MUMBAI, India, Oct. 12 -- Intellectual Property India has published a patent application (202621101553 A) filed by Dr. D. Y. Patil Institute Of Technology, Pimpri, Pune -; and Dnyaan Prasad Global University, School Of Technology And Research on August 22, 2026, for Adaient – Ai-Powered Industrial Troubleshooting System Using Rag Architecture.
Inventors include Dr Madhavi Repe; Omkar Nikam; Shubham Linge; Atharva Taras; and Kaustubh Bhoir.
The application for the patent was published on October 09, 2026, under issue no. 41/2026.
Abstract: The current invention deals with an AI driven industrial troubleshooting system by using Retrieval-Augmented Generation (RAG) architecture which is implemented on a web based maintenance support platform. The objective of this system is to automatically answer maintenance technician's natural language troubleshooting questions with accurate and source cited answers generated with a locally hosted large language model after retrieving necessary information from indexed PDF equipment manual documents. The main intention of the current invention is to optimize the performance and accuracy of industrial maintenance with zero reliance on manual document search to reduce average troubleshooting time from 42 minutes per issue to 15 minutes. The retrieval process uses semantic vector search technology where chunks from the PDF manual documents are converted into 384-bit dimension vector embeddings using all-MiniLM-L6-v2 Sentence-Transformer and stored in a FAISS vector database which is capable of providing sub millisecond cosine similarity retrieval. The smart application infrastructure consists of a FastAPI web server and a FAISS vector index and the Llama 3.2 (3 billion parameters) large language model is run on the user's system with the Ollama inference engine and communicated with others via RESTful APIs and there is no reliance on any external cloud services. The control system involves role-based access control and EV owner equivalent Superadmin, Admin and Technician role based on real-time data governance along with a detailed audit log of each user interaction. The system uses a dynamic query processing mechanism considering the document content, query semantic and the retrieved context to ensure better answer quality and relevancy. The smart user interface enables the technicians to query and analyze the system's source-cited answers on a secure web chat interface. The evaluation shows 94% of retrieval relevance, 92% of answer correctness and 98% of source attribution. The system solves the current industrial maintenance system challenges like the problem of non-access to information, long manual document search time, loss of knowledge when an experienced technician retires and can support integration in the smart factory solution to provide analytics based on query trends, manual utilization and system usage. The proposed system supports 50 and above concurrent user sessions and this invention can be utilized in different industries and it's also capable of managing a number of user sessions without performance reduction on conventional server hardware.
Disclaimer: Curated by HT Syndication.
