Introduction to Predictive Maintenance Introducing the speaker as a data scientist at an Italian startup specializing in artificial intelligence and predictive maintenance. Discussing the importance of machines, types of maintenance programs (run-to-fail, preventive, and predictive), benefits of predictive maintenance in today's digital era.
Challenges in Predictive Maintenance Explaining the challenges faced during data gathering and preprocessing such as sensor selection, data collection frequency/aggregation, handling missing data/downtime periods. Also addressing feature engineering/selection difficulties including time-based features vs frequency-based features.
Workflow for Predictive Maintenance Describing the workflow steps: Data Gathering & Preprocessing (sensor selection/collection/aggregation/labeling), Feature Engineering & Selection (time/frequency based approaches), Model Selection (anomaly detection or regression models). Explaining how these steps are applied using real-world examples from NASA Army research lab dataset.
Technical Aspects of Anomaly Detection Models 'Elaborating on anomaly detection methods like Auto Regressive models for time series forecasting with separate treatment per sensor; discussing challenges encountered during model deployment.'
Evaluation Metrics and Future Work 'Addressing evaluation metrics used for anomaly detection models such as F1 score; mentioning future work involving over-sampling labeled positive shutdown/faulty instances.'