1

Wall System Identification for Stud Finding

D. Thomas, K. Helbert, A. Loomis, D. Ajilo, F. Grant
Abstract To identify the location of a stud, many rely on
taping a wall to hear a fluctuation in pitch, which is manual
and imprecise. This paper presents a stud finding technique
that measures the mechanical properties of common
drywall panels to find the location of a stud using a
stochastic binary input. This device uses four piezoelectric
sensors to capture the frequency response created by the
solenoid driven by stochastic binary input function.
Through the implementation of system identification
techniques, we are able to determine the frequency response
of the wall and the device can then identify the studs
location. Additionally, a first generation prototype of a
portable device is presented which is capable of capturing
and analyzing the frequency response of a wall to detect the
studs location.
Index Terms Stud Finder, System Identification,
Piezoelectric measurements
I. INTRODUCTION
Stud finding is a problem facing many people who want to hang
that perfect picture on the wall or install shelving. Many
existing stud finders work by magnetically detecting metal,
which works decently if the stud is steel representing many
commercial buildings; However, residential installations are
typically wooden, so these stud finders struggle, only having the
screws to detect [1-2]. Modern capacitive stud finders can more
easily detect these wooden studs based on density measurement
changes, but with the new downside of false positive from pipes
and wiring that can give similar readings and lead to dangerously
nailing through such conduits [3]. Oftentimes, the user resorts to
simply knocking on the wall and listening for fluctuations in
pitch, effective for quickly finding the general vicinity of stud,
but unreliably for accurately locating the center. This results in
many unsightly additional nail holes in a wall that need to be
fixed.
This paper investigates the use of system identification to
measure changes in mechanical properties of a wall to safely,
precisely, and conveniently find the location of a stud. This
improves the experience of many construction and home
improvement projects by eliminating false identification of a
studs location. This is accomplished by inducing a physical
vibration to a wall via a force input from a solenoid to the wall,
and then measuring the resulting propagation through
piezoelectric sensors surround the epicenter. Using this output,
we applied system identification techniques to model the
frequency response of a wall with a stochastic binary input. We
hypothesized that due to the change in wall stiffness, resulting

from the periodic nature of the walls structure, a change in the

impulse response temporally across the wall could be observed.
This would ideally result in the ability to tell the user when the
device is positioned at the edge of a stud and subsequently the
center of the stud. Our device goes beyond other devices because
we are fundamentally measuring the material properties of a
wall this measurement should also provide insight into its load
capacity, or potential material degradation due to moisture.
II. SYSTEM OVERVIEW
This implementation of a stud finder uses a binary stochastic
input imparted to a wall and measured through four piezoelectric
sensors placed in an X pattern around the epicenter. Due to the
change in the material properties across measurement zone
caused by the structural changes in the wall, the piezoelectric
devices have different frequency responses to the same input.
Using system identification techniques, the signals were
analyzed to detect the edge and presence of a stud within a wall
as the stud finder is horizontally traversed across the wall. The
to utilization of a relative measurement the ability for this device
to detect studs is not dependent on absolute wall properties, but
a change in wall properties across the measurement area.
The stud finder is designed to operate under three different
conditions operating over the cavity, and spanning/straddling
the stud, Figure 1. Because of the three different positions the
piezoelectric elements experience the location of the stud can be
found.

Figure 1: Stud finding methodology used by the piezoelectric

stud finder

Figure 2: A. Isometric rendering of piezoelectric stud finder B. Exploded view of device

III. MECHANICAL DESIGN
The mechanical design of the stud finder went through several
iterations to improve the vibrational isolation of the stochastic
input from the piezoelectric sensors. The device integrates the
sensing, amplification, and input into one handheld device,
Figure 2A, that can be used on any wall to detect studs.
The solenoid, part number SH-T1939L, sits in the center of the
structure, Figure 2B, while the center of each sensor is located
50mm radially from the center of the solenoid. The structure is
then placed on the wall, and the wall is stochastically knocked
by the solenoid. Since the sensors, equivalent to part number
CEB-20D64 by CUI Inc., are equidistant from the input point,
the vibration wave propagating from the solenoid contact point
will take the same amount of time to reach each of the sensors
in a uniform wall. Non-uniformity of the wall, in the form of a
discrete wall stud, registers as a different impulse response to the
nearby sensor. The stiffness of the composite drywall-and-stud
combination is distinguishable from the drywall alone, allowing
this configuration to detect studs as the device passes over a
portion of wall with either a vertical or a horizontal stud behind
it.

In order to design a single, enclosed device that both perturbs

and measures the impulse response of a wall, it is important to
isolate the device itself from unwanted vibration. Specifically,
the portion of the device housing the solenoid must be
mechanically isolated from the sensors, and the sensors from
each other, such that vibration cannot propagate through the
device itself to give a false reading to the piezoelectric sensors.
To achieve this, highly compliant vibration damping mounts
were used to connect the solenoid housing to the sensor support
structure, Figure 2.
The sensor support structure is connected to each sensor with a
less compliant damping mount, isolating each piezoelectric
sensor and ensuring the sensors are preloaded to the wall with
consistent contact throughout the measurement process.
IV. ELECTRONIC DESIGN

Figure 4: Flow diagram of control scheme

Figure 3: Picture of stud finder in use on demonstration wall

In the present configuration, the measurement process is

initiated when stochastic signals are generated in MATLAB and
output through an Arduino to drive the solenoid knocker. The
piezoelectric sensors are then read as inputs to an oscilloscope
and the data saved for processing, Figure 4. In a fully integrated

3
implementation, a microcontroller will drive the stochastic
input, onboard, and also perform the signal processing that
conveys the result of the measurement to an LED array that
indicates where a stud is relative to the device.
The solenoid driving board includes transistor and fly back
diode connected to the knocking solenoid in addition to the
power distribution ports, Figure 5. A 9V battery is sufficient to
drive the solenoid with an enough force, approximately 1.4 N
with a 2mm stroke length, to provide useable data to the
piezoelectric sensors, while not damaging the wall. Currently,
both the operational amplifier circuits and the solenoid draw
directly from the 9V power supply. The computer driving its
input signal presently powers the Arduino.

V. EXPERIMENTAL SETUP
An experimental wall was built for testing the device. The wall
consists of an 87x121x1.6 cm piece of drywall supported by a
wooden frame with a 38.24 mm wooden stud in the center, 406
mm (16) away from the side beams as per standard
construction codes. In the initial experimental set-up, the wall
was propped up against a lab bench and the Wall Knocker
device was manually braced against the wall while data was
collected on a two-channel oscilloscope. This configuration
introduced noise to the measurement system due to the inherent
instability of a manually applied normal force. In addition, the
input data had to be collected through the Arduino, which could
not collect data at the same sampling rate as the scope. This
complicated signal processing and severely limited the options
for data analysis.

Figure 5: Power amplification circuit diagram

Four piezoelectric sensors radially located about the solenoid
measure the impulse response of the wall (as illustrated in
Figure 2). The raw output of the piezoelectric sensors in this
configuration is on the millivolt scale, and the signal itself
noisy, so a first order RC filter and an operational amplifier are
implemented to reduce high frequency noise and amplify the
piezoelectric signal prior to data collection, Figure 3. The
Zener diode protects the operational amplifier from substantial
voltage spikes. The LM358 is a dual operational amplifier in an
8-pin package. Therefore, there are four operational amplifier
circuits configured on two boards that each share an operational
amplifier IC[4-5]. The individual operational amplifier
schematic is shown in Figure 3.

Figure 3: Operational Amplifier Circuit Diagram

Figure 4: Final design of the piezoelectric element stud finder

Once the final iteration, Figure 4, of the device was complete,
a new experimental set-up was implemented. Points were
marked along the width of the wall at 28 mm, 50 mm, 80 mm,
and 100 mm from the center of the stud on either side. Three of
the piezoelectric sensors of the wall knocker and the solenoid
actuator input were connected to a four-channel oscilloscope.
The actuator was controlled with MATLAB through an
Arduino. The entire device was powered with a 9V power
supply. One of the piezoelectric feet of the wall knocker was
centered at each of the demarcated points along the wall, Figure
5, and the response of the system to a stochastic binary input
was measured at each position. During all of these tests, the wall
was laid on the floor and a 1 kg weight was placed on the top
of the device to avoid the noise introduced by earlier
experimental configurations. A total of three trials were run for
each position in this configuration.

Figure 5: Position of the piezoelectric sensors

relative to the center of the stud. The measured
distance, d, was taken from the center of right most
piezo, P2, to the center of the stud.

assumption, a white noise input signal was convolved with the

estimate of the impulse response function to get the
corresponding output. This output was then deconvolved from
the stochastic binary input signal to get the inverse of the
impulse response. The measured output signal was then
convolved with the inverse impulse response to get an estimate
of the intermediate output, u(t). The stochastic binary input was
then plotted against u(t) to estimate N(). For the Wiener system
assumption, the input was convolved with the estimate of the
impulse response function to produce an estimate of the
intermediate output, u(t). The measured output was plotted
against u(t) to estimate N(). The goal was to iterate through
these calculations to obtain accurate estimates of the impulse
response function and u(t), but in both cases the estimates of N()
were unrecognizable as functions. This suggests that the system
is highly non-linear, and more advanced non-linear system
identification is required to relate the input and output signals.
The coherence squared was plotted and showed a quick
departure from unity, which confirms the highly non-linear, and
potentially noisy, nature of the system.

Figure 7: A. Hammerstein system non-linear model B.

Wiener system non-linear model

Figure 6: Comparison of responses to impulses imparted in

the cavity of the wall vs. directly onto the stud.

As the system could not be accurately represented by a linear

frequency response plot, the raw data was analyzed using Fast
Fourier Transform (FFT). It was determined from the frequency
content of the data that a sampling rate of 2.5 kHz would be
sufficient, Figure 9.

VI. SIGNAL PROCESSING

A difference can be observed in the raw output data as the device
is moved from above a cavity to above a stud, and the response
appears to be an approximately second-order impulse response
with some non-linearity, Figure 6. In an attempt to isolate the
non-linearity in the system, the input and output were analyzed
as a Hammerstein and Wiener system to determine the statistic
nonlinearity function N(), as shown in Figure 7.
Using MATLAB, the auto-correlation of the stochastic binary
input signal was deconvolved from the cross-correlation of the
input and output signal using Toeplitz matrix inversion to
calculate an estimate of the linear impulse response function
between the output and input. For the Hammerstein system

Figure 8: Power spectrum analysis

5
A power spectrum, Figure 8, was then generated using the auto
correlation of the standardized raw data in MATLAB. This
power spectrum was subsequently divided into bins according
to the observed peaks. It was determined that bins ranging from
0500 Hz would be sufficient as the frequency content seemed
to drop off after 500 Hz.
The peak magnitudes and their corresponding frequencies
within each bin were plotted against the center-to-center
distance of the right most piezo from the stud, Figure 10.
VII. RESULTS
Our results show trends that certain frequencies experience
significant decrease in magnitude as our device approaches a
stud. This occurs specifically around vibrations of about 230
Hz, and the amplitude drop offs are illustrated in the, Figure
10B. This also occurs with vibrations of about 90Hz, albeit
seemingly producing more peaks as the different piezoelectric
sensors cross the stud boundary, Figure 10A.

However, this is not yet conclusive as our system was found

to be extremely non-linear in nature, as confirmed by poor
values of our coherence squared function. We believe these
non-linearities could stem from the mechanical coupling of our
piezoelectric sensors, significant differences in the sensitivity
of the individual piezoelectric sensors, position of the solenoid
impact relative to the stud, and the inherent wall characteristics.
Determining proximity to the stud by calculating the time delay
in a reflected signal off of the stud was also inconclusive as we
believe the vibrations attenuated well before a perturbation
from the solenoid traveled to the stud and back through the
gypsum.
VIII. CONCLUSION
While the device and measurement method has demonstrated
the ability to detect changes in frequency response due to a
given impulse, future work would include calibrating such a
device to known characteristics and consolidating the
electronics into a handheld portable device. We imagine this
device to incorporate LED indicators for feedback and a

6
bluetooth connection in order to utilize common smartphones
as an easy and accessible display / control input source.
Recommended improvements to our methodologies include
calibrating the piezoelectric sensors individually as the variance
in
inherent
characteristics
were
larger
than
anticipated. Additionally, using a Lorentz force actuator as
compared to our solenoid implementation might give more
repeatable impulses from a known displacement over a
solenoids binary on / off force. While this would potentially
sacrifice maximum impact force, it would allow for calculation
of the transfer function of the vibration directly from the
mechanical impulse rather than the applied electric impulse,
removing the solenoids
non-linearities from the
system. Alternatively, placing a piezoelectric sensor directly
between the solenoid striker and the wall would also result in
more accurately measuring when the impulses occur. Faster
frequencies of impulses are also required in order to excite the
high modes of vibration within the wall and glean further
insight into the physical characteristics.
Ultimately, a direct displacement measurement is most likely
required in order to ascertain the most usable information from
our device. Piezoelectric sensors can measure vibration
responses, but not displacements as can LVDT, hall effect
sensors, linear potentiometers, or speckle interferometry
methods. These distance output measurements should be able
to give a clearer insight into the walls impedance
characterization when perturbed with a similar knocking
mechanism and processed through the system identification
methods outlined above, however, might be more difficult to
completely isolate from the solenoid impulse.
Finally, we envision our devices capabilities to ultimately
applied in order to deterministically measure any given wall
locations maximum loading force (eg. Will this area of the wall
support a TV or a heavy picture frame here?). By comparing
readings of different areas of the wall to known characteristics,
this could help inform the owner of important but non-obvious
features of the wall, such as: material, thickness, water damage,
insulation characterization, and invisible cracks.
ACKNOWLEDGMENTS
The authors would like to thank Prof. Ian Hunter for his
insightful suggestions during our class presentations and
extensive course lectures. Seyed Mirvakili and Ashley Raynal
for answering all of our technical question, help with procuring
equipment, and report suggestions.
REFERENCES
[1] E. J. Ebner, B. J. Huffer, Magnetic Stud Fastener Finder,
2015, US9069028 B2.
[2] V. Cook, Studfinder and Laser Level Tool, 2004,
USD498151 S1.
[3] M. Lake, The New York Times 2001.

[4] R. Mancini, Op Amps for Everyone: Design Reference,

Texas Instruments, 2003.
[5] Active Low Pass Filter, Basic Electronics Tutorials 2013.
Dale Thomas is a PhD student studying
Mechanical
Engineering
at
the
Massachusetts Institute of Technology and
earned his Bachelors of Science degree in
Marine Systems Engineering from Maine
Maritime Academy. His research has been
focused on utilizing flow chemistry for the
manufacturing of pharmaceuticals for ondemand use and rapid synthesis of
peptides.
Kendall Helbert is a senior at MIT
studying Mechanical Engineering with a
concentration in Product Design. He has
done research in additive manufacturing
methods through the Mechanosynthesis
Group and plans on joining the Apple
Softgoods team after graduation.

Debbie Ajilo is a first year Masters

student studying Mechanical Engineering
at MIT where she also earned her
Bachelors Degree in Mechanical
Engineering. Her current research is
focused on monitoring the performance of
manufacturing
machines
using
inexpensive and noninvasive sensing
techniques.
Fiona Grant is a junior at MIT studying
Mechanical Engineering. She has done
research with the GEAR Lab on solarpowered desalination systems, and she
is interested in pursuing graduate
studies in control systems and product
design.

Amy Loomis is a Masters student at

MIT in the CADLab group, with a BS
in Mechanical Engineering and Minor
in Electrical and Computer Engineering
from WPI. Her research focuses on
team information management in the
product design process and she plans on
working at Altitude, Inc. upon
graduation.