Deep-Learning-Based Approach to Anomaly Detection Techniques for Large Acoustic Data in Machine Operation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Automotive Engineering
• Acoustic Signal Processing

Background:

• Increasing demand for automated maintenance in vehicles due to workforce shortages.
• In-cabin noise significantly impacts driver and passenger emotional satisfaction.
• Limitations of current anomaly detection methods in handling noisy environments.

Purpose of the Study:

• To develop an efficient deep learning-based anomaly detection (AD) algorithm for vehicles.
• To address challenges posed by noisy acoustic measurements in vehicular environments.
• To improve the accuracy and speed of detecting abnormal machine operations.

Main Methods:

• Collection of acoustic data using a large microphone array within a vehicular environment.
• Development of a novel AD algorithm designed to overcome noise interference.
• Training and evaluation of multiple AD models using simulated noisy data with five distinct error types.

Main Results:

• The proposed AD algorithm demonstrates robustness against noisy acoustic measurements.
• High accuracy, exceeding 90%, was achieved in detecting various machine operation anomalies.
• The method effectively utilizes noise signals to improve detection system reliability.

Conclusions:

• The developed deep learning approach offers an efficient and accurate solution for in-vehicle anomaly detection.
• This technology can enhance vehicle safety and user experience by identifying malfunctions early.
• The algorithm's ability to perform in noisy conditions represents a significant advancement over existing methods.