Factory Copilot AI Chatbot for Smart Manufacturing

Plant managers and floor operators faced severe bottlenecks when trying to access actionable manufacturing insights.

While operational data was continuously generated across the plant, retrieving it was difficult because critical information was fragmented across disconnected systems, SQL databases, REST APIs, and thousands of PDFs.

To get an answer, users had to manually navigate complex dashboards or file a report ticket with IT and wait for a response.

The Factory Copilot AI assistant had to meet several strict requirements:

• Real-time retrieval and user experience. The goal was to develop a solution capable of real-time data retrieval delivered through an intuitive Angular-based UI. It was expected to streamline KPI analysis, production order tracking, device configuration help, and machine status monitoring.
• Natural language processing. We needed to build a chatbot that would enable natural language queries across SQL databases, PDF documents, REST APIs, and SAP systems simultaneously.
• Security and scalability. It was critical to implement role-based security and encryption, paired with an architecture built to scale for large-scale industrial deployments.

Following the provided requirements, Akveo developed Factory Copilot, an AI-powered chatbot seamlessly integrated into the Factory Intelligence platform.

Leveraging Azure OpenAI's GPT-4, Semantic Kernel, and Azure AI Search, Factory Copilot acts as a conversational layer over the platform’s central orchestrator.

Users can use the copilot to:

• Query plant documentation
• Analyze machine telemetry
• Check alerts and alert statistics
• Query Power BI production data
• Get information from different plant systems

If a request is unclear, the copilot asks follow-up questions before running the query.

Core capabilities:

• Real-time KPI analysis. Users can ask questions about production KPIs (e.g., “What was the KPI for machine Y last week?”) and get answers based on live plant data and analytics.
• Production order tracking. The chatbot can retrieve recent production orders and monitor activity across connected process chains.
• Device configuration assistance. The copilot helps users configure and operate specific equipment by searching plant documentation, equipment manuals, and technical PDFs.
• Machine status and predictive maintenance monitoring. Users can check machine conditions, alerts, and predictive maintenance information in real time. When a potential issue is detected, the system can provide early maintenance recommendations.
• Data integration across sources. Factory Copilot interacts simultaneously with SQL databases, PDF documents, Azure IoT Hub telemetry pipelines, embedded Power BI reports, SAP, and REST APIs to deliver comprehensive responses directly inside the UI application.

Microservice Architecture

The Factory Intelligence platform is built on a modular microservice architecture. This approach helps ensure high performance and scalability.

Separate services handle different tasks and communicate through Azure Service Bus. When an issue occurs, such as a new machine alert, the relevant services are notified and can respond at the same time without blocking each other.

The table below contains a short overview of the platform's services.

All services reference a centralized MongoDB database. As a result, plant data remains consistent and immediately available across the entire system.

Technical Implementation

Factory Copilot is designed to handle highly complex data tasks while keeping the user experience simple.

To make this possible, we implemented the following technical solutions:

Plant vocabulary

The system understands the plant's specific industrial vocabulary, for example, machines and equipment, alert rules, and signals and telemetry. These terms are stored in Azure AI Search. The copilot checks them before sending a query, which helps avoid incorrect or misunderstood requests.

Smart data queries

We built a DAG-based tool engine that finds the best way to get information from different data sources. It can combine data from multiple sources and process the results in the right order.

Dynamic visualizations

The system automatically chooses the most suitable way to show the results. Depending on the data, it can present information as charts, tables, and other relevant visualizations.

Security

The copilot uses role-based access control and encryption. Users can only access the data and analytics they are authorized to see.