Sensor-Based Predictive Maintenance with Reduction of False Alarms-A Case Study in Heavy Industry

Area of Science:

• Industrial Engineering
• Data Science
• Machine Learning

Background:

• Undesirable events in industrial settings are often poorly represented in historical data, hindering traditional predictive maintenance.
• Existing systems can overwhelm operators with false alarms, especially in cold start conditions.

Purpose of the Study:

• To propose a novel predictive maintenance method for identifying undesirable events in industrial machinery.
• To address the challenge of limited historical data and reduce false positive alarms in real-time monitoring.

Main Methods:

• Analysis of vibration and temperature data from wireless sensors in crushers and gantries.
• Application of outlier identification methods to multidimensional feature vectors.
• Development of a false positive alarm reduction methodology incorporating data adaptation, dispatcher interaction, and eXplainable Artificial Intelligence (XAI).

Main Results:

• The proposed method achieved an average reduction of 90.25% in false alarms compared to stand-alone outlier detection.
• The methodology proved effective in cold start scenarios and was successfully implemented in a real industrial facility.

Conclusions:

• The developed predictive maintenance method offers a robust solution for identifying undesirable events, particularly in systems with limited historical data.
• The reduction in false alarms enhances user trust and system adoption, crucial for effective industrial monitoring.