despite notable classification accuracy, critical evaluation metrics—such as False Positive Rate (FPR), inference latency, and computational resource consumption—were frequently underreported or omitted, thereby impeding real-world applicability, especially in edge computing environments with limited resources. To address this deficiency, a unified benchmarking framework was proposed, incorporating precision-recall analysis, latency profiling, and memory usage evaluation. Furthermore, strategic directions for future research were outlined, including the integration of federated learning to preserve data privacy and the development of lightweight hybrid models tailored for edge deployment. This review provides a data-driven foundation for the design of scalable, resource-efficient, and privacy-preserving IDPS solutions within next-generation IIoT smart metering systems.

Accurate smoke detection in complex industrial environments, such as chemical plants, remains a significant challenge due to the inherently low contrast, transparency, and weak texture features of smoke, which often exhibits blurred boundaries and diverse spatial scales. To address these limitations, YOLOv8n-AM, an enhanced lightweight detection framework belonging to the YOLO (You Only Look Once) series, was developed by integrating advanced architectural components into the baseline YOLOv8n model. Specifically, the conventional Spatial Pyramid Pooling-Fast (SPPF) module was replaced with an Attention-based Intra-scale Feature Interaction (AIFI) Convolution Synergistic Feature Processing Module (SFPM), i.e., AIFC-SFPM, enabling more effective semantic feature representation and an improvement in detection accuracy. In parallel, the original convolutional module was optimized using a Multi-Scale Downsampling (MSDown) module, which reduces model redundancy and computational overhead, increasing the detection speed. Experimental evaluations demonstrate that the YOLOv8n-AM model achieves a 1.7% improvement in mean Average Precision (mAP), accompanied by a 9.1% reduction in Giga Floating-point Operations Per Second (GFLOPs) and a 15.4% decrease in parameter count when compared to the original YOLOv8n framework. These improvements collectively underscore the model’s suitability for real-time deployment in resource-constrained industrial settings where rapid and reliable smoke detection is critical. The proposed architecture thus provides a computationally efficient and high-precision solution for safety-critical visual monitoring applications.

Click fraud is a deceptive malicious strategy that relies on repetitive mimicking of human clicking on online advertisements, without actual intention to complete a purchase. This fraud can result in significant financial loses for both advertising companies and marketers, and at the same time destroying their public images. Nevertheless, detection of these illegitimate clicks is very challenging as they closely resemble to authentic human engagement. This study examines the utilization of artificial intelligence approaches to detect deceptive clicks, by identifying subtle correlations between the timing of the clicks, taking into account their geographical or network sources and linked application sources as indicators to separate legitimate from malicious activity. This study highlights the application of recurrent neural networks (RNNs) for this task, keeping in mind that the process of selection and tuning of the model's hyperparameters plays a vital role in the performance. An adapted implementation of crayfish optimization algorithm (COA) was consequently proposed in this paper, and used to optimize RNN models to enhance their general performance. The developed framework was evaluated utilizing actual operational datasets and yielded encouraging outcomes.