Knowledge Graph–Based Root Cause Analysis for Intelligent Manufacturing

Introduction With the rapid development of big data and artificial intelligence, new technologies are increasingly empowering the manufacturing sector, especially in quality and reliability where massive raw data enables AI applications. This article shares Daguan Data’s knowledge‑graph‑based root‑cause analysis approach.

1. Development of Intelligent Manufacturing Intelligent manufacturing integrates smart machines with expert knowledge to perform analysis, judgment, reasoning, and decision‑making, extending human labor and transforming traditional automation. In China, the surge in intelligent manufacturing is evident from rising revenues despite declining employee numbers.

2. Quality and Reliability Engineering Quality and reliability are critical to manufacturing, affecting product quality and process stability. Core activities include product planning, design, and production, with examples from chip design (Huawei HiSilicon, Qualcomm) and fabrication (TSMC).

Product planning: define new product types, collect FMEA/FA documents, understand failure mechanisms.

Product design & development: DFMEA, PFMEA, risk assessment, process design.

Production: trial runs, failure localization, expert discussion, FA/FTA updates.

Customer feedback: rapid failure identification and improvement.

3. Root Cause Analysis Methods Root cause analysis (RCA) seeks the fundamental reasons for failures, preventing recurrence. Traditional methods include fishbone diagrams, Pareto analysis, 5‑Why, genetic algorithms, classic machine‑learning (random forest, SVM), deep learning (CNN, RNN), and Bayesian/causal inference.

Challenges of AI‑Based RCA AI‑driven RCA faces difficulties such as data quality, model interpretability, and integration with existing workflows.

4. Knowledge Graphs and RCA Knowledge graphs, combined with causal inference, enable a shift from perception to cognition intelligence, linking disparate manufacturing data (products, equipment, R&D, production) to uncover hidden relationships.

The knowledge‑graph architecture consists of five layers: schema design, construction, storage, application, and user interface. It supports knowledge computation (path analysis, deductive reasoning, community detection) and knowledge inference (e.g., Bayesian statistics for fault tree analysis).

5. Application Cases

5.1 R‑GCN for Multi‑Stage Manufacturing Attribution A relational graph convolutional network (R‑GCN) models complex manufacturing processes, mapping process dependencies and fine‑grained factors into a knowledge graph to assist engineers in fault attribution.

5.2 Daguan Data Knowledge‑Graph Platform The platform integrates semantic parsing, QA, user profiling, and recommendation to enable interactive fault analysis, e.g., answering "What are possible causes of increased battery internal resistance?" and providing personalized knowledge.

6. Q&A

Q1: How does Daguan Data use knowledge graphs to find commonalities across manufacturing industries?

A1: By leveraging academic and industrial research, building generic pre‑trained models, and employing human‑machine collaboration to ingest enterprise data securely.

Q2: Best practices for building an RCA system?

A2: Organize internal data, choose algorithms suited to data quality and efficiency, and maximize human‑machine synergy.

Q3: Transferring knowledge between domains?

A3: Transfer is feasible between similar domains (e.g., automotive OEMs) but limited across disparate fields like chips.

Q4: Convincing customers to adopt AI‑based knowledge‑graph solutions?

A4: Demonstrate added value, ensure accuracy through AI‑human verification, and provide measurable improvements.

Thank you for attending the session.