Sameer Prabhu, Ravdeep Kour, Pierre Dersin, Ramin Karim
Ore processing in the mining industry depends on the efficient and continuous operation of large, heavy and interconnected machinery or assets throughout the processing plant. A reduction in asset performance can lead to downtime, production loss, safety risks, and an increase in maintenance costs. Within the broader transition towards intelligent asset management, the convergence of digital technologies, data analytics, and machine learning has positioned data-driven approaches as a central enabler for efficient maintenance of industrial assets. Prognostics and Health Management (PHM) is an enabler of modern maintenance strategies and has demonstrated substantial benefits in industries such as aerospace, automotive, railway, and manufacturing. Processing machines in mining exhibit similar characteristics, including complex multicomponent interactions, dynamic behaviour, substantial mechanical stress, and operation under harsh conditions. However, the mining processing environment also presents challenges like heterogeneous and abrasive raw materials, interdependent components and cascading effects, and domain-specific machines. These factors motivate the need to understand how PHM methods have been explored and applied to machines within the mining processing domain. The objective of this paper is to systematically review data-driven PHM approaches applied to mining processing machines published since 2015. The review synthesizes research studies according to the three PHM tasks, fault detection, diagnostics, and prognostics, focusing on associated data modalities, modelling methods, equipment types, and health indicators. The observations reveal that existing PHM research in mining processing machines is predominantly focused on component-level fault detection and diagnostics tasks, vibration-based sensing, and supervised learning approaches. System-level health management, where prognostics information is translated into maintenance or operational decision-making, is rarely addressed. Additional gaps include insufficient exploration of multimodal sensing and approaches suitable for scarce fault data. By consolidating trends and highlighting research needs, this review provides a timely reference for advancing PHM in mining processing machines and supporting improved system availability and operational continuity.
Saiful Islam Sagor, Tania Haghighi, Md Rahatuzzaman, Anasheh Khecho, Minhaj Nur Alam, Jacob Zorn, Erina Baynojir Joyee
General-purpose large language models (LLMs) often struggle to generate reliable responses in specialized engineering domains due to limited domain grounding and insufficient exposure to structured technical knowledge. This study investigates practical strategies for adapting a foundation LLM to the additive manufacturing (AM) domain in order to improve answer accuracy, relevance and usability for expert-level question answering. AM knowledge is distributed across heterogeneous sources such as academic literature, manufacturer documentation, technical standards and procedural guides. Although general LLMs demonstrate strong linguistic capabilities, they frequently fail to retrieve and contextualize such domain-specific information. Two common approaches to address this limitation are domain-specific fine-tuning and retrieval-augmented generation (RAG). A curated AM corpus is constructed and used to evaluate three configurations based on LLaMA-3-8B: (1) a pretrained baseline model, (2) a RAG system that retrieves relevant document chunks from a vector database and (3) a model fine-tuned on raw domain text using quantized low-rank adaptation (QLoRA). Performance is evaluated using 200 expert-designed AM questions assessed by mechanical engineering experts for accuracy, relevance and overall preference. The RAG model consistently outperforms the baseline. Among the 200 questions, 75.5% of RAG responses are judged more accurate, 85.2% are preferred overall and 90.8% are rated more relevant than baseline responses. In contrast, fine-tuning on raw AM text reduces performance, producing more accurate answers in only 5.6% of cases and more relevant answers in 32.5% of cases. These results indicate that retrieval-augmented approaches provide a more effective pathway for adapting LLMs to specialized engineering domains than naive fine-tuning on unstructured technical data. RAG grounds responses in retrieved source documents at inference time, enabling higher accuracy and relevance without requiring model retraining.
Altynay Smagulova, Aleksandar Vakanski, John W. Merickel, Isshu Lee, Charith Purushotham, Yalei Tang, Alberto E. Cerpa, Fei Xu, Rongjie Song
Experimental tensile testing of sub-sized nuclear structural materials is often limited by the high cost, time requirements, and scarcity of available data. These constraints lead to sparse and imbalanced datasets that challenge the development of reliable predictive models. This study investigates the effectiveness of machine learning (ML) algorithms and data augmentation strategies for predicting tensile properties of sub-sized Stainless Steel 316 specimens. A dataset comprising 396 tensile testing records and 45 input features describing material composition, processing history, irradiation conditions, specimen geometry, and testing parameters was assembled from the literature. Nine ML algorithms were evaluated for predicting ultimate tensile strength, yield strength, total elongation, and uniform elongation. Two data augmentation methods, Generative Adversarial Networks (GAN) and Synthetic Minority Oversampling Technique for Regression with Gaussian Noise (SMOGN), were applied at multiple augmentation ratios. Model performance was assessed using five-fold cross-validation with Pearson correlation coefficient (r), coefficient of determination (R 2), and root-mean-square error (RMSE). Tree-based ensemble methods achieved the highest predictive performance across most tensile properties. Random Forest and Extreme Gradient Boosting (XGBoost) attained Pearson correlation coefficients exceeding 0.98 for yield strength and uniform elongation. Data augmentation effects were property dependent: GAN augmentation generally improved predictions of strength-related properties, while SMOGN yielded greater benefits for elongation-related properties. The best augmentation configurations improved predictive accuracy and model robustness, particularly for sparse regions of the feature space. Physics-Informed Neural Networks incorporating empirical constraints underperformed conventional ML approaches, exhibiting approximately 10% lower correlation and higher prediction errors. The results demonstrate that combining suitable ML algorithms with property-specific data augmentation strategies enables accurate prediction of tensile properties from limited experimental data. The differing effectiveness of GAN and SMOGN suggests that augmentation strategies should be selected according to the physical characteristics of the target property. The proposed framework provides a practical approach for addressing data scarcity in nuclear materials research and supports accelerated qualification and evaluation of nuclear structural materials.
Robotic automation is a key technology that increases the efficiency and flexibility of manufacturing processes. However, one of the challenges in deploying robots in novel environments is finding the optimal base pose for the robot, which affects its reachability and deployment cost. Yet, existing research on automatically optimizing the base pose of robots has not been compared. We address this problem by optimizing the base pose of industrial robots with Bayesian optimization (BO), exhaustive search (ES), genetic algorithms (GAs), and stochastic gradient descent (SGD), and we find that all algorithms can reduce the cycle time for various evaluated tasks in synthetic and real-world environments. Stochastic gradient descent shows superior performance with regard to the success rate, solving more than 90 % of our real-world tasks, while genetic algorithms show the lowest final costs. All benchmarks and implemented methods are available as baselines against which novel approaches can be compared.
Lars Leyendecker, Ana Maria Gonzalez Degetau, Katharina Bata, Jessica Emonts, Angela Schmitz, Robert H. Schmitt
In production engineering, the identification of optimal process parameters is essential to advance product quality and overall equipment effectiveness. Optimizing and adapting process parameters through experimental design is relevant for different phases of the life cycle of a production process: (i) design and development of new processes, (ii) failure analysis and optimization, and (iii) adaptation and calibration in series production. Existing experimental design approaches tend to be inefficient because they comprise static, non-adaptive methodologies that separate experiment design from execution and analysis. Instead, Bayesian Optimization (BO) offers an adaptive and data-efficient methodology for experimental design termed Bayesian experimental design (BED). In BED, the selection of an experiment is re-evaluated in each iteration based on previous experiment results according to an acquisition function that aims to maximize the informational content of each experiment. However, the configuration of BO algorithms for specific optimization problems requires extensive knowledge of both BO and process characteristics. The mean and covariance functions of the surrogate model, the acquisition function, and initial data sampling must be individually configured and significantly influence overall optimization performance, preventing widespread adoption in production engineering practice. To guide the configuration of BO algorithms for optimizing production processes, in this paper, we perform an extensive benchmark study with a total of 15,360 experiments. We evaluate the performance of a variety of BO algorithm configurations (including kernels, acquisition functions, and initial sampling sizes) on a total of eight optimization problems with a noiseless and a noisy variant each. The performance and robustness analysis reveals significant performance differences between individual BO algorithm configurations. The results of our benchmarking serve as empirical references based on which we derive actionable guidelines for the application of BED in production engineering.
