A Multiplier-Free Convolution Neural Network Hardware Accelerator for Real-Time Bearing Condition Detection of CNC...

Area of Science:

• Industrial Engineering
• Machine Learning
• Embedded Systems

Background:

• Bearing failures are a primary cause of industrial machinery malfunction, accounting for 41-44% of operational breakdowns.
• Effective predictive maintenance strategies are essential to prevent costly downtime and equipment damage.

Purpose of the Study:

• To develop a lightweight neural network for efficient bearing condition detection.
• To enable real-time monitoring on resource-constrained hardware like FPGAs.
• To reduce memory footprint and power consumption for industrial applications.

Main Methods:

• Implementation of a lightweight neural network with 8.69K parameters on a field-programmable gate array (FPGA).
• Utilization of incremental network quantization and fixed-point operation techniques.
• Real-time data processing at 48,000 samples per second.

Main Results:

• Achieved a 63.49% memory saving compared to 32-bit floating-point operations.
• Operated on a minimal power budget of 342 mW.
• Demonstrated a high detection accuracy of 95.12% for bearing conditions.

Conclusions:

• The developed lightweight neural network system is effective for real-time bearing condition detection.
• The system offers significant advantages in memory efficiency, low power consumption, and high accuracy.
• This approach is highly suitable for predictive maintenance in industrial settings, preventing machinery failures.