AI Quality Defect Analysis Platform
An intelligent Quality Defect analysis system for automotive wire harness manufacturing, powered by RAG and large language models. It covers defect registration, 8D report generation, PSS problem solving, and knowledge base management, providing one-stop AI-assisted decision support from defect intake through root cause analysis.
Industry Challenges
Inefficient Document-Based Workflows
8D reports rely on manually completed Excel templates, causing version confusion, cumbersome approvals, difficult case retrieval, and experience that cannot be captured and reused.
Slow Root Cause Analysis
Engineers depend on individual experience to identify root causes, leading to slow onboarding, inconsistent results, and limited semantic search.
Fragmented Knowledge
Defect knowledge is scattered across systems and individuals, search depends on memory, and similar issues recur without effective prevention.
Key Features
End-to-End Quality Defect Lifecycle Management
Supports defect registration, S/A/B/C severity grading, classification, and source Traceability, with a complete record from discovery through closure. A soft-delete recycle bin prevents accidental data loss.
Collaborative Online 8D Reports
Manages the complete D1 through D8 workflow online, with collaborative editing and stage progression. AI automatically generates root cause analyses and recommended actions.
PSS Problem-Solving Worksheet
A structured problem-solving workflow tracks every stage from issue identification through standardization, with a Champion index for locating historical cases.
AI Assistant
An embedded AI chat panel supports quality consultations, root cause reasoning, and solution recommendations using MiniMax and DeepSeek.
Intelligent RAG Search
A two-stage filter combines broad vector retrieval with precise LLM validation to match similar historical defects and avoid semantic false positives.
Knowledge Base Capture
Defect cases are archived automatically with 5M1E classification (Man, Machine, Material, Method, Environment, and Measurement) and process-stage tags, enabling ChatGPT-style retrieval.
QMIS Data Migration
Imports a complete legacy QMIS migration package, matches users, creates missing accounts, and normalizes severity levels with no data loss.
AI-Assisted Root Cause Diagnosis
Generates an internal 8D report from the defect description, including probable root causes, 5M1E classification, and long-term actions supported by historical experience and AI reasoning.
Quality Dashboard
Visualizes defect statistics, severity distribution, and trends, with month-over-month analysis for a timely view of quality performance.
Project Background
Quality Defect management in automotive wire harness manufacturing has long relied on paper-based 8D reports and scattered Excel files, creating version conflicts, slow approvals, and inaccessible historical knowledge. Built on PostgreSQL and pgvector with MiniMax and DeepSeek LLMs, AI-QPA digitizes and augments the complete workflow from defect registration and analysis to knowledge capture.
Technology Stack
Feature Details
End-to-End Quality Defect Lifecycle Management
Covers defect registration, classification, S/A/B/C severity assessment, and source Traceability across customer complaints, internal findings, audit findings, and suppliers. Multi-criteria filtering, bulk actions, and a soft-delete recycle bin protect data while allowing recovery or permanent deletion.
Collaborative Online 8D Reports
Designed around the standard D1–D8 process: D1 Build the Team → D2 Describe the Problem → D3 Implement Interim Containment → D4 Identify Root Causes → D5 Define Permanent Corrective Actions → D6 Implement and Validate → D7 Prevent Recurrence → D8 Recognize the Team. Supports collaborative editing, stage progression, and version traceability.
AI Assistant
The embedded AI chat panel supports open-ended quality questions. Powered by MiniMax and DeepSeek LLMs, it generates internal 8D reports with root cause reasoning, 5M1E classification, and targeted long-term action recommendations. It combines historical cases with AI reasoning to ensure reliable analysis.
Two-Stage Intelligent RAG Search
An innovative two-stage filter first uses pgvector for broad candidate retrieval with a relaxed threshold, then applies an LLM for precise relevance validation, distinguishing semantically similar but fundamentally different issues such as a broken wire and a fractured weld. This preserves recall while greatly improving precision.
PSS Problem Solving and Knowledge Capture
The PSS worksheet covers the complete closed loop from problem identification and current-state assessment through target setting, cause analysis, corrective action, and standardization. The Champion index locates comparable historical cases, while automatic 5M1E classification and semantic search make experience reusable.
Seamless QMIS Data Migration
Upload a complete ZIP migration package exported from the legacy QMIS system, preview record counts and user matches, then import with one confirmation. Users are matched by username or display name, missing accounts are created with a default password, and severity levels are normalized automatically (for example, A–Critical → A).