Welcome to the Anomalous Sound Detection of Industrial Machines repository

This repository contains the source code for the Anomalous Sound Detection of Industrial Machines.

This work is based on the

Getting Started

1. Create virtual environment & install python packages/dependencies

Run the following commands to install all the dependencies into a virtual environment.

Linux/Mac Bash:

python3 -m venv venv

source venv/bin/activate
python3 -m pip install --upgrade pip
pip install -r requirements.txt

2. Download dataset

To download all the dataset (.zip files) from ZENODO, run:

sh grab_data.sh

After downloading, you'll find the dataset in the dataset directory, or the directory specified in config.yaml for dataset_dir.

2a. Create a sample dataset (optional)

This creates a sample dataset from the entire dataset to speed up development. Instead of experimenting on the entire dataset. Run: python mk_sample_dataset.py

After downloading, you'll find the sample dataset in the sample dataset directory, or the directory specified in config.yaml for sample_dataset_dir.

3. Preprocessing

Run the preprocess.py script to convert the audio wav files to mel spectrograms and save the spectrograms to file as numpy arrays. All parameters related to the spectrogram can be changed in the config.

python preprocess.py

4. Training

Run the train.py script with classifier/anomaly_detector argument to train a classifier or anomaly detector.

Classification

python train.py --model classifier

Anomaly detection

python train.py --model anomaly_detector

After training, you'll find the training results in the results & logs directories, or the directories specified in config.yaml for results and log, respectively.

Other results, such as training, validation and testing data used can be found in the dataset directory, or the directory specified in config.yaml for dataset.

5. Inference

Similar to the training, run the inference.py script with classifier/anomaly_detector argument to infer for classification or anomaly detection.

Classification

python inference.py --model classifier

Anomaly detection

python inference.py --model anomaly_detector

After inference, you'll find the inference results in the results & logs directories, or the directories specified in config.yaml for results and log, respectively.

Dependencies

This source code was developed on Ubuntu 18.04.3 LTS (Bionic Beaver).

Software packages

• wget
• unzip
• python==3.6.9

Python packages

Please refer to requirements.txt for a full list of python packages and their respective versions.

References

[1] Harsh Purohit, Ryo Tanabe, Kenji Ichige, Takashi Endo, Yuki Nikaido, Kaori Suefusa, and Yohei Kawaguchi, “MIMII Dataset: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection,” arXiv preprint arXiv:1909.09347, 2019. URL: https://arxiv.org/abs/1909.09347

[2] Harsh Purohit, Ryo Tanabe, Kenji Ichige, Takashi Endo, Yuki Nikaido, Kaori Suefusa, and Yohei Kawaguchi, “MIMII Dataset: Sound for Malfunctioning Industrial Machine Investigation and Inspection,” in Proc. 4th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE), 2019.

[3] Choi, Keunwoo, Deokjin Joo, and Juho Kim. "Kapre: On-gpu audio preprocessing layers for a quick implementation of deep neural network models with keras." arXiv preprint arXiv:1706.05781 (2017).

[4] Dataset (https://zenodo.org/record/3384388)