chapter eight

8 Integrating inference and intelligence

 

This chapter covers

  • The importance of a solid data foundation for intelligence
  • Detecting anomalies and faults
  • Forecasting the future with classic machine learning approaches
  • Generative and agentic AI

So far, we have visualized the data we collected about the physical system our digital twin represents, enabling us to build descriptive and informative capabilities so we can see the system’s current state and use that information to make better decisions. In this chapter, we’ll look at how we can integrate artificial intelligence (AI) and machine learning (ML) into a digital twin to enable diagnostic, predictive, and decision-support capabilities.

We have looked at how to sense changes in a physical system, communicate and store those changes, and build a digital representation of reality through data contextualization. This solid data foundation is important because AI and machine learning models are only as reliable as the data on which they are built. In practice, many AI failures in digital twin projects stem not from poor algorithms but from a poor data foundation. At its core, a digital twin is a data-driven representation of a physical system, and AI amplifies the value of that data rather than replacing it.

8.1 A practical data pipeline

8.1.1 Cleaning and normalization

8.1.2 Time alignment and resampling

8.1.3 Feature extraction

8.1.4 Storage

8.2 Anomaly and fault detection

8.2.1 Rule-based detection and its limitations

8.2.2 Statistical approaches

8.2.3 Unsupervised learning

8.3 Predictive modeling and machine learning

8.3.1 Training a forecasting model

8.3.2 Applying AutoML

8.3.3 Supervised learning

8.4 Generative AI

8.4.1 Retrieval-augmented generation

8.4.2 Agentic AI

8.4.3 Building an agent for the home digital twin

8.4.4 When agents complement or replace control strategies

8.4.5 Multi-agent systems

8.5.3 Cost