Mohammad Modarres

Hossein Jahandideh, Ardavan Asef-Vaziri, Mohammad Modarres

In this paper, we present a heuristic procedure for designing a loop material flow pattern on a given facility layout with the aim of minimizing the total material handling distances. We present an approximation of the total material handling costs and greatly drop the required computational time by minimizing the approximation instead of the original objective function.

Sina Abbasi, Mohammad Reza Modarres, Mohammad Taher Pilehvar

With new large language models (LLMs) emerging frequently, it is important to consider the potential value of model-agnostic approaches that can provide interpretability across a variety of architectures. While recent advances in LLM interpretability show promise, many rely on complex, model-specific methods with high computational costs. To address these limitations, we propose NormXLogit, a novel technique for assessing the significance of individual input tokens. This method operates based on the input and output representations associated with each token. First, we demonstrate that during the pre-training of LLMs, the norms of word embeddings effectively capture token importance. Second, we reveal a significant relationship between a token's importance and the extent to which its representation can resemble the model's final prediction. Extensive analyses reveal that our approach outperforms existing gradient-based methods in terms of faithfulness and offers competitive performance in layer-wise explanations compared to leading architecture-specific techniques.

Saman Marandi, Yu-Shu Hu, Mohammad Modarres

In this paper, we present a novel diagnostic framework that integrates Knowledge Graphs (KGs) and Large Language Models (LLMs) to support system diagnostics in high-reliability systems such as nuclear power plants. Traditional diagnostic modeling struggles when systems become too complex, making functional modeling a more attractive approach. Our approach introduces a diagnostic framework grounded in the functional modeling principles of the Dynamic Master Logic (DML) model. It incorporates two coordinated LLM components, including an LLM-based workflow for automated construction of DML logic from system documentation and an LLM agent that facilitates interactive diagnostics. The generated logic is encoded into a structured KG, referred to as KG-DML, which supports hierarchical fault reasoning. Expert knowledge or operational data can also be incorporated to refine the model's precision and diagnostic depth. In the interaction phase, users submit natural language queries, which are interpreted by the LLM agent. The agent selects appropriate tools for structured reasoning, including upward and downward propagation across the KG-DML. Rather than embedding KG content into every prompt, the LLM agent distinguishes between diagnostic and interpretive tasks. For diagnostics, the agent selects and executes external tools that perform structured KG reasoning. For general queries, a Graph-based Retrieval-Augmented Generation (Graph-RAG) approach is used, retrieving relevant KG segments and embedding them into the prompt to generate natural explanations. A case study on an auxiliary feedwater system demonstrated the framework's effectiveness, with over 90% accuracy in key elements and consistent tool and argument extraction, supporting its use in safety-critical diagnostics.

Mohammad Reza Modarres, Sina Abbasi, Mohammad Taher Pilehvar

Advances in dataset analysis techniques have enabled more sophisticated approaches to analyzing and characterizing training data instances, often categorizing data based on attributes such as ``difficulty''. In this work, we introduce RepMatch, a novel method that characterizes data through the lens of similarity. RepMatch quantifies the similarity between subsets of training instances by comparing the knowledge encoded in models trained on them, overcoming the limitations of existing analysis methods that focus solely on individual instances and are restricted to within-dataset analysis. Our framework allows for a broader evaluation, enabling similarity comparisons across arbitrary subsets of instances, supporting both dataset-to-dataset and instance-to-dataset analyses. We validate the effectiveness of RepMatch across multiple NLP tasks, datasets, and models. Through extensive experimentation, we demonstrate that RepMatch can effectively compare datasets, identify more representative subsets of a dataset (that lead to better performance than randomly selected subsets of equivalent size), and uncover heuristics underlying the construction of some challenge datasets.

Sergio Cofre-Martel, Enrique Lopez Droguett, Mohammad Modarres

Sensor monitoring networks and advances in big data analytics have guided the reliability engineering landscape to a new era of big machinery data. Low-cost sensors, along with the evolution of the internet of things and industry 4.0, have resulted in rich databases that can be analyzed through prognostics and health management (PHM) frameworks. Several da-ta-driven models (DDMs) have been proposed and applied for diagnostics and prognostics purposes in complex systems. However, many of these models are developed using simulated or experimental data sets, and there is still a knowledge gap for applications in real operating systems. Furthermore, little attention has been given to the required data preprocessing steps compared to the training processes of these DDMs. Up to date, research works do not follow a formal and consistent data preprocessing guideline for PHM applications. This paper presents a comprehensive, step-by-step pipeline for the preprocessing of monitoring data from complex systems aimed for DDMs. The importance of expert knowledge is discussed in the context of data selection and label generation. Two case studies are presented for validation, with the end goal of creating clean data sets with healthy and unhealthy labels that are then used to train machinery health state classifiers.

Sergio Cofre-Martel, Enrique Lopez Droguett, Mohammad Modarres

Deep learning (DL) has become an essential tool in prognosis and health management (PHM), commonly used as a regression algorithm for the prognosis of a system's behavior. One particular metric of interest is the remaining useful life (RUL) estimated using monitoring sensor data. Most of these deep learning applications treat the algorithms as black-box functions, giving little to no control of the data interpretation. This becomes an issue if the models break the governing laws of physics or other natural sciences when no constraints are imposed. The latest research efforts have focused on applying complex DL models to achieve a low prediction error rather than studying how the models interpret the behavior of the data and the system itself. In this paper, we propose an open-box approach using a deep neural network framework to explore the physics of degradation through partial differential equations (PDEs). The framework has three stages, and it aims to discover a latent variable and corresponding PDE to represent the health state of the system. Models are trained as a supervised regression and designed to output the RUL as well as a latent variable map that can be used and interpreted as the system's health indicator.