PNNL-19104

Prepared for the U.S. Department of Energy

under Contract DE-AC05-76RL01830

Self-Correcting HVAC Controls:

Algorithms for Sensors and
Dampers in Air–Handling Units

N Fernandez
MR Brambley
S Katipamula

December 2009
 DISCLAIMER

This report was prepared as an account of work sponsored by an agency of the

United States Government. Neither the United States Government nor any
agency thereof, nor Battelle Memorial Institute, nor any of their employees,
makes any warranty, express or implied, or assumes any legal liability or
responsibility for the accuracy, completeness, or usefulness of any
information, apparatus, product, or process disclosed, or represents that its
use would not infringe privately owned rights. Reference herein to any
specific commercial product, process, or service by trade name, trademark,
manufacturer, or otherwise does not necessarily constitute or imply its
endorsement, recommendation, or favoring by the United States Government or
any agency thereof, or Battelle Memorial Institute. The views and opinions of
authors expressed herein do not necessarily state or reflect those of the United
States Government or any agency thereof.

PACIFIC NORTHWEST NATIONAL LABORATORY

operated by
BATTELLE
for the
UNITED STATES DEPARTMENT OF ENERGY
under Contract DE-AC05-76RL01830

Printed in the United States of America

Available to DOE and DOE contractors from the

Office of Scientific and Technical Information,
P.O. Box 62, Oak Ridge, TN 37831-0062;
ph: (865) 576-8401
fax: (865) 576-5728
email: [email protected]

Available to the public from the National Technical Information Service,

U.S. Department of Commerce, 5285 Port Royal Rd., Springfield, VA 22161
ph: (800) 553-6847
fax: (703) 605-6900
email: [email protected]
online ordering: http://www.ntis.gov/ordering.htm
 PNNL-19104

Self-Correcting HVAC Controls:

Algorithms for Sensors and Dampers in
Air-Handling Units

N Fernandez
MR Brambley
S Katipamula

December 2009

Prepared for
the U.S. Department of Energy
under Contract DE-AC05-76RL01830

Pacific Northwest National Laboratory

Richland, Washington 99352

 
Summary 
This report documents the self-correction algorithms developed in the Self-Correcting Heating,
Ventilating and Air-Conditioning (HVAC) Controls project funded jointly by the Bonneville Power
Administration and the Building Technologies Program of the U.S. Department of Energy. The
algorithms address faults for temperature sensors, humidity sensors, and dampers in air-handling units
and correction of persistent manual overrides of automated control systems. All faults considered create
energy waste when left uncorrected as is frequently the case in actual systems. The algorithms are
presented in the form of a highly integrated set of flowcharts and include processes for:

• fault detection,
• fault isolation
• fault characterization, and
• fault correction,

which are suitable for automation in computer software.

The four processes are accomplished in the algorithms using passive observational fault detection,
proactive tests for fault isolation (when needed), additional proactive testing for fault characterization for
some faults, and formulation of a mathematical compensation for the fault to correct for the presence of
the fault and permit continued operation of equipment. The processes flowcharts express the algorithms
as rules based on fundamental physical and engineering principles together with knowledge of the
physical configuration of the HVAC components and their relationships in systems. Models provide the
analytic redundancy for fault tolerance or correction in the absence of redundant physical components.

iii
Acknowledgments 
The work leading to the interim accomplishments reported in this document are the results of work at
Pacific Northwest National Laboratory funded by the Bonneville Power Administration and the Building
Technologies Program of the U.S. Department of Energy, which the authors gratefully acknowledge.

iv
Nomenclature 
AFDDC automated fault detection, diagnosis, and correction
corr corrected
Δ change
ε error
FPMI Factory Plant Management Interface
H enthalpy
HVAC heating, ventilating and air conditioning
KI integral constant in PI or PID controller
MA mixed air
n current time step number
N total number of time steps
OA outdoor air
OAF outdoor-air fraction
PI proportional-integral type controller
PID proportional-integral-derivative type controller
PNNL Pacific Northwest National Laboratory
RA return air
RH relative humidity
SCC Self-Correcting Controls
t time
T temperature
W humidity ratio
 

v
 

vi
Contents 
Summary .......................................................................................................................................... iii 

Acknowledgments ............................................................................................................................ iv 

Nomenclature .................................................................................................................................... v 

Figures ............................................................................................................................................ viii 

Tables ................................................................................................................................................ x 

1 Introduction ................................................................................................................................... 1 

2 Algorithm Flowcharts.................................................................................................................... 2 

2.1 Passive Diagnostics ............................................................................................................. 6 

2.2 Proactive Diagnostics .........................................................................................................14 

2.3 Fault Correction .................................................................................................................20 

2.4 Decision to Run Passive and Proactive Diagnostics ..........................................................31 

2.5 The Role of Tolerances ......................................................................................................34 

References ........................................................................................................................................35 

 
 

vii
Figures 
Figure 1: Top-Level Flowchart for Self Correcting Controls for Air Handlers ................................ 3 

Figure 2: Flowchart for the Process of Storing Data for Detection of a Hunting Outdoor-Air
Damper ...................................................................................................................................... 4 

Figure 3: Summary Flowchart for Passive Diagnostics for Sensor and Damper Faults ................... 7 

Figure 4: Flowchart for Automatic Fault Correction: Override Manual Control.............................. 8 

Figure 5: Passive Diagnostics for Detecting Temperature Sensor Faults ......................................... 9 

Figure 6: Flowchart for Passive Diagnostics for a Fault Associated with the Minimum Occupied
Position Too Open or Too Closed ............................................................................................10 

Figure 7: Flowchart for Passive Detection of Relative Humidity Sensor Faults .............................11 

Figure 8: Flowchart for Automatic Detection and Correction of Damper Hunting .........................13 

Figure 9: Passive Diagnostics: Automatic Detection and Correction of Hunting - Example ..........14 

Figure 10: Top Level Flowchart for Proactive Diagnostics.............................................................15 

Figure 11: Flowchart for Proactive Diagnostics for RH Sensors.....................................................17 

Figure 12: Flowchart for Proactive Diagnostics for Temperature Sensor and Damper Faults (Test
1) ..............................................................................................................................................18 

Figure 13: Flowchart for Proactive Diagnostics for Temperature Sensor and Damper Faults (Test
2) ..............................................................................................................................................19 

Figure 14: Top-level Flowchart for Fault Correction ......................................................................21 

Figure 15: Flowchart for Correction of a Biased RA Temperature Sensor .....................................23 

Figure 16: Flowchart for Correction of a Biased MA Temperature Sensor.....................................24 

Figure 17: Flowchart for Correction of a Biased OA Temperature Sensor .....................................25 

Figure 18: Flowchart for Correction of a Biased OA Relative Humidity Sensor ............................26 

Figure 19: Flowchart for Correction of a Biased MA Relative Humidity Sensor ...........................27 

Figure 20: Flowchart for Correction of a Biased RA Relative Humidity Sensor ............................28 

Figure 21: Flowchart for Correction of the Outdoor-air Damper Minimum Occupied Position
Too Open ..................................................................................................................................29 

Figure 22: Flowchart for Correction of the Outdoor-air Damper Minimum Occupied Position
Too Closed ...............................................................................................................................30 

Figure 23: An Example of the Process of Bisecting the Control Signal Interval to Correct an
Outdoor-air Damper Minimum Occupied Position that is Too Open .......................................32 

Figure 24: Flowchart of the Process for Deciding to Initiate Passive Diagnostics ..........................32 

viii
Figure 25: Flowchart of the Process for Deciding to Initiate Proactive Diagnostics .......................33 

 
   

ix
Tables 
Table 1: Nested Structure of Self-Correcting Controls Flowcharts .................................................. 2 

Table 2: Proactive Tests Temperature Sensor Test Results .............................................................20 

x
1  Introduction 
This report documents the self-correction algorithms developed in the Self-Correcting HVAC Controls
project funded jointly by the Bonneville Power Administration and the Building Technologies Program of
the U.S. Department of Energy. The algorithms address faults for temperature sensors, humidity sensors,
and dampers in air-handling units. The algorithms are presented in the form of a highly integrated set of
flowcharts and include processes for:

• fault detection
• fault isolation
• fault characterization, and
• fault correction.
These processes are sufficiently complex and intertwined that clear separation of them into separate
flowcharts is not entirely possible; therefore, some flowcharts contribute to more than one of these
processes and address faults with more than one type of physical component (e.g., temperature sensor and
damper faults).

The algorithms detect/correct the soft faults and detect/report the hard faults listed below:

Temperature Sensor Faults

Biased mixed-air (MA) sensor, soft
Biased outdoor-air (OA) sensor, soft
Biased return-air (RA) sensor, soft
Erratic mixed-air sensor, hard
Erratic outdoor-air sensor, hard
Erratic return-air sensor, hard

Damper Faults
Outdoor-air damper minimum occupied position is too open or closed, but damper is fully modulating,
soft
Dampers hunt, soft
Damper stuck fully open, completed closed or between, hard
Outdoor-air damper does not modulate to fully open (100% OA), hard
Outdoor-air damper does not modulate to completely closed (100% RA), hard

Relative Humidity (RH) Sensor Faults

Biased mixed-air RH sensor, soft
Biased outdoor-air RH sensor, soft
Biased return-air RH sensor, soft
Erratic mixed-air RH sensor, hard
Erratic outdoor-air RH sensor, hard
Erratic return-air RH sensor, hard

System Level Fault

Automatic control overridden too long, soft

1
2  Algorithm Flowcharts 
This section presents the series of nested flowcharts that describe the self-correction algorithms (see Table
1 for an overview of the flowcharts and their relationships). Fault correction requires first detecting that a
fault has occurred, identifying or isolating the specific fault, and characterizing the fault. Therefore, self-
correction cannot be implemented in isolation from these other processes, and the flowcharts capture all
of them. All of the flowcharts are identified in Table 1.

Table 1: Nested Structure of Self-Correcting Controls Flowcharts

Self‐Correcting Controls        
   • Run Passive Diagnostics?          
   • Passive Diagnostics          
      – Automatic Fault Correction: Override Manual Control 
      – Passive Diagnostics: Temperature Sensors 
      – Passive Diagnostics: Minimum Occupied Position 
      – Passive Diagnostics: Relative Humidity Sensors 
      – Passive Diagnostics: Automatic Detection and Correction of Hunting 
      – Find Current Bias in MA RH Sensor    
      – Find Current Bias in OA RH Sensor    
      – Find Current Bias in RA RH Sensor    
   • Run Proactive Diagnostics?       
   • Proactive Diagnostics          
      – Proactive Diagnostics: RH Sensor Fault    
      – Proactive Diagnostics: Temperature Sensor and Damper Faults 1 
      – Proactive Diagnostics: Temperature Sensor and Damper Faults 2 
   • Fault Correction             
      – Biased Mixed‐Air Temperature Sensor    
      – Biased Return‐Air Temperature Sensor    
      – Biased Outdoor‐Air Temperature Sensor 
      – Biased Mixed‐Air RH Sensor       
         ο Biased MA RH Sensor: Collect Data for two unique MA state 
         ο Biased MA RH Sensor: Reconfigure RH control algorithm 
      – Biased Outdoor‐Air RH Sensor    
         ο Biased OA RH Sensor: Collect Data for two unique OA state 
         ο Biased OA RH Sensor: Reconfigure RH control algorithm 
      – Biased Return‐Air RH Sensor       
         ο Biased RA RH Sensor: Collect Data for two unique RA state points 
         ο Biased RA RH Sensor: Reconfigure RH control algorithm 
   • Store Data for 'Hunting' Passive Diagnostics    

2
The top-level flowchart (see Figure 1) shows the overall process schematically including the economizer
control system and the automated fault detection, diagnosis, and correction (AFDDC) processes, which

Figure 1: Top-Level Flowchart for Self Correcting Controls for Air Handlers

3
execute simultaneously and communicate with one another. The economizer control system loop is
enclosed in the red, dashed box, and the AFDDC loop is enclosed by the dashed, green boundary. The
control system loop is programmed into an economizer controller, and consists of a proportional-integral
(PI) control loop that recalculates on a fixed time interval, referred to as the system time step. In addition
to using its own sensors to provide measured data for control of the dampers and the cooling coil, the
control system loop sends its sensor readings to the AFDDC loop. The control system also passes to the
AFDDC process a time history of damper control signals, which is used for detecting a hunting damper.

Hunting actuators are characterized by undamped (or poorly damped) control response, which causes the
actuator to oscillate indefinitely about the desired position. Figure 2 shows the process of creating a
database for the time history of damper control signals. This process could be implemented in the control
code with the results passed to the AFDDC process. In the figure, the damper’s control signal is a scaled
value from 0 to 100, where the value of 0 corresponds to the lowest current or voltage signal provided to
the dampers and the value of 100 is the highest. When operating properly, a signal of 0 should actuate the
connected dampers to a position where the outdoor-air damper is fully closed and the recirculation (i.e.,
return-air) damper is fully open. A signal of 100 should actuate the linked dampers to fully open the
outdoor-air damper and fully close the recirculation damper.

Figure 2: Flowchart for the Process of Storing Data for Detection of a Hunting Outdoor-Air Damper

4
The AFDDC process involves taking the virtual sensor readings from the control system, then following a
sequential process of passive fault detection, proactive fault diagnosis, and fault correction. Virtual
sensor readings are the readings intended to be used by the AFDDC to modify the actual readings from
the sensors. This modification includes accounting for previously detected and corrected biases by
subtracting the bias function. In the passive detection process, measured values from the sensors are
processed through a series of rules designed to isolate a set of measurements that either violate
thermodynamic laws or do not conform to the expected operational characteristics of the system. These
tests are referred to as passive tests because they rely entirely on observational data collection with no
interruption or interference with normal system operation. When a fault is detected in the passive stage,
proactive diagnostics are run to isolate the sensor or actuator causing the fault. The proactive diagnostic
process involves taking physical control of the system away from the automated control loop and
performing a series of tests specifically designed to isolate the detected fault (usually to one component or
piece of equipment).

In developing these algorithms, the assumption has been made that only one fault occurs at a time. Faults
are assumed to occur infrequently enough that any previously occurring faults can be detected and
corrected before the onset of another fault. This imposes some limitations on applicability of the
algorithms. The assumption should be valid, provided that the entire AFDDC process can be completed
in the order of seconds or minutes. The probability of two faults occurring precisely simultaneously or
within seconds or minutes of one another is very small, unless they are causally linked so that the
occurrence of one precipitates the other. However, as the time required to perform the AFDDC process
increases, the probability of a second fault occurring increases in proportion to the increase in time.
Moreover, the AFDDC process will not resolve and correct faults in a system when it is first applied, if
the system has several pre-existing faults. The process would best be applied to systems that have been
commissioned or thoroughly serviced before installation of a system that implements the AFDDC process
on it. Having said that, it is still open to question whether the process might still be useful for detecting,
isolating and correcting multiple pre-existing faults by successively applying the procedure to one fault at
a time. This question should be addressed in future research. Furthermore, the algorithms documented in
this report are specific to air handlers using enthalpy-based control of economizing or dry-bulb
temperature-based control but with additional sensors installed to measure the humidities of the air
streams. The humidity sensors are necessary to provide sufficient redundancy to fully isolate all potential
faults. 1

Any ‘soft faults’ (or automatically correctable faults) that are diagnosed in the proactive tests are
immediately corrected with the correction processes. Physical (“hard”) faults, such as stuck dampers, can
be detected and alerts of their occurrence provided to building operators, but these faults must be
manually repaired.

The three main processes within the AFDDC process--fault detection, fault isolation and fault correction-
may run serially, or may run in alternate loops where initiation is governed by the decisions to run the
passive or proactive tests. These decisions will be described in more detail later.

1
Another option would be to use an economizer with redundant temperature sensors. Such cases will be
documented in later research.

5
2.1 Passive Diagnostics
An overview of the algorithm for passively detecting faults is shown in Figure 3. Before any of the
individual passive tests are done, the algorithm checks to ensure that the system is not in “manual
control.” In manual control, the user suspends automatic control, taking control of the system away from
the automatic control loop either to perform maintenance or often because the user is not satisfied with
the way the system is operating automatically. If the system is in manual control, the normal set of
passive tests is skipped, but a fault correction algorithm is applied, as shown in Figure 4. The algorithm
resets the control system back to automatic, if manual control has been enabled for longer than a specified
time limit (the variable manualTimeLimit in Figure 4). This is done to compensate for the user having
forgotten to return the system to automatic control or as a mechanism to force the user to find a better way
to correct problems with the system than resorting to manual control. Manual control generally corrects
problems only temporarily and degrades system performance over the long term, thus our decision to
default to automatic control.

If the system is not under manual control, all four individual passive tests are run in sequence until a fault
is detected or until all of the tests have been run without detecting any faults. The first of these, the
passive temperature sensor test, is shown in Figure 5. This test reveals whether a temperature sensor fault
has led to the measurement of a thermodynamically impossible condition, where the mixed-air
temperature is higher or lower than both the outdoor-air and return-air temperatures. This does not,
however, reveal which temperature sensor is faulty.

Figure 6 shows the passive diagnostic algorithm for a fault in the minimum position of the outdoor-air
damper when the indoor space served by the air handler is occupied, which we refer to as the minimum
occupied position. This is the damper position that is set to provide the minimum outdoor-air ventilation
required for the building during occupied hours. This position generally should be maintained when the
building is occupied and outdoor conditions are not compatible with air-side economizing. Correct
outdoor-air damper positioning is checked under these conditions by comparing the fraction of outdoor air
in the mixed-air stream, determined from measured values of the outdoor-, return-, and mixed-air
temperatures, to the expected outdoor-air fraction (OAF) corresponding to the minimum occupied
position (i.e., OAF required to meet the minimum ventilation rate for the indoor space served). Because
this test of damper position relies on measurements by temperature sensors, an incorrect value for the
OAF could result from either a fault in positioning the damper or a faulty temperature sensor. To resolve
this ambiguity and isolate the fault, proactive diagnostics are used.

Figure 7 shows the passive diagnostic algorithm for detecting faults in relative humidity sensors. The
logic in this algorithm is similar to that used for passive diagnostics for temperature sensor faults (in
Figure 5). Unlike the algorithm for temperature sensors, however, the algorithm for relative humidity
sensors employs three tests to detect potential faults. Each of these tests is based on the physical
requirement that the mixed-air stream must have values for its temperature and moisture content that are
between those of the outdoor-air and return-air streams. The first test checks whether the mixed-air
humidity ratio determined from the measured values of the mixed-air relative humidity and temperature is
between the humidity ratios of the outdoor- and return-air streams. The second test checks whether the
enthalpy of the mixed air, calculated from its dry-bulb temperature and humidity, is between the outdoor-
air and return-air enthalpies determined from measurements of their dry-bulb temperature and humidity.

6
 

Figure 3: Summary Flowchart for Passive Diagnostics for Sensor and Damper Faults

7
Figure 4: Flowchart for Automatic Fault Correction: Override Manual Control

8
 

Figure 5: Passive Diagnostics for Detecting Temperature Sensor Faults

9
  

TMA − TRA
OAF =
TOA − TRA

Figure 6: Flowchart for Passive Diagnostics for a Fault Associated with the Minimum Occupied
Position Too Open or Too Closed

10
 

Figure 7: Flowchart for Passive Detection of Relative Humidity Sensor Faults

11
The third test determines whether the measured relative humidity of the mixed air is equal to or greater
than (i.e., not less than) the measured values of relative humidity for the outdoor air and return air. If any
of these conditions is violated, the algorithm concludes that a relative humidity sensor fault exists and a
global flag for this fault is activated.

Figure 8 provides a flowchart for the passive diagnostic algorithm for automatically detecting and
correcting “hunting” dampers.” Damper hunting is characterized by inadequately damped oscillations in
damper position about the required damper position. The algorithm uses a history of the value of damper-
signal developed in the process shown in Figure 2, in which the values of the damper signal are passed to
the AFDDC process from the control system. The history is recorded in a database set up to contain the
damper signal history for the N most recent control time steps. The value of N necessary for successful
detection of damper hunting is not known and will be determined through laboratory tests.

The algorithm for detection of hunting uses changes in the sign of the time derivative of damper signal to
determine the number of damper oscillations. Each time the time derivative of the damper signal changes
from positive to negative or from negative to positive, an oscillation is counted as occurring. The
dampers are then determined to be hunting if the magnitude of the oscillation is above a specified
threshold, called the high amplitude threshold (HighAmpThreshold), and the number of oscillations in the
time period corresponding to N time steps exceeds a sign change threshold (SignChangeThreshold).
Appropriate values for these thresholds will be determined in laboratory tests, balancing the ability to
detect hunting with the rate of occurrence of false positive indications of hunting. False positive
indications of the damper-hunting fault can result from normal oscillations caused by changing outdoor-
air temperatures and changes in set points, thus the thresholds for filtering out small numbers of
oscillations and oscillations of small amplitude.

When the dampers are determined to be hunting, the correction algorithm activates. To correct damper
hunting, the integral constant (KI) in the proportional-integral (PI) controller commonly used in HVAC
control applications is reduced by 10% 2 . A value of KI set too high (relative to the proportional constant)
can lead to hunting. Therefore, reducing this value should decrease oscillation. Empirical procedures
have existed for many years for setting (or “tuning) the values of the constants in PI and proportional-
integral derivative (PID) controllers, such as the Zeigler-Nichols method. (Ziegler and Nichols 1942,
Åström and Hägglund 2006) Use of such methods generally involves significant user interaction, and
they, therefore, are not suitable for automatic correction algorithms. Instead, we use a relatively small
incremental adjustment in KI in steps to increase damping and reduce hunting oscillations until acceptable
damper behavior is obtained. The selected incremental reduction of 10% in the integral constant is
intended to be conservative, and if the system still exhibits hunting behavior after adjustment, the
AFDDC process again detects hunting at a later time, and another 10% correction is applied. This
process continues to increment KI downwards by 10% until the dampers no longer hunt.

2
Laboratory tests will be used to investigate the optimal value for this correction, and it can be programmed as a
variable in the software implementing this algorithm.

12
 

Figure 8: Flowchart for Automatic Detection and Correction of Damper Hunting

13
A graphical example of hunting is shown in Figure 9, where the damper control signal (SignalOAt) is
plotted as a function of the number of time steps (n). For the example, a high amplitude threshold of 5 is
assigned for the damper signal (the damper signal is a normalized value from 0 to 100), a sign change
threshold of 4 is used, and N = 30 for the hunting detection algorithm. A theoretical damper signal is
plotted in the light blue points, with red, highlighted points at the times steps where the algorithm counts
a sign change. The signal oscillates about a value of the damper signal of approximately 30. Sign
changes are detected for the first three sign changes, until the amplitude of oscillation becomes
sufficiently damped that the amplitude no longer exceeds the high amplitude threshold, preventing further
sign changes from being counted. Using the values for the thresholds specified, this type of oscillation is
just below the criterion necessary to characterize it as hunting. If the sign change threshold selected had
instead been assigned a value of 2 instead of 4, the algorithm would have classified this behavior as
hunting. This illustrates that the thresholds can be adjusted to empirically set limits for the kinds of
oscillation signatures considered to be hunting.

Figure 9: Passive Diagnostics: Automatic Detection and Correction of Hunting – Example

2.2 Proactive Diagnostics

With the exception of the two faults that can be immediately isolated and corrected (manual control and
damper hunting), insufficient information can be collected passively to isolate faults. To isolate other
faults, proactive tests are performed by overriding automatic control and placing the system in specific
known states. These states are chosen based on the relationships they establish between key variables,
enabling additional analytic relationships to be used to isolate specific faults and the components in which
they occur. The overall proactive diagnostic process is shown in Figure 10. As shown, the proactive
process consists of three primary subprocesses: isolation of a relative humidity sensor fault and two
processes associated with isolating temperature sensor and damper faults. Descriptions of these
subprocesses follow.

14
 

Figure 10: Top Level Flowchart for Proactive Diagnostics

15
Figure 11 shows the algorithm for the proactive tests to isolate RH sensor faults. This test is performed if
and only if a general relative humidity fault was detected in the passive diagnostic process. The test
involves completely closing the outdoor-air damper and then comparing the measured values of the
mixed-air and return-air relative humidities. If the sensors are operating properly, their values should be
equal, because the mixed-air stream under the test condition is composed entirely of return air. If the
measured values are equal, we conclude that the mixed-air and return-air RH sensors are both operating
properly and, therefore, because the passive process detected a fault in one of the RH sensors, by
elimination, the outdoor-air sensor is faulty. If the measured values differ, a second proactive test is
performed in which the outdoor damper is opened fully and the measured values of the mixed-air and
outdoor-air RH are compared. If the values are equal, we conclude that the mixed-air and outdoor-air
sensors are operating properly, and the return-air sensor is faulty. If both tests fail, by exclusion, the
mixed-air sensor must be the one at fault.

If a fault is detected in the passive tests for either the temperature sensors or the minimum occupied
position, the proactive test for temperature and damper faults is run. We choose to do this because a) if
the fault was detected in the passive test for correct minimum occupied position, the test result could have
been caused by either a temperature sensor fault or a damper fault and b) if the fault was detected in the
passive test for temperature sensors, even though that fault must have been with a temperature sensor, this
algorithm will isolate the specific faulty sensor. Moreover, this algorithm will first ensure that the
dampers are operating well enough to diagnose and correct a biased sensor fault. If the test reveals that
the dampers are not operating adequately, a damper fault is flagged, and that fault must be corrected prior
to successful automatic correction of the temperature sensor fault.

The process for proactively testing for temperature sensor and damper faults is shown in Figure 12. The
test involves the same general approach as that used in the RH sensor test. The outdoor-air damper is first
closed completely, and the measured values of the mixed-air temperature and the return-air temperature
are compared. Then, the outdoor-air damper is opened completely, and the measured values of the
mixed-air temperature and the outdoor-air temperature are compared. A number of secondary checks are
performed to determine whether the dampers are faulty. Four different outdoor-air damper faults can be
identified: dampers completely stuck, dampers that won’t modulate to a completely open position (100%
outdoor air), dampers that won’t modulate to a completely closed position (100% return air), and dampers
that modulate fully but do not achieve the desired minimum occupied position.

After the outdoor-air damper has been modulated to both the fully open and fully closed positions in this
proactive test, a check is made whether the value of the mixed-air temperature sensor changed from one
damper position command to the other. If it does not change significantly, the dampers are either
completely stuck or the temperature difference between the outdoor air and return air is too small to
induce a significant change in the mixed-air temperature when the damper positions are changed. In this
case, the algorithm concludes that the test is inconclusive and needs to be repeated under different
conditions. The process then requires waiting for the difference in the measured values for outdoor-air
temperature and return-air temperature to exceed a fixed threshold before the temperature sensor and
damper proactive diagnostic process is repeated. When conditions are appropriate, proactive test 2 for
temperature sensor and damper faults is performed (see Figure 10 and Figure 13). In the second test
(Figure 13), if the measured mixed-air temperature does not change significantly between conditions with

16
 

Figure 11: Flowchart for Proactive Diagnostics for RH Sensors

17
 

Figure 12: Flowchart for Proactive Diagnostics for Temperature Sensor and Damper Faults (Test 1)

18
 

Figure 13: Flowchart for Proactive Diagnostics for Temperature Sensor and Damper Faults (Test 2)

19
the outdoor-air damper fully open and fully closed, a stuck-damper fault exists. If the mixed-air
temperature does change significantly, the original test (Test 1) was run when the values of the actual
outdoor-air and return-air temperatures were too close to one another. The second test, then, should have
a large enough difference between the outdoor-air and return-air temperatures to characterize the
effectiveness of the dampers.

Two more criteria are applied to isolate dampers that modulate but do not open or close completely.
When the outdoor-air damper is shut completely, the mixed-air conditions should equal the return-air
conditions. Because a temperature sensor fault may exist (otherwise the temperature sensor and damper
proactive diagnostic process would not be activated), the temperature sensor measurements cannot be
relied upon in determining whether the conditions of the mixed- and return-air streams are identical.
Under the assumption that only one fault exists at a time, however, the relative humidity sensors can be
trusted as working reliably and used to test for a damper that does not close completely. If at steady state,
the measured RH of the mixed air differs from the RH of the return air when the outdoor-air damper is
signaled to close completely, the outdoor-air damper has not closed completely or leaks, allowing a
significant amount of outdoor air to flow into the mixing box. By similar logic, if at steady state, the RH
of the mixed air differs from the RH of the outdoor air when the outdoor-air damper is commanded to
open fully (and the recirculation damper is commanded to close completely), the outdoor-air damper has
not opened completely or the recirculation damper has not closed completely (because the two dampers
are mechanically linked).

If the dampers pass all the proactive tests, then the equality of the mixed-air temperature and outdoor-air
temperature is checked to distinguish among faults of the outdoor-air temperature sensor, mixed-air
temperature sensor, return-air temperature sensor, and minimum occupied position of the outdoor-air
damper. The conditions leading to each of these faults are shown in Table 2.

2.3 Fault Correction

An overview of the algorithm for fault correction is shown in Figure 14. The process shown initiates
each of the subprocesses for correcting a specific fault based on the findings from the passive and
proactive diagnostics. The algorithm shown also provides detection of a type of inconsistent diagnosis.
The inconsistency may arise if the passive temperature sensor test detects a fault that is later diagnosed as
a minimum occupied position fault (within the temperature and damper proactive diagnostics; see Figure
12, Figure 13 and Table 2). This result indicates that the temperature sensors are not exhibiting a

Table 2: Proactive Tests Temperature Sensor Test Results

Measured Conditions* Fault

TMA1 ≠ TRA1 RHMA1 = RHRA RHMA2 = RHOA TMA2 = TOA Return-air temperature sensor fault
TMA1 ≠ TRA1 RHMA1 = RHRA RHMA2 = RHOA TMA2 ≠ TOA Mixed-air temperature sensor fault
TMA1 = TRA1 RHMA1 = RHRA RHMA2 = RHOA TMA2 ≠ TOA Outdoor-air temperature sensor fault
TMA1 = TRA1 RHMA1 = RHRA RHMA2 = RHOA TMA2 = TOA Minimum occupied position fault
* Subscript 1 indicates measurements for tests with the outdoor-air damper completely closed; subscript 2 indicates
measurements for tests with the outdoor-air damper fully open.

20
 

Figure 14: Top-level Flowchart for Fault Correction

21
significant fault during the proactive tests. In this situation, the diagnostics would be reset and passive
diagnostic processing would begin again.

Figure 15, Figure 16 and Figure 17 depict the processes for correcting faults in the return-, mixed-, and
outdoor-air temperature sensors, respectively. Before correction, the fault is characterized. Values in the
time series of measurements from the faulty sensor are compared directly to values in a corresponding
time series of measurements from one of the two reliable temperature sensors. If the difference between
the measured values from the two sensors is nearly constant over time, the fault is categorized as a sensor
bias, which is correctable. In this case, measured values from the faulty sensor are corrected by
subtracting the average difference between the faulty and reliable sensor from the measured value of the
biased sensor to provide a virtual temperature sensor point, which can be used in place of values directly
from the faulty sensor (e.g., for use in control). If the difference between the time series of measurements
for the faulty sensor and good sensor is not nearly constant over time, the sensor has a time dependent
fault, which in some cases may be correctable but algorithms for doing so have not yet been developed.
For sensors with faults other than simple constant biases, we currently recommend replacing the faulty
sensor.

The processes for correcting biased outdoor-, mixed-, and return-air RH sensors are shown in Figure 18,
Figure 19, and Figure 20, respectively. The correction process for each RH sensor is identical to the
correction process for a corresponding temperature sensor – only with the corresponding RH sensors and
RH measurements taking the place of temperature sensors and temperature measurements.

The algorithms for correcting minimum occupied position faults are shown in Figure 21 (Minimum
Occupied Position Too Open) and Figure 22 (Minimum Occupied Position Too Closed). The actual
determination of whether the minimum occupied position is too open or closed is made in the passive
diagnostics (see Figure 6), and the proactive diagnostics determine if the fault was indeed caused by the
sensor. If the minimum occupied position is corrected in the process, according to Figure 21 or Figure
22, the damper has already been determined to correctly modulate fully open and completely closed. This
is, therefore, considered a soft fault, wherein the damper position (modulated via the control signal) has
gone out of calibration with respect to the expected OAF.

The minimum occupied position of the outdoor-air damper is corrected by using the bisection method. In
the case where the minimum occupied position is too open, the current control signal for the minimum
occupied position is set as the upper bound and a control signal of 0 is set as the lower bound. The
control signal interval between these two signals is then bisected by taking the average of the two. Thus,
when the minimum occupied position is too open and the current minimum occupied position control
signal is 10 (based on a normalized full range from 0 to 100), the first value from bisecting the signal
interval is 5. The system then commands the damper with the new value of the signal, and the OAF is
calculated based on the new measured temperatures at steady state. When the OAF is greater than the
expected (desired) OAF at minimum occupied position, the lower bound is reset to the bisected value. If
the OAF is greater than the expected minimum OAF, the upper bound is reset to the value previously
obtained as the bisected value with the lower bound of zero and the interval bisected again. This process

22
 

Figure 15: Flowchart for Correction of a Biased RA Temperature Sensor

23
 

Figure 16: Flowchart for Correction of a Biased MA Temperature Sensor

24
 

Figure 17: Flowchart for Correction of a Biased OA Temperature Sensor

25
 

Figure 18: Flowchart for Correction of a Biased OA Relative Humidity Sensor

26
 

Figure 19: Flowchart for Correction of a Biased MA Relative Humidity Sensor

27
 

Figure 20: Flowchart for Correction of a Biased RA Relative Humidity Sensor

28
  

TMA − TRA
OAF center =
TOA − TRA

Figure 21: Flowchart for Correction of the Outdoor-air Damper Minimum Occupied Position Too Open

29
 TMA − TRA
TOA − TRA

Figure 22: Flowchart for Correction of the Outdoor-air Damper Minimum Occupied Position Too
Closed

30
is repeated iteratively, with the upper and lower bounds of the bisection eventually converging to the
command signal required to produce the desired minimum OAF.

Figure 23 illustrates the bisection process for the example of correcting a ‘Minimum Occupied Position
Too Open’ fault. In this example, the existing (faulty) damper signal of 38 is yielding a minimum
occupied OAF of 0.46, while the desired minimum occupied OAF is 0.23. The damper signal of 38 is
initially set as the upper bound (Signalhigh with corresponding OAFhigh) with a damper signal of 0 set as
the lower bound (Signallow with corresponding OAFlow). Each of the iterations is then numbered, until
iteration 4 yields an OAF sufficiently close to the desired minimum occupied OAF.

For a ‘Minimum Occupied Position Too Closed’ fault, the process is performed initially using the faulty
damper signal as the lower bound and 100 as the upper bound on the control signal interval that is
bisected.

2.4 Decision to Run Passive and Proactive Diagnostics

Two decision boxes appear in the top-level self-correcting air-handler controls flowchart that direct the
passive fault detection and proactive fault diagnostic processes to execute.

Passive fault detection is run if three criteria identified in Figure 24 are met. The first criterion requires
that at least 5 minutes pass since the passive diagnostics were last run. Although this time constraint
could be changed, the value of 5 minutes seems reasonable for ensuring that passive diagnostics are run
frequently enough to detect and diagnose faults in reasonable time, yet not so frequent as to perform
unnecessary processing. If the time between passive diagnostic tests becomes too short, when the process
is executed in software, the program could crash because the available memory limit is exceeded. The
second criterion requires that no recent inconclusive temperature and damper proactive tests are still
pending completion. Under such a circumstance, the proactive test program would be waiting for
conditions to change sufficiently for performance of another proactive test, therefore, performing passive
diagnostics before the proactive process is concluded would be useless. The final criterion requires that
no undiagnosed and/or uncorrected faults that were previously detected by passive diagnostics be waiting
for proactive diagnostic tests. This could be true if the passive tests detected a fault, but the proactive
tests for that fault were not yet performed. This situation might occur if software implementing the self-
correcting controls provides modes where the proactive diagnostic processes can be initiated according to
a schedule or manually by an operator, rather than immediately by the software. This situation might also
occur when a hard fault is detected, which requires physical repair or replacement of a component. If the
hard fault interferes with performance of passive or proactive diagnostics, then the system would require
an indication that the repair has been completed before resuming operation.

The decision to run proactive diagnostics (see Figure 25) relies on several criteria. If there has been a
recent inconclusive proactive temperature and damper test, the algorithm determines whether the
difference between the outdoor-air and return-air temperatures has changed sufficiently to run the
proactive diagnostics again. If there has not been a recent inconclusive proactive test, the algorithm
checks to ensure that there is currently a fault identified by the passive diagnostics that requires isolation,
and if so, decides based on the mode of software operation whether to run the proactive diagnostics.

31
Figure 23: An Example of the Process of Bisecting the Control Signal Interval to Correct an Outdoor-air
Damper Minimum Occupied Position that is Too Open

Figure 24: Flowchart of the Process for Deciding to Initiate Passive Diagnostics

32
 

Figure 25: Flowchart of the Process for Deciding to Initiate Proactive Diagnostics

33
2.5 The Role of Tolerances
Tolerances are estimated ranges of uncertainty in the measured values from sensors. Each of the
temperature and relative humidity sensors in an air handler is subject to uncertainties in its recorded
values that arise for a variety of reasons. Tolerances capture the uncertainty associated with normal
deviations of sensor readings about the true value of the property measured. To conserve space, explicit
equations including tolerances were not provided in the flowchart boxes; however, anywhere a decision
depends on the relative value of two measurements, tolerances are applied. The propagation of tolerances
through the rules used for fault detection, isolation and characterization helps to rule out false conclusions
regarding the existence of a fault, the specific location of a fault, and characteristics of a fault. An
example of application of tolerances is in evaluation of whether readings from a sensor are different or
equal. Application of the sensor tolerances defines a range of the difference in readings within which the
two readings are not considered significantly different (i.e., they are treated as equal) and only when the
difference in readings is outside this range are the two values considered unequal. Tolerances are
typically expressed as the nominal sensor reading +/- the tolerance.

Consider the decision box in Figure 5, which asks the question “Is TMA > (TRA AND TOA)?” Here, TMA,
TRA, and TOA each have tolerances that define their permissible ranges. So it is not sufficient for the
measured value of TMA to simply be larger than TRA and TOA. TMA must be significantly greater than TRA,
enough to account for the tolerances in TMA and in TRA. The same is true for the relationship between TMA
and TOA. Because of the direction of the inequality, the actual algorithm including tolerances is

Is measuredTMA - toleranceTMA > (measuredTRA + tolerance TRA AND measuredTOA+ tolerance TOA)?

If the tolerances for all three sensors are the same (i.e., toleranceTMA = toleranceTRA = toleranceTOA =
toleranceT, then the overall decision rule becomes:

Is measuredTMA – 2*toleranceT > (measuredTRA AND measuredTOA)? If so, the actual value of TMA is
concluded to be greater than the actual values of TRA and TOA.

A more complicated version of an applied tolerance is found in Figure 6, where a decision box asks “Is
abs(OAF-OAFexp,min) < OAFERRthreshold?” In this equation, OAFERRthreshold represents an implied tolerance.
Because OAF is not measured directly, but is a calculated value dependent on three temperature
measurements (see the equation in the top box of Figure 6), the tolerances of the three measured
temperatures are propagated through the calculation of OAF, providing a range of permissible values of
OAF. According to the rules for propagation of uncertainty from Lindberg (2000), the value of

OAFERRthreshold , based on identical tolerances for the temperature sensors is given by

2 ⋅ (1 + OAF ) ⋅ toleranceT
OAFERRthreshold =
TOA − TRA
. (1)

Tolerances are assigned to all measured variables used in the AFDDC process and propagated through all
equations and decisions that use them.

34
References
Åström, K.J. and T. Hägglund. 2006. Advanced PID Control, pp. 159-169. ISA—Instrumentation,
Systems and Automation Society, Research Triangle Park, North Carolina.

Ziegler, J.G. and N.B. Nichols. 1942. “Optimal Settings for Automatic Controllers.” Transactions
ASME 64, pp. 759-768.

Lindberg, V. 2000. “Uncertainties and Error Propagation- Part One of a Manual on Graphing, Uncertainty
and the Vernier Caliper.” http://www.rit.edu/cos/uphysics/uncertainties/Uncertaintiespart1.html.

35