Industrial Intelligence: Current Status, Talent Requirements, Challenges, and AI Application Process

Using the DataFunTalk platform and the author’s experience in Fortune‑500 manufacturing, this talk explores AI implementation in industrial settings from the perspectives of "flow" and "fusion".

1. Current Status of Industrial Intelligence

Industrial AI projects often face limited high‑quality data, complex model building, and difficulty finding collaborative partners, making data acquisition, modeling, and application challenging.

2. Talent and Skills

Successful projects require a blend of specialists: algorithm developers (few but highly skilled), product managers (cross‑functional coordinators), data scientists (problem‑solving and statistical expertise), and data engineers (pipeline and security skills). Each role participates at different project phases, from initial scoping to deployment.

3. Significance of AI in Manufacturing

AI serves as a tool to improve decision‑making efficiency; its core value lies in addressing manufacturers’ concerns such as safety, cost, quality, efficiency, and profitability.

4. DIKW Model

The Data‑Information‑Knowledge‑Wisdom hierarchy explains how raw sensor data (D) is transformed into actionable insights (W) in industrial contexts.

5. General AI Application Workflow

Based on the CRISP‑DM process, the stages include business understanding, data understanding, data preparation, modeling, evaluation, and deployment.

6. Industrial AI Landscape

Despite rapid AI advances, AI accounts for less than 4% of industrial output; typical applications involve smart products, smart factories, smart management, supply‑chain optimization, and monitoring.

7. Project‑Stage Talent Involvement

Industry experts define goals, data scientists refine problems and design solutions, data engineers implement pipelines, and all collaborate during analysis and validation.

8. AI Development History & Machine‑Learning Process

AI has evolved through behaviorism, symbolism, and connectionism. Machine‑learning projects follow the six‑step CRISP‑DM methodology, with emphasis on data preparation and model evaluation.

9. Demand and Pain Points – Edge‑Cloud Architecture

Industrial AI relies on a three‑layer architecture: edge devices (real‑time sensor acquisition), edge side (aggregation and micro‑services, latency 0.1‑100 ms), and cloud side (high‑performance compute for model training). Communication speeds vary from milliseconds (edge) to minutes (cloud).

10. Exploration – Edge Lab Setup

An experimental lab uses Raspberry Pi for data collection, Jetson Nano for inference, and K3s for container orchestration. Open‑source components include ActiveMQ, Kafka, Flink, Redis, InfluxDB, Elasticsearch, Grafana, Docker, and K3s.

11. Data Acquisition – OPC UA & MQTT

OPC UA extracts data from PLCs and other devices, while MQTT provides lightweight publish/subscribe messaging for IoT scenarios.

12. Real‑Time Computation – Kafka & Flink

Kafka offers high‑throughput, fault‑tolerant streaming; Flink provides stateful stream and batch processing with APIs for Java/Scala, Table/SQL, and Python (PyFlink).

13. Time‑Series Storage – Apache IOTDB

Time‑series databases address massive write throughput, efficient compression, and fast aggregation queries, which relational databases cannot handle effectively.

14. Service Design – DDD & Micro‑services

Domain‑Driven Design (DDD) helps decompose complex AI systems into bounded contexts, enabling micro‑service implementation. Communication uses RESTful APIs for external calls and gRPC (protobuf) for internal high‑performance RPC.

15. Visualization – Grafana & Matplotlib

Grafana provides dashboards for time‑series data; Matplotlib enables custom Python plots and real‑time animation.

Conclusion

Industrial AI should focus on solving concrete manufacturing problems, balancing the "flow" of data, talent, and requirements with the "fusion" of AI techniques into production processes. Practical advice includes iterative experimentation, top‑down design, and addressing challenges of timeliness, reliability, explainability, and generalization.