An innovative and automated building diagnostics based on high-level correlation rules and data fusion, using severity indicators to diagnose and identify the possible causes.
Indoor lighting fault detection and diagnosis using a data fusion approach
1. Indoor lighting fault detection and
diagnosis using a data fusion approach
An innovative and automated building diagnostics based on high-level
correlation rules and data fusion, using severity indicators to diagnose
and identify the possible causes
Francesco Marino
francesco.ingmarino@gmail.com
3. Introduction
The building sector is responsible for more than 40% of European energy consumption [1].
Poorly maintained and improperly controlled building equipment wastes an estimated 15% to
30% of energy*. To prevent this waste an appropriate ENERGY MANAGEMENT is needed,
through widespread automated DIAGNOSTICS and CONTROL ACTIVITIES.
*EURIMA - European Insulation Manufacturers Association
4. Introduction
ENEA Research Center’s Smart Village
In the ENEA research center the realization of a demonstrative prototype of Smart Village is
under experimentation. Many smart activities are carried out, with an energy on demand
approach: SMART BUILDING, SMART LIGHTING, SMART MOBILITY.
5. What is a Smart Building?
A Building Automation System (BAS) is an example of a distributed control system
The control system is a computerized, intelligent network of electronic devices designed to
monitor and control the mechanical, electronics and lighting systems in a building
A building controlled by a BAS is often referred to as a SMART BUILDING
6. Focus on the F40 building
Located in the ENEA Research Center, it can be considered as an example of many old
structures belonging to the public administration
3 floors
Constituted by offices and labs
Highly equipped with sensors and actuators
Smart Building
7. The Goal
Implementation of an innovative FDD (Fault Detection and Diagnostics)
method based on data fusion for Smart Buildings
8. Innovative aspects
Applicable to buildings network
Energy and economic savings without making structural changes to the existing building but
only acting on the management of its power consumption
Scale economy with the monitoring of the main working parameters of all the buildings
10. Pre-processing indices of energy and environmental monitored data, for each of potential
anomaly that can occur in a building, represents the symptoms. A situation assessment is done
through the aggregation of the preprocessing previously specified.
Methodology description
Pre-processing Situation assessment
S1a. Simultaneous lights switched on of
an abnormal number of electrical utilities
S30. Large number of occupied rooms
compared to working hours
P1a. Outlier electrical power
P1b. Outlier electrical energy
P8a. Abnormal electricity trend than the
historical one (mon/tue – thur/fri/weekend)
P8b. Abnormal electricity trend than the
previous week
P24c. Percentage of occupied rooms
P51. Working Schedule
11. The cause under exam is obtained combining the explained situations and also gives an
important indication of the building energy performances and users’ behavior.
Methodology description
Situation assessment Cause
S1a. Simultaneous lights switched on of
an abnormal number of electrical utilities
S30. Large number of occupied rooms
compared to working hours
C13a. Indoor lights switched on in relation
to occupied rooms and working hours
12. From sensors data, preprocessing indices is defined. P1a is fuzzified with a sigmoid expression
to normalize the severity of the electrical peak detected, P51 with an inverse gaussian function.
Definition of the diagnostics rule
P1a P8aP1b P8b P51P24c
S1a S30
C13a
Fuzzy LogicFuzzy Logic
Sensors Data
Data Fusion
S1a = P1a OR P1b OR (0.7 * P8a + 0.3 * P8b)
Data Fusion
S30 = P51 AND P24c
Data Fusion
C13a = 0.7 * S1a + (1 - 0.7) * S30
14. The Peak Detection process
SFunction definition
Local analysis
window’s width
S is a peak function, which associates a score (a non-negative real number) S(i, xi, T) to each
element of an uniformly sampled time-series T, containing N values. It computes the average of
the maximum among the signed distances of xi from its k left neighbors and the maximum
among the signed distances of xi from its k right neighbors [3].
S(k,i,xi,T) - m’ > h * s’
15. A 24 hours time span resulted the optimum between speed and reliability in detecting faults.
The analysis takes in consideration 144 samples of power consumption per day (each one is
acquired every 10 minutes). With a new sample, the chronologically oldest is eliminated.
The Peak Detection process
Samples time series dimension
16. Diagnostics results
On 29/3/2013, a fault was
manually generated to tune the
parameters of the membership
functions and to reveal
effectively and in a reliable way
the anomalous event.
A detailed notification of the detected faults was sent to the energy manager, where the values
reveal their critical issues.
17. Diagnostics recap
Hour Number of faults Average value Max value
1 0 0 0
2 1 0.902 0.902
... ... ... ...
6 30 0.774 0.788
7 1 0.760 0.760
... ... ... ...
17 1 0.787 0.787
18 33 0.816 0.842
... ... ... ...
Faults for 2nd floor in detail
Critical faults
Systematic faults due to cleaning services
18. Diagnostics recap
Floor Number of faults Average value Max value
0 102 0.716 0.857
1 57 0.764 0.852
2 72 0.799 0.902
Faults for each floor
19. Future directions
● Keep going on collecting data in order to have a more accurate framework of building’s
energy situation providing eventually operations of maintenance.
● Study of new diagnostic rules applied to conditioning related to both electrical and thermal
measurements. These aspects with the lighting one are the most important factors that
affect energy consumption and continually require to be monitored.
● This diagnostics method will be applied to buildings network with the precise aim at
monitoring more and diverse structures thinking in a future that it would be adopted in a
large scale, in small districts or cities.
20. References
1. IEA (International Energy Agency), Electricity/Heat Data for IEA Europe, 2008.
1. Khan, I., Capozzoli, A., Corgnati, S. & Cerquitelli, T., Fault Detection Analysis of Building
Energy Consumption Using Data Mining Techniques, Energy Procedia 42, pp. 557 – 566,
2013.
1. Palshikar, G.K., Simple Algorithms for Peak Detection in Time-Series, Proc. of 1st Int. Conf.
Advanced Data Analysis, Business Analytics and Intelligence, 2009.
21. Full paper available at:
https://drive.google.com/open?id=0B9stuYwPgesya1NvZnlUNkIwYjQ