Vibrational Analysis agent- coral protocol

The Bearing Vibrations Analyst is an AI-driven multi-agent system built on Coral MCP runtime to support predictive maintenance in rotating machinery. It integrates a set of modular agents that work together to process vibration data, detect bearing faults, and generate human-readable reports for maintenance engineers. The workflow begins with the Interface Agent, which provides a simple entry point for users. Engineers can upload vibration recordings in MP3 format along with an optional natural language prompt (e.g., “Check the status of the bearing”). This agent coordinates the full diagnostic pipeline by invoking specialized modules: a DataAgent to load and convert audio signals into mel-spectrograms, a PreprocAgent to normalize input, a ModelAgent (powered by a trained CNN classifier) to predict fault types, and a DecisionAgent to map predictions into actionable maintenance steps. Fault categories include Normal, Ball fault, Inner Race fault, and Outer Race fault. Once the diagnostic decision is produced, the Explain Agent generates a detailed, contextualized report. Leveraging Nebius AI’s large language models (e.g., gpt-oss-20b), the Explain Agent transforms raw predictions and decisions into clear, actionable guidance tailored for maintenance engineers. This removes the need for specialized vibration analysts to interpret raw signals, making the system usable by non-experts on the shop floor. Deployed as native STDIO agents with Coral SDK, both Interface and Explain Agents can be registered in Coral Studio and invoked through a shared registry. The system supports modular extension, meaning that more diagnostic agents (e.g., for gearboxes, pumps, or turbines) can be added without rewriting the pipeline. Overall, the Bearing Vibrations Analyst accelerates condition monitoring, reduces unplanned downtime, and empowers maintenance teams with clear AI-driven fault diagnostics and recommendations.