Investigation of Change Detection and Forecasting Methods for Predictive Maintenance
Abstract
In this project, the research team evaluated the applicability of neural networks to the detection of abnormalities in equipment and developed a method using simulation and experimental data. The team evaluated how neural networks classify or estimate data deviating from the training data. Conclusions were drawn for both the classification and regression problems. Neural networks were applied to acoustic data of a pump, and the detectability of abnormalities was evaluated. The research focused on the development of algorithms for forecasting the movement of advanced images such as the human lung and to test the feasibility of an organ motion monitoring algorithm to detect anomalies during treatment in near real time. All work was performed with the available images in our database. In the case of forecasting using PCA/MSSA, further calculation time reduction can be achieved by using this method in combination with other methods such as optical flow by which only few components can be tracked for the tracking of the entire region of displacement in the image. The combination is also believed to reduce the noise appearing in longer sequences.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 21, 2018
- Accession Number
- AD1057267
Entities
People
- Manabu Tsunokai