Providing Fault Detection from Sensor Data in Complex Machines That Build the Smart City.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Complex machines in smart cities and Industry 4.0 generate vast amounts of data.
• Limited memory and processing power in these devices pose challenges for data analysis.
• Predictive maintenance is crucial for preventing failures and optimizing machine operation.

Purpose of the Study:

• To propose a layered methodology for automatic fault detection and predictive maintenance in complex machines.
• To reduce data volume and optimize information extraction through indicator generation.
• To enhance the efficiency of data processing for fault identification.

Main Methods:

• A layered structure for data collection, filtering, and indicator extraction.
• Utilizing Kullback-Leibler divergence for visualizing operational data differences.
• Employing a neural network for failure classification.

Main Results:

• Successfully visualized differences between normal and failure operation data using Kullback-Leibler divergence.
• Achieved 90% accuracy in correctly classifying simulated failures with a neural network.
• Demonstrated the proposed approach's effectiveness on a cash counting machine prototype.

Conclusions:

• The proposed layered methodology effectively enables automatic fault detection in complex machines.
• The approach optimizes data processing by generating informative indicators.
• This methodology supports predictive maintenance strategies in Industry 4.0 applications.