Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Benchmark of Causal vs Correlation AI for Predictive Maintenance

Krishna Taduri, Shaunak Dhande, Giacinto Paolo, Saggese, Paul Smith

TL;DR

Addresses predictive maintenance under extreme cost asymmetry by benchmarking eight approaches, including a causal inference pipeline, on a CNC dataset. The causal DAG–based pipeline (L5) with derived physical features achieves the highest annual cost savings (USD 1,157,500, 70.2%) and the highest precision (92.1%) with only 5 false positives, outperforming the best correlation approach by about USD 80k. It also dramatically reduces false alarms (97% fewer) and generalizes robustly across splits due to causal features that capture invariant mechanisms like thermal stress, power dissipation, and overstrain. The work supports domain-knowledge-guided causal AI as a practical, interpretable, and financially advantageous approach for predictive maintenance, while noting limits such as dataset scope and DAG construction.

Abstract

Predictive maintenance in manufacturing environments presents a challenging optimization problem characterized by extreme cost asymmetry, where missed failures incur costs roughly fifty times higher than false alarms. Conventional machine learning approaches typically optimize statistical accuracy metrics that do not reflect this operational reality and cannot reliably distinguish causal relationships from spurious correlations. This study evaluates eight predictive models, ranging from baseline statistical approaches to formal causal inference methods, on a dataset of 10,000 CNC machines with a 3.3% failure prevalence. The formal causal inference model (L5) achieved estimated annual cost savings of 1.16 million USD (a 70.2 percent reduction), outperforming the best correlation-based decision tree model (L3) by approximately 80,000 USD per year. The causal model matched the highest observed recall (87.9 percent) while reducing false alarms by 97 percent (from 165 to 5) and attained a precision of 92.1 percent, with a train-test performance gap of only 2.6 percentage points. These results indicate that causal AI methods, when combined with domain knowledge, can yield superior financial outcomes and more interpretable predictions compared to correlation-based approaches in predictive maintenance applications.

A Benchmark of Causal vs Correlation AI for Predictive Maintenance

TL;DR

Addresses predictive maintenance under extreme cost asymmetry by benchmarking eight approaches, including a causal inference pipeline, on a CNC dataset. The causal DAG–based pipeline (L5) with derived physical features achieves the highest annual cost savings (USD 1,157,500, 70.2%) and the highest precision (92.1%) with only 5 false positives, outperforming the best correlation approach by about USD 80k. It also dramatically reduces false alarms (97% fewer) and generalizes robustly across splits due to causal features that capture invariant mechanisms like thermal stress, power dissipation, and overstrain. The work supports domain-knowledge-guided causal AI as a practical, interpretable, and financially advantageous approach for predictive maintenance, while noting limits such as dataset scope and DAG construction.

Abstract

Predictive maintenance in manufacturing environments presents a challenging optimization problem characterized by extreme cost asymmetry, where missed failures incur costs roughly fifty times higher than false alarms. Conventional machine learning approaches typically optimize statistical accuracy metrics that do not reflect this operational reality and cannot reliably distinguish causal relationships from spurious correlations. This study evaluates eight predictive models, ranging from baseline statistical approaches to formal causal inference methods, on a dataset of 10,000 CNC machines with a 3.3% failure prevalence. The formal causal inference model (L5) achieved estimated annual cost savings of 1.16 million USD (a 70.2 percent reduction), outperforming the best correlation-based decision tree model (L3) by approximately 80,000 USD per year. The causal model matched the highest observed recall (87.9 percent) while reducing false alarms by 97 percent (from 165 to 5) and attained a precision of 92.1 percent, with a train-test performance gap of only 2.6 percentage points. These results indicate that causal AI methods, when combined with domain knowledge, can yield superior financial outcomes and more interpretable predictions compared to correlation-based approaches in predictive maintenance applications.

Paper Structure

This paper contains 14 sections, 1 figure, 7 tables.

Table of Contents

  1. Contributions
  2. Introduction
  3. Dataset and Experimental Design
  4. Target Definition and Rationale
  5. Methodology
  6. Success Metrics
  7. Correlation-Based Models (L0-L3)
  8. Probabilistic and Causal Approaches
  9. Validation Strategy
  10. Results and Analysis
  11. Detailed Model Analysis
  12. Comparative Analysis: Causal vs. Correlation
  13. Generalization and Overfitting Assessment
  14. Conclusion

Figures (1)

  • Figure 1: Causal directed acyclic graph (DAG) for CNC machine failures. The graph encodes domain knowledge about physical mechanisms: exogenous variables (sensor readings and machine type) affect endogenous variables (temperature differential, power, and overstrain) which directly cause machine failures.