Project: (PwC) Enhancing Anomaly Detection: Integrating User Explainability
Description
PwC developed an unsupervised Transformer-based anomaly detection tool to enhance insights into machine functionality in factories by analyzing machinery timeseries sensor data. However, the current solution lacks explainability for why certain time windows are flagged as anomalous. Root cause algorithms, such as Bayesian inference, Association Rule Learning, or similar techniques, could be employed to uncover the reasons behind these flagged timeseries.
During your thesis, you will work with real-world data from at least one large company, and potentially incorporating new data as well. Your task will be threefold:
Dive into the wonderous world of neural networks to improve the model's architecture if deemed necessary
Implement algorithms that enhance the explainability of flagged time windows
Step into the role of a consultant by designing an end-to-end framework that factory engineers can use to optimize their operations
By the end of this period, you will have developed a deep understanding of complex Machine Learning algorithms, coded them yourself, and been challenged to apply them to real-world scenarios. This approach reflects how we tackle projects in the Data & Analytics team at PwC.
References
Kromwijk, T. J. (2022) “A Self-Supervised Learning Approach for Anomaly Detection of Industrial Systems” Student thesis: Master
Tuli, S., Casale, G., & Jennings, N. R. (2022). Tranad: Deep transformer networks for anomaly detection in multivariate time series data. arXiv preprint arXiv:2201.07284.
Details
- Supervisor
-
Bart Engelen
- External location
- PWC
- Interested?
- Get in contact
- Link
- Thesis