Abstract
We are building a failure prediction algorithm with sequences of event code data from devices
using a deep learning model called LSTM (Long Short‐Term Memory). LSTM models are widely
used to predict the sequence of words in word embedding technique. The principle idea of this
algorithm is very similar to next word prediction in your cell phone when you are sending
messages. The algorithm has the ability to predict the event codes leading up to the failure and
resulting time of the failure. The algorithm is fed a list of the last five event codes on the specific
device and can predict the next event codes leading up to the next failure by applying the LSTM
model recursively. The data used to train this algorithm are telemetry data which contains event
code data, time when the event happened and repair data, which is used to cut the sequence of
event code data. We treated one sequence leading to a failure (failure date) as one paragraph of
words. Employing this algorithm, we can accurately predict the next sequence of event codes
and the resulting time until the next failure.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
Recommended Citation
INC, HP, "A FAILURE PREDICTIVE ALGORITHM USING SEQUENCE OF EVENT CODES WITH A DEEP LEARNING MODEL (LSTM)", Technical Disclosure Commons, (January 24, 2020)
https://www.tdcommons.org/dpubs_series/2894