This repo contains the work of the demo paper presented in The 21st International Conference on Pervasive Computing and Communications (PerCom 2023).
The demo video is available here: https://iotgarage.net/projects/demos/HakanPerCom2023Demo
Initial steps:
-
Download the dataset from Kaggle or IEEEDataPort:
1.1. Link for the Kaggle: https://www.kaggle.com/datasets/hkayan/industrial-robotic-arm-anomaly-detection
1.2. Link for the IEEEDataPort: https://ieee-dataport.org/documents/casper-context-aware-anomaly-detection-system-industrial-robotic-arms
-
Extract the
zipfile.
The dataset contains four files: * A .csv file that consists of accelerometer, gyroscope, and magnetometer data of an arm that accomplishes a repetitive task, captured via Nicla Sense ME. * Two .csv files (one per industrial arm) that consist of built-in arm parameters such as joint current, and velocity values etc. * A .pcap file that contains the TCP/IP traffic between the arms and the controller PC.
For this work, you will only need right_arm.csv and nicla.csv.
The whole test is around 24 hours.
We modify the joint velocity of the right arm after 14.4 hours (corresponding to 60%).
You should replace the file paths with your own paths.
The cells starts with %%time are expected to take longer running times depending on your setup. As the training, takes long time, we provide the saved model for convenience.
We run this experiment on a data science workstation with NVIDIA RTX A6000.
We provide a conda environment to test the work.
Please check the following guide to install conda: https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html
After installing conda, create our environment via (set the path accordingly):
conda env create -f environment.yml -p /home/user/miniconda3/envs/env_name
Then follow the casper.ipynb to recreate the work.
In case of a related error downgrade black==22.12.1.dev44+g196b1f3 to black==22.12.0.