generative-model-fdd

Accompanying code for our paper

A. Mitrevski and P. G. Plöger, “Data-Driven Robot Fault Detection and Diagnosis Using Generative Models: A Modified SFDD Algorithm,” in 30th International Workshop on Principles of Diagnosis (DX), Klagenfurt, Austria, 2019.

Short description

The repository includes:

• black_box_data: Data collected from a ROPOD platform using the black box data logger at https://github.com/ropod-project/black-box (both nominal and faulty data). The subdirectories of black_box_data contain descriptions of the conditions under which the data were collected as well as descriptions of the recorded data items
• sw_faults_current.ipynb: Illustrates the anomaly detection procedure using a restricted Boltzmann machine (RBM) as described in the paper. The notebook includes our implementation (a very naive, non-optimised version) of a deep belief network, the procedure for training correlation dependency models, as well as evaluation on the faulty data
• sw_faults_current_gmm.ipynb: This notebook is the same as the previous one, except that the RBM models are replaced by Gaussian Mixture Models (GMMs)

Requirements

The code in the provided notebooks depends on the packages specified in requirements.txt.