Fault Diagnosis from Raw Sensor Data Using Deep Neural Networks Considering Temporal Coherence.

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Conventional fault diagnosis relies on manual feature extraction and signal processing, which is expertise-intensive and ignores temporal data coherence.
• Existing methods often struggle with noise reduction and feature selection, impacting diagnostic accuracy and efficiency.

Purpose of the Study:

• To propose a novel fault diagnosis model using Deep Neural Networks (DNN) that directly processes raw time series sensor data.
• To overcome the limitations of conventional methods by eliminating the need for manual feature engineering and incorporating temporal data coherence.

Main Methods:

• A Deep Neural Network (DNN) model was developed to directly analyze raw time series sensor data without preprocessing.
• The DNN was trained on sensor data until minimal cost function value, followed by testing on new data for classification accuracy.
• The model leverages the temporal coherence of sensor data for enhanced fault diagnosis.

Main Results:

• The proposed DNN model achieved 100% classification accuracy in identifying bearing faults.
• The model demonstrated effectiveness in recognizing various types of bearing faults.
• The approach successfully integrated temporal coherence into the fault diagnosis process.

Conclusions:

• The DNN-based fault diagnosis model is highly effective and accurate for machinery health monitoring.
• This method offers a significant advancement over traditional approaches by automating feature extraction and utilizing temporal data characteristics.
• The model shows promise for ensuring machinery safety through intelligent condition monitoring.