A Novel LSTM Pipeline to Detect Anomalies in Manufacturing Production (Datasets)

This dataset includes the relevant data for the journal article titled 'A Novel LSTM Pipeline to Detect Anomalies in 

Manufacturing Production'. In this paper, we present a novel anomaly detection method using a semi-supervised LSTM 

forecasting approach to highlight process anomalies in a complex, real-world dataset in an automotive manufacturing 

setting. This data includes two time-series subsets, each with 5000 labeled observations. Both subsets were recorded 

using an inbuilt torque-time sensor within a DC nut runner tool used to fasten nuts onto various parts throughout the 

assembly line. The resultant torque time data was labeled by a test engineer and domain expert using the methods 

outlined in the paper.

Labels Key:

1  : Normal Waveform - Waveform appears normal. No action required.

-1 : True Anomaly - Waveform is anomalous and implies some process error has occoured. Action required.

2  : Anomaly No Concern - Waveform is anomalous but upon further inspection no action required. 

3  : Rehit - No process has occoured. Repeat process. These can be removed in pre-processing.