A quality traceability method in a manufacturing process based on multi-source data parallel processing

By using the Transformer framework that integrates variable tokens and patch tokens in parallel with dual tokens, combined with MPC closed-loop regulation and Neo4j-PROV traceability, the challenges of cross-variable dependency and local detail capture in the manufacturing process are solved, achieving high-precision prediction and rapid traceability, and meeting compliance requirements.

CN121744170BActive Publication Date: 2026-06-30NORTHEASTERN UNIV CHINA +1
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NORTHEASTERN UNIV CHINA
Filing Date
2026-03-02
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing Transformer models struggle to capture both global dependencies across variables and details of local segments during the manufacturing process. The fragmentation of the process leads to untraceable prediction biases, and the lack of a PROV-DM traceability framework fails to meet compliance requirements.

Method used

A Transformer framework that uses dual-token parallel fusion of variable tokens and patch tokens, combined with MPC closed-loop regulation and Neo4j-PROV tracing, is adopted to achieve parallel processing of multi-source data and end-to-end tracing.

Benefits of technology

It improves prediction accuracy, reduces traceability query latency to within 100ms, and achieves integrated manufacturing quality management with high precision, high transparency, and high governance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121744170B_ABST
    Figure CN121744170B_ABST
Patent Text Reader

Abstract

This invention provides a quality traceability method for manufacturing processes based on parallel processing of multi-source data, relating to the fields of quality control and data traceability technology. It employs a dual-perspective parallel self-attention mechanism using variable tokens and patch tokens, enabling the model to simultaneously capture cross-variable global coupling relationships and local time-series microstructural features. Parallel decomposition of patch and variable dimensions significantly reduces memory and latency overhead. The entire process—collection → cleaning → embedding → attention → fusion → prediction → MPC control → evaluation → feedback—is written to Neo4j in real time, combined with a Time-Tree hierarchical index, achieving a complete lineage path query latency of less than 100ms at a scale of millions of nodes. Defect detection is moved from after sampling inspection to during process execution, reducing production line rework rates. RMSE and SMAPE are continuously monitored, and the system automatically generates feedback when thresholds are exceeded. This invention achieves breakthroughs in four dimensions: high precision, low resource consumption, high transparency, and strong governance, providing a directly implementable and sustainably evolving new industry paradigm for high-dimensional heterogeneous frequency time series prediction.
Need to check novelty before this filing date? Find Prior Art