V. Plevris, A. Jimenéz Rios* and G. Papazafeiropoulos, "Chatbots put to the test in math and logic problems: A comparison and assessment of ChatGPT-3.5, ChatGPT-4, and Google Bard", AI, 4(4), Article ID 4040048 (DOI: 10.3390/ai4040048), 2023.

Abstract:
In an age where artificial intelligence is reshaping the landscape of education and problem solving, our study unveils the secrets behind three digital wizards, ChatGPT-3.5, ChatGPT-4, and Google Bard, as they engage in a thrilling showdown of mathematical and logical prowess. We assess the ability of the chatbots to understand the given problem, employ appropriate algorithms or methods to solve it, and generate coherent responses with correct answers. We conducted our study using a set of 30 questions. These questions were carefully crafted to be clear, unambiguous, and fully described using plain text only. Each question has a unique and well-defined correct answer. The questions were divided into two sets of 15: Set A consists of “Original” problems that cannot be found online, while Set B includes “Published” problems that are readily available online, often with their solutions. Each question was presented to each chatbot three times in May 2023. We recorded and analyzed their responses, highlighting their strengths and weaknesses. Our findings indicate that chatbots can provide accurate solutions for straightforward arithmetic, algebraic expressions, and basic logic puzzles, although they may not be consistently accurate in every attempt. However, for more complex mathematical problems or advanced logic tasks, the chatbots’ answers, although they appear convincing, may not be reliable. Furthermore, consistency is a concern as chatbots often provide conflicting answers when presented with the same question multiple times. To evaluate and compare the performance of the three chatbots, we conducted a quantitative analysis by scoring their final answers based on correctness. Our results show that ChatGPT-4 performs better than ChatGPT-3.5 in both sets of questions. Bard ranks third in the original questions of Set A, trailing behind the other two chatbots. However, Bard achieves the best performance, taking first place in the published questions of Set B. This is likely due to Bard’s direct access to the internet, unlike the ChatGPT chatbots, which, due to their designs, do not have external communication capabilities.

Keywords:
Chatbot; AI; logic; mathematics; ChatGPT; GPT-3.5; GPT-4; Google Bard.

T.G. Wakjira, A. Abushanab, M. Shahria Alam, W. Alnahhal* and V. Plevris, "Explainable Machine Learning-Aided Efficient Prediction Model and Software Tool for Bond Strength of Concrete with Corroded Reinforcement", Structures, 59, Article ID 105693 (DOI: 10.1016/j.istruc.2023.105693), 2024.

Abstract:

The bond strength between concrete and reinforcement is crucial for the composite action and serviceability of reinforced concrete (RC) structures. However, it is vulnerable to deterioration from the corrosion of reinforcement bars, especially in marine structures. Thus, a precise and reliable model for the bond strength in corrosive environments is necessary to evaluate the serviceability and structural performance of corroded RC members. This study employs explainable machine learning (ML) techniques to assess the bond strength between concrete and corroded bars. Eight ML models are developed to establish the best predictive model for bond behavior, considering seven input parameters: corrosion level (CL), steel yield strength, compressive strength of concrete, concrete cover-to-bar diameter ratio, bar diameter-to-bonded length ratio, reinforcement type, and test type. The super learner (SL) model, integrating three ML models, outperforms other models and analytical methods with a large R² value (98% on the test set) and minimal statistical errors. The SHapley Additive exPlanation (SHAP) technique identifies CL as the most influential parameter on bond strength, while the reinforcement and test types have the least effect. Finally, a user-friendly graphical user interface (GUI) tool is established to facilitate the practical implementation of the developed model and support accurate bond strength prediction in concrete with steel reinforcement under corrosive environments.

Keywords:
Machine learning; Bond strength; Concrete; Corrosion; SHAP; Graphical user interface.

R. Ehtisham, W. Qayyum, C.V. Camp, V. Plevris*, J. Mir, Q.Z. Khan and A. Ahmad, "Computing the Characteristics of Defects in Wooden Structures using Image Processing and CNN", Automation in Construction, 158, Article ID 105211 (DOI: 10.1016/j.autcon.2023.105211), 2023.

Abstract:

Wood, a time-honored construction material prized for its exceptional properties, has been in use for millennia. Its enduring popularity is attributed to its remarkable strength, aesthetic appeal, and favorable environmental footprint. However, wooden structures are susceptible to various defects and imperfections that pose threats to their structural integrity, durability, and safety. These issues encompass knots, cracks, warping, twisting, decay, insect infestations, and more, all of which, if left unaddressed, can culminate in structural failures. Thus, a comprehensive strategy involving inspection, maintenance, and remediation is indispensable for safeguarding wooden structures. Traditional manual inspections, while effective, are characterized by their resource-intensive nature, entailing significant time and cost investments. This study presents a pioneering approach that leverages Convolutional Neural Networks (CNNs) and Image Processing techniques to revolutionize the assessment of damage in wooden structures using digital imagery. Initially, CNNs are employed to categorize images into three fundamental classes: cracks, knots, and undamaged sections. Subsequently, Image Processing techniques are harnessed to compute precise characteristics of these defects, including parameters such as crack length, width, angle, and the extent of the defective area within knots. The Inception-ResNet-V2 pre-trained model is utilized, fine-tuned and validated with a robust dataset comprising 9000 wooden defect images, evenly distributed across the three aforementioned categories. A prudent division allocates 70% of the dataset for model training, with the remaining 30% reserved for validation. Following successful training, the model demonstrates an impressive overall accuracy of 92% when classifying an independent test set comprising 100 new images. To illustrate the model's performance, two images from each damage category are selected and tested to compute the characteristics of the defects. The quantification error for crack angle is only 0.15%, while it is 0.99% for crack length, and 2% for crack width, demonstrating the high performance of the model. The practical implications of this work are profound. By automating defect assessment in wooden structures, our approach offers significant advantages to industry professionals. It expedites inspections, reduces labor costs, and enhances the accuracy of defect quantification.

Keywords:
Wooden structures; Characteristics of defects; CNN; Image processing.