IndustryAssetEQA
Neurosymbolic Operational Intelligence for Industrial Asset Maintenance
Click to view presentation
IndustryAssetEQA: Neurosymbolic Operational Intelligence for Embodied Question Answering in Industrial Asset Maintenance
Project Description: IndustryAssetEQA is a neurosymbolic operational intelligence system for embodied question answering in industrial asset maintenance. The system addresses a critical limitation of LLM-based maintenance assistants: they often generate fluent but generic explanations that are weakly grounded in telemetry, lack verifiable provenance, and fail to support counterfactual or action-oriented reasoning in safety-critical environments. IndustryAssetEQA combines episode-centric telemetry representations with a Failure Mode and Effects Analysis Knowledge Graph (FMEA-KG) to support grounded reasoning over industrial assets. The system is evaluated across four industrial asset types, including rotating machinery, turbofan engines, hydraulic systems, and cyber-physical production systems. Compared with LLM-only baselines, IndustryAssetEQA improves structural validity, counterfactual accuracy, and explanation entailment, while reducing severe expert-rated overclaims from 28% to 2%. Code, datasets, and the FMEA-KG are available at this link.
AssetOpsBench
Real-World Benchmark for AI-Driven Task Automation in Industrial Asset Management
Click to open demo
AssetOpsBench: A Real-World Evaluation Benchmark for AI-Driven Task Automation in Industrial Asset Management
Project Description: AssetOpsBench is a real-world evaluation benchmark for AI-driven task automation in industrial asset lifecycle management. The project targets complex operational workflows such as condition monitoring, maintenance planning, and industrial decision support, where traditional AI and machine learning systems often solve narrow tasks in isolation. AssetOpsBench provides a unified framework for orchestrating and evaluating domain-specific agents for Industry 4.0. It includes a multimodal ecosystem with four domain-specific agents, a curated dataset of 140+ human-authored natural-language queries grounded in real industrial scenarios, and a simulated CouchDB-backed IoT environment. The benchmark introduces automated evaluation metrics for comparing architectural trade-offs between Tool-As-Agent and Plan-Executor paradigms, while also supporting systematic discovery of emerging failure modes. The platform has seen broad community adoption, with 250+ users and more than 500 submitted agents through the public benchmarking demo. Mode and Effects Analysis Knowledge Graph (FMEA-KG) to support grounded reasoning over industrial assets. The system is evaluated across four industrial asset types, including rotating machinery, turbofan engines, hydraulic systems, and cyber-physical production systems. Compared with LLM-only baselines, IndustryAssetEQA improves structural validity, counterfactual accuracy, and explanation entailment, while reducing severe expert-rated overclaims from 28% to 2%. More details in link.
CausalTrace
Neurosymbolic Causal Analysis Agent for Smart Manufacturing
Click to view lecture
CausalTrace: A Neurosymbolic Causal Analysis Agent for Smart Manufacturing
Project Description: CausalTrace is a neurosymbolic causal analysis module integrated into the SmartPilot industrial CoPilot for smart manufacturing. The project addresses a key limitation of existing AI systems in industrial environments: many tools provide predictions but remain black-box systems that do not connect prediction, explanation, causal reasoning, root-cause analysis, and intervention support in a unified decision-support workflow. CausalTrace performs data-driven causal analysis enriched with industrial ontologies and knowledge graphs. It supports causal discovery, counterfactual reasoning, and root-cause analysis while enabling real-time operator interaction through transparent and explainable decision support. The system was evaluated using multiple causal assessment methods and the C3AN framework, covering robustness, intelligence, and trustworthiness. In an academic rocket assembly testbed, CausalTrace achieved strong agreement with domain experts and high root-cause analysis performance, including MAP@3 of 94%, PR@2 of 97%, MRR of 0.92, and Jaccard similarity of 0.92. It also achieved a C3AN evaluation score of 4.59/5, demonstrating reliability for live deployment.
CausalPulse: Multi-Agent Causal Diagnostics Copilot for Smart Manufacturing
Project Description: CausalPulse is an industry-grade multi-agent copilot for real-time, trustworthy, and interpretable root-cause analysis in smart manufacturing. The system unifies anomaly detection, causal discovery, and causal reasoning within a neurosymbolic architecture built on standardized agentic protocols. Unlike traditional analytics pipelines that treat anomaly detection, causal inference, and root-cause analysis as separate stages, CausalPulse integrates these capabilities into a modular human-in-the-loop workflow for scalable industrial diagnostics. The system is being deployed in a Robert Bosch manufacturing plant and integrates with existing monitoring workflows for production-scale operation. Evaluations on public Future Factories data and proprietary Planar Sensor Element datasets demonstrate high reliability, achieving overall success rates of 98.0% and 98.73%, respectively. Per-criterion evaluations show strong performance in planning and tool use, self-reflection, and collaboration, while runtime experiments report end-to-end diagnostic latency of 50-60 seconds with near-linear scalability. These results demonstrate CausalPulse’s potential as a production-ready causal diagnostics copilot for next-generation manufacturing.
SmartPilot: Agent-Based CoPilot for Intelligent Manufacturing
Project Description: In the fast-changing landscape of Industry 4.0, achieving efficiency, precision, and adaptability is crucial for optimizing manufacturing operations. While foundational models have shown promise in delivering such advanced solutions in certain domains, they often face challenges in industrial applications due to certain limitations. This work introduces SmartPilot, a custom, compact, and neurosymbolic co-pilot designed to achieve the above metrics in modern manufacturing processes. By leveraging a domain-specific, right-sized solution optimized for edge devices, SmartPilot enhances real-time decision-making capabilities on the factory floor. The system employs an agent-based architecture to predict anomalies, forecast demand, and provide domain-specific question-and-answer support. These agents operate within a robust architecture that ensures seamless interoperability with existing systems. This solution ensures the trustworthiness of manufacturing processes while optimizing for performance, reliability, safety, and efficiency. SmartPilot is currently deployed in two manufacturing facilities focused on rocket assembly and Vegemite production. The dataset, codes to reproduce the results, and supplementary materials are available at this link.
NSF-MAP: Neurosymbolic Multimodal Fusion for Robust and Interpretable Anomaly Prediction in Assembly Pipelines
Project Description: In modern assembly pipelines, identifying anomalies is crucial in ensuring product quality and operational efficiency. Conventional single-modality methods fail to capture the intricate relationships required for precise anomaly prediction in complex predictive environments with abundant data and multiple modalities. This project proposes a neurosymbolic AI and fusion-based approach for multi-modal anomaly prediction in assembly pipelines. We introduce a time series and image-based fusion mechanism that leverages decision-level fusion techniques. link.