Cost-Effective Vibration Analysis through Data-Backed Pipeline Optimisation

Area of Science:

• Engineering
• Data Science

Background:

• Vibration analysis is crucial for machinery failure mode classification.
• Current signal processing pipelines are computationally intensive and unsuitable for IoT devices.

Purpose of the Study:

• To propose a data-driven methodology for optimizing signal processing pipeline complexity.
• To assess the impact of reduced sampling rates and observation times on failure classification performance for IoT applications.

Main Methods:

• Developed a data-driven approach to optimize and justify signal processing pipeline complexity.
• Evaluated machinery failure classification performance using the MAFAULDA dataset with reduced sampling rates (down to 1 kHz) and short observation windows (0.1 s).

Main Results:

• A statistically significant drop in classification performance was observed with decreased sampling rates and shorter observation times.
• Effect sizes of these reductions were small to medium, indicating potential trade-offs.

Conclusions:

• The proposed optimization methodology enables efficient design of IoT vibration analysis systems.
• Findings suggest that lower sampling rates and shorter observation windows can be viable, facilitating the use of cost-efficient sensors and promoting accessible IoT/Edge-based vibration analysis.