1. Designing a tool to support fault diagnosis
and help with predictive maintenance of
HVAC electrical actuators
AUTODIAG
J. Vachaudez
23/11/2017
: :
2. Outline
Part 1 : Context
Part 2 : Data Extraction
Part 3 : Test Bench
Part 4 : Data Manipulation & Visualization
Part 5 : Imbalance Detection - A Machine Learning
Example
2 : :
6. Maintenance types
Types of maintenance :
Reactive maintenance : Repair when it is broken
Preventive maintenance : Repair at regular intervals
Predictive maintenance : Follow the state of the equipement
All together . . .
Proactive maintenance : Operate in the best conditions
6 : :
7. Maintenance types
Types of maintenance :
Reactive maintenance : Repair when it is broken
Preventive maintenance : Repair at regular intervals
Predictive maintenance : Follow the state of the equipement
All together . . .
Proactive maintenance : Operate in the best conditions
6 : :
11. Unit conversion
x = X · sin(ωt) (1)
dx
dt
= v = ω · X · cos(ωt) (2)
dv
dt
= a = −ω2
· X · sin(ωt) (3)
• Displacement :
µm Pk − Pk
• Speed : mm/s RMS
• Acceleration : g s Pk
t
d
9 : :
12. Unit conversion
x = X · sin(ωt) (1)
dx
dt
= v = ω · X · cos(ωt) (2)
dv
dt
= a = −ω2
· X · sin(ωt) (3)
• Displacement :
µm Pk − Pk
• Speed : mm/s RMS
• Acceleration : g s Pk
t
d
9 : :
13. Unit conversion
x = X · sin(ωt) (1)
dx
dt
= v = ω · X · cos(ωt) (2)
dv
dt
= a = −ω2
· X · sin(ωt) (3)
• Displacement :
µm Pk − Pk
• Speed : mm/s RMS
• Acceleration : g s Pk
t
d
9 : :
14. Units – Order
Définition
The order number may be defined by the quotient of the analyzed
frequency and the reference shaft rotation frequency.
#ordre =
Fpeak[Hz]
Frot[Hz]
• The order number is dimensionless but often represents as a X
• It represents the number of times that the phenomenon occurs per
revolution
10 : :
15. Units – Order
Définition
The order number may be defined by the quotient of the analyzed
frequency and the reference shaft rotation frequency.
#ordre =
Fpeak[Hz]
Frot[Hz]
• The order number is dimensionless but often represents as a X
• It represents the number of times that the phenomenon occurs per
revolution
10 : :
36. Measurements Database Extraction
Need to create a tool, able to extract all measurements from the MHM
database, one at a time.
Results
• ≈ 10 days of extraction 24/7
• ≈ 15000 text files
• ≈ 8 GB of text files
27 : :
37. Measurements Database Extraction
Need to create a tool, able to extract all measurements from the MHM
database, one at a time.
Results
• ≈ 10 days of extraction 24/7
• ≈ 15000 text files
• ≈ 8 GB of text files
27 : :
44. Equipment Database
Database creation
Needs
• Create an equipment database ;
• Bearing and belt informations for each equipments ;
• Other informations about equipment can be useful (criticity,
type,. . .)
• Need to easily edit the equipment database.
34 : :
45. Equipment Database
Database creation
Needs
• Create an equipment database ;
• Bearing and belt informations for each equipments ;
• Other informations about equipment can be useful (criticity,
type,. . .)
• Need to easily edit the equipment database.
Figure 9 – Equipments sheet header
34 : :
46. Equipment Database
Database creation
Needs
• Create an equipment database ;
• Bearing and belt informations for each equipments ;
• Other informations about equipment can be useful (criticity,
type,. . .)
• Need to easily edit the equipment database.
(a) Manufacturer (b) Types
Figure 9 – Manufacturer and Types drop list
34 : :
51. Data Visualization
Development of a Bokeh application in order to visualize data.
Goal
• Select data based on :
Geographic location (customer, area, equipment)
Fault diagnose in the report
Type of measurement
• Compute indicators on each selected measurement
• Visualize data
39 : :
60. Supervised learning
SVM
y
x
Figure 14 – Support Vector
Machines
Caracteristics
• Binary classifier
• Find the best separating
hyperplane that separates
the data
45 : :