Delta: Our new graph-based algorithm for root cause analysis

Announcing Delta, our new graph-based algorithm for root cause analysis in manufacturing.

Manufacturers increasingly rely on off-the-shelf machine learning algorithms like Random Forests for root cause analysis. However, these algorithms often fail to identify the actual causes of production issues because they focus on predictive relationships rather than causal effects. This can result in misleading conclusions and oversight of critical relationships when investigating root causes of production issues (e.g., quality losses or downtime). We are excited to introduce Delta, our new graph-based algorithm, which leverages causal AI to overcome these limitations.

Delta is fully integrated into our Analyst software for root cause analysis. Our Analyst software supports process engineers in identifying the causes of production losses across thousands of process parameters. By interpreting data from various sources, including machine setpoints, sensor readings, and operational KPIs, Delta offers a comprehensive view of production processes. It empowers manufacturers to move far beyond simple correlation analysis, and thus to uncover the true root causes that affect operational KPIs like quality and efficiency.

Delta achieves state-of-the-art performance across benchmarks, and significantly outperforms existing causal algorithms. It performs up to four times better than existing causal algorithms (e.g., NOTEARS) on public benchmarks. Additionally, it has up to 10x faster runtime, which enables the analysis of high-dimensional datasets with thousands of production variables.

Highlights:

• Delta seamlessly integrates into the EthonAI Analyst platform, offering powerful root cause analysis for manufacturers.
• Delta comprehends data related to various production types, including unit-based, batch-based, and time-based features.
• Delta can handle many data types, like continuous, binary, and even categorical.
• Delta can identify complex root cause chains and nonlinear interaction effects.
• Delta outperforms competing causal algorithms by up to four times according to industry benchmarks.
• Delta’s runtime is significantly faster, providing results up to 10x times faster than existing causal algorithms

Delta is available for all EthonAI Analyst users starting today. To learn more about how our Analyst software can help you identify causes of production losses, book a demo with our team.