Big Machinery Data Preprocessing Methodology for Data-Driven Models in Prognostics and Health Management.

Area of Science:

• Reliability Engineering and Big Data Analytics
• Prognostics and Health Management (PHM)
• Machinery Health Monitoring

Background:

• Advancements in sensor networks and big data analytics enable analysis of big machinery data.
• Internet of Things (IoT) and Industry 4.0 generate rich datasets for PHM frameworks.
• Existing data-driven models (DDMs) often lack validation on real operating systems and sufficient data preprocessing.

Purpose of the Study:

• To present a comprehensive, step-by-step pipeline for preprocessing monitoring data from complex systems for DDMs.
• To bridge the knowledge gap concerning data preprocessing for PHM applications in real-world machinery.
• To emphasize the role of expert knowledge in data selection and label generation for robust model training.

Main Methods:

• Development of a formal and consistent data preprocessing pipeline tailored for PHM applications.
• Integration of expert knowledge for critical steps like data selection and health state labeling.
• Validation through two case studies involving real operating system data.

Main Results:

• Successful creation of clean datasets with distinct healthy and unhealthy labels.
• Demonstration of the pipeline's effectiveness in preparing data for machinery health state classifiers.
• Highlighting the necessity of structured preprocessing for reliable DDMs in industrial settings.

Conclusions:

• A standardized data preprocessing pipeline is crucial for applying DDMs to real machinery monitoring data.
• Expert knowledge significantly enhances the quality and relevance of data for PHM.
• The proposed pipeline facilitates the development of accurate machinery health classifiers, advancing reliability engineering.