¶ UC19 - Extension of Digital Reference Ontology for Smart Sustainable Business Planning and Incident Handling
This use case extends the Digital Reference Ontology in two main directions: traceability and sustainability. The goal is to enable quality incident prediction, root-cause traceability, sustainable business planning, and cross-company collaboration along semiconductor supply chains. Graph-based data structures, GNNs, and deep reinforcement learning are used to predict quality incidents, support planning, reduce bullwhip effects, and include carbon footprint and other sustainability indicators in decision-making. The expected benefits include fewer incidents with unknown root causes, reduced quality-related delays, less engineering effort for root-cause analysis, improved throughput efficiency, and more measurable and optimizable sustainability.
| # |
Technology Name |
Description |
| 1 |
Production Trace Algorithm |
The production trace algorithm extracts production traces from lot-level manufacturing event data. These traces often include lot splits and merges, which introduce uncertainty that is quantified by the algorithm. The algorithm supports decision-making in applications such as root-cause analysis and risk assessment. |
| 2 |
Supply Chain Simulation Model |
The supply chain simulation model is used to test and validate supply chain master plans generated by linear programming models and stochastic optimization models. The model covers the complete semiconductor supply chain, including supply uncertainty, such as lead time and yield, and demand uncertainty. The result operates in a Python environment. |
| 3 |
Linear Programming Model |
The linear programming model generates supply chain master plans and serves as a baseline for stochastic optimization models and the master planning heuristic. It mimics the supply chain master planning engines used by NXP. The model operates in a Python environment and requires a Gurobi license. |
| 4 |
Stochastic Optimization Models |
Stochastic optimization models are developed to increase plan stability in supply chain master planning. The results operate in a Python environment and may require a Gurobi license. |
| 5 |
Supply Chain Master Planning Heuristic |
The supply chain master planning heuristic provides a computationally efficient alternative to the linear programming model. It supports rapid scenario modeling and operates in a Python environment. |
| 6 |
LLM for Root Cause Analysis |
The LLM-based framework for root-cause analysis is used to query knowledge graphs with SPARQL queries. A natural language question is provided to the model, which generates and selects the best query. The selected query can then be executed on the knowledge graph to answer the question. |
| 7 |
LLM for Supply Chain Management |
The LLM-based framework for supply chain management is used to query master data with SQL queries. A natural language question is provided to the model, which generates and selects the best query and retrieves the corresponding data from the database. |
| 8 |
Single and Multi-Stage Quality Control Models |
These models focus on quality detection and quality prediction. They support improved product quality and operational decision-making. |
| KPI # |
Related Technology |
KPI Target |
| 1 |
Production Trace Algorithm |
Enable extraction of production traces from datasets with lot-level manufacturing events, including lot splits and merges. Depending on the use case, this can reduce the number of resources to inspect by 10%. |
| 2 |
Supply Chain Simulation Model |
Validate supply chain master plans under uncertainty, achieving at least 75% of products with a statistically valid uncertainty modeling approach. |
| 3 |
Linear Programming Model |
Generate baseline master plans for one product group within 5 minutes. |
| 4 |
Stochastic Optimization Models |
Improve plan stability under uncertainty by reducing rescheduling frequency by at least 20% compared to LP-based plans. |
| 5 |
Supply Chain Master Planning Heuristic |
Enable rapid scenario modeling with at most 20% deviation from LP model outputs and at least 20% reduction in computation time. |
| 6 |
LLM for Root Cause Analysis |
Automate root-cause identification with at least 80% accuracy in retrieving relevant knowledge graph insights, reducing engineering analysis time by at least 50%. |
| 7 |
LLM for Supply Chain Management |
Enable natural language access to master data with at least 80% query accuracy and at least 25% reduction in data retrieval time. |
| 8 |
Single and Multi-Stage Quality Control Models |
Reduce quality incidents with unknown root cause by 20% and reduce quality-related manufacturing delays by 10%. |
- Industrial process phases made more effective and precise through integrated and tailored AI algorithms, methods, and tools.
- 20% improvement in prediction precision.
- 25% improvement in classification precision.
- Reduced manufacturing costs in the long term.
- Improved overall productivity of the manufacturing factory by approximately 20%.
- Providing documented methods for dynamic application of tools to improve sustainability.
- Improved performance and faster time-to-market.
The use case contributes to more effective and precise industrial processes by extending the Digital Reference Ontology with traceability, quality, planning, and sustainability-related capabilities. The Production Trace Algorithm supports root-cause analysis and risk assessment by extracting uncertain production traces from lot-level manufacturing event data. The supply-chain simulation model, LP model, stochastic optimization models, and planning heuristic support more robust and efficient master planning under uncertainty. The LLM-based tools improve accessibility of knowledge graphs and master data through natural language interfaces, reducing engineering analysis time and data retrieval time. The single and multi-stage quality control models support prediction and detection of quality issues, helping to reduce incidents with unknown root causes and quality-related manufacturing delays. Together, these results support better planning, improved throughput, faster incident handling, and more sustainable decision-making across semiconductor supply chains.
| AI Toolbox Tool |
Use within the Use Case |
| NL2SPARQL Semantic Error Detection Framework |
Used to support the knowledge-graph querying workflow for root-cause analysis. It helped validate and detect errors in SPARQL queries generated from natural-language questions before they were executed on the Digital Reference Ontology. |
| Automatic Ontology Learning Framework |
Used to support the extension and refinement of the Digital Reference Ontology with traceability, sustainability, quality, planning, and incident-handling concepts. |
| Embedding Tool |
Used to represent ontology content, production traces, incident descriptions, and supply-chain knowledge in a semantic search space. It supported retrieval of related cases, concepts, and knowledge graph elements. |
| Reranker Tool |
Used to improve the precision of retrieved ontology, traceability, root-cause, and supply-chain information before it was used by LLM-based analysis components. |
| Tabular Prediction Tool |
Used to support quality prediction, supply-chain planning analysis, and structured-data modelling based on lot-level manufacturing events, master data, lead time, yield, demand, and planning indicators. |
| Explainability Tool |
Used to interpret prediction and decision-support models for quality control, incident handling, and supply-chain planning. It helped make AI-supported root-cause and planning recommendations more transparent for domain experts. |
| Device-Level Quality Detection Tool |
Used to support quality detection and quality prediction by linking device-level quality issues to production traces, process conditions, and root-cause-related manufacturing data. |