Hitachi researchers have just released a sound dataset for malfunctioning industrial machine investigation and inspection (MIMII). The dataset consists of over 100 GB in industrial factory sounds (valves, pumps, fans, slide rails) in various real-world factory settings (contamination, leakage, rotating unbalance, rail damage). The ‘broken’ factory sounds could make for perfect sampling material for your next techno track or apocalyptic sci-fi soundtrack.
The MIMII dataset has been released to assist the machine-learning community to develop automated facility maintenance. This helps factories monitor their production lines with the development of acoustic detection techniques. With the advancement of the Internet of Things and the low cost of sensor technology (like microphones), factory owners are looking to leverage this machine-learning technology to identify malfunctioning machinery before it malfunctions.
Each type of machine audio sample is broken down by abnormal/normal type functioning (5-10 seconds in length). The samples were recorded by an 8 channel microphone array with 16 kHz sampling rate and 16 bit per sample. Each machine audio dataset is available at -6 DB, 0 DB, and +6DB.
Example Industrial Machine Audio Samples
The dataset is released under Attribution-ShareAlike 4.0. You are free to share (copy and redistribute the material in any medium or format), adapt (remix, transform, and build upon the material for any purpose – even commercially) with attribution.