How LLMs Transform Industrial Configuration Software: Architecture & Use Cases

Introduction

On June 1 at the Guangzhou Port‑Zhongshan‑Macao Cloud Summit, the industrial configuration intelligent edition (visual engine for production control) was announced, combining traditional industrial SCADA software with large language model (LLM) capabilities to provide a Copilot‑style AI assistant that enhances creativity, lowers product barriers and improves productivity.

Key Features Demonstrated

Industrial knowledge Q&A

Industrial configuration application generation

Smart drawing – basic shape creation

Smart drawing – property modification

Smart script generation

Smart material generation

Technical Architecture

Client Layer

The client layer hosts the Copilot UI, registers LLM skills, and executes actions returned by the LLM. It interacts with the Maliang Engine, which provides the core configuration capabilities.

Agent Layer

The agent layer acts as a proxy for LLM services, handling HTTP requests, streaming responses, and building prompts. It is implemented with Python/Django and SDKs for OpenAI and Alibaba Qwen.

LLM Layer

Both commercial (Alibaba Qwen) and research (OpenAI GPT‑3/3.5/4) models are used. Model selection, temperature, and top‑p settings are tuned for stability and determinism.

Prompt Engineering

Prompt templates are defined for knowledge Q&A, script generation, SVG drawing, property modification, and application recommendation. Few‑shot examples and exception handling are included to guide the model.

Code Examples

import spacy
nlp = spacy.load("zh_core_web_lg")
text = "把背景颜色改为红色"
doc = nlp(text)

def props_prompt(input):
    example = '''
    输入示例1：帮我把字体大小改成12
    输出示例1：{ "fontSize":"12"}
    '''
    # prompt construction omitted for brevity
    return ret

Implementation Insights

Key lessons include setting low temperature for deterministic output, handling inconsistent JSON formats from LLMs, filtering unsupported properties, and adapting prompts for different models.

Conclusion

The integration of LLMs into industrial configuration software demonstrates how AI can lower the barrier for non‑programmers, accelerate development, and enable natural‑language interaction with complex automation systems.