Adapting LLM agents to domain-specific tool calling remains notably brittle under evolving interfaces. Prompt and schema engineering is easy to deploy but often fragile under distribution shift and strict parsers, while continual parameter-efficient fine-tuning improves reliability at the cost of training, maintenance, and potential forgetting. We identify a critical Lazy Agent failure mode where tool necessity is nearly perfectly decodable from mid-layer activations, yet the model remains conservative in entering tool mode, revealing a representation-behavior gap. We propose Activation Steering Adapter (ASA), a training-free, inference-time controller that performs a single-shot mid-layer intervention and targets tool domains via a router-conditioned mixture of steering vectors with a probe-guided signed gate to amplify true intent while suppressing spurious triggers. On MTU-Bench with Qwen2.5-1.5B, ASA improves strict tool-use F1 from 0.18 to 0.50 while reducing the false positive rate from 0.15 to 0.05, using only about 20KB of portable assets and no weight updates.
Carlos Rafael Catalan, Lheane Marie Dizon, Patricia Nicole Monderin
et al.
Over-reliance on AI systems can undermine users' critical thinking and promote complacency, a risk intensified by the emergence of agentic AI systems that operate with minimal human involvement. In software engineering, agentic coding assistants (ACAs) are rapidly becoming embedded in everyday development workflows. Since software engineers (SEs) create systems deployed across diverse and high-stakes real-world contexts, these assistants must function not merely as autonomous task performers but as Tools for Thought that actively support human reasoning and sensemaking. We conducted a formative study examining software engineers' cognitive engagement and sensemaking processes when working with an ACA. Our findings reveal that cognitive engagement consistently declines as tasks progress, and that current ACA designs provide limited affordances for reflection, verification, and meaning-making. Based on these findings, we identify concrete design opportunities leveraging richer interaction modalities and cognitive-forcing mechanisms to sustain engagement and promote deeper thinking in AI-assisted programming.
Emotion recognition through facial expressions is crucial in fields like healthcare, entertainment, and education, offering insights into user experiences. In online learning, traditional methods fail to capture students’ emotions effectively. This research introduces a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model to recognize learning emotions (interest, boredom, and confusion) during online lectures. A custom dataset was constructed by mapping action units from FER2013, CK+48, and JAFFE datasets into three learning-related categories. Images were preprocessed (grayscale conversion, resizing, normalization) and divided into training and testing sets. The CNN layers extract spatial facial features, while the LSTM layers capture temporal dependencies across video frames. Evaluation metrics included accuracy, precision, recall, and F1-score. The model achieved 98.0% accuracy, 97% precision, 98% recall, and 98% F1-score, surpassing existing CNN-only methods. This advancement enhances online learning by enabling personalized support and has applications in education, psychology, and human–computer interaction, contributing to affective computing development.
Jesús Gerardo Ávila-Sánchez, Manuel de Jesús López-Martínez, Valeria Maeda-Gutiérrez
et al.
The Cutting Development Chamber (CDC) design is presented as an innovative solution to crucial human challenges, such as food and plant medicinal production. Unlike conventional propagation chambers, the CDC is a much more comprehensive research tool, specifically designed to optimize plant reproduction from cuttings. It maintains precise control over humidity, temperature, and lighting, which are essential parameters for plant development, thus maximizing the success rate, even in difficult-to-propagate species. Its modular design is one of its main strengths, allowing users to adapt the chamber to their specific needs, whether for research studies or for larger-scale propagation. The most distinctive feature of this chamber is its ability to collect detailed, labeled data, such as images of plant growth and environmental parameters that can be used in artificial intelligence tasks, which differentiate it from chambers that are solely used for propagation. A study that validated and calibrated the chamber design using cuttings of various species demonstrated its effectiveness through descriptive statistics, confirming that CDC is a powerful tool for research and optimization of plant growth. In validation experiments (<i>Aloysia citrodora</i> and <i>Stevia rebaudiana</i>), the system generated 6579 labeled images and 67,919 environmental records, providing a robust dataset that confirmed stable control of temperature and humidity while documenting cutting development.
Engineering machinery, tools, and implements, Technological innovations. Automation
In this study, a rubber vibration isolator is designed for certain aerospace equipment, and a finite element simulation is carried out to obtain the modal frequency and random vibration response, and to verify the accuracy of the design. The test verifies that there is no amplification of vibration within 100 Hz; the damping efficiency values of vertical and horizontal random vibration are, respectively, 42.12% and 40.54%; and the impact isolation rate is more than 80%. The test results show that the vibration isolation buffer effect of the isolator is satisfactory and meets the design requirements.
Enhancing the sustainability of manufacturing systems requires reducing product defects through effective management of risks that impact product quality. A crucial component in minimizing defects is the adoption of robust risk management strategy. This study examines risk mitigation in the tofu production process to reduce product defects, by employing the House of Risk (HOR) framework to prioritize mitigation efforts. Data were collected through observations, in depth interviews, and focus group discussions, following the two-step HOR methodology. The analysis identified 12 risk events and seven risk agents, along with six prioritized mitigation strategies, based on the Aggregate Risk Potential (ARP) ranking of the identified risk agents. The highest-priority strategy involves developing standardized work instructions for the tofu production process. This study offers practical insights for companies seeking to lower defect rates, thereby supporting the sustainability of their manufacturing systems.
Enrique Pujada-Gamarra, Daniel Lavayen-Farfán, Davy Olivera-Oliva
et al.
In recent years, ecofriendly and renewable energy solutions have gained relevance mainly to lessen the effects of climate change. Governments and companies across the world have commitments to reduce fuel consumption and emissions as part of the 2030 Sustainable Development Goals. Solar energy systems have great importance as a renewable energy source; however, they often have large space requirements to be effective, e.g., large areas covered by solar panels, as well as low efficiency and strong dependance on the weather. On the other hand, origami, the art of folding paper, can be a source of inspiration for new technologies and solutions for modern problems. In this paper, origami-inspired solar panels are presented as a potential solution for naval and mining operations. Prototype panels are manufactured based on the Miura-Ori pattern. Using this pattern, the photovoltaic modules can be folded by just one movement, thus reducing their footprint by up to 90%. The prototype photovoltaic modules are then tested on land and on board a vessel, where their efficiency and resistance can be tested. It is shown that naval and mining operations, where fuel consumption can be extremely high and available space is a major constraint, benefit greatly from this kind of development.
The automotive industry generates vast amounts of data from sensors, telemetry, diagnostics, and real-time operations. Efficient data engineering is critical to handle challenges of latency, scalability, and consistency. Modern data lakehouse formats Delta Parquet, Apache Iceberg, and Apache Hudi offer features such as ACID transactions, schema enforcement, and real-time ingestion, combining the strengths of data lakes and warehouses to support complex use cases. This study presents a comparative analysis of Delta Parquet, Iceberg, and Hudi using real-world time-series automotive telemetry data with fields such as vehicle ID, timestamp, location, and event metrics. The evaluation considers modeling strategies, partitioning, CDC support, query performance, scalability, data consistency, and ecosystem maturity. Key findings show Delta Parquet provides strong ML readiness and governance, Iceberg delivers high performance for batch analytics and cloud-native workloads, while Hudi is optimized for real-time ingestion and incremental processing. Each format exhibits tradeoffs in query efficiency, time-travel, and update semantics. The study offers insights for selecting or combining formats to support fleet management, predictive maintenance, and route optimization. Using structured datasets and realistic queries, the results provide practical guidance for scaling data pipelines and integrating machine learning models in automotive applications.
Zeynep G Akdemir-Beveridge, Arash Zaghi, Connie Syharat
Creativity is essential in engineering education, enabling students to develop innovative and practical solutions. However, assessing creativity remains challenging due to a lack of reliable, domain-specific tools. Traditional assessments like the Torrance Tests of Creative Thinking (TTCT) may not fully capture the complexity of engineering creativity. This study introduces and validates the Engineering Creativity Assessment Tool (ECAT), designed specifically for engineering contexts. ECAT was tested with 199 undergraduate students who completed a hands-on design task. Five trained raters evaluated the products using the ECAT rubric. Exploratory and confirmatory factor analyses supported a four-factor structure: fluency, originality, cognitive flexibility, and creative strengths. Reliability was high, convergent and discriminant validity were examined using TTCT scores, revealing moderate correlations that support ECATs domain specificity. ECAT offers a reliable, valid framework for assessing creativity in engineering education and provides actionable feedback to educators. Future work should examine its broader applicability across disciplines and instructional settings.
Khorolsuren Tuvshinbayar, Nonsikelelo Sheron Mpofu, Thomas Berger
et al.
Possibilities to perform 3D printing directly on textile fabrics have been investigated intensively during the last decade. Usually, fused deposition modeling (FDM) printing with often inexpensive 3D printers is applied in these experiments. Several studies revealed the influence of textile fabrics, FDM polymers and printing parameters, indicating that not all combinations of fabrics and printing materials are suitable for this task. Recently, first approaches to use stereolithography (SLA) or PolyJet Modeling (PJM) directly on textile fabrics have been reported. Here, the first comparison of the adhesion forces reached by FDM and SLA printing on different woven fabrics is shown, revealing significantly better adhesion for SLA printing.
Textile bleaching, dyeing, printing, etc., Engineering machinery, tools, and implements
Takahiro EINAGA, Hiroki SAKAKIMA, Asuka HATANO
et al.
American football is characterized by its intense collision, which can induce serious injuries such as concussions and head trauma. Therefore, the evaluation of the safety of players is mandatory. In this study, we have developed a new finite element analysis model to evaluate the head kinematic response in American football. The novelty of our model is the incorporation of the effect of cervical muscle strength and pre-impact cervical muscle activation on head kinematics response. This model is based on the head-neck model extracted from the Total HUman Model for Safety (THUMS). The finite element analysis of the hit movement, which is frequently seen during the American football practices, is performed with four different cervical muscle conditions; (i) early pre-impact activation, (ii) late pre-impact activation, (iii) weak muscle strength, and (iv) No pre-impact activation. We also performed an experiment to measure head kinematics response and torso rotation in the hit movement of American football players by mouthguard sensors and video motion capture system for model validation. The four conditions parameter study shows that the time history of head angular velocity with pre-impact muscle activation is better agreement with our experimental results. Therefore, it is important to model the pre-impact cervical muscle activation in the finite element analysis for evaluating the head kinematic response accurately. We also found that higher cervical muscle strength and earlier cervical muscle activation decrease the head angular acceleration at 20-35 ms after the head collision. This suggests that, in the case of low-intensity head collision, strengthening cervical muscles and activating them before the head impact by anticipating the collision can reduce head angular acceleration.
Mechanical engineering and machinery, Engineering machinery, tools, and implements
Precise position tracking is crucial for autonomous vehicle development. Technological innovations offer more accurate sensors, but the ideal combination is still debated. This study, inspired by a motorsport project, analyzes errors in simpler positioning solutions, examining environmental interference and dynamic movement impacts. It discusses real-world testing challenges and explores improving precision through sensor combinations. Using high-accuracy GPS and accelerometers with LIDAR, radar, or cameras enhances navigation, especially in complex environments. The study emphasizes that the future of autonomous vehicle localization depends on technological advancements, sensor integration, and intelligent algorithms.
The current technology landscape lacks a foundational AI model for solving process engineering calculations. In this work, we introduce a novel autonomous agent framework leveraging Retrieval-Augmented Instruction-Tuning (RAIT) to enhance open, customizable small code language models (SLMs) for these calculations. By combining instruction tuned code SLMs with Retrieval-Augmented Code Generation (RACG) using external tools, the agent generates, debugs, and optimizes code from natural language specifications. Our approach addresses the limitations of the current lack of a foundational AI model for specialized process engineering tasks and offers benefits of explainability, knowledge editing, and cost-effectiveness. Additionally, we curate custom datasets of chemical and process engineering problems and solutions to overcome data scarcity. Experimental results show that our framework matches the performance of large-scale proprietary models on benchmark datasets, proving its effectiveness and usability.
Jinqi Luo, Tianjiao Ding, Kwan Ho Ryan Chan
et al.
Large Language Models (LLMs) are being used for a wide variety of tasks. While they are capable of generating human-like responses, they can also produce undesirable output including potentially harmful information, racist or sexist language, and hallucinations. Alignment methods are designed to reduce such undesirable outputs via techniques such as fine-tuning, prompt engineering, and representation engineering. However, existing methods face several challenges: some require costly fine-tuning for every alignment task; some do not adequately remove undesirable concepts, failing alignment; some remove benign concepts, lowering the linguistic capabilities of LLMs. To address these issues, we propose Parsimonious Concept Engineering (PaCE), a novel activation engineering framework for alignment. First, to sufficiently model the concepts, we construct a large-scale concept dictionary in the activation space, in which each atom corresponds to a semantic concept. Given any alignment task, we instruct a concept partitioner to efficiently annotate the concepts as benign or undesirable. Then, at inference time, we decompose the LLM activations along the concept dictionary via sparse coding, to accurately represent the activations as linear combinations of benign and undesirable components. By removing the latter ones from the activations, we reorient the behavior of the LLM towards the alignment goal. We conduct experiments on tasks such as response detoxification, faithfulness enhancement, and sentiment revising, and show that PaCE achieves state-of-the-art alignment performance while maintaining linguistic capabilities.
Yuri Ivanov, Sergey Zhiganov, Mikhail Gorkavyy
et al.
It is suggested that the use an ensemble of deep neural networks can determine the spatial position of the operator using keypoints with a multicamera sensor system. The advantage of the algorithm is the use of a multicamera system that allows keypoints to be linked to the local coordinate system of an industrial robotic complex. The testing of this work was made on the basis of modern embedded computing hardware and software. The effectiveness of the proposed approach is demonstrated even when only a subset of key points is found in the frame, as well as when they partially overlap. A software module in Python has been developed for detecting and localizing key points of the operator and industrial manipulator. The proposed approach will make it possible to plan the robot’s trajectories for the safe execution of joint operations in one workspace. The developed algorithm will be used to predict the operator’s actions in the workspace and detect abnormal situations and possible intersections in the trajectories of the collaborative robot.
Georgii Konoplev, Artur Kuznetsov, Aleksandr Frorip
et al.
A novel simple optical sensor based on fast protein liquid chromatography was developed and tested for monitoring end stage renal disease (ESRD) patients treated with continuous ambulatory peritoneal dialysis (CAPD). The device provides direct determination of proteins and lower molecular weight metabolites in effluent peritoneal dialysate using ultraviolet (UV) photometric detection at the wavelengths 285 nm or 260 nm with deep ultraviolet light-emitting diodes. The sensor was calibrated with bovine serum albumin and nucleotides standard solutions. Chromatograms of peritoneal dialysate samples taken from a group of 28 ESRD patients were processed and approximated by a set of split-Gaussian functions. All chromatograms show three overlapping peaks: the first one represents proteins; the other two peaks probably correspond to mid- and low molecular weight metabolites. Strong correlation was reveled between the area of the first peak and total protein concentration determined by a standard biochemical assay, this makes possible estimation of peritoneal protein loss with a reasonable precision less than 15%. The area of the second peak correlated with dialysate optical density at a wavelength 355–365 nm, associated with the UV absorption of advanced glycation end (AGE) products. The third peak correlated with the optical density of the eluate at a wavelength 255–265 nm, associated with the UV absorption of purines and pyrimidines. Thus, we demonstrated the possibility of estimation of proteins and lower molecular weight metabolites in effluent peritoneal dialysate with the compact and affordable chromatographic optical sensor.
With the increasing predilection for renewable energy sources across the world, the novel idea of a miniature version of a grid, called a microgrid, has emerged. The efficiency and sustainability of a power grid increase by integrating distributed energy resources (DERs). However, designing an optimum protection scheme has become a substantial challenge due to bi-directional power flows and varying fault levels in the microgrid with distributed energy resources (DERs). The existing protection strategies are not capable of dealing with the different operational states and natures of DERs. Therefore, modifications to the conventional protection schemes are required to benefit from the advantages of de-centralized power generation. Optimum co-ordination between the protection devices (PDs) is needed to achieve fast, secure, and reliable protection of the system. This paper proposes a protection philosophy for a renewable-based AC microgrid and validates its resilience by analyzing the response of the system in different faulty scenarios. Moreover, a genetic algorithm (GA) is used to optimize the proposed protection scheme to achieve a cost-effective, resilient, reliable, and long-term solution for sustainable power generation.
Ali Ilghami Kkhosroshahi, Mohammad Bejani, Hadi Pourali
et al.
This study aims to identify the optimal regression techniques for downscaling among ten commonly used methods in climatology, including SVR, LinearSVR, LASSO, LASSOCV, Elastic Net, Bayesian Ridge, RandomForestRegressor, AdaBoost Regressor, KNeighbors Regressor, and XGBRegressor. For the Köppen climate classification system, including A (tropical), B (dry), C (temperate), and D (continental), synoptic station data were collected. Furthermore, for the purpose of downscaling, a general circulation model (GCM) had been utilized. Additionally, to enhance the performance of downscaling accuracy, mutual information (MI) was employed for feature selection. The downscaling performance was evaluated using the coefficient of determination (DC) and root mean square error (RMSE). Results indicate that SVR had superior performance in tropical and dry climates and LassoCV with RandomForestRegressor had better results in temperate and continental climates.