Automatic tools for fault detection and diagnostic of HVAC systems for hotel and office
Ph.D., H. Vaezi-Nejad, M. Jandon, Ph.D., J.C. Visier,
CSTB (Centre Scientifique et Technique du Bâtiment)
B. Clémençon, F. Diot, J.M. Jicquel
EDF/ARIPA (Electricité De France)
Faults and malfunctioning of HVAC systems in building can lead to important
waste of energy, decrease of user’s comfort and deterioration of building and
its facilities. These faults are sometime difficult to detect with present tools and
can remain in the building during long period of time.
The aim of our survey is to develop automatic Faults Detection Diagnostic
(FDD) tools for helping building managers or building engineers to improve
their task of Buildings facilities supervision.
Faults and malfunctioning of HVAC systems in building can lead to important waste of energy,
decrease of user’s comfort and deterioration of building and its facilities. These faults are sometime
difficult to detect with present tools and can remain in the building during long period of time (several
months or years).
The aim of our survey is to develop automatic Faults Detection Diagnostic (FDD) tools for helping
building managers or building engineers to improve their task of buildings’ facilities supervision.
For this survey we have first held an inquiry among buildings managers and maintenance experts to
rank the HVAC systems faults to study on.
Then we have developed FDD tools with the help of Building Management System (BMS) experts
and building managers. We have tested these tools with simulated data and real data collected from
BMS of two test buildings.
Today, we are in the validation phase, testing the tools in-line in the two buildings:
- In the first building, an hotel building, the manager use the FDD tool,
- In the second building, a commercial building, an engineer of the maintenance teams uses the
The next phase will be to develop more generic and robust tools that can be installed in a set of hotel
and office building.
This work has been done in the framework of International Energy Agency Annex 34 “Computer-
aided Evaluation of HVAC System Performance: the Practical Application of Fault Detection and
Diagnosis Techniques in Real Buildings” .
In the field of FDD there is two main approaches: top-down or bottom-up approaches  and .
With the bottom-up approach, the system or equipment to survey is break down into subsystems or
elementary elements in order to identify the causes of malfunctioning. Then, different mathematical
methods (usually based on the comparison of non-faulty model of the element with the real element)
can be used in order to detect the faults of the element . This approach is usually exhaustive,
detailed, make easy the following task of diagnosis (definition of the cause of a fault) and is mainly
maintenance team user oriented. But on the other hand, the detailed work can produce large number of
cases with the difficulty to find a solution or to define the main problem to solve.
The top-down approach is based on the global analyse of the building. The main faults of the building
are defined according to different criteria: waste of energy, discomfort for occupants, deterioration of
building and its facilities, difficulty for the users to find the faults with traditional systems (BMS,
vibration analysis, …), etc. Then, mathematical approach similar to the bottom-up approach can be
used to detect fault  and . This approach allows a gradual work (from important to less important
faults) on different faults in the building, an effective definition and orientation of the work on faults
that cause serious consequences in the building and is mainly building manager user oriented. But on
the other hand, the global view makes difficult any detailed diagnosis.
Our FDD method is based on a top-down approach and a development oriented from the beginning to
the end-users needs.
For our development we have followed the seven next phases:
1) Definition of the major faults for managers of hotel and office building equipped of convectors
and fan coil unit. This list of fault has been established by interviews of a large number of building
managers and building engineers.
2) Expert analysis of the faults and selection of those that had the most important consequences
according to fourth criteria:
- Energy consumption and its costs,
- Deterioration of facilities,
During this phase, we have selected two buildings, a hotel and an office building also with two
different BMS' supplier. We have begun the monitoring of these buildings and asked their
managers to archive measurements that could serve us in the development of our faults detection
3) Development of FDD method based on the expert rules. We have begun to develop for each fault
a book of specification in order to establish needs and constraints for the fault detection methods.
4) Assessment of FDD method with buildings simulators for hotel and office buildings. We have
realised for this phase building simulators  in order to have databases in normal and faulty
conditions of the buildings.
During this phase we have established and structured the databases of hotel and office building.
The state of our databases is summarised in the following table.
Building Type Equipment State Exploitation
Real Data Started Used to validate
Altiport since 12/12/98 the FDD tool
Hotel Used to assess
Simulated Convector Normal case
fault detection method
Data + 10 Faulty case
Fan Coil Unit Not realized
Real Data Started Used to validate
Fan Coil Unit
EDS since 18/02/99 the FDD tool
Used to assess
Convector Normal case
Office fault detection method
Simulated + 7 Faulty case
Used to assess
Fan Coil Unit Normal case
fault detection method
+ 12 Faulty case
Table 1. State of the databases used for developing FDD tools .
5) Development of user interfaces for our FDD tools.
6) Evaluation of FDD tools off-line with the real building databases. During this phase we have
tested and adapted the FDD methods and the FDD users’ interfaces.
7) Evaluation of FDD tools in-line in the building with the help of end-users.
The last phase will be to assess fault detection software on other sites and to develop more generic
FDD tools that can be implemented in a set of hotel and office buildings.
The following graph summarises our development process of FDD tools:
Inventory &Analyse of
faults to detect
Implement faults 2 Fit interface according
to detect to users demands
Selection of faults to treat
with BMS suppliers
4 3 Implement 5
Assesment of FDD Development of FDD Development of user
methods with simulation methods interface
Test the Adjust FDD
Adjust methods Correct and
Correct simulations in site
or models methods
Off-line validation with real
Implement Required link
7 Optional link
On-Line validation in
Figure 1. FDD development process chart.
Buildings and databases description
The main features of our two buildings are the use of electricity as main source of energy and the
individualisation of comfort conditions.
Rooms and offices of these buildings are equipped therefore of individual systems of heating and air-
conditioning: electric convectors in rooms of the hotel and fan coils unit in offices.
These systems are equipped of intelligent room controllers. They allow the transfer toward the BMS
central unit of information useful for the managers to survey its buildings. These information can also
be archived on the central unit in order to permit verifications and the further balances on facilities.
It is therefore possible to have a large number of information on the supervisor: indoor temperatures,
orders of actuators, states, etc. in each room of the building. This important quantity of information is
difficult to analyse by the building manager who has daily just a short time to use the central unit.
The aim of our survey is to develop automatic FDD tools for helping building managers in order to
facilitate and to improve their usual diagnosis task of the state of their buildings and facilities.
For the hotel, we have chosen to monitor 11 rooms. These 11 rooms are distributed on the north front,
the south front and in the different floors. The selected rooms are presented on the next synoptic.
Figure 2. BMS synoptic of the hotel building representing the rooms and different facilities.
For the hotel building we monitor the following measurements.
Level Monitored data
The outside temperature, cyclic ratio of the limiter, the cyclic ratio
of the south and north floor heating system
The indoor temperature, the indoor temperature setpoint, the
electricity demand for each convector.
Hot water tanks The running permission of heating for the tank, the hot water
(to produce sanitary hot water) storage temperature, the hot water consumption.
The powers subscribed for each rate, the global consumption of
Electricity energy meters
electric energy during off-peak hours, full hours and peak hours
Level Monitored data
and the electric energy consumption for the heating.
The running permission of lighting for 2nd, 3rd, 4th and 5th floor
and the running permission of the Jacuzzi.
Table 2. Database of the hotel building.
Office Building data Base
For the office building, we have chosen to monitor 10 different offices. These 10 offices are
distributed on the Southwest, Northeast front and in the 1st and 2nd floor. Offices selected on the 1st
floor are presented on the next graphic.
Figure 3. Plan of the second floor of the office building with the selected offices.
For the office building we monitor the following measurements.
Level Monitored data
Building The outside temperature
The indoor temperature, the indoor setpoint temperature, the state
Offices of the fan coil unit (start/stop), the percentage of hot and of cold
demand and ventilation speeds, the change/over state.
The supply temperature, the supply setpoint temperature, the state
AHU of the supply and return fan, the control signal of the hot and cold
coils valves, the control signal of electrical coils.
Electricity energy meters The powers subscribed for each rate, the global consumption of
Level Monitored data
electric energy for the heating and the air-conditioning (during off-
peak hours, full hours and peak hours).
The departure temperature, the departure setpoint temperature, the
Heat Pump Unit
return temperature, the change/over state, the water flow rate.
The running permission of heating for the tank, the storage water
Hot water storage tank
Table 3. Database of the office building.
All those information monitored by the BMS are treated by the FDD tools and the results of the
analysis are presented to the building managers. Thus, the building managers can decide if any action
need to be implemented: maintenance task, using BMS for further investigation and more detailed
diagnostic, waiting for more FDD results before decision, …
Fault detection method
The FDD method we have used is based on the following approach.
1) To define the specific symptom for each fault to detect
The definition of a specific symptom for each fault implies to determine the variable or the
parameter of the installation that can highlight the presence of this fault by an abnormal behaviour.
Example of symptom: temperature too low in occupation.
2) To define conditions of validity for each fault in order to limit the false alarms
The definition of validity conditions for each fault implies to determine what are modes of the
building for which the presence of a symptom means the existence of the fault.
Example of validity condition: to validate the detection of low temperature during occupation we
need to be in occupation mode, not in boost mode, to be in normal outside temperature mode (not
too low), etc.
3) To define for each fault the likely causes.
The definition of likely causes is an "up-to-date" procedure that is gained in the time. It consists to
the characterisation of reasons that produces a given fault.
Example of causes: the heating system is manually switched off, open window, control is out of
We have applied this procedure to the set of faults defined in the following list.
Type of Impact Detection
Equipment Faults or symptoms Damaging Complexity
Building Comfort Cost Time Space
boost during the high tariff hours O.B. 0 + to +++ 0 delayed global low
electric heater too frequently used O.B./H + +++ 0 delayed local/global low
filter's fouling O.B./H + + +++ delayed local/global low/high
simultaneous funtionning of heating
O.B./H + +++ 0 delayed local low
late boost (during eating or cooling
O.B. +++ ++ 0 delayed local/global low/high
heaters and overheating during occupancy period O.B./H ++ +++ 0 delayed local/global low
underheating during inoccupancy
O.B./H 0 0 +++ delayed local/global low
abnormal functioning during
ventilation/ O.B./H i +++ + delayed global low
storage heating during the high tariff
H 0 +++ 0 delayed global low/high
Hot Water derogation has no effect H +++ i 0 on line global high
on line /
lack of hot water H +++ i 0 global high
H hotel +++ major impact
O.B. offices building ++ medium impact
+ low impact
0 no impact
i indirect impact
Table 4. List of faults to detect.
Presentation of the FDD Tool
The FDD tools are composed of six main modules presented in the following flow chart.
Module 1 Transfert Data from BMS Database to
Module 2 Eliminate inconsistant values and filter
data with a moving average window
Estimation of operating modes
Estimation of FDD thresholds
FDD rules application
Apply FDD rules
Suggest likely fault causes
Figure 4. Main modules of the FDD tools.
The module “Acquisition Interface” ensures the transfer of data between the BMS databases and FDD
database. This module needs to be adapted different type of BMS (different types of databases).
The module “Data Filtering” eliminates inconsistent data and filtered data with a moving average
The module “Mode estimation” helps to predict the different running modes of the building and its
facilities (occupation/non-occupation modes, boost mode, heating/cooling modes, etc.).
The module “Threshold estimation” calculates the thresholds for FDD rules according to the user
sensibility choices, the different setpoints or estimated modes.
The module “FDD rules application” detect the different faults according to expert rules, the
thresholds and the modes estimated by the previous modules.
Finally, the module “Diagnostic application” suggest likely diagnostic for the detected faults.
The FDD tools are developed with C language for the calculation part. The user interface is at the
present builds with MS Excel.
In the first view (first window) of the FDD tools, the building manager can know quickly if there is
any fault detected, where are the faults (location in the building) and the seriousness of the detected
Selection of the
View of the hotel
View of the time period to
An orange signalet Hot A click on a box give
access to specific A red signaled room
Water Tank indicate a
graphs that help the indicate a serious fault
A click on this box give low important fault
user to understand the
access to details about
the detected faults
Figure 5. First window of the FDD tool for hotel building.
If the building manager needs more details, he can get information about the time of the detection of
the faults and the type of detected faults (explanation about the faults).
Fault detection Explanation about
week the detected fault
Number of fault
Type of detected
Access to specific detected for a
graphs that help the room or a facility
user to understand the
Figure 6. Second window of the FDD tool for hotel building.
We have developed two FDD tools, FDD_Hotel and FDD_Office, for our two different buildings
(hotel and office buildings) and users. In the hotel building, the user is the manager and in the office
building, an engineer of the maintenance teams.
In the two cases we have tried to adapt the tools to the end-users’ demands:
- Easy to use.
- Presentation of the results on a long period of time :
- 4 weeks for the hotel managers who works on a weekly-based period. The HVAC equipment
supervision is a secondary task for the hotel manager as her main job is to give to her client a
- 1 month for the maintenance engineer who generally use the tools more often (several times
during the month period). But the maintenance engineer has to work on different buildings
and he needs at the end of each month a global result in other to plan its different tasks.
- Possibility to select or unselect the faults to detect and to set the sensibility of fault detection
method (in order to view only high, medium or low important faults). These procedures help the
user to organise and prioritise its maintenance tasks and to decrease the rate of false alarms.
- Possibility to access to details for a better understanding of the fault detection process.
Today, and after a first phase of off-line validation we are in the phase of testing the tools in line in the
The next steps will be to test the tools in on a larger sample of buildings in order to have more generic
tools that could be implemented in a set of buildings.
(1) Dexter A L, Benouarets M. 1996. A generic approach to identifying faults in HVAC plants.
ASHRAE Transactions, 102(1), 550-556.
(2) Hyvärinen, J. and al. 1996 Building Optimisation and Fault Diagnosis (BOFD) source book
document - IEA Annex 25). VTT, Finland, ISBN 952-5004-10-4.
(3) Isermann R., 1984, Process fault detection based on modelling and estimation methods - A survey,
Automatica, Vol. 20, n°4, 387-404.
(4) Li X., Visier J.C., Vaezi-Nejad H., A Neural Network Prototype for Fault detection and Diagnosis
of Heating Systems, Ashrae Winter meeting 1997, Philadelphia, in Ashrae Transactions (to be
(5) Pakanen J., Dexter A. L., and al 2001, Computer-aided Evaluation of HVAC System Performance:
the Practical Application of Fault Detection and Diagnosis Techniques in Real Buildings, source
book document, To appear in IEA source book.
(6) Rossi T M and Braun J E. 1997. A statistical, rule-based fault detection and diagnostic method for
vapor compression air conditioners. International Journal of HVAC&R Research, 3(1), 19-37.
(7) Vaezi-Nejad H., Jandon M., Visier J.C., Clemençon B, and al, Real Time Simulation of a Building
with Electrical Heating System or Fan Coil Air Conditioning System, C, CLIMA 2000,
(8) Visier J C., Vaezi-Nejad H., Corrales P. 1999. “A fault detection tool for school buildings.”
ASHRAE Transactions. 105(1): pp. 543-554.