Core Concepts
The author compares non-neural Machine Learning and Deep Learning approaches for failure prediction, highlighting the superiority of Deep Learning for complex data sets with diverse patterns.
Abstract
The content discusses the impact of varying historical data sizes on predicting machine failures using different algorithms. It emphasizes the effectiveness of Deep Learning over traditional Machine Learning methods in certain scenarios.
The study evaluates three industrial cases involving an industrial wrapping machine, a blood refrigerator, and a nitrogen generator. It formulates the problem as a binary classification task to predict system failures based on past data.
Key findings include the importance of reading window and prediction window sizes on model performance. Results show that Deep Learning approaches outperform traditional Machine Learning methods when dealing with complex data sets with diverse patterns preceding failures.
The study also highlights that increasing historical data does not always lead to better results in failure prediction tasks. The research aims to provide insights into predictive maintenance strategies using advanced machine learning techniques.
Overall, the content underscores the significance of selecting appropriate algorithms and parameter settings for effective failure prediction in industrial maintenance systems.
Stats
Results reveal Deep Learning superiority over non-neural Machine Learning approaches for failure prediction only for complex data sets with more diverse anomalous patterns.
Increasing historical data does not necessarily yield better results.
Six algorithms (logistic regression, random forest, support vector machine, LSTM, ConvLSTM, and Transformers) are compared using multivariate telemetry time series.
The results indicate that the dimension of the prediction windows plays a crucial role in failure prediction performance.
DL approaches outperform ML approaches significantly only for complex data sets with more diverse patterns.
All methods lose predictive power when the horizon enlarges because temporal correlation tends to vanish.