Water Leakage Prediction [AI + Noise Logger]
1. Objectives
Using AI (Deep Learning) to predict “potential water leakage” by analyzing audio file record by “Noise Logger”.
2. Background
Existing
practice relies on staffs to check information captured by Noise Logger and
enter information in excel as shown below, it relies on experience of
individual staff to check and determine whether there is potential leakage or
not.
Figure 1 Excel
file with data captured by Noise Logger
3. Technical details
Using “Deep
Learning” which is one type of AI (by using “neural network”) to build an
engine which analyze audio files (training set) captured by Noise Logger,
perform grouping of files, define types of category/group (with information provided
by staff), and to predict “potential water leakage” of new audio file (test
set) captured by Noise Logger.
1.
Develop
C# module to convert “audio file” (sound clip) to image file
Figure 2. Convert sound clip to image file
2.
Develop
Python module to train Neural network based on approach of Deep Learning, to
classify image files to different categories (groups)
3. User
helps to define nature of different image category/group (eg: normal, potential
leakage, etc)
4. Based on result
of training set, develop Python module to check and predict nature (eg: normal,
potential leakage) of new audio file (sound clip) captured by noise logger
5.
Similar
technique can also be used to check “data abnormality” of time series data of “data
logger” and “DMA” (please refer to Appendix 3, figure 5)
Assumption
1.
Available of hardware (Noise
logger or audio recording device) which can record audio file of pipeline on
site
2.
Available training set of audio
files (around 50 - 100 sample files): Around 50 samples from pipeline which has
leakage, and 50 samples from pipeline which is normal
Integrated with existing IT system
After “engine” is available with
proper training set and result of accuracy, can implement below to predict “potential water leakage”
based on audio file collected by “Noise Logger”.
1.
Front-end
staff record audio file of pipeline on site (by using Noise
Logger or audio recording device), assume audio file is in .WAV format.
2.
Through
a simple Web UI:
l Staff upload .WAV file to SFTP server of this project, with Facility ID of pipeline
as a key
l Click a button of the
Web UI to start analysis
3.
Web
UI returns result of analysis – about possibility of water leakage
4. Implementation plan
Hardware Platform
l
Beside sourcing of “Noise
Logger” of “Audio Recording Device”, there is no hardware investment in Phase I
(for Phase II development, assume mobile phone is used so that user can get
result of analysis on site)
l
Assume using resource available
from “Private
Cloud”
Software Development (in-house, around 6 .. 9 months)
l
Design script to upload audio
file to SFTP folder of “Private Cloud”
l
Develop software (C#) to
convert .WAV file to spectrogram file (ie: image file)
l
Develop software (Python) to
train “Deep Learning Engine”
l
Develop software (Python) to
predict possibility of water leakage.
l
Develop simple Web UI (.NET
ASPX, hosted in IIS) to upload WAV file, return result of prediction, maintain
historical records, etc
l
Result will be saved to a
“MS-SQL Express Server”
Remark: All components above
is either free-of-charge or “open-license”
Appendix 1
Block Diagram: Build Deep Learning engine based on training set
(based
on training set)
Appendix 2
Block diagram : Use “Deep Learning Engine”
to analyse audio file, and to predict “potential water leakage”
Noise
Logger and predict potential water leakage
Appendix
3
Usage of similar approach to detect
abnormality of data of “data logger” or “DMA/PMA”, and to create alert alarm
automatically. Below are typical image which may related to abnormal scenario.
Figure 5. Image file of time series data (Data Logger, DMA/PMA)
Comments
Post a Comment